Expert Systems with Applications 39 (2012) 10921104

Contents lists available at SciVerse ScienceDirect

Expert Systems with Applications

journal homepage: www.elsevier.com/locate/eswa

Analysis of facility location model using Bayesian Networks

Ibrahim Dogan
Department of Industrial & Manufacturing Engineering, Wayne State University, 4815 Fourth Street, Detroit, MI 48202, USA

a r t i c l e i n f o a b s t r a c t

Keywords: In this study, we propose an integrated approach that combines Bayesian Networks and Total Cost of
Facility location Ownership (TCO) to address complexities involved in selecting an international facility for a manufactur-
Bayesian Networks ing plant. Our goal is to efciently represent uncertain data and ambiguous information, and to unite
Total Cost of Ownership (TCO) them to improve the quality of the decisions. Bayesian Networks provide a framework to elicit informa-
tion from experts, and provide a structure guide to efcient reasoning, even with incomplete knowledge.
Our method is presented in a hierarchical structure so that it can be decomposed into the forms of more
manageable units. We consider many tangible and intangible facility location criteria, then these criteria
are grouped into few numbers of factors. These factors are then combined to form a cost perspective
using the essentials of TCO.
2011 Elsevier Ltd. All rights reserved.

1. Introduction holistic view that addresses numerous qualitative and quantitative

The location decision describes, given a set of alternatives, the
problem of where to locate an organizations facility (Melo, Nickel,
& Saldanha-da-Gama, 2009). Since this decision is strategically
important, it has a signicant impact: the companys overall suc-
cess is directly related to the sound decisions on facility location.
Over the long-term, location decisions affect every aspect across
the organization, so that any poor location decision will result in
excessive costs, poor customer service, disappointed workforce,
and failure of the organization strategy as well as loss of competi-
tive advantage (Canbolat, Chelst, & Garg, 2007; Stevenson, 2009).
In our study, we focus on the global facility location decision for
a manufacturing rm. The numbers of companies that adopt glob-
alization as an organizational strategy have increased in last two
decades. Organizations are enthusiastic to work overseas for sev-
eral reasons: expanding current market, spreading foreign ex-
change risks, attracting talented workforce, working with capable
suppliers, and learning from foreign customers, suppliers, compet-
itors and foreign research centers, as well as savings due to re-
duced taxes, labor, transportation, and labor costs, incentives and
capital subsidies (Ferdows, 1997). In consequence, the question
of identifying a best geographic location for facilities has gained
attention from both researchers and businesses.
In this study, we propose a novel and integrated approach that
combines a Total Cost of Ownership (TCO) approach with a Bayes-
ian Network for selecting an international facility for a manufac-
turing rm. We analyze the facility location problem from a
Tel.: +1 313 577 9665; fax: +1 313 577 8833.
E-mail address: dogan@wayne.edu
criteria and their effects on different cost items in probabilistic
cause-effect relations. Bayesian Networks facilitate modeling of
causality in an efcient, intuitive, and transparent way, and the
available graphical probabilistic structure allows the user to ac-
count for uncertainty and ambiguous domain knowledge within
the problem context.
Cost elements that are selected as nal measures for comparing
different alternatives are not limited to investment costs unique to
the location; included are all the costs related to the life cycle of
the facility that can vary based on the selected facility. This Total
Cost of Ownership approach (TCO) aims to evaluate true cost of
the location decision and provide a better estimate of total costs
while comparing different alternatives. This approach provides us
an opportunity to combine numerous qualitative and quantitative
factors by considering the effects on different cost elements that
are critical to the location decision. The proposed combined tech-
nique will simultaneously tackle both monetary and non-mone-
tary criteria and combine them to create a thorough and accurate
measure that takes into consideration all of the costs that are rel-
evant to these selection criteria.
Signicant amount of facility location criteria are uncertain and
can easily uctuate before and after decisions have been made
(Snyder, 2006). Therefore, it is critical to clearly model the uncer-
tainty in each facility location criteria and their effects on different
cost elements. A key problem has been addressing the inuence of
each criterion on the long horizon, so an explicit representation of
uncertainty has gained signicant importance. Failure to precisely
represent and understand these uncertainties results in strategic
mistakes that are difcult to overcome. Human subjective judg-
ment is critical in this task, especially if the decision makers need
to make an assessment when there is limited and incomplete data.

0957-4174/$ - see front matter 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2011.07.109
 I. Dogan / Expert Systems with Applications 39 (2012) 10921104
Therefore, there is a requirement for a complete and efcient
method to encode human belief to analyze different alternatives
under this uncertain and complex environment. The proposed
method allows exploring and directing judgment and managers
opinions by following a systematic approach while globally consid-
ering all the relations among factors, as well as between factors
and cost measures. Bayesian Networks have a high potential to
efciently facilitate incorporating, representing, and propagating
the uncertain and ambiguous information.
A Bayesian Networks (BN) approach offers several distinctive
features that are useful in our facility location decision analysis.
First, BN builds on causality that provides useful insight and intu-
itive understanding. This feature is particularly important in our
case since it will assist to model the relationships between facility
selection criteria and their effects on cost elements. Second, BN al-
lows us to incorporate subjective information of the decision mak-
ers. This domain information is considerably useful when the data
is limited, and human experience is signicant. Third, BN can com-
bine historical data with human knowledge while drawing conclu-
sions for the problem at hand. In a facility location decision, some
of the information and degree of uncertainty can be learned from
historical data such as currency and ination rates. However, hu-
man inputs are required for a signicant part of the selection crite-
ria and their effects on the cost elements. BN can unite these two
different sources of information. Fourth, BN allows us to model
uncertainty by dening probability distributions over the random
variables. Modeling uncertainty by plain probability assessments
provides compactness and explanations for variability and risks in-
volved in decisions. Finally, BN facilitates reasoning by providing
efcient inference algorithms. Even when the information is
incomplete and the knowledge of an expert is limited, BN allows
us to draw conclusions even in poorly dened environments. This
feature also allows the sensitivity analysis to understand the de-
gree of variability that caused by the different selection criteria.
This provides explanations for the responsibility of each factor
for the companys overall success.
The proposed approach consists of mainly four steps: one, iden-
tify the selection criteria, factors and cost elements that are signif-
icant for global location decision; two, determine the structure by
building relations among the selection criteria and factors, and
among factors and cost elements; three, construct conditional
and unconditional probabilities and collect alternatives; four, make
decisions. Fig. 1 shows the model elements in a hierarchical view.
The uncertainties related to facility location are conceptualized by
the data and human judgment. For example, the uncertainty in-
volved in current and future ination rate and demand may be ad-
dressed by analyzing the historic data. The overall mean and
variability can be parameterized by the existing data. On the other
hand, for some factors the decision maker may have inadequate
information due to a lack of historical data. However, the incom-
plete knowledge of the decision maker due to vagueness in the
selection criteria is still critical and needs to be included in the
Fig. 1. Hierarchical view of model.
1093
analysis. BN has the potential to combine existing data with
incomplete expert knowledge.
The remainder of this paper is structured as follows. In Section 2,
the relevant research is presented concerning the global facility
location. Section 3 provides a short overview of BN. The proposed
model and its details are explained in Section 4. The model is then
tested and a manufacturing sector illustrative example is pre-
sented in Section 5. The last section presents a summary and con-
clusion of the proposed method.
2. Literature review
For years, the complexities and risk of facility location has at-
tracted a great deal of interest from both academia and industry,
generating a wide variety of approaches and methods, and each
having its own objectives and attributes. (For details, readers are
referred to comprehensive reviews in facility location by Owen
and Daskin (1998), Goetschalckx, Vidal, and Dogan (2002), Bhutta
(2004), Klose and Drexl (2005), Meixell and Gargeya (2005), Snyder
(2006), Sahin and Sral (2007), ReVelle, Eiselt, and Daskin (2008),
Melo et al. (2009), and Farahania, Sei, and Asgaria (2010)).
Our focus is global facility location modeling. The literature for
this can generally be divided into two groups: one that presents
and describes facility location factors, importance and steps of
international facility decisions (Canel, Khumawala, Law, & Loh,
2001); one that attempts to quantify and model the problem.
In the rst group of studies, researchers discuss the overall envi-
ronment in facility location decisions. MacCormack, Newman, and
Roseneld (1994) highlight the importance of qualitative factors in
global site selection. They analyze the effects of technological ad-
vances and new production systems on shaping the future strate-
gies of global corporations. They conclude that these expansions
will increase the requirement for exible and decentralized plants
so the sophistication of the location owing to regional infrastruc-
ture and quality of labor will be vital on location decision. Hoffman
and Schniederjans (1994) developed a two-stage model that uses
goal programming for global site selection. In the rst stage,
numerous environmental factors are identied that are country-
specic, and in the second stage, goal programming is used to se-
lect a site that has close t with the organizations overall strategy.
Ferdows (1997) highlights the importance of foreign facilities to
utilize a source of competitive advantage. He emphasizes the fact
that if managers consider manufacturing as a competitive advan-
tage, they have gained more from their overseas factories by
achieving low costs and high productivity. MacCarthy and
Atthirawong (2003) provide a survey and explanatory study to de-
scribe the importance of different global facility location criteria
using the Delphi study. They identied cost, infrastructure, labor
characteristics, government-political, and economic as top signi-
cant factors in the global facility location decision. Lee and Wil-
helm (2010) provide explanations to the facility location problem
from an international economics perspective by considering and
relating comparative advantage, competitive advantage, and com-
petitiveness of country; and discuss the measures in two popular
competitiveness reports, the Global Competitiveness Report
(GCR) and World Competitiveness Yearbook (WCY).
In the second group, different quantitative models are used
including stochastic programming, mixed-integer optimization,
transportation, multi-attribute utility theory, analytical hierarchy
process, analytical network process, goal programming models,
and combined approaches including fuzzy logic and sets. Min
(1994) used the multi-attribute utility theory approach to selecting
an international supplier, taking into account many quantitative and
qualitative criteria such as trade restrictions, cultural and communi-
cation barriers, quality, risk, and nancial and service performance.
1094 I. Dogan / Expert Systems with Applications 39 (2012) 10921104
Yurimoto and Masui (1995) developed a decision support system for
global plant location, analyzing the decision making progress in two
stages: the selection of countries, and the selection of sites within
the country. They conduct surveys to identify and rank the selection
criteria, and developed the analytical hierarchy process model for
decision process. Huchzermeier and Cohen (1996) formulated a sto-
chastic dynamic programming for evaluating global manufacturing
alternatives under exchange risk. Badri (1999) analyzed the facility
location problem in a multi-objective setting that takes into consid-
eration the political situation, global competition, government
regulations, and economic factors. The analytical hierarchy process
and goal programming are combined to solve both facility location
and allocation problem. Canel and Khumawala (1997) developed a
mathematical model to solve the deterministic un-capacitated
multi-period facility location model using a branch and bound
procedure. Vidal and Goetschalckx (2000) formulated a mixed inte-
ger programming to show how uncertainties in demand, exchange
rate, supplier reliability, and lead time affect a global logistics sys-
tem. In another study, Canel et al. (2001) provide a heuristic
procedure to solve the deterministic facility location model with
prot maximization. The model considers the issues such as ex-
change rate, export incentives, tariffs, taxes, and ownership that
are signicantly important for global site selection decisions. Canel
and Das (2002) formulate a mixed-integer programming model to
integrate global manufacturing and marketing operations by facility
location decision. The objective of this mathematical model is to de-
cide on the location of manufacturing facilities, the production
quantities at each facility, and shipping quantities from each facility
to market by including operating, transfer, shortage and holding
costs as well as considering incentives, market demand and capacity
components.
One of the critical challenges and complexities in global facility
location models is the uncertainty of the factors. Several studies at-
tempt to model the facility location decision by including uncer-
tainty for more realistic scenarios. Lowe, Wendell, and Hu (2002)
discuss the effects of exchange and ination rates on the network
design of a global manufacturing rm. They propose a two-phase
approach to evaluate different network congurations coupled
with scenario planning and scenarios for possible future real ex-
change rates. They consider different cost items including packag-
ing, direct labor, raw material, overhead, supplies, building,
development, waste treatment, maintenance, and depreciation.
Canbolat, et al. (2007) suggest an integrated approach combining
decision tree and multi-attribute utility theory analysis for a man-
ufacturing company for his global facility location decision. This
approach has mainly two phases; in the rst phase, the risk prole
regarding to the cost structure of each country is obtained by using
decision tree considering uncertainties regarding the ination rate,
exchange rate and productivity improvement. In the second phase,
the best alternative is selected using MAUT by considering the cost
structure as well as factors related to quality, stability and geo-
graphical location of the each alternative.
In an uncertain environment, some of the factors may not be di-
rectly measurable, so the human judgment is critical. Researchers
have developed models with different techniques combining with
fuzzy logic to incorporate the ambiguous knowledge of modelers
into models such as incomplete information, unquantiable infor-
mation, partial facts and non-obtainable information (Chou, Hsu, &
Chen, 2008; Kroemer, Kroemer, & Kroemer-Elbert, 1999). For
example, Kahraman, Ruan, and Dogan (2003) use different fuzzy
multi-attribute group decision making methods such as group
decision proposed by Blin, fuzzy synthetic evaluation, Yagers
weighted goals method, and fuzzy analytical hierarchy process.
Chou et al. (2008) developed a fuzzy multi-criteria decision making
model for the international tourist hotel selection. Guneri, Cengiz,
and Seker (2009) employ analytical network process with fuzzy
logic for shipyard selection to reduce the ambiguity and subjectiv-
ity. Although the several advantages of techniques that use com-
bined fuzzy logic are listed in the literature, the reasoning under
uncertainty is not based on plain probability distributions. Further-
more the building of a fuzzy system in facility location problem is
demanding; it requires identication of many appropriate rules
and tuning of many membership functions for inference, rendering
the inference process opaque to the decision maker.
Our approach employs Bayesian Networks (BN) to incorporate
uncertainty and incomplete knowledge by explicitly considering
the probability distributions. Furthermore, modeling with BN takes
natural causal interpretation into consideration. This property
facilitates the analysis of the problem in a hierarchical formmod-
eling the relations between selection criteria and factors, as well as
between factors and different cost elements. We employ BN in dif-
ferent contexts for knowledge representation and drawing conclu-
sions in uncertain environments such as root causes of process
variation (Dey & Stori, 2005), operations management (Ahn &
Ezawa, 1997; Neil, Fenton, & Tailor, 2005), multi-criteria decision
making (Watthayu & Peng, 2004), biomedicine and health-care
(Hanson & Marshall 2001; Lucas, van der Gaag, & Abu-Hanna,
2004; Maglogiannis, Zaropoulos, Platis, & Lambrinoudakis,
2006), information technology (IT) implementation (Lauria &
Duchessi, 2006) and sustainability (Castelletti & Soncini-Sessa,
2007; Pollino, Woodberry, Nicholson, Korb, & Hart, 2007;
Ticehurst, Newham, Rissik, Letcher, & Jakeman, 2007).
3. Bayesian Networks
Bayesian Networks are directed acyclic graphs (DAGs) where
the directed cycles are not allowed. The random variables are rep-
resented as nodes (vertices), and the arcs (edges) between nodes
capture the dependence or casual relations. BN merges the princi-
ples of graph theory, where a complex system can be represented
by smaller and manageable parts, and probability theory, where
these parts are combined in a consistent way so that efcient rea-
soning becomes feasible (Murphy 2001). Let V = {X1, X2, . . . , XN} be
the set of variables in a BN whose structure and semantics dene
conditional independence. An edge from Xi to Xj indicates an inu-
ential relation that is the value of variable Xj, depending on the va-
lue of the variable Xi. A node without any parent is called root node,
and a node without any child is called leaf node. Descendants of a
node include its children, childrens children and so on. The rela-
tionships among variables are quantied by conditional probabil-
ity distributions. If there is a link from variable Xi to Xj, Xi is
called parent of Xj and Xj is called child of Xi. Let pa(Xi) is the set
of all parents of variable Xi. A conditional distribution is then at-
tached to variable Xi conditioned on pa(Xi) : P(Xi|pa(Xi)).
The joint distribution of all the variables in set V is given by
Y
n
AV PX 1 ; X 2 ; . . . ; X N PX i jpaxi 1
i1
BN variables can be discrete, continuous, or mixed which includes
both discrete and continuous variables. For discrete variables, each
variable Xi in the set V has nite number of mutually exclusive
states that denes the status of the variable. The conditional distri-
butions are named Conditional Probability Tables (CPT) due to the
table formation of conditional distributions.
An example of BN with eight variables is shown in Fig. 1. The
corresponding decomposition of joint distribution of variables is
given by
PX 1 ; . . . ; X 8 PX 1 PX 2 PX 3 PX 4 jX 1 ; X 2 PX 5 jX 2 ; X 3
PX 6 jX 3 PX 7 jX 4 ; X 5 PX 8 jX 5 ; X 6
 I. Dogan / Expert Systems with Applications 39 (2012) 10921104 1095

P(X2) modular, in other words it has ability to structure a complex system

2 by simpler parts.
One of BNs basic tasks is belief propagation; that is, to update
P(X3) the marginal probabilities of variables P(Xi) after observing the val-
P(X1)
3
ues of some variables. This observed and known information is
1
called evidence and is denoted by e. For example in Fig. 2, the prob-
ability distribution of variable X4 given the value of all variables ex-
cept X4(e = {X1, X2, X3, X4, X5, X6, X7, X8}) is calculated as
5
P(X6|X3) PX 1 ; X 2 ; X 3 ; X 4 ; X 5 ; X 6 ; X 7 ; X 8
P(X5|X2,X3) PX 4 je P
4 X 4 PX 1 ; X 2 ; X 3 ; X 4 ; X 5 ; X 6 ; X 7 ; X 8
6
P(X4|X1,X2)
This conditional probability can be calculated more efciently by
exploiting conditional independencies in BN using Eq. (1). With
BNs neat representation ability, this can be rewritten as
P(X7|X4,X5) 7 8 P(X8|X5,X6)
PX 4 jX 1 ; X 2 PX 7 jX 4 ; X 5
PX 4 je P
Fig. 2. A sample Bayesian Networks. X 4 PX 4 jX 1 ; X 2 PX 7 jX 4 ; X 5

It is obvious that to calculate the joint distribution, unconditional 4. Model

distributions of P(X1), P(X2), P(X3) and conditional distributions of
PX 4 jX 1 ; X 2 ; PX 5 jX 2 ; X 3 ; PX 6 jx3 ; PX 7 jX 4 ; X 5 ; PX 8 jX 5 ; X 6 must be
specied. One of BNs advantages is that it requires less parameters
than the conventional nave method. For example, if we assume
that each variable is a binary variable in Fig. 2, the method needs
at least 255 (28  1) parameters, however, a BN approach can work
with only 21 parameters: three for unconditional probabilities, four
for each of the four variables with two parents, and two for one var-
iable with one parent. As depicted in this small example, BN com-
pactly and economically forms the joint distribution. It is
In this study, we approach the selection of a suitable, global
facility location for a manufacturer from a total cost perspective.
The decision makers goal is to select the facility that minimizes
the total cost. We consider costs that are not only related to the
acquisition of the place at the instant of decision, such as land
and investment costs, but also the costs that are related to possess-
ing and using the facility to satisfy the market demand during its
life cycle, such as labor, manufacturing, material, energy, and
nancial costs. While analyzing and evaluating the costs associated

C5
C8 C25 C28
C3 C12 C15 C22

C6 C20 C29
C2 C9 C26
C11 C14 C18 C31
C4 C23
C17 C21
C30
C1 C7 C19 C27
C10 C13 C32
C24
C16

F3 F8 F10
F7 F9 C33
F1
F2 F4 F5
F6

F11

F15
F13
F16
F14

Cost1 Cost2 Cost3 Cost4 Cost5 Cost6 Cost7 Cost8 Cost9 Cost10 Cost11 Cost12

C34
C35

F12
C36
Total
Cost

Fig. 3. Bayesian Networks for facility location.

1096 I. Dogan / Expert Systems with Applications 39 (2012) 10921104

Table 1
Facility location criteria, factors and their states.

Node label Factor/states [1 2] Node label Criteria States [1 2]

F1 Quality of labor [high low] C1 Labor skill [Skilled unskilled]
C2 Motivation of workforce [Motivated unmotivated]
C3 Availability of workforce [Available unavailable]
F2 Quality of suppliers [High Low] C4 Technological capability of suppliers [High Low]
C5 Alternative suppliers [Sufcient insufcient]
C6 Suppliers reliability and responsiveness [Adequate inadequate]
F3 Demographics [sufcient insufcient] C7 Population [Sufcient insufcient]
C8 Language skills [Sufcient Insufcient]
C9 Unemployment rate [High Low]
F4 Geographical Location [adequate inadequate] C10 Land availability and price [Low high]
C11 Quality and availability of raw materials [Adequate inadequate]
C12 Climate [Acceptable unacceptable]
F5 Quality of Life [sufcient insufcient] C13 Standard of living [High low]
C14 Education system [Sufcient insufcient]
C15 Health system [Sufcient insufcient]
F6 Financial efciency [high low] C16 Banking and nancial services [Sufcient insufcient]
C17 Credit availability [Sufcient Insufcient]
C18 Financial risk level [Low High]
F7 Quality of transportation [efcient inefcient] C19 Quality of air/water transportation [Sufcient insufcient]
C20 Quality of railway/road transportation [Sufcient Insufcient]
C21 Quality of distribution infrastructure [Sufcient insufcient]
F8 Government efciency [efcient inefcient] C22 Political stability [Stable unstable]
C23 Bureaucracy [Low High]
C24 Business legislation [Satisfactory unsatisfactory]
F9 Quality of infrastructure [sufcient insufcient] C25 Technological infrastructure [Sufcient insufcient]
C26 Scientic infrastructure [Sufcient insufcient]
C27 Basic infrastructure [Sufcient insufcient]
F10 Regulatory [unrestricted restricted] C28 Environmental regulations [Unrestricted restricted]
C29 Tax structure [Unrestricted restricted]
C30 Custom duties [Low high]
F11 Social and cultural factors [sufcient insufcient] C31 Cultural barriers [Low high]
C32 Community attitudes towards business and industry [Positive negative]
C33 Unionization [Unrestricted restricted]
F12 Economic performance [sufcient insufcient] C34 Currency stability [Stable unstable]
C35 Economic growth rate [Positive negative]
C36 Ination [Low high]

with facility location, we linked these cost elements to different Table 2

factors and criteria such that if the level of these factors changes, Direct factors and their states.
then these cost elements also vary signicantly. Therefore, this
Node label Factor States
new approach provides a plausible explanation for variation in
the total cost. Furthermore, this holistic view of total cost approach F13 Proximity of suppliers [Close remote]
F14 Proximity of market [Close remote]
will allow the manufacturing organization to make the more accu-
F15 Tax rate [Low high]
rate evaluations in identifying the responsibility of each factor on F16 Labor wage [Low high]
different cost elements.
Fig. 3 shows the Bayesian Network structure for the proposed
model. All the discrete variables in the model including criteria
we aim to see the direct impact of these signicant factors on dif-
and factors are binary, either they are in the preferred state (attrac-
ferent cost elements. This separation will also assist the subject
tive) or less preferred state (unattractive) for the facility location
matter experts to more accurately and compactly evaluate the
standpoint.
relationships and conditional probability distributions. The states
Literature review and subject matters expert interviews are
of these additional factors are listed in Table 2. The list of criteria
conducted to identify a set of complete criteria for global locations
and costs items that are considered in this paper is substantially
decisions. These criteria are evaluated under different subgroups
larger and more complete than the similar studies. Our methodol-
(factors). There are mainly 37 criteria (C1 to C37) that are grouped
ogy denes and structures the location problem more efciently
into 12 factors (F1 to F12). These factors are quality of labor, quality
due to its holistic view.
of suppliers, demographics, geographical location, quality of life,
nancial efciency, quality of transportation, government ef-
ciency, quality of infrastructure; regulatory, social and cultural fac- 4.1. Factors
tors, and economic performance. The factors and related criteria
are listed in Table 1. The factors from F13 to F16 represent the ele- Factors are formed to reduce the complexity of the model. These
ments of proximity of suppliers, proximity of market, labor wage, factors explain and represent the numerous selection criteria.
tax rate, and size of the market demand, respectively. These are ac- Although these factors cannot be observed directly, they can be in-
cepted in the literature as critical parameters for facility location. ferred through the selection criteria that are connected to these
Therefore, they are not included in any subgroup; by doing so, variables. Decision makers now need to relate the reduced number
 I. Dogan / Expert Systems with Applications 39 (2012) 10921104
of variables to the cost elements instead of the several selection
criteria. By doing so, the number of parameters to construct the
graphical model is signicantly reduced. Another advantage of
using factors is that experts can easily and efciently direct their
domain knowledge to the model and can combine with historical
data.
4.1.1. Quality of labor
This factor includes the criteria of availability of labor skill level,
motivation of workforce, and availability of the workforce. The low
labor cost is one of the important factors for the organizations who
search places to locate their facilities internationally due to its di-
rect impact on quality and productivity. However, organizations
who want to achieve their long-term objectives need to include
other measures. Especially in certain industries, the workforces
knowledge capacity becomes a more critical issue than labor
wages. If the workforce knowledge capacity is inadequate, the
company needs to plan for some additional employee training.
4.1.2. Quality of suppliers
This factor includes suppliers technological capability, alterna-
tive suppliers availability, suppliers reliability, and responsive-
ness. The success of the companies in the eld is directly related
to incorporation of suppliers who have the technological capability
to do specic jobs, who are reliable, and responsive to the changes
in the supply chain and the requirements of buyer. It is also impor-
tant for global companies to reach and maintain alternative suppli-
ers to mitigate the risks due to less control and limited knowledge
on global operations.
4.1.3. Demographics
Demographic factors include population, language skills, and
unemployment rate. From these measures, population and unem-
ployment rate will directly affect the availability of the workforce.
Language is seen as one of the biggest barriers and has an adverse
effect on efcient operations. Communication efciency within the
organization will increase the motivation of the workforce. The
best communication is achieved when the workforce speaks the
same language. The lack of a common language communication
platform will require additional language training to build healthy
communication channels.
4.1.4. Geographical location
This factor includes the criteria of land availability and its price,
quality and availability of raw materials, and climate. The facility
location involves a big investment and long term commitment of
resources. The cost of acquiring land for a facility is one of the ma-
jor cost elements in this investment. Signicant climate uctua-
tions and severe weather conditions disrupt the business as well
as affect the human efciency.
4.1.5. Quality of life
This factor includes standard of living, and education and health
systems. Highly motivated employees transfer their energy, crea-
tivity, and commitment to organizations, resulting in increased
quality products, improved operations, and reduced overall costs.
Researchers show that the employees motivation is directly asso-
ciated with the quality of life; if the people work in an environment
that provides a better standard of living, education and health sys-
tems, they are eager to contribute more to their organizations.
4.1.6. Financial efciency
This factor includes the efciency of banking and nancial ser-
vices, credit availability, and nancial risk level of the country. A
countrys nancial efciency has a huge impact on the health
of the nancial structure of the businesses. A simple and
1097
straightforward way to access funds and credit sources increases
opportunities for new investments and reduces nancial costs.
4.1.7. Quality of transportation
This factor includes quality of air, water, railway and road trans-
portation, and quality of distribution infrastructure. Quality of
transportation affects the whole organization in different perspec-
tives, including the cost and time of transporting the goods, and
travel costs of employees (Leitham, McQuaid, & Nelson, 2000). Poor
transportation infrastructure affects the performance of the orga-
nization and this bad performance results in increased costs.
4.1.8. Government efciency
This factor includes political stability, bureaucracy and business
legislation. It is accepted in the literature that well-established
government policies towards business contribute to the organiza-
tions and economy in different perspectives. These policies with
uncorrupted bureaucracy, modest taxation, and a stable political
environment are among the key drivers to attract more foreign
capital.
4.1.9. Quality of infrastructure
This factor includes technological, scientic and basic infra-
structure. These factors have been seen as enablers for successful
organizations. Reliable technological, scientic and basic infra-
structure such as electricity, ports, water, education, and health
systems may help the organization develop and maintain innova-
tive and efcient operations and quality products. If the corpora-
tions need to facilitate in a country that does not meet the
standards of infrastructures, the organization needs to invest more
to reduce the challenges of these key drivers. Several authors have
showed the evidence that high quality public services can result in
more investments of organizations (Bartik, 1985).
4.1.10. Regulatory
This factor includes environmental regulations, tax structure,
and custom duties. In some areas, strict environmental regulations
impede the economic growth of the organizations. Many underde-
veloped countries offer tax incentives and reduced custom duties
to draw more foreign investments. Companies get the advantage
of these lower tax burdens by signicantly reducing the costs. Bar-
tik (1985) shows that high tax rates discourage new businesses.
4.1.11. Social and cultural factors
This factor includes cultural barriers, community attitudes to-
wards business and industry, and unionization. Cultural and social
barriers challenge effective communication, regulate the successful
solutions to the problems, and increase the tension in organiza-
tions. In some countries, labor union restrictions may create seri-
ous problems to the organizations. According to the economic
theory, strict unions rules impose higher wages at the expense
of higher costs, and that increase in the costs cause lower invest-
ment (Hirsch, 2007).
4.1.12. Economic performance
This factor includes currency stability, economic growth rate,
and ination. Business wants to make investments in stable and
growing markets that provide lower ination rates with stable cur-
rency. Volatile markets have adverse effects on the whole
organizations.
4.1.13. Other factors
We also include the other variables such as proximity of suppli-
ers, proximity of market, tax rate and labor wage into the model.
Due to signicant effects of these factors on cost, they are directly
1098 I. Dogan / Expert Systems with Applications 39 (2012) 10921104

Table 3 4.2.8. Manufacturing cost

Cost elements. Includes assembly and fabrication of materials, long-term oper-
Node label Cost elements ations, and equipment.
Cost1 Quality related cost
Cost2 Labor cost 4.2.9. Taxes, tariffs, and insurance
Cost3 Construction and land cost These include insurance of materials, equipment and facilities,
Cost4 Material cost as well as costs associated with any legal responsibility such as
Cost5 Overhead cost
Cost 6 Development cost
taxes, custom costs and tariffs.
Cost7 Financial Cost
Cost8 Manufacturing cost 4.2.10. Energy cost
Cost9 Taxes, tariffs, insurance This includes any cost incurred as a result of consumption of an
Cost10 Energy cost
energy resource during manufacturing and running the business
Cost11 System and integration cost
Cost12 Transportation cost such as electricity, gas, and oil cost.

4.2.11. System and integration cost

This cost include the supplier relationship management.
linked to the different cost elements over which they have
inuence. 4.2.12. Transportation cost
Any cost related to freight, packaging, customer service, han-
4.2. Cost elements dling, tariffs warehousing, duties, import fees, and outbound costs.

The costs examined in this analysis can be grouped in three cat-
egories: costs incurred once such as land and investment; costs
that are time-dependent and xed such as labor, and costs that
vary with the size of the business such as material and logistics.
To combine these different items to make up the total cost, we as-
sume ten years of planning horizon and calculate and consider the
costs during this period. Table 3 lists these different cost items in-
cluded into analysis and their descriptions are as follows.
4.2.1. Quality related cost
Quality costs include failure as well as appraisal and prevention
costs. Failure costs are losses due to production of defective items
associated with scrap, rework, downtime, disposition, warranty ex-
penses, return and replacement of the products. Appraisal costs are
mainly related to the inspection of materials and products, and the
prevention costs are the costs of process control, quality improve-
ment programs, process design and review, and quality planning.
4.2.2. Labor cost
This is associated with the cost of all employees including
wages, hiring and training costs, taxes and benets.
4.2.3. Construction and land cost
These are the investment costs including acquisition of land,
construction of the facilities such as plants, and warehouses, pro-
curing the necessary equipment, machines and vehicles.
4.2.4. Material cost
This includes all direct and indirect materials used during the
production of products.
4.2.5. Overhead cost
These are not directly linked to products but necessary to run
the business and that are not accounted for in other cost elements,
such as advertising, legal fees, ordering costs, communication bills,
and accounting fees.
4.2.6. Development cost
These are product and process designs including testing and
evaluations, prototyping, tooling, design documentation, and pro-
cess planning costs.
4.2.7. Financial cost
These are the costs of funding the companys assets including
capital, interest, leasing, and other nancial expenditures.
4.3. Conditional probabilities
After constructing the graphical network, the next step is to
determine all the conditional and unconditional (prior) distribu-
tions. Causal relations are represented by arcs in the graph and
the strengths of these relations are assessed by conditional proba-
bilities. This includes the conditional probability of each factor
(child) given the criteria (parent), the conditional probability of
each cost element (child) given the factors (parent) and the condi-
tional probability of total cost (child) given the all costs elements
(parent) and factor of economic performance (F12). Since all the
factors and criteria are discrete variables, the rst conditional
probability distributions are depicted as a Conditional Probability
Table (CPT). For example, Fig. 4a is an example for this type of rela-
tionship. The probability of quality of labor factor given the labor
skill, motivation of workforce, and availability of workforce criteria
are shown as a CPT in Fig. 4b. This CPT is constructed for all possi-
ble values of each criterion and factor. Since all the variables are
binary, there are 16 congurations. The subject matter experts
need to identify only the rst eight of the 16 possible combinations
where the following eight combinations are the complementary of
the rst eight. Each probability associated with the possible cong-
urations of the parents quanties the plausibility of the state. Sim-
ilarly, Fig. 4c is an example of the second type of the relationship.
We modeled the cost elements as continuous variables which are
conditionally distributed as Gaussian distribution. This assumption
requires identifying the mean and variance parameters of labor
costs for all possible combinations of discrete factors of quality of
labor, quality of suppliers and labor wage (Fig. 4d). The specica-
tion of the last relation is the most demanding for the number of
parameters. This is the identication of conditional distribution
of total costs for all possible combinations of parent nodes includ-
ing one discrete factor of economic performance and all continuous
cost elements. Since the parent nodes involve both discrete and
continuous nodes, the model requires identication of the mean
and variance of total cost which is conditionally distributed as
Gaussian distribution, for all combination of discrete variables as
well as the weights due to the parent continuous variables. Appen-
dices A and B depict all the conditional probability distributions
among criteria and factors, and factors and cost elements.
The probabilities in our model reect the decision makers be-
lief regarding the likelihood of the occurrence of an event. The
higher probability values represent the higher degree of certainty
of specic state, while on the other hand small differences in prob-
ability of states shows the higher ambiguity and uncertainty.
 I. Dogan / Expert Systems with Applications 39 (2012) 10921104 1099

Movaon
of Availability
CPT (P(F1|C1,C2,C3)) P(Cost2|F1, F3,F16)
Labor Skill Workorce of
(C1) (C2) Worforce
(C3)

Quality of
Labor
(F1)

Quality of Quality of Labor

Suppliers Wage
Labor
(F3) (F16)
(F1)

Labor Cost
(Cost2)

Fig. 4. Sample relations among selection criteria and factors (a), and cost element and factors (b), and their conditional probability distributions (c) and (d).

Table 4
Location alternatives data for tangible criteria.

Criteria Measure Global rank (state rule) State State

rule
Turkey Mexico S. Korea China Turkey Mexico S. Korea China
Availability of workforce Labor Force 25 12 22 1 6 20 2 1 2 1
Population Population 17 11 23 1 6 20 2 1 2 1
Unemployment rate Unemployment rate (largest to smallest) 72 151 160 153 675 1 2 2 2
Education system Education expenditure 81 25 75 123 675 2 1 2 2
Quality of the educational system 79 115 47 52
Quality of management schools 81 49 104 72
Health system HDI (Human Development Index) 84 52 26 81 675 2 1 1 2
Banking and nancial services Financial market sophistication 80 73 58 81 675 2 1 1 2
Credit availability Ease of access to loans 75 85 80 89 675 1 2 2 2
Financial risk level Government surplus/decit 73 46 28 56 650 2 1 1 2
Quality of air/water transportation Quality of port infrastructure 78 82 36 61 650 2 2 1 2
Quality of air transport infrastructure 54 56 21 80
Quality of railway/road transportation Quality of roads 48 57 14 50 650 2 2 1 1
Quality of railroad infrastructure 63 66 8 27
Bureaucracy No. of procedures required to start a business 26 75 85 117 650 1 2 2 2
Time required to start a business 9 78 49 99
Business Legislation Efciency of legal framework in settling disputes 83 94 62 43 650 2 2 2 1
Efciency of legal framework in challenging regs 71 80 69 57
Technological infrastructure Mobile telephone subscriptions 71 89 61 104 675 1 1 1 2
Internet users 54 73 9 72
Personnel computers 80 54 17 81
Broadband internet subscribers 49 50 7 52
Scientic infrastructure Innovation 69 78 11 26 650 2 2 1 1
Basic infrastructure Quality of overall infrastructure 62 71 20 66 650 2 2 1 2
Tax structure Extent and effect of taxation 121 91 72 32 650 2 2 2 1
Custom duties Tariff barriers 46 97 86 120 675 1 2 1 2
Burden of customs procedures 98 86 42 41
Ination Ination 88 38 34 46 650 2 1 1 1
Tax rate Total tax rate 77 98 35 124 650 2 2 1 2
Labor wage Gross domestic product per capita 53 54 37 88 P75 2 2 2 1

5. Illustrative case
There are four alternatives for the proposed location: Turkey,
Mexico, South Korea, and China. The information regarding facility
location criteria for each location alternatives are divided into two
groups. The rst group represents the criteria where the state of the
individual criterion is known or can be easily determined. The data
sources for this group are the Global Competitiveness Report
(Schwab, 2009), Central Intelligence Agency, Index Mundi (WWW
resource 1), and Human Development Report (WWW resource 2;
WWW resource 3). Table 4 lists all the criteria and their measures
in this group. The state of each criterion is identied according to
global ranking of the selected facility location. The ranking of the
countries for different measures is available in the data sources.
The decision rule on the state determination is identied by subject
matter experts. The threshold value in the decision rule is the rank-
ing, where the experts believe that if the selected facility is higher
than this ranking, the alternative location has a competitive advan-
tage for this criterion. Therefore the status of the state will be iden-
tied as the one that has the most preferred level of the criterion.
1100 I. Dogan / Expert Systems with Applications 39 (2012) 10921104

Table 5 Table 6
Belief of experts for intangible location selection criteria for each location alternative. Mean and standard deviation of total cost (in millions $).

Criteria Turkey Mexico Korea China Location Mean Standard deviation

[1 2] [1 2] [1 2] [1 2]
Turkey $ 612.5 43.47
Labor skill [0.8 0.2] [0.7 0.3] [0.8 0.2] [0.7 0.3] Mexico $ 603.2 41.01
Motivation of workforce [0.7 0.3] [0.7 0.3] [0.7 0.3] [0.8 0.2] S. Korea $ 602.5 38.81
Technological capability of [0.8 0.2] [0.7 0.3] [0.7 0.3] [0.8 0.2] China $ 576.3 42.58
suppliers
Alternative suppliers [0.9 0.1] [0.7 0.3] [0.6 0.4] [0.9 0.1]
Suppliers reliability and [0.6 0.4] [0.7 0.3] [0.7 0.3] [0.6 0.4]
responsiveness 1
Language skills [0.4 0.6] [0.5 0.5] [0.6 0.4] [0.5 0.5]
Land availability and price [0.7 0.3] [0.7 0.3] [0.4 0.6] [0.7 0.3]
Quality and availability of raw [0.9 0.1] [0.8 0.2] [0.7 0.3] [0.9 0.1] 0.8
materials

Probability
Climate [0.9 0.1] [0.9 0.1] [0.8 0.2] [0.8 0.2]
0.6
Standard of living [0.7 0.3] [0.6 0.4] [0.9 0.1] [0.6 0.4]
Quality of distribution [0.6 0.4] [0.7 0.3] [0.7 0.3] [0.6 0.4]
infrastructure 0.4
Political stability [0.4 0.6] [0.7 0.3] [0.8 0.2] [0.8 0.2] Turkey
Environmental regulations [0.9 0.1] [0.8 0.2] [0.4 0.6] [0.7 0.3] Mexico
0.2
Cultural barriers [0.7 0.3] [0.8 0.2] [0.7 0.3] [0.6 0.4] S.Korea
Community attitudes towards [0.8 0.2] [0.9 0.1] [0.7 0.3] [0.6 0.4] China
business and industry 0
Unionization [0.8 0.2] [0.8 0.2] [0.6 0.4] [0.8 0.2] 450 500 550 600 650 700 750
Currency stability [0.6 0.4] [0.6 0.4] [0.8 0.2] [0.8 0.2] Total Cost (In Millions)
Economic growth rate [0.9 0.1] [0.8 0.2] [0.6 0.4] [0.9 0.1]
Proximity of suppliers [0.9 0.1] [0.9 0.1] [0.7 0.3] [0.7 0.3] Fig. 5. Cumulative probabilities for total cost.
Proximity of market [0.7 0.3] [0.9 0.1] [0.6 0.4] [0.6 0.4]

For example, the criterion of nancial risk level of each alternative
country is measured by government surplus/decit, and the rank-
ing of Turkey, Mexico, Korea, and China are 73, 46, 28 and 56,
respectively. The experts identied the threshold ranking as 50;
any countrys ranking that is less than 50 will be identied as state
one. Therefore the states of Mexico and Korea for nancial risk level
criterion are identied as state one, and Turkey and China as state
two (least preferred level). If there is more than one measure for
any criterion, the decision threshold value is compared to the aver-
age ranking of all measures that explain the selection criterion. For
example, the state of education system is identied by three mea-
sures: education expenditure, quality of the educational system,
and quality of management schools. Turkey in this criterion is
determined by the average value of the 81, 79 and 81, and this aver-
age is compared to threshold value of 50.
The second group includes criteria that are not directly measur-
able or may uctuate during the planning horizon. The inherent
nature of uncertainty in these intangible criteria increases the
requirement of inputs from human subjective judgment and do-
main specic information of the experts. Table 5 depicts the beliefs
of experts regarding the status of each criterion. For example, the
expert has assigned a probability of 0.8 to represents his high belief
for the labor skill that is in its attractive level for facility location A.
Since all criteria have two states, experts only need to evaluate the
probability of the attractive level of each criterion, and the proba-
bility of the complement state can be determined automatically.
5.1. Results
The facility location problem was analyzed using Bayes Net
Toolbox for Matlab developed by Murphy (2004). This toolbox
has several strengths, such as allowing exact and approximate
inference algorithms, parameter and structure learning. Overall
ranking of the alternatives based on the total cost are depicted in
Table 6. The countries are compared on both the mean and stan-
dard deviation of the resulted total cost. China has the least mean
total cost with $ 576.3 million; on the other hand, South Korea has
the least standard deviation with 38.81.
The combined effect of both mean and standard deviation of to-
tal costs can be represented in a cumulative risk prole. Since the
calculated total cost has a normal distribution, the cumulative dis-
tribution function, F, can be simply used to identify the situations
where one alternative location place is preferred to another. Fig. 5
compares four location alternatives for the cumulative probability
distributions. It is clear from the gure that China is superior to
other countries for every possible cost value; in other words, China
has the stochastic dominance to other countries. This means that
the probability of getting a cost is never better in other countries
than in China. Conversely Turkey always has a higher cost than
the other countries. Note that from the gure, the preference be-
tween Mexico and S. Korea is changing at the cost of $ 590 million;
for the lower values of costs smaller than $ 590 million, the prob-
ability of observing a cost less than or equal to this cost value is
higher in S. Korea than in Mexico.
The probability of states was calculated using the given the evi-
dence in Table 4 and belief of experts in Table 5. Fig. 6 compares
the probability of being in attractive (1) and unattractive (2) states
for each factor and facility location. This gure assists decision
makers to see the strength and weakness of each alternative. For
example, in China, the probability of being in attractive states is
higher for quality of labor (F1), quality of suppliers (F2), demo-
graphics (F3), geographical location (F4), quality of transportation
(F7), government efciency (F8), social and cultural factors (F11)
and economic performance (F12). On the other hand, factors of
quality of life (F5), nancial efciency (F6), quality of infrastructure
(F9) and regulatory (F10) have higher probabilities for the unattrac-
tive states than the attractive states.
Fig. 7 depicts the expected costs for different cost elements due
to the status of each location alternative. Note that major cost ele-
ments among these are material cost, construction and land cost,
manufacturing costs and labor cost. China, ranked best overall based
on the total cost, may be expected to perform better than the com-
petitors in material, construction and land, labor, quality related,
system and integration, taxes, tariffs, and insurance cost items.
A straightforward sensitivity analysis is performed to under-
stand the possible cost range for any facility location. A facility
location within its attractive states for all factors is compared to
a facility location within its untractive states for all factors. Table 7
compares the results of these scenarios to the current case of alter-
native China. The average total cost of alternatives will range from
 I. Dogan / Expert Systems with Applications 39 (2012) 10921104 1101

Fig. 6. Strengths and weaknesses of alternatives; the probabilies of being in attractive (1) and untractractive (2) states of facility locations.

Fig. 7. Comparison of alternatives for different cost elements.

Table 7 well as the belief of experts. Therefore the uncertain knowledge re-
Comparison of scnearios. sults in higher variability.
Mean Standard deviation
Attractive states $ 480 6
Unattractive states $ 769 9.13
6. Conclusion
Current information (China) $ 576 42.58
In this study, we analyze the problem of selecting a global man-
ufacturing facility location under conditions of uncertainty and
$ 480 to $ 769 million. It is important to note that both attractive information ambiguity. Using a Bayesian Networks approach offers
and unattractive states scenarios represent the known case so that a high potential for representing this ambiguous knowledge and
they result in reduced variation in total costs. However the current for performing reasoning under uncertainty. We evaluate alterna-
information case for China includes both known information as tive facility locations from a cost perspective, taking into account
1102 I. Dogan / Expert Systems with Applications 39 (2012) 10921104
a number of quantitative and qualitative criteria. Using a combina-
tion approach of Bayesian Network with Total Cost of Ownership
(TCO) provides a realistic measure for comparing facility location
alternatives.
The proposed method is structured as a hierarchy. The top level
consists of 36 supplier selection criteria. Those criteria then explain
and measure 12 factors in the second level, with a nal level that
evaluates 11 cost items based on the state of each factor in the sec-
ond level. This hierarchical modeling assists decision makers in
guiding and structuring the problem while selecting the best loca-
tion. The problem is decomposed into smaller and modular parts
that can be solved more efciently. The insights obtained using this
approach are also more intuitive.
One of the key features of the proposed method is that it com-
bines data with domain knowledge. Decision makers can then
incorporate their soft beliefs regarding the status of the each alter-
native in an efcient way. The Bayesian Network is constructed by
using expert knowledge with additional inputs from extensive lit-
erature in facility location decisions. The conditional probabilities
and their parameters are identied using both expert knowledge
and the data.
Appendix A. Conditional Probability Tables
With an increasing number of alternatives and factors, the
involvement of a large amount of uncertainty around facility loca-
tion becomes more complex. Human judgment obviously plays a
key role in the decision making, particularly around problems that
involve a number of qualitative factors. Our goal was to create a tool
that provides a complete and efcient method for accurately encod-
ing human beliefs into a model that can analyze different alterna-
tives within the complex environment. The proposed method
allows exploring and directing judgment, opinions and belief of
managers by following a systematic approach while globally consid-
ering all the relations among factors, and between factors and
objectives.
Another advantage of the Bayesian Networks approach is its ca-
sual representation of various facility location factors. Bayesian
Networks relate factors to each other by parental representation
in a graph so that transparency in reasoning is possible. The deci-
sion makers have a better understanding and interpretation of fac-
tors by this graphical representation.
The proposed method identies the signicant cost elements
along with the status of each location in terms of different factors.
This study provides opportunities for decision makers to identify
the strengths and weakness of the each facility location place.
 I. Dogan / Expert Systems with Applications 39 (2012) 10921104 1103

Appendix B. Conditional distribution of cost items

References Ferdows, K. (1997). Making the most of foreign factories. Harvard Business Review,
75(2), 7388.
Goetschalckx, M., Vidal, C. J., & Dogan, K. (2002). Modeling and design of global
Ahn, J.-H., & Ezawa, K. J. (1997). Decision support for real-time telemarketing
logistics systems: A review of integrated strategic and tactical models and
operations through Bayesian network learning. Decision Support Systems, 21,
design algorithms. European Journal of Operational Research, 143, 118.
1727.
Guneri, A. F., Cengiz, M., & Seker, S. A. (2009). A fuzzy ANP approach to shipyard
Badri, M. (1999). Combining the AHP and GP for global facility locationallocation
location selection. Expert Systems with Applications, 36(4), 79927999.
problem. International Journal of Production Economics, 62(3), 237248.
Hanson, C. W., & Marshall, B. E. (2001). Articial intelligence applications in the
Bartik, T. J. (1985). Business location decisions in the U.S. Estimates of the effects of
intensive care unit. Critical Care Medicine, 29(2), 427435.
unionization, taxes, and other characteristics of states. Journal of Business and
Hirsch, B. T. (2007). What do unions do for economic performance? In J. T. Bennett &
Economic Statistics, 3, 1422.
B. E. Kaufman (Eds.), What do unions do? A twenty-year perspective
Bhutta, K. S. (2004). International facility location decisions: A review of the
(pp. 193237). New Brunswick, NJ: Transaction Publishers.
modelling literature. International Journal of Integrated Supply Management, 1(1),
Hoffman, J., & Schniederjans, M. (1994). A two-stage model for structuring global
3350.
facility site selection decisions. International Journal of Operations and Production
Canbolat, Y. B., Chelst, K., & Garg, N. (2007). Combining decision tree and MAUT for
Management, 14(4), 7996.
selecting a country for a global manufacturing facility. Omega, 35, 312325.
Huchzermeier, A., & Cohen, M. A. (1996). Valuing operational exibility under
Canel, C., & Das, S. R. (2002). Modeling global facility location decisions: Integrating
exchange rate risk. Operations Research, 44(1), 100113.
marketing and manufacturing decisions. Industrial Management and Data
Kahraman, C., Ruan, D., & Dogan, I. (2003). Fuzzy group decision making for facility
Systems, 102, 110118.
location selection. Information Sciences, 157, 135153.
Canel, C., & Khumawala, B. M. (1997). Multi-period international facilities location:
Klose, A., & Drexl, A. (2005). Facility location models for distribution system design.
An algorithm and application. International Journal of Production Research, 35(7),
European Journal of Operational Research, 162, 429.
18911910.
Kroemer, K., Kroemer, H., & Kroemer-Elbert, K. (1999). Ergonomics. New Jersey:
Canel, C., Khumawala, B. M., Law, J., & Loh, A. (2001). Algorithm for the capacitated,
Prentice-Hall.
multi-commodity multi-period facility location problem. Computers and
Lauria, E. J. M., & Duchessi, P. J. (2006). A Bayesian belief network for IT
Operations Research, 28(5), 411427.
implementation decision support. Decision Support System, 42, 15731588.
Castelletti, A., & Soncini-Sessa, R. (2007). Bayesian networks and participatory
Lee, C., & Wilhelm, W. (2010). On integrating theories of international economics in
modelling in water resource management. Environmental Modelling and
the strategic planning of global supply chains and facility location. International
Software, 22(8), 10751088.
Journal of Production Economics, 124(1), 225240.
Chou, T. Y., Hsu, C. L., & Chen, M. C. (2008). A fuzzy multi-criteria decision model for
Leitham, S., McQuaid, R. W., & Nelson, J. D. (2000). The inuence of transport on
international tourist location selection. International Journal of Hospitality
industrial location choice: a stated preference experiment. Transportation
Management, 27(2), 293301.
Research Part A, 34, 515535.
Dey, S., & Stori, J. A. (2005). A bayesian network approach to root cause diagnosis of
Lowe, T. J., Wendell, R. E., & Hu, G. (2002). Screening location strategies to reduce
process variations. International Journal of Machine Tools and Manufacture, 45(1),
exchange rate risk. European Journal of Operational Research, 136, 573590.
7591.
Lucas, P. J. F., van der Gaag, L. C., & Abu-Hanna, A. (2004). Bayesian networks in
Farahania, R. Z., Sei, M. S., & Asgaria, N. (2010). Multiple criteria facility location
biomedicine and health-care. Articial Intelligence in Medicine, 30(3), 201214.
problems: A survey. Applied Mathematical Modelling, 34(7), 16891709.
1104 I. Dogan / Expert Systems with Applications 39 (2012) 10921104
MacCarthy, B. L., & Atthirawong, W. (2003). Factors affecting location decisions in
international operationsA Delphi study. International Journal of Operations &
Production Management, 23, 794818.
MacCormack, A. D., Newman, L., III, & Roseneld, D. B. (1994). The new dynamics of
global manufacturing site selection. Sloan Management Review, Summer, 6979.
Maglogiannis, I., Zaropoulos, E., Platis, A., & Lambrinoudakis, C. (2006). Risk
analysis of a patient monitoring system using Bayesian network modeling.
Journal of Biomedical Informatics, 39(6), 637647.
Meixell, M. J., & Gargeya, V. B. (2005). Global supply chain design: A literature
review and critique. Transportation Research Part E: Logistics and Transportation
Review, 41, 531550.
Melo, M. T., Nickel, S., & Saldanha-da-Gama, F. (2009). Facility location and supply
chain management A review. European Journal of Operational Research, 196(2),
401412.
Min, H. (1994). International supplier selection: a multi-attribute utility approach.
International Journal of Physical Distribution & Logistics Management, 24(5),
2433.
Murphy, K. (2004). The BayesNet toolbox for Matlab. Available from http://
bnt.sourceforge.net.
Murphy, K. P. (2001). A introduction to graphical models. Available from
www.ai.mit.edu/murphyk/papers.html.
Neil, M., Fenton, N. E., & Tailor, M. (2005). Using bayesian networks to model
expected and unexpected operational losses. Risk Analysis, 25(4), 963972.
Owen, S. H., & Daskin, M. S. (1998). Strategic facility location: A review. European
Journal of Operational Research, 111, 423447.
Pollino, C. A., Woodberry, O., Nicholson, A., Korb, K., & Hart, B. T. (2007).
Parameterisation and evaluation of a Bayesian network for use in an
ecological risk assessment. Environmental Modelling and Software, 22(8),
11401152.
ReVelle, C. S., Eiselt, H. A., & Daskin, M. S. (2008). A bibliography for some
fundamental problem categories in discrete location science. European Journal
of Operational Research, 184, 817848.
Sahin, G., & Sral, H. (2007). A review of hierarchical facility location models.
Computers & Operations Research, 34, 23102331.
Schwab, K. (Ed.). (2009). The global competitiveness report 20092010. Geneva:
World Economic Forum.
Snyder, L. V. (2006). Facility location under uncertainty: A review. IIE Transactions,
38, 537554.
Stevenson, W. J. (2009). Operations management (10th ed.). New York: McGraw-Hill.
Ticehurst, J. L., Newham, L. H. T., Rissik, D., Letcher, R. A., & Jakeman, A. J. (2007).
Bayesian network approach for assessing the sustainability of Coastal Lakes in
New South Wales, Australia. Environmental Modelling and Software, 22(8),
11291139.
Vidal, C., & Goetschalckx, M. (2000). Modeling the impact of uncertainties on global
logistics systems. Journal of Business Logistics, 21(1), 95120.
Watthayu, W., & Peng, Y. (2004). A Bayesian network based framework for multi-
criteria decision making. MCDM 2004, Whistler, BC, Canada (pp. 611).
Yurimoto, S., & Masui, T. (1995). Design of decision support system for overseas
plant location in EC. International Journal of Production Economics, 41, 411418.
www resource 1. Available from http://hdr.undp.org/.
www resource 2. Available from http://www.indexmundi.com/.
www resource 3. Available from http://www.oecd.org/.