Professional Documents
Culture Documents
This manuscript has been reproduced from the microfilm master. UMI
films the text directly from the original or copy submitted. Thus, some
thesis and dissertation copies are in typewriter face, while others may be
from any type of computer printer.
In the unlikely event that the author did not send UMI a complete
manuscript and there are missing pages, these will be noted. Also, if
unauthorized copyright material had to be removed, a note will indicate
the deletion.
UMI
A Bell & Howell Information Company
300 North Zeeb Road, Ann Arbor MI 48106-1346 USA
313/761-4700 800/521-0600
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
A METHODOLOGY FOR DESIGNING EFFECTIVE VALUE CHAINS: AN
R C. Baker
Supervising Professor
J. H. Semple
H. W. Corley
W. E. Pinney
FaizulHuq
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Copyright by Srinivas Talluri 1996
AH Rights Reserved
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
A METHODOLOGY FOR DESIGNING EFFECTIVE VALUE CHAINS: AN
by
SRINIVAS TALLURI
of the Requirements
DOCTOR OF PHILOSOPHY
December 1996
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UMI Number: 9718559
UMI
300 North Zeeb Road
Ann Arbor, MI 48103
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ACKNOWLEDGMENTS
This dissertation could not have been written without the guidance and supervision of
Dr. R. C. Baker. Dr. Baker encouraged and challenged me in developing new research ideas
throughout my doctoral program. He always had an open door for constructive intellectual
discussions, which brought out the best in me. I cherish the countless hours he had spent with
All the members of my committee, Dr. Baker, Dr. Semple, Dr. Corley, Dr. Pinney, and
Dr. Huq have provided me with valuable suggestions and comments which made this
dissertation work so much more interesting and thought provoking. I would like to thank Dr.
Semple for introducing me to the world of data envelopment analysis; Dr. Pinney for acting as
my mentor and leading me through the program; Dr. Huq for his support and advise; Dr.
Corley who instigated my interest in applied operations research. Special thanks to Dr. Sarkis
I dedicate this dissertation to my wife, Manogna, who has been extremely supportive,
patient and caring at all times. If not for her invaluable encouragement and help I would not
I would like to specially thank Amma and Nanna who taught me the value of higher
education, and gave me all the possible support in the world that could be given to a child. I
iv
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
will always remember the inspiration and guidance provided by my brothers, Chammanna and
Rajee, throughout my education, and well wishes from Sunhha, Subha, Suraj, and Nitya .
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
ABSTRACT
Publication No.__________
Value chains are being envisioned by many experts as the organizational forms of the
contribute their expertise to the value chain. These new forms o f organizations are considered
as the solution for quick introduction of variety of products with high quality and short lead
times. Although several conceptual value chain models have been proposed by researchers,
literature lacks formal decision making models for value chain design and improvement. Such
decision models would be very useful for managers involved in value chain design.
programming model for effectively designing value chains. A combination of new extensions
in data envelopment analysis and integer programming are utilized in the solution procedure.
vi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
This research also provides effective benchmarks for improving poor performers in
every category of value chain processes. Benchmarks are identified by utilizing a clustering
algorithm. The poor performers can use these targets to understand the policies and
participating in the value chain. This technique evaluates the performance of business
vii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
TABLE OF CONTENTS
ACKNOWLEDGMENTS.....................................................................................................iv
A B S T R A C T ......................................................................................................................... vi
LIST OF ILLUSTRATIONS................................................................................................. x
LIST OF TA B L ES................................................................................................................ xi
Chapter
1. IN TR O D U C TIO N ................................................................................................. 1
2. LITERATURE R E V IE W ....................................................................................... 7
3. M ETHODOLOGY......................................................................................................36
viii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4. RESULTS AND ANALYSIS......................................................................................... 55
5. CONCLUSIONS........................................................................................................ 72
Appendix
A. T A B L E S ................................................................................................................75
B. COMPUTER PR O G RA M S..................................................................................101
B IB L IO G R A PH Y ......................................................................................................... I l l
ix
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
LIST OF ILLUSTRATIONS
Figure Page
1. Internal N e t w o r k ....................................................................................................... 9
3. Dynamic N e tw o rk ......................................................................................................11
4. Virtual Corporation.................................................................................................... 14
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
LIST OF TABLES
Tables Page
xi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
19. Range of Optimality for the Integer Programming Problem..................................98
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CHAPTER 1
INTRODUCTION
The high volatility of customer demand in the product and service industries has
working towards meeting customer oriented performance measures such as high product
quality, short lead times, high product variety, and low cost. It is a formidable task for large
or small organizations to accomplish all the aforementioned goals in a timely and efficient
manner. Large organizations are often very complex, and small organizations suffer from a
scarcity o f resources. The network organizations (NOs), virtual corporations (VCs), and
value-added partnerships (VAPs) are envisioned by many experts as the solution for quick
introduction o f a variety of products with high quality and low costs (Byme 1993, Goldman
1994, Iacocca Institute 1991, Johnston and Lawrence 1988, Snow et aL 1992). We
collectively refer to NOs, VCs, and VAPs as new value chain forms (NVCFs).
Snow et aL (1992) classified NOs into three categories: internal network, stable
network, and dynamic network organizations. They have also identified managerial roles that
are critical to the success of these organization forms. They concluded that competitors of the
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
2
perform only those functions for which the company has, or can develop, expert
skills.
outsource those activities that can be performed more quickly, more effectively, or
In order to be globally competitive, the basic tenets of other NVCFs such as VCs and
VAPs suggested in the literature are in harmony with the above mentioned criteria (Byme
1993, Goldman 1994, Iacocca Institute 1991, Johnston and Lawrence 1988, Porter 1993,
these NVCFs is the selection of competent and compatible partners (Byme 1993, Goldman
1994, Henderson 1990, Iacocca Institute 1991, Johnston and Lawrence 1988, Oliver 1990,
Porter 1993, Presley et aL 1995, Sheridan 1993, Slowinski 1992, Snow et aL 1992). These
new forms o f organizations cannot work effectively without highly efficient and compatible
partners. Although various theoretical models have been proposed in the literature, no formal
decision making models have been suggested for designing these NVCFs. In designing these
value chains, not only is there a need for strong compatibility among the participating business
processes, but more importantly, participants must be very effective in what they contribute
individually and as a group. The partner selection process is thus extremely critical for
value chains, consider a simple scenario in which a NVCF must be formed with three types of
business processes (A, B, and C). If there are 10 potential candidates for the type A process,
10 potential candidates for the type B process, and 8 potential candidates for the type C
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
business process, then the total number of combinations under consideration is 800. To
evaluate these combinations and identify the most desirable one(s) can be an extremely tedious
and time consuming process. More importantly, it is a daunting task for the decision maker to
incorporate into one decision model both the internal and external decision variables
associated with various process type candidates. Internal decision variables are the
performance measures related to the individual business process types. For example,
operating costs, number of employees, work-in-process levels, flow times, utilization, etc., are
some of the internal performance measures that can be utilized for evaluating manufacturing
different business process type candidates. For example, costs associated with the formation
of a value chain linkage and cultural compatibility of participating business processes are some
programming approach for designing effective value chains. Phase 1 uses a filtering technique
based on the internal decision variables of the candidate processes. This technique identifies
good and poor performers in each type of business process. The good performers are
considered in phase 2, and the poor performers are filtered out. This process has two
advantages:
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
4
the number of combinations that are to be considered in phase 2 are
problem.
The techniques used in phase 1 of the decision making process are data envelopment
other new extensions in DEA that effectively discriminate between good and poor performers.
Phase 2 utilizes a binary (0-1) goal programming model for selecting an effective combination
of good performers (identified in phase 1) to participate in the formation of the NVCF. This
model evaluates various value chain combinations with respect to compatibility measures.
The second goal of this dissertation is identifying effective benchmarks for improving
the poor performers in every category of business processes. Benchmarks are identified by
applying a clustering algorithm to the results obtained from DEA. Managerially, this is
extremely important, because the poorly performing candidates must understand the policies
and procedures they need to adopt in order to become competitive and participate in similar
The third goal of this dissertation is proposing a new tool for continually improving
the business processes selected in the formation of the value chain. This new technique is a
decision model used in formation of the value chain selects participating candidates that are
effective independently and as a group, there is a need for continuous process improvement.
The proposed technique evaluates the performance o f a participating business process over
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Several business processes types can be considered in designing a value chain. This
dissertation emphasizes on the key business processes related to the area of supply chain
management. The four business process types that are critical in this context are: supplier
process, design process, manufacturing process, and distribution process. The characteristics
of each o f these processes and the related critical performance measures are discussed in detail
in the literature review section. It should be emphasized that, although the decision making
process in this dissertation is conducted with four business process types, it can be extended
to other types.
five sections. Section 1 deals with the review o f various NVCFs, proposed theoretical
models, and real life examples of such new forms. Section 2 describes the supply chain
processes (supplier process, design process, manufacturing process, and distribution process)
and addresses the characteristics and critical performance attributes of each of them. Section 3
provides an introduction to DEA and extensions in DEA. Sections 4 and S review goal
Chapter 3 is subdivided into five sections. Section 1 describes the proposed decision
making process involved in designing an effective value chain. Section 2 describes the
identification o f effective benchmarks for improving candidates with poor operating practices.
Section 3 proposes a new continuous process improvement technique for evaluating and
improving the performance of participating candidates. Section 4 presents the data selection
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
6
Chapter 4 exemplifies the methodology by using a semi-realistic example with four
business process types: supplier process, design process, manufacturing process, and
distribution process.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CHAPTER 2
LITERATURE REVIEW
In the late 1950s and early 1960s, management theorists proposed decentralization of
large companies in order to better manage the workplace. Subsequently, in the 1970s, more
and more firms disaggregated their operations for reasons of economics and efficiency. This
finally lead to the development of the new organization forms or value chains o f today, such
In their preliminary work, Miles and Snow (1978), discussed the effects of market
subsequent work, Miles and Rosenberg (1982) extended the market concepts to internal work
teams. Miles and Snow (1984) introduced the concept of external groups, which they termed
business processes each contributing what it does best to the network. The NOs gained
importance in the mid 1980s and were described in more detail by Miles and Snow (1986) and
Thorellie (1986).
Conceptual and empirical articles in the area of NOs are currently receiving increased
attention. These include the effect of networks on labor relations and human
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
8
resources development (Miles 1989). Miles suggested that managers in a NO must have both
technical/managerial expertise. And managers and workers must involve in continuous skills
development to be current and competitive. He also presented new forms and roles of unions
in a NO. According to which, the primary role of a union must be to work towards upgrading
worker skills to make them flexible and better marketable. Lawless and Moore (1989) studied
the application of dynamic networks in private and public industries. Although these types of
networks are more common in private industries, their studies provided evidence of such
network forms in public services such as fire fighting and health. Kensinger and Martin (1991)
suggested financing policies, procedures, and strategies in NOs. Managerial processes for
designing, operating, and caretaking a network were suggested by Snow et al. (1992) and
Snow et aL (1992) illustrated three types of NOs: they are internal, stable, and
dynamic. In an internal network, firms own most of their assets in the business, and they do
not become involved in outsourcing. Figure 1 depicts an internal network In this type of
network, various divisions are formed within the organization which specialize in a particular
component o f business. For example, in the 1980s, General Motors significantly reduced its
subassembly. According to Snow, a well formed internal network results in reducing resource
network, a set of vendors support the lead firm. These vendors either provide inputs to the
firm or distribute its outputs. Figure 2 depicts a stable network. For example, BMW can be
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
considered to be operating with a stable network. It is estimated that 55% to 75% of total
production costs at BMW are incurred from outsourcing. Certain divisions at BMW called
the research groups are constantly working towards identifying best vendors to link with.
Stable network results in higher quality products and decreased response times.
Designers Producers
Brokers
Suppliers Distributors
Dynamic networks are formed by a group o f independent companies. The lead firm, in
this network, identifies potential partners who own a large or sometimes the entire portion of
the assets in the network. Each partner contributes its 'core competency1to the value chain.
Figure 3 depicts a dynamic network. In some cases, the lead firm contributes the core skill,
such as manufacturing in the case of Motorola, research and development in the case of
Reebok, and design and assembly in the case of Dell Computers. It is also possible for the
lead firm to undertake a pure broker role in designing and forming a network of highly
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
10
efficient business processes. For example, at Lewis Galoob Toys, a handful of key executives
Supplier] Supplier]
CORE
FIRM
Supplier] Supplier]
Thus, Galoob acts as a pure broker in forming an effective value chain of highly competent
to Snow et aL (1992), dynamic networks provide both specialization and flexibility. Each
component o f the network practices its 'core competency1, and the brokers act as a catalyst in
organizations are: network design, network operation, and network caretaking. The critical
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
communication and networks.
Designers Producers
Brokers
Suppliers Distributors
Snow et aL (1992) suggested three broker roles that are important to the success of
the NOs, they are architect, lead operator, and caretaker. The architect and lead operators
role overlap considerably and may be performed by the same individual or a group. The
primary job o f these roles is to identify and connect highly efficient and compatible firms into
an operating network. Snow argues that the partner selection process in stable and dynamic
networks is very complicated, because the business processes are mostly extemaL Although
the architecture of these networks, new managerial roles, labor relations and human resources
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
development, financing strategies are discussed in the literature, no formal decision making
A caretaker monitors the operation of the network after it is formed. Essentially, the
job involves identifying and implementing new technology developments for improving
and maintaining good relations among the participants. Although specific functions of the
caretaker are suggested, the tools and techniques that aid the caretakers job have not been
VCs are very similar to NOs. They are an alliance of independent business processes
or enterprises each contributing 'core competencies' in areas such as, design, manufacturing,
and distribution to the corporation (Byrne 1993, Goldman 1994, Iacocca Institute 1991,
Porter 1993, Presley et aL 1995, Sheridan 1993). VCs are formed in the event of a market
opportunity and would be dissolved when the opportunity passes. Similar to broker-led
dynamic network organizations, they do not own any of the individual business processes that
design, produce, and market the product. However, the VCs always indulge in temporary
Computer and Sony Corp. engaged in a similar temporary alliance to manufacture PowerBook
notebooks. TelePad Corp. collaborated with more than two dozen partners in bringing its
pen-based computer to market. IBM, Apple Computer, and Motorola have become involved
generation of computers. This idea of temporary alliances has gained prominence in service
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
13
industries also. For example, InterSolve Group Inc., a Dallas-based management consulting
firm, assembles 'just-in-time' talent to solve problems or implement strategies for a large
number o f clients. After completing the assignment, the consulting team dissolves. In one of
their projects for First Interstate Bankcorp., the expert team saved nearly $14 million from
its key capability. It will mix and match what it does best with the best of other companies.
Byre suggests that experts in the area o f VCs emphasize cultural compatibility among
participating firms and state of the art communication modes. Goldman (1994) defines a VC
capabilities.
function together.
Porter (1993) argues that one of the critical drawbacks of a VC is that the employees
in a VC must split their loyalty between the VC and the home company. He proposes a key
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
14
change in designing VC in that it needs to have its own employees. This helps in building
Presley et al. (1995) suggested an architecture for the VC. Their architecture
categorizes business processes in a VC into three types. Category 1 business processes are
collectively owned to develop enterprise objectives, strategies, tactics, and plans. Category 2
and 3 business processes are individually owned and managed to market, design, produce, and
distribute product. Figure 4 depicts the VC architecture proposed by them. Although, they
emphasize that the participating business processes must be agile, no formal decision models
Category 1 business
processes collectively
.PWngd______________
Design
Distribute
Category 2 & 3 business processes
individually owned and managed
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
15
According to Johnston and Lawrence (1988), the VAPs are " a set of independent
companies that work closely together to manage the flow o f goods and services along the
entire value-added chain." Each company in a VAP focuses on performing a single step of the
value-added chain. The primary difference between VAPs and VCs is that, the firms in a VAP
develop lasting ties with others in the value-added chain. VAPs envelop some of the
advantages of vertically integrated companies in that the managers in a VAP work toward the
common goal of making the whole VAP competitive. For example, Japanese auto companies
are a typical form of VAPs, producing only about 20% o f the value of their automobiles.
Similarly, Chryslers resurgence may be attributed to the creation of VAPs with its suppliers.
Busby and Fan (1993). The concept of extended enterprises is very similar to VAPs.
According to Busby and Fan, in an extended enterprise there is a great extent of cooperation
between firms operating at different points of the value chain. Thus, it differs from the
scale.
designing these NVCFs is the selection of highly efficient and compatible partners. Some of
the compatibility issues addressed in the literature concerning the linkages among the
standards, frequent interactions, trust, and goals in alignment with the vision of the value chain
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
16
For effective communication, the business processes in a value chain must incorporate
state of art communication strategies such as electronic data interchange (EDI) into their
operating practices. At the same time technical hardware and software costs, communication
costs, new equipment costs, standard costs, new employee costs, etc., associated with the
interactions among the participants in order to respond quickly to the changing needs of the
customer, the participating candidates must be geographically close to each other. This
assists not only in improving the frequency of interactions but also helps in decreasing
transportation costs. Culture is one of the most vital ingredients for a successful partnership.
It influences the behavior, values, and goals of the employees. The participating business
processes must have a high degree of cultural compatibility. An empowered work force,
Total Quality Management (TQM), and concurrency are some of the critical attributes used in
Since the concept of NVCFs is fairly new, related literature and conceptual models are
still at an embryonic stage and thus the area has high research potential It is evident that
without highly effective and compatible partners, these new forms of organizations cannot be
successfiiL Thus, not only is there a need for strong compatibility among the participating
business processes, but more importantly they need to be very efficient in what they contribute
individually and as a group. The partner selection process is thus very important in the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
17
A business process is that which transforms inputs into outputs. The transformation
process is governed by controls, which are generally policies and procedures followed in the
process. The actual transformation is accomplished by mechanisms, which are the tools used
Supply chain can be defined as a network of individual business processes that produce
goods and services to the end customer (Christopher 1994). According to Christopher
(1994), in a supply chain the individual organizations focus on their 'core business process',
where they have a differential advantage. The different types of NVCFs addressed in the
previous section can be considered as supply chains. The key business processes in a supply
chain are supplier process, design process, manufacturing process, and distribution process
(Christopher 1994, Davis 1993, Herald et aL 1993, Lee and Billington 1993, Presley et al
1995, Snow et aL 1992, Turner 1993). Snow et aL (1992) have also considered design
process as a critical phase in a supply chain. Each of these business processes and related
high levels of performance and in satisfying end customer needs. From a just-in-time (JIT)
manufacturer relationship (Bailes and Kleingsorge 1992, Schonberger 1986). Some of the
service, and accurate order processing (Christopher et al. 1979, Kleinsorge et al. 1992).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
18
Other intangible measures that can be used for evaluating supplier performance include
responsiveness of the suppliers personnel, and the credibility o f supplier in offering terms and
promise dates (Christopher et al. 1979, Grantzin et aL 1986, LaLonde et al. 1988).
satisfaction, product and service quality, and production costs. According to Bandyopadhyay
(1990), the design of the product should be developed based on three principles: design for
high product variety at low cost, design for quality, and design for high productivity and
process include modularity of the design (minimize the number of parts in a design),
manufacturability of the design (minimize the number of operations required to produce the
product design), and lead time for design development (Rosenthal 1992, Stevenson 1993).
Modularity refers to the simplicity o f the design. Modular designs have relatively flat bill of
materials. This allows for high product variety, since the product will be easier to
manufacture. Manufacturability o f the design refers to the ease with which the design can be
manufactured. This aids in decreasing the production costs o f the product. Other intangible
performance measures o f a design process include customer satisfaction and customer appeal
(Stevenson 1993).
Manufacturing process is the 'core' of supply chain processes. It is the module that
adds maximum value to the product. The primary function o f manufacturing is in utilizing raw
materials and sub-components to produce the final product with respect to the design
specifications. Some o f the key functional processes involved in manufacturing are: master
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
planning (CRP), and shop floor control (SFC). In general, the key performance measures
considered in a manufacturing process are setup times, flow times, work-in-process levels,
worker and machine utilization (Shafer and Bradford 1995, Vakharia 1986, Wemmorlov and
Hyer 1987).
order taking, merchandising, and physical transfer of products and services. Thus, the
distribution process envelops marketing, sales, and logistics of product movement. Physical
distribution is an integral part o f the logistics system that deals only with the physical
movement of the products from the seller to the customer. According to Magee et al (1985),
the critical performance variables of a physical distribution process are: speed of delivery,
reliability, and degree of immediate availability of the item. Availability and speed are related,
because immediate availability usually results from quick delivery. However, prompt delivery
can also be achieved without availability. More warehouses at key locations or large markets,
fast transportation modes, and rapid order processing systems improve delivery times and
and quality of information must persist between the carriers and the distribution department
(Lieb et aL 1988). Service reliability can be improved by maintaining reasonable safety stocks
at key warehouses. Operational variables that measure service reliability include stockouts per
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
20
homogeneous units in the presence of multiple inputs and outputs. The efficiency is defined as
the ratio o f the weighted sum o f outputs to inputs. DEA has been used to compare the
efficiencies o f homogeneous set of decision making units (DMUs) such as schools, hospitals,
If there are n DMUs each with m inputs and s outputs, the relative efficiency of a
s / m
Max v & tp / ujxjp
H / 7=4
k = 1 through s
j = 1 through m
i = 1 through n
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
21
The above model is solved n times in order to evaluate the relative efficiencies o f all
the DMUs. The DMU being evaluated each time is referred to as the target DMU. The model
allows the target DMU to effectively select its own optimal weights in order to maximize its
output to input ratio, but at the same time the constraint set prevents the output to input ratios
of each o f the n DMUs computed with these weights from exceeding 1. A relative efficiency
score of 1 indicates that the DMU under consideration is ratio efficient, whereas a score less
than 1 indicates that it is ratio inefficient. DMUs with a relative efficiency score of 1, and
with optimal slack variables o f zero in every solution are considered to be Pareto-Koopmans
The above fractional program (ratio model) can be converted to a linear programming
(LP) problem, where the optimal value of the objective function indicates the relative
Max ^ v*y*p
M
m
s.t UJXJP = 1
The weighted sum of the inputs for the target DMU is forced to 1 and the constraint
set is linearized, allowing for the conversion of the fractional programming problem to a LP
problem.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
22
The relative efficiency computed from problem (2.1) is referred to as the simple
efficiency score. In the evaluation of simple efficiency, (2.1) allows for complete weight
flexibility. This may result in identifying a DMU to be efficient based on some unrealistic
weighting scheme (Dyson and Thannassoluis 1988, Wong and Beasley 1990). DMUs with
such extreme weights have the potential of being 'false positive' candidates. A 'false positive'
candidate achieves a high relative efficiency score by weighing heavily on few favorable inputs
and outputs while completely ignoring the other inputs and outputs. Such DMUs are only
performing well with respect to few operating characteristics and are not indulging in good
overall practices. Figure 5, illustrates a two dimensional graphical example in which DMUs W
and Z are efficient but possible 'false positive' candidates, since their performance is highly
dependent on a single output. DMUs X and Y are efficient and also good overall performers.
Whereas, DMUs A and B are inefficient but possibly better overall performers than DMUs W
and Z. Therefore, simple efficiency score alone is insufficient when differentiating between
DMUs indulging in good overall practices and those that are not. Also, having an efficiency
score of 1 may not be managerially significant in each and every case. Two important
remedial measures that have been suggested in the DEA literature to encounter this problem
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
23
0 2/I
4 Z
0,/I
Figure 5. Illustration of Good and False Positive Performers.
Cross-Efficiencies in DEA
Cross-efficiency in DEA can be defined as the efficiency score a DMU achieves when
evaluated with the optimal weights (input and output weights obtained from (2.1)) o f another
DMU. Thus, for each DMU there will be exactly n-\ cross-efficiencies, which can be
represented in a cross-efficiency matrix (CEM). In the CEM, shown in figure 6, the element
(Ojj) in the f h row and j h column of the matrix provides the cross-efficiency score o f DMU j
when using the optimal DEA weights of target DMU /. A DMU with several high cross
efficiencies along its column in a CEM is considered to be a good overall performer, and a
DMU with several low cross-efficiencies can be regarded as a poor performer (Boussofiane et
aL 1991). The column means can be computed as an index for effectively differentiating
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
24
between good and poor performers. Thus, the performance of the DMUs can be ranked
DMU
1 2 3 n
Target DMU
1 0 0 ,2 0,3 0 ,n
n 0 nI On O ni Onn
A A A A
mean . 2 3
A limitation with the CEM evaluated from (2.1) weights is that optimal weights
obtained from this problem may not be unique. This condition occurs if multiple optimal
solutions exist. For this reason the CEM constructed from (2.1) weights is considered to be
an arbitrary evaluation. Thus, the usefulness of CEM in identifying good overall performers
can be undermined, since multiple CEMs can be constructed with different sets of optimal
weights. This ambiguity can be resolved by using formulations proposed by Doyle and Green
(1994). These formulations can be categorized into aggressive and benevolent approaches. In
these formulations Doyle and Green not only maximize the efficiency of the target DMU, but
they also take a second goal into account. This second goal, in the case of aggressive
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
25
formulation, is to minimize the efficiency of the composite DMU constructed from the other
/z-l DMUs. The outputs and inputs of the composite DMU are obtained by summing the
corresponding outputs and inputs of all the other n- 1 DMUs except the target DMU. The
weights obtained from this formulation make the efficiency of the target DMU the best that it
can be, and all other DMUs efficiencies as low as possible. Thus, the CEM evaluated from
these weights is more meaningful, and can effectively discriminate between good and poor
performers. The aggressive formulation proposed by Doyle and Green (1994), shown below,
S ( Vt Z y * )
t* h tP
s-t Z ( / Z xJi ) = 1
H *P
s w
X -Z u^ j - 0 Vi * p
fed M
2 4 r Z uf t p = 0 (2-2)
fed M
Vk,Uj>0 V k andj
where:
s _________ ___
^ ( v* yu ) is the weighted output of composite DMU.
fe d t*p
m_________ ___
Z ( UJ Z XJ ) k weig^ted input of composite DMU.
M *P
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
26
The benevolent formulation uses the same set of constraints as (2.2) with the
exception that the efficiency of the composite DMU is maximized. Doyle and Green (1994)
approach to identifying unique weights is a two step approach. In the initial step, simple
efficiency scores are obtained for all DMUs. In the subsequent step the efficiency of the
composite DMU is minimized (aggressive formulation) by using the simple efficiency score of
Rousseau and Semple (1993) approached the same problem as a two-person ratio
efficiency game. In their formulation, Player I is the minimizing player and represents the
target DMU. The target DMU selects its own nonnegative optimal weights in achieving a
virtual ratio o f exactly 1. Player II is a central evaluator (composite DMU). The payoff is
single virtual ratio. This formulation provides unique weights in a single step as opposed to
maximize target DMUs efficiency and at the same time minimize the efficiencies o f all other
DMUs in some sense. The only difficulty in considering the later goal is that the sum of the
surrogate measure for this sum of efficiencies. Since the actual objective function as discussed
above is non-linear in nature, any of the surrogate measures do not accurately portray the goal
of minimizing the efficiencies of all other -l DMUs. Thus, there is every possibility that the
optimal weights obtained from these formulations may be compromised, when minimizing the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
27
efficiency of composite DMU. Thus, there is a need for a formulation that overcomes this
problem.
performers and poor performers. Thompson et aL (1986) introduced the concept of weight
restrictions in DEA, which they called as assurance regions (ARs). According to them, ARs
can be classified into cone-ratio ARs and non-cone-ratio ARs. Polyhedral cone-ratios are
usually expressed in intersection form, as shown below, or in sum form (Chames et al 1990).
Non-cone-ratio ARs can be expressed as linked-cones which link inputs and outputs, and also
defining bounds on the input-output weights that reflect the relative importance of different
inputs and outputs. These weight restrictions are represented in the form of linear inequalities
and appended to the ratio DEA model shown in (2.1). The basic form of ARs are:
Dyson and Thanassoulis (1988) illustrated the use of weight restrictions in evaluating
local-authority rates departments. Wong and Beasley (1990) used a method that is based on
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
28
proportions for restricting weight flexibility in DEA. Although weight restrictions effectively
identify good overall performers, there are certain problems associated with them. That is the
bounds utilized in them are subjective in nature, and are a result of managerial knowledge and
expert opinion. In certain cases, it is a daunting task for the decision maker to come up with
DEA for benchmarking purposes is a recent innovation. Even though most of the
applications o f DEA may be considered as some form of benchmarking evaluation, the two
techniques have not, until recently, been explicitly linked. The literature has focused on DEA
as applied and tested for external benchmarking evaluation approaches. From a benchmarking
methodology viewpoint, it has been utilized for selection of partners for benchmarking in the
telecommunications industry (Collier and Storbeck 1993) and for travel management
benchmarking within various organizations (Bell and Morey 1995). Collier and Storbeck used
the standard DEA approach that calculated technical' efficiencies for determining
benchmarking partners. Bell and Morey applied the allocative DEA (ADEA) approach for
selection of benchmarking partners. ADEA estimates the cost expenditures of a unit that
would produce: at least the same level of all outputs, do so in no easier an operating
environment, deliver at least the same level of service, and achieve the lowest possible cost
(Banker and Mainderatta 1988). The emphasis o f ADEA application is on the identification of
appropriate benchmarking partners that use a different mix of resources that are more cost
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
29
Additional work on external benchmarking and DEA has been used for the banking
and finance industry (Barr and Seiford 1994), and the grocery industry (Athanassopoulos and
Ballantine 1995) . In these works they show that DEA can be used effectively as a
performance analysis tool, similar to the traditional gap analysis or ratio analysis used in
benchmarking- There has been less emphasis on the use of DEA for internal benchmarking. In
a relatively comprehensive search, only a single paper, by Schefcyzk (1993), was found to
reference sets are used by inefficient DMUs as 'models' for improvement. Usually, a linear
combination of DMUs in the reference set is the target that an inefficient DMU uses to
become efficient. A difficulty with these conventional reference sets is that an inefficient
DMU and its reference set may not be inherently similar in their operating practices and
characteristics. Thus, in many cases these reference targets are unattainable goals for the
inefficient DMUs. Literature has suggested other more appropriate methods for identifying
models for poorly performing DMUs based on the cross-efficiencies (Doyle and Green
1994).
Cluster analysis, principal components and multi-dimensional scaling are some of the
techniques that can be utilized to classify units into similar groups or clusters based on some
criteria. Doyle and Green (1994) have suggested that these methods are more appropriate for
identifying reference sets. The advantage of these techniques is that reference targets that are
inherently similar to the test DMU are provided. Also, in the traditional DEA benchmarking
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
30
only the inefficient DMUs are provided with benchmarks for improvement, but in this case
benchmarks are identified for some of the efficient DMUs that may be 'false positive'.
(TWA) was proposed by Chames et aL (1985). TWA assesses the performance (efficiency)
o f a DMU over time by treating it as a different unit in each time period. Usually, a
reasonable number of time periods are blocked as a window, and the input and output values
of the DMU in each of the time periods are used in the efficiency evaluation. A new window
is then formed by dropping the earliest period from the previous window and adding a new
period. The efficiency evaluations are re-assessed, and this process is repeated over the
specified time horizon. The advantage o f the TWA is that it allows for monitoring the
performance of a DMU over time, thereby providing managers critical information on the
behavior o f the process or unit under consideration. The efficiency variation over time can
occur due to changes in resource levels and/or operating practices. Figure 7 depicts an
example o f the TWA with a five period moving window. Symbol e,j represents the relative
1 through 5 are utilized. Window 2 is then formed by dropping period 1 and adding new
Although the TWA monitors the unit performance over time, its use is limited from a
process improvement standpoint. For example, consider a situation where the earliest period
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
31
dropped from a window has a high relative efficiency score, then the efficiency score of the
new period added may be misleadingly high since it is being compared to other periods
without the period with high efficiency. Thus, the TWA may overestimate the relative
efficiency scores, making it unsuitable for accurate performance evaluation and process
improvement. One would like to have periods with good operating practices, rather than
misleadingly efficient periods, for evaluating new periods and aiding in continuous process
improvement.
Period
1 2 3 4 5 t
window k . . #
organization can often have multiple objectives that need to be optimized, some o f which can
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
32
be something other than profit or cost. For solving such types of problems with multiple
GP model was first introduced by Chames and Cooper (1961) for dealing with
infeasible linear programming problems. The technique has been expanded to deal with
planning, medicare planning, etc (for more information on GP see Lee 1972). The approach
o f GP is to establish a specific numerical goal for each of the objectives, and obtain a solution
that minimizes the weighted sum o f deviations of these objectives from their respective goals.
The goals can be one-sided or two-sided. In the case of a lower one-sided goal, a lower limit
that one does not want to fall under is established. Upper one-sided goal sets an upper limit
that may not be exceed. A two-sided goal sets a specific target that one does not want to miss
on either side.
Goal programming models can be categorized into linear goal programming, integer
goal programming, and non-linear goal programming. Although goal programming reflects
complex reality and allows for multidimensionality of the objective function, it has some
limitations. Primarily, the decision maker must either set priorities to goals, which in some
cases is difficult to do; or assign weights to deviations, which in most cases is subjective in
nature.
mathematical methods, numbering in hundreds, that can be used to find out which objects in a
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
33
set are similar." In the late 1970s and early 1980s several hundred articles in cluster analysis
have been published. Most of these applications utilized methods of hierarchical cluster
analysis. The steps involved in hierarchical cluster analysis are: 1) identification of the data
matrix, where in the columns represent objects to be clustered and the rows are the attributes
that describe the objects; 2) optional standardization of the data matrix, which allows for
the similarity among all pairs of objects; 4) application of clustering method to process the
values of resemblance coefficient, which results in a tree diagram, or dendrogram, that depicts
the hierarchy of similarities among all pairs of objects. Clusters can be directly read off from
Resemblance Coefficients
coefficients that are most widely used are: the Euclidean distance coefficient (eJk); the
coefficient of shape difference (Zjt); the cosine coefficient (c7*); the correlation coefficient (ry*);
The Euclidean distance coefficient between two objects j and k is expressed by the
equation:
172
ejk= 2>-~)2
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
34
ejk and its variations such as the average Euclidean distance coefficients {djk) can be
applied in situations where size displacements among the data profiles are important (Fanizza
and Bogyo 1976). These coefficients are dissimilarity coefficients, where higher values
Zjk is a dissimilarity coefficient suggested by Penrose (1933). Its values range from 0
to oo. Boyce (1969) used zjk to measure resemblance between the skulls of hominoids. This
coefficient can be used when differences in shapes instead of sizes is desired. Outside of
Cjk has been widely applied in geology (Harbaugh and Merriam 1968, Imbrie and Purdy
1962, Thompson 1972, Bell 1976). In mathematics of linear algebra the usual method of
comparing two vectors is to compute the cosine of the angle between them, which can be
used to index their agreement. This similarity coefficient is defined over the range -1 to 1.
rjk is a similarity coefficient defined over the range >1 to 1. The coefficient rjk is widely
^ n
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
35
When size displacements should not be allowed to affect similarity, coefficient such as can
djk and bjk are dissimilarity coefficients defined over the range 0 to 1. These
coefficients are mostly applied in ecology. For more information on the applicability of these
Clustering Methods
methods involve a sequence of steps that shrink the resemblance matrix in forming clusters of
similar objects. At the start of the steps in the clustering method, each object is considered as
being a separate cluster. Each o f the clustering steps will merge the two most similar clusters
that exist at the start of the step, thereby decreasing the number of clusters by one. There are
several clustering methods that have been widely applied. Some of them are: the unweighted
pair-group average method; the single linkage clustering method; the complete linkage
clustering method; Wards minimum variance clustering method; the centroid method.
The unweighted pair-group average method is probably the most widely applied
technique in the above mentioned group. We utilize this clustering method for identifying
effective benchmarks for improvement. In this method, initially, the two objects with highest
similarity coefficient value are grouped into a single cluster. In updating the resemblance
matrix, the method computes similarity between any two clusters by averaging the similarities
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CHAPTER 3
METHODOLOGY
The methodology suggested in this section accomplishes three goals. The first goal
involves the design of an effective value chain. The participating business processes are
assumed to be independent with each performing a specific function of the value chain. A
new extension in DEA called the "pairwise efficiency game formulation" is proposed and used
combination of business processes in the formation of the value chain. The second goal
involves the identification of effective benchmarks for the poorly performing candidates.
Cluster analysis and cross-efficiency evaluations are used together for selection of appropriate
benchmarks for improvement. The third goal involves continuous improvement of the
selected value chain processes. A new technique called the "modified window analysis
Figure 8 depicts the decision making process involved in the formation of a NVCF.
The decision makers in this framework are top executives acting as brokers. These brokers
act as agents in selecting efficient and compatible partners. For more information on broker-
lead dynamic organizations, see Snow et aL (1992). hi the event of an identified market
opportunity for a product, the brokers are expected to respond by creating a NVCF
36
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
37
and delivering the product at low cost, high quality, and short lead times. Although the
OSIB'ERS
Phase 1 o f the decision making process is initiated with the identification of the
required business process types by the brokers followed by the consideration of potential
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
38
candidates for each business process type. The key input/output (I/O) measures for each
business process type are then selected. These I/O measures must be in alignment with the
objectives, goals, policies, and procedures o f the NVCF. The inputs should encompass any
resources utilized by the business processes, and the outputs should include a range of
performance and activity measures (Boussofiane et al 1991). Data on all the identified I/O
measures are then collected, and DEA is used to evaluate all potential candidates of each
process type.
Problem (2.1) is utilized to evaluate the simple efficiency scores of supplier, design,
efficiency score alone is insufficient in discriminating between good and poor performers. It is
possible that some inefficient DMUs can in fact be better overall performers than other
efficient DMUs. This is because some of the efficient candidates identified by DEA may be
'false positive', Le., they can achieve a relative efficiency score of 1 by indulging in an
unrealistic weighting scheme. These 'false positive' candidates do not have good overall
practices, and only perform well with respect to few favorable inputs and outputs. They
obtain a high relative efficiency by heavily weighing those few inputs and outputs that are
most beneficial to themselves. These types of candidates are ineligible for participating in a
NVCF. To identify and select good overall performers under each category of business
processes a CEM is utilized. The CEM for each case is evaluated by using the optimal
weights obtained from problem (3.1), which we call as the "pairwise efficiency game
formulation."
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
39
Vi * p
s
Min v*y*,
jH
m
St 2 ui xi' = 1
X
x m
S V* V * /-Z >*A - 0
fc4 x
2 V ^ - I tjCjp = 0 ( 3 . 1)
vt Uj > 0 1/ k and j
where:
Similar to Doyle and Green (1994) formulation, this model identifies optimal weights
in a two-step procedure. In the initial step the simple efficiency scores are evaluated by
utilizing (2.1). In the subsequent step, instead of minimizing the efficiency of the composite
DMU, the target DMU minimizes the efficiencies of each of the n- 1 DMUs separately. This
formulation allows for pairwise comparison of the DMUs, where the target DMU selects
optimal weights that maximize its simple efficiency score and at the same time minimize the
efficiency score of each competitor separately. Since, only two DMUs are compared each
time in (3.1), the restriction that all other DMUs efficiencies being less than or equal to 1 is
removed. The primary advantage of this formulation is that the optimal weights obtained will
not be compromised, which may happen when minimizing the efficiency of the composite
DMU.
In the proposed formulation, the optimal weights of a target DMU may vary
depending on the competitor. In essence, the target DMU can involve in multiple strategies
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
40
(optimal solutions), i.e each time, it emphasizes on its strengths that are weaknesses for a
specific competitor. The obtained optimal weights are used to evaluate the CEM. The
columns means o f the CEM can be utilized to differentiate between good and poor
performers. Good performers with high cross-efficiency means are selected and considered in
the phase 2 of the decision making process, and the poor performers with low mean cross
efficiency scores are filtered out. The mean cross-efficiency score level that makes the
distinction between good and poor performers is set according to the policies and procedures
o f the NVCF, and requires knowledge on the criticality of the particular business process type
in the value chain. The managers or brokers must make an astute decision in identifying these
cutofflevels.
DEA and the cross-efficiency extensions reduce the complexity o f the problem by
minimizing the number of combinations that are to be analyzed in phase 2 of the decision
making process. Also, this procedure selects good performers to be considered in phase 2 of
the analysis, Le., candidate processes indulging in good overall practices, which is a key for
Following is a list of I/O measures of the four business process types, which will be
Supplier process
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
41
Outputs: number of shipments to arrive on time (# OT)
Design process
Manufacturing process
number of workers (# W)
Distribution process
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
42
These input-output measures are not intended to be exhaustive, but they are some
general measures addressed in the related areas o f literature (Christopher et al. 1979,
Kleinsorge et al. 1992, Leib et al. 1988, Magee et al. 1985, Rosenthal 1992, Shafer and
Bradford 1995, Stevenson 1993, Vakharia 1986, Wemmerlov and Hyer 1987).
The selected candidates of each business process type are combined into all possible
combinations. For example, if two candidates from each of the four business process types
are selected in phase 1, then the total number o f combinations are 24 . These sixteen
are evaluated with respect to compatibility criteria. Some of the compatibility issues
addressed in the literature concerning the linkages among the participating business processes
are: effective communication networks, cultural compatibility, frequent interactions, trust, and
goals in alignment with the vision of the enterprise (Gilbert et aL 1994, Henderson 1990,
Oliver 1990, Slowinski 1992). By incorporating some of the factors mentioned above into the
decision making process, phase 2 selects an effective combination of partners, thereby insuring
that the participating candidates are efficient individually and as a group. For example,
assume the following goals have been identified by the brokers as important in the formation
of the NVCF.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
43
These are technical hardware and software costs for effective communication, order
processing costs, new equipment costs, new employee costs and other costs
participating candidates must be geographically close to each other. This assists not
only in improving the frequency of interactions but also helps in decreasing the
transportation costs.
The inception time is defined as the time taken for a NVCF to be operational To
capture the market opportunity at the earliest, it is necessary to have relatively low
inception times. Also, low inception times decrease the amount of lost sales.
Culture is one o f the most vital ingredients for a successful partnership. It influences
the behavior, values, and goals of the employees. An empowered work force,
Total Quality Management (TQM), and concurrency are some of the critical
compatibility criteria are determined. Conversion factors between cost and the remaining
three goals distance, time, and culture are utilized as the weights in the objective function of
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
44
the model. The conversion scale between cost ($) and distance (miles) can be identified by
estimating the cost incurred in covering a mile of distance. The relationship between cost ($)
and inception time (days) can be determined by estimating the cost of lost sales for each day
o f delay in the inception process. The intangible goal (culture) can be operationalized by
estimating the cost involved in building a high degree o f cultural compatibility. These costs
may be in the form of training and empowering the work force, building teams, implementing
Although the binary GP formulation, shown as (3.2), involves four business processes
types and four goals, it is extendible to any number of processes and goals. The combination
that mmimi7.es the sum o f the weighted deviations from the best goal compatibility measure
a b e d
st Z Z Z Z x *u =1
M jm1 JU1 /-I
a b e d
Cijkl XijkJ " V / Cjno1
W j *4 JM M
a b e d
ZZZZ ^
t*l jM JW /I
Vi tmm
a b e d
Z Z Z Z ^ ijU x yu ~ v * ~ ^min
tl j*4 /=!
XgU = 0 or 1
Vi, v2, v3, and v4 > 0 (3.2)
where:
/ = I through a.
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
45
j = I through b.
k = 1 through c.
I = 1 through d.
cmin is the cost bound derived from the minimum cost combination.
dmin is the distance bound derived from the minimum distance combination.
tmin is the time bound derived from the minimum inception time combination.
knin is the cultural bound derived from the minimum cultural cost
combination.
w/, w2, vvj, and w4 are the weights derived from the conversion factors,
vi, v2, v3, and v4 are the deviations from the best goal compatibility measure.
The solution to the above binary goal programming model identifies an effective
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
46
(Athanassopoulos and Ballantine 1995, Banker and Mainderatta 1988, Barr and Seiford 1994,
Bell and Morey 1995, Collier and Storbeck 1993, Sarkis and Talluri 1995, Schefcysk 1993,
reference DMUs, called peer group or peer facet. The inefficient DMU can benchmark with
the facet members. Thus, by comparing with these facet members the inefficient DMU needs
to either decrease its inputs for the same level of outputs, or increase outputs for the same
level o f inputs. However, as explained earlier, a linear combination o f DMUs in the reference
set is the target that an inefficient DMU uses to become efficient. A difficulty with these
conventional reference sets is that an inefficient DMU and its reference set may not be
inherently similar in their operating practices and characteristics. Thus, in many cases these
reference targets are unattainable goals for the inefficient DMUs. Literature has suggested
other more appropriate methods such as cluster analysis, multi-dimensional scaling, principal
component analysis for identifying benchmarks for poorly performing DMUs based on the
In this dissertation, cluster analysis will be used in identifying the benchmark DMUs.
In a CEM, the cross-efficiencies under the column of a DMU represent how the particular
DMU is appraised by other DMUs. Computing the correlation coefficient between a pair of
columns tells us how similarly those two DMUs are appraised by their peers. A high positive
correlation coefficient indicates that the two DMUs are inherently similar with respect to their
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
47
resemblance matrix and executing a clustering method yields clusters with inherently similar
DMUs. The DMU with the highest column mean in a given cluster can be used as a primary
This technique will be applied to all candidates in each of the four business process
types. The identification of benchm arks for improving poor performers provides useful
information to the managers. They will obtain information on their existing practices and who
they need to target to improve their performance. Thus, DEA and the cross-efficiency
extensions allow for effective benchmarking, which is critical for the poorly performing
Snow et aL (1992) suggested that afier the formation of a network, the caretakers
responsibilities begin. The primary responsibility of the caretaker is to make sure that the
participating candidates are continually improving their operations, keeping abreast with
technological changes, and maintaining good working relations with other participants.
Although the decision making process involved in the formation of the NVCF
effectively selects a combination of partners that are efficient individually and as group, there
is always scope for continuous process improvement. In this dissertation, we present a new
tool that aids management (caretaker) to evaluate and improve the productivity of the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
48
The changes in business process efficiency over time can occur due to assignable
drivers and/or natural drivers. Assignable drivers are defined by operating characteristics of
the process. For example, in a m anufacturing process, assignable drivers are number of
costs, lot sizing, and other variables that affect process efficiency and productivity. Natural
drivers are exogenous events, such as seasonal factors, customer demand patterns, inflation,
etc. It is not usually practicable to constrain the occurrence of the natural drivers, but it is
within the scope o f management planning and control to identify the optimal values of the
assignable drivers and implement them. However, since there can be many assignable factors
that affect the performance, it is a difficult task to consider all factor level combinations
simultaneously in determining the optimal factor levels. Thus, management must prioritize the
factors and consider few factors at a time in identifying best operating practices for continuous
improvement.
al (1985). As explained earlier *window analysis', which we refer to as TWA may not be
appropriate for continuous process improvement. A variation of the TWA called modified
window analysis (MWA) is used for continuous process improvement. The following section
To m ake the window analysis suitable for continuous process improvement, instead of
dropping the earliest period from a given window, the MWA drops the period with lowest
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
49
mean cross-efficiency score. This period is identified from the CEM, which is evaluated using
optimal weights obtained from the pairwise efficiency game formulation. This modification
allows for the comparison of a new period with periods of good overall practices. From a
process improvement standpoint this analysis is superior to the TWA, because the relative
efficiency for a new period is challenged against the 1)681' of the previous periods, thus
providing valuable information on performance and scope for improvement of the business
process. The operating practices of the period with the highest mean cross-efficiency score in
a window are implemented in the new period. This process is continued through the entire
time horizon. Figure 9 depicts the MWA with five periods in each window.
Period
1 2 3 4 5 . t
window k
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
50
(assuming that it has the lowest mean cross-efficiency score) is dropped out from the analysis.
This process is repeated until the last period in the planning horizon is added. This technique
can be applied individually to the four selected value chain process in a specified time horizon.
proposed MWA over the TWA in evaluating and improving the performance of a DMU over
time. Two inputs and two outputs are considered in the illustration. The hypothetical data
The results of the DEA assessments of window 1 (W l), containing five periods, are
efficient, and inefficient in periods 2 and 4. Periods 2 and 4 achieved relative efficiency scores
of 0.750 and 0.727, respectively. TWA evaluations are then conducted in window 2 (W2) by
dropping period 1 from Wl and adding a new period 6. DEA identified periods 3, 5 and 6 to
be efficient in W2, and periods 2 and 4 to be inefficient. Thus, the TWA identified new period
6 to be efficient.
For comparison purposes, the MWA assessments are performed by dropping period 2
from W l (which resulted in lowest mean cross-efficiency score o f 0.551) and adding the new
period 6. The mean cross-efficiency scores are shown in the CEM, table 3, which is evaluated
by using weights obtained from the pairwise efficiency game formulation. MWAs W2
evaluation results, shown in table 2, identified periods 1, 3, and 5 to be efficient, and periods 4
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
51
and 6 to be inefficient. The new period 6 achieved a relative efficiency score of 0.850. It is
evident from this example that TWA identified period 6 to be efficient when in fact it was
inefficient with a relative efficiency score o f 0.850. This has occurred because of dropping an
efficient period, period 1, in the TWA and comparing the new period 6 with other periods,
which resulted in the overestimation of period 6s efficiency score. There is good possibility of
this happening whenever an efficient period is dropped, and the added new period is compared
to other periods in the window. MWA overcomes this problem by dropping the period with
the lowest cross-efficiency mean score. This modification not only provides a more accurate
portrayal o f the DMUs performance, but also proves to be extremely useful in process
improvement. This is because from a process improvement standpoint one would like to have
periods with good operating practices evaluating the new periods and providing effective
benchm ark s for improvement. In this illustration, period 1, which achieved the highest mean
performance. This process can be continued for the entire planning horizon.
The data utilized in this dissertation is obtained from previously published data sets
and by random number generation from predefined distributions. The supplier data set is
selected from a case study that utilized DEA for monitoring customer-supplier relationship
(Kleinsorge et al 1992). The supplier inputs considered are total cost of shipments (TC) and
number o f shipments per month (# S). The tangible outputs utilized in the study are number
o f shipments to arrive on time (# OT) and number of bills received from the supplier without
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
52
errors (# B). The intangible outputs considered are ratings for service-quality experience (E)
and service-quality credence (C). Experience and credence describe separate dimensions of
Reliability refers to the degree of commitment of the supplier in meeting the demands and
deadlines o f the customer. Responsiveness is the ability of the supplier to address the
customers concerns in an efficient manner Access is the ability to contact the supplier for
m aking the required changes to meet the customer needs. Courtesy refers to the customers
future performance of the supplier. Credence measures involve degree of trust in the supplier
providing an arbitrary scale from 0 to 50 points for each of the aforementioned six service-
quality determinants. Therefore, the highest score that can be achieved for experience is 300
(six ratings o f 50). The service-quality credence allows for maximum score of 100 (two
ratings o f 50). They checked the sensitivity of the DEA results to the scale of 50, for the
qualitative variables, by considering scaling by multiples of 10. Their results did not change
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
53
It must be pointed out that Kleinsorge et al. (1992) evaluated the performance of a
single supplier over eighteen periods. They applied DEA by considering the supplier as a
different entity in each time period. However, in this dissertation we are considering this data
set as eighteen different suppliers. This assumption does not result in any changes in the DEA
evaluations.
The manufacturer data is obtained from an efficiency study of alternate plant layouts
conducted by Shafer and Bradford (1995). They utilized simulation and DEA for evaluating
alternate plant layouts. The cell formation procedure developed by Vakharia and Wemmerlov
(1990) was used by them in generating forty seven layouts. The inputs utilized are number of
workers (# W) and number of machines (# M). The outputs included average work-in-
process (WIP), average flow time (FT), and resource utilization (UTL). In our analysis these
The physical distribution data is selected from a carrier performance study performed
by Dresner and Xu (1995). They considered thirteen carriers and evaluated the effect of
customer service variables on customer satisfaction and profitability of the carriers. The input
considered from the data set is the operating costs per dollar revenue (OC). The outputs
utilized are percentage of on-time deliveries (OTD), service level (SL), and percentage of
accurately handled customer orders (AHO). Service level can be defined as the percentage of
The data for the design process is obtained by generating random numbers from
predefined distributions. The inputs considered are number of employees (# E) and average
operating costs/period (OC). Outputs used are average number of parts in the design (# P),
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
54
average number of operations per product (# O), and average lead time for design
development (LT). For the continuous variables, RAND function in EXCEL is used to
generate random values from predefined uniform distributions. For discreet variables,
The DEA and GP models are solved using LINDO (Linear, Interactive, and Discrete
Optimizer) Release 5.0. In phase 1 of the decision making process, a total of 2,927 linear
programs have been solved for selecting an initial set of candidates for the participation in the
value chain. In order to solve these problems iteratively three FORTRAN programs were
written to call LINDO subroutines. The initial program computes the simple efficiency
scores; the second program evaluates the optimal weights from the pairwise efficiency game
formulation; the third program computes the CEM. These programs are included in the
appendix.
In phase 2 o f the analysis, the GP model was solved by using the branch and bound
algorithm in LINDO. Cluster analysis was performed through a personal computer based
statistical software package. Finally, a total of 200 linear programs were solved for MWA
using LINDO.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CHAPTER 4
The simple efficiency scores of the eighteen suppliers are evaluated by utilizing
problem (2.1). The supplier data and the relative efficiency scores are provided in table 4.
DEA identified suppliers 2, 3, 4, 6, 13, 14, 15, and 18 to be efficient with a relative efficiency
score of 1. The remaining ten suppliers with relative efficiency scores of less than 1 are
considered to be inefficient. As discussed earlier, since the simple efficiency scores obtained
from (2.1) may not accurately portray the overall performance of a DMU, a CEM is evaluated
to discriminate between good and poor performers. This CEM is generated by using the
factor weights obtained from the pairwise efficiency game formulation shown as problem
(3.1). Table 5 depicts the supplier CEM of the order 18x18. The column means of the CEM
An index which measures the 'false positiveness1 of the DMUs can also be used to
differentiate between good and poor performers. The false positive index, FPL shown as
(4.1), relates to the percentage increment in efficiency that a DMU achieves when moving
from peer-appraisal to self-appraisal. The FPI is similar to the maverick index suggested by
Doyle and Green (1994). Peer-appraisal for a DMU is how it is rated by other DMUs, a
measure for this is the column mean computed from CEM. Self-appraisal is the simple
55
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
56
where:
The higher the value o f FPIp, the more false positive is DMU p. A relatively efficient
DMU with a high FPIp value can be considered to be a strong 'false positive1DMU that is not
From the CEM in table 5, it is evident that simple efficient suppliers 6, 15, and 18
resulted in the highest mean scores of 0.914, 0.894, and 0.925 respectively. These three
suppliers can be considered as good overall performers, since they achieved relatively high and
consistent cross-efficiency scores along their columns in the CEM. The standard deviation of
the columns can be considered as a measure for consistency. Suppliers 6, 15, and 18 resulted
in relatively low standard deviations of 0.084, 0.094, and 0.086 respectively. A low standard
deviation represents that the cross-efficiency scores obtained for a DMU are close to the mean
score. These suppliers can also be considered as least 'false positive', since their FPI values of
9.44%, 11.91%, and 8.09% respectively, are the smallest in the entire data set. The supplier
FPI values are shown in table 6. Based on these evaluations suppliers 6, 15, and 18 are
considered for the phase 2 o f the decision making process, and the remaining fifteen suppliers
It is interesting to note that supplier 17, which was considered to be inefficient from
problem (2.1) evaluations, is in fact a better overall performer than efficient suppliers 2, 3, and
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
57
4. Supplier 17 resulted in a mean score o f 0.817, whereas suppliers 2, 3, and 4 achieved mean
scores of 0.793, 0.762, and 0.791, respectively. Thus, it is possible for an inefficient DMU to
be a better overall performer than some o f the efficient DMUs. Although suppliers 13 and 14
resulted in higher mean scores o f 0.849 and 0.833 respectively, their cross-efficiencies are not
as consistent as the selected suppliers 6, 15, and 18. This conclusion can be reached by
closely observing the CEM columns and the corresponding standard deviations.
The data for the design process and the simple efficiency scores are shown in table 7.
The data in this case is generated form the random number generator functions RAND and
RANDBETWEEN in EXCEL software. The RAND function returns a random value from a
uniform distribution with specified lower and upper limits, and the RANDB ETWEEN
function returns a random integer value from a user specified range. Fifteen data points with
two inputs and three outputs are generated using these functions. The number o f employees
functions with limits of $20,000 to $40,000 is used for the operating costs (OC); 6 to 12 for
and 20 to 40 days for average lead time for design development (LT).
In DEA, large value of an output is better than small, and small value o f an input is
preferred to large. For the design process, since lower values of outputs # P, # O, and LT
indicate better performance, these three variables are transformed by taking an inverse before
the application of DEA From table 7, it is evident that designers 9 and 11 are the only two
efficient ones with relative efficiency score of 1. Designer 7 also resulted in a high simple
efficiency score of 0.994. The designer CEM shown in table 8 indicates that designers 9 and
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
58
11 with cohinm means o f0.950 and 0.944 respectively are indulging in good overall practices.
They also exhibited low standard deviations of 0.056 and 0.064. These designers also
achieved lowest FPI values of 5.26% and 5.93% in the entire data set, the FPI results are
depicted in table 9.
Although designer 7 exhibited a high relative efficiency score, its mean cross efficiency
score is only 0.577 with the highest standard deviation of 0.169. Its FPI value is 72.27%,
which is the largest in the entire data set. Therefore, designer 7 can be considered as a strong
'false positive' candidate. None o f the other designers resulted in significant mean scores.
Based on this analysis, designers 9 and 11 are considered for the phase 2 of the evaluation
process.
Before evaluating the forty seven manufacturers, the outputs, average WIP level and
average FT are transformed by taking an inverse for the same reasons mentioned in design
process evaluations. O f the forty seven manufacturers, initial DEA assessments identified
manufacturers 9, 10, 14, 22, and 47 to be efficient. The input-output data and the efficiency
scores are shown in table 10. Problem (3.1) is then utilized to identify the weights used in the
CEM evaluation. From the manufacturer CEM, shown in table 11, it is evident that
manufacturers 14, 15, and 22 exhibited several relatively high cross-efficiency scores along
their columns in the CEM. These manufacturers are indulging in good overall practices with
relatively high mean cross-efficiency scores o f0.783, 0.724, and 0.761 respectively. Also, the
standard deviations of the cross-efficiencies for these manufacturers are relatively low, there
by indicating consistency in performance. Table 12 depicts the FPI values for all the
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
59
manufacturers Of the entire set, manufacturers 14, 15, and 22 resulted in the lowest FPI
score o f 0.96, it proved to be a better overall performer than the efficient manufacturers 9, 10,
and 47, which resulted in low mean cross-efficiency scores of 0.126, 0.251, and 0.257
respectively. In fact, based on the mean scores manufacturer 9 is ranked forty fourth in the
entire set. DMUs 9, 10, and 47 can be considered to be highly 'false positive' with poor
operating practices. The FPI value for manufacturer 9 is 693.65%, which is the highest in the
data set. Manufacturers 10 and 47 also resulted in high FPI values o f 298.41% and 289.11%
respectively. Based on these results manufacturers 14, 15, and 22 are considered for the
data and the simple efficiency scores are shown in table 13. From the CEM shown in table 14,
it is evident that distributors 1, 2, 5, 9 and 10, with mean cross-efficiency scores of 0.944.
0.948, 0.933, 0.962, and 0.984 respectively, are indulging in good overall practices. They
also exhibited low column standard deviations o f0.008, 0.007, 0.009, 0.023, and 0.010 in that
order. The FPI values, shown in table 15, of these five distributors are relatively low when
compared to other distributors. These five distributors are considered for the phase 2 of the
0.915, it is not selected for phase 2 since it resulted in a FPI value o f 4.81%, which is the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
60
4.2 Phase 2 o f Value Chain Design
designers, 3 manufacturers, and 5 distributors have been selected for consideration in the
x 5, which equals to 90. Table 16 provides the compatibility data that is used in the integer
programming problem.
The cost associated in forming the linkage for each of the 90 combinations is
is assumed that these are the estimated costs incurred per year for each combination. The
distance measure is related to the transportation distance associated with each combination. It
is the sum o f distances between supplier and manufacturer, and manufacturer and distributor.
These distances are randomly generated by RANDBETWEEN function with limits of 200 to
the distances from combination 46 to 90 are repeated, Le., combinations 1 and 46, 2 and 47,,
, . . , 45 and 90 have the same distance metric. This is because these pairs of combinations
have the same supplier, manufacturer, and distributor associated with them. The inception
times are obtained from RANDBETWEEN function with limits of 30 to 90 days. The final
metric is the estimated cost per year for each of the combinations in terms of conducting
training programs and seminars, forming teams, implementing TQM, etc., for building a better
culture and vision for the value chain. This metric is used as a proxy for cultural
compatibility, the lower the cost per year the higher is the compatibility among the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
61
The (0-1) goal programming problem is shown in table 17. The objective function
minimizes the deviation variables (v, s) form the best goal compatibility measure. The
coefficients for deviation variables v2 (distance) and v3 (inception time) are the conversion
factors between cost and miles, and cost and days, respectively. The conversion factor
between cost and miles is obtained by assuming the average transportation cost to be
20,000, which is $800/mile/year. This is used as the coefficient of v2 in the objective function.
The conversion factor between cost and days is obtained by assuming the product life
span to be 5 years, annual demand of 20,000 units, and product price of $ 100/unit. The total
estimated sales would then be 5 x 20,000 x 100, which is $10 million for 5 years. This
amounts nearly to $5,480/day for 5 years (assuming 365 days/year), which can be considered
as the cost of lost sales for each day delay in the inception process. However, since all the
delay in the inception process occurs in the first year, the cost of lost sales must be spread
evenly across the entire five year period. For example, if there is a 10 day delay in the
inception process this accounts for lost sales of $54,800 which occurs in the first year. Since
the objective is to minimize the cost/year, $54,800 must be divided by 5 years to get the
amount of lost sales per year. Therefore, the conversion factor that is to be used to factor this
into above situation is $5,480/5, which is $ 1,096/day/year. This factor is used as the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
62
In the integer programming model, shown in table 17, the first constraint is the hard
constraint which allows for selection of only one combination. The other four constraints are
the soft constraints with right hand side values corresponding to the best goal compatibility
measures. These values are selected by identifying the minimum values in each of the four
compatibility measures across the ninety combinations. The values identified are $110,482,
226 miles, 31 days, and $20,092 for constraints 2, 3,4, and 5 respectively.
The optimal solution to the integer programming problem is combination 48, which
consists o f designer 11, supplier 6, manufacturer 14, and distributor 5. The objective function
vahie for this least cost combination is $48,731. These candidates are utilized in the formation
of the value chain. Table 18 lists all the ninety combinations arranged in the ascending order of
the Z-vahies. If combination 48 could not be selected for other reasons beyond the scope of
the models applied in this methodology then the next best combination can be selected from
table 18. Combination 3 with designer 9, supplier 6, manufacturer 14, and distributor 5 is the
Table 19 provides the sensitivity analysis for the coefficients o f deviations in the
objective function. The ranges of optimality for w\ (coefficient of vt) and w4 (coefficient of v4)
have no meaning because Vi and v2 are represented in dollars. However, the ranges of
optimality for w2 (coefficient of v2) and w3 (coefficient of v3) are of importance to the
managers in evaluating the strength of the final solution. These ranges are:
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
63
When other coefficients in the objective function are kept constant, as long as w2
changes in the specified range, the objective function value changes, but the optimal solution
will still be combination 48. The left side limit does not have any meaning, because w2 cannot
take negative values. Ifw2 increases to more than 17039.25 then the new optimal solution is
combination 24. Similarly, other coefficients held constant, as long as w3 changes in the
aforementioned range, the objective function value changes, but the optimal solution will
remain to be combination 48. In this case also the left side limit does not have any meaning,
because w3 cannot take any negative values. If w3 increases to more than 27733.67 then the
new optimal solution will be combination 89. This type of post optimality analysis provides
the decision maker with information relating to the validity of the conversion factors utilized
The suppliers, designers, manufacturers, and distributors that have been filtered out in
the phase 1 o f the evaluation process need to improve their operating practices in order to
become competitive and participate in future value chains. To accomplish this goal they need
has certain limitations that have been addressed in the methodology section, cluster analysis is
utilized to identify appropriate targets for improving poorly performing candidates in each
The corresponding CEM in each of the four business process types is used as the data
matrix for clustering. Each DMU is considered as a variable, and the corresponding cross
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
64
efficiency scores in the column of the CEM are used as cases. The distance metric utilized is
Pearsons r, where Pearson r is the correlation coefficient. The correlation coefficient between
pairs o f columns in the C EM is evaluated, and a clustering method based on unweighted pair-
Figure 10 depicts the results of cluster analysis for supplier processes. A total of five
clusters are identified. These five clusters are formed by specifying a threshold correlation
coefficient o f 0.70, Le., the dendogram is truncated at Pearsons r o f 0.70. The numbers at
each branch represent the correlation coefficient at which two clusters can be grouped
Suppliers 4, 1, 11, 16, 10, and 7 are grouped into cluster A. In this cluster, supplier 4
has the highest cross-efficiency mean score of 0.791. It can be used as the primary benchmark
for improvement by other six suppliers in cluster A. These six suppliers must identify the
operating practices o f supplier 4, which make it superior to them. Supplier 13 in cluster B can
dendogram that there is a weak correlation with a correlation coefficient of 0.32 between
clusters A and B.
Supplier 15 has the highest cross-efficiency mean score o f 0.894 in cluster C. It can be
used as the primary target for improving the performance o f suppliers 14, 17, 2, 9, and 8.
in cluster C. However, it should be noted that the correlation between clusters C and D is
relatively weak with a correlation coefficient of 0.58. Similarly suppliers 6, 12, and 5 in
cluster D can utilize supplier 18, with a cross-efficiency mean score of 0.925, as a benchmark
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
for improvement Supplier 3 in cluster E does not have a good target for improvement. It
can use supplier 18 in cluster C as a target, but one has to be cautious about the low
0.70
DMU 4 - 0.791
DMU 1 - 0.654
DMU 11 - 0.594 0.32
DMU 16-0.529
DMU 10-0.511
0.28
DMU 7-0.430
D M U 13 - 0.849
DMU 15 - 0.894
DMU 14-0.833
DMU 17-0.817 0.58
DMU 2-0.793
DMU 9-0.672
DMU 8 - 0.615 0.45
DMU 18-0.925
DMU 6 - 0.914
DMU 12-0.748
DMU 5 - 0.681
D M U 3 - 0.762
As one goes from left to right in the dendrogram, it is evident that the similarity index
decreases and the dissimilarity increases. At a correlation coefficient o f 0.28 all the suppliers
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
can be grouped into a single cluster. Managers have to be astute in identifying the threshold
0.56
DMU 11 - 0.944 0.08
DMU 2-0.369
DMU 1-0.357
0.13
DMU7-0.577
DMU 5-0.559 0.35
DMU8-0.500
DMU 3 - 0.488
DMU 4-0.503
DMU 13-0.450
DMU 15 -0.450
DMU 14 -0.417
DMU 6-0.402
0.44
DMU9-0.950
0.05
DMU 12-0.484
DMU 10-0.565
Figure 11 illustrates the results of cluster analysis for the design process. After
specifying a threshold level of 0.56, a total of six clusters have been identified. Designer 11 in
cluster F has the highest cross-efficiency mean score o f 0.944. It can be used as a benchmark
designers 5, 8, and 3. Similarly, designers 13, 15, 14, and 6 in cluster H can use designer 4 as
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the primary rary r for Hnpro\aaK- The designers * dusters G and H caa uuhre designer 11
to try and improxx; in their oxxn duster first. The analogy for this is that one has to team how
to walk before runnaig. Designer 12 m duster J can use designer 0 in chtster I as the primary
benchmark, because this is the only deserter that comes dose to being similar to designer 12
in the entire data set. Designer 10 in duster K does not haxe a good target to benchmark
again <a Designer 9 is the closest that it can benchmark with at a correlation coefficient ot'
0.05.
The forty seven manufacturers are segregated into four dusters at a threshold
correlation coefficient of 0.70. The results o f cluster analysis are shown in figure 12. Cluster
A contains eighteen manufacturers which are all relatively poor performers. This is evident
from the low cross-efficiency mean scores shown in cluster A. Manufacturer 10 has the
highest cross-efficiency mean score of 0.251 in this cluster. The other manufacturers in
cluster A can use manufacturer 10 as the primary benchmark and manufacturer 21. with a
mean score o f 0.667 in cluster M, as a secondary benchmark for improvement. Basically, the
manufacturers in cluster L do not have any good performers to benchmark against, because
there are all extremely 'false positive' and require improvement across many dimensions. Hie
mean FPI value for the manufacturers in this cluster is 342.78%. Managcrially, it is difficult to
improve these processes because they are performing very well with respect to few inputs and
outputs, and are extremely poor with respect to many inputs and outputs.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
68
0.70
DMU 13-0.398
DMU 20 - 0.495
DMU 45 - 0.364
DMU 4 4-0307
D M U 2-0.249
Similarly, in cluster N, manufacturer 13 can be used by manufacturers 20, 45, 44, and 2 as a
Cluster O is the most efficient o f all the clusters in the manufacturing processes
The analysis of the thirteen distributors resulted in four clusters at a threshold level of
0.84. The results are depicted in figure 13. Cluster P has distributors 1 and 5. Distributor 5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
can utilize 1 as the primary benchmark, and distributor 9 in cluster Q as the secondary
benchmark. In cluster Q, distributor 9 is the most efficient one with a cross-efficiency mean
score o f 0.962. Distributors 12, 4, 11, 7, 13, and 8 can use distributor 9 as the primary
benchmark for improvement. In cluster R, distributor 6 can use 2 as a primary source for
0.84
DMU 1-0.944
DMU 5-0.933
-0.48
DMU 9-0.962
DMU 12-0.892
DMU 4-0.882
DMU 11- 0.881
DMU 7-0.862
DMU 13-0.853
DMU 8-0.798
0.48
DMU 10 -0.984
DMU 3-0.915
Cluster S has distributors 10 and 3. Distributor 10 is the most overall efficient one in
the entire data set, with a mean score of 0.984. Distributor 3 can use it for improving farther
benchmarks for poorly performing business processes. However, it does not prescribe the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
70
actual changes these processes need to perform in order to become more efficient. It is for
the management to identify these changes and implement them in a cost effective manner.
process over tim e and aiding in continuous process improvement. We assume that the process
is deterministic in nature in that the majority of variation in the outputs occurs due to changes
in the input variables. The random variation of the process is assumed to be minimal and is to
an extent controlled by setting large time periods, and obtaining the mean values of the
Initially various configurations within the scope of the management are considered.
For illustration purposes a maximum of twenty six machines and seventeen workers are
considered available. A total of ten configurations with changing worker and machine levels
are investigated in the analysis. Each of these ten configurations are implemented in a time
period, and are collectively considered as window 1. It is assumed that apart from the input
levels all the other factors that affect the process productivity such as, scheduling rules,
sequencing rules, lot size levels, etc., are held constant. The hypothetical data and the simple
efficiency scores o f the DMU in each of the ten periods are shown in table 20.
It is evident from the DEA analysis that periods 1,2, 3, 6, 7, 8, and 9 are efficient with
relative efficiency score of 1. The periods 4, 5, and 10 are inefficient with relative efficiency
scores o f 0.847, 0.800, and 0.801 respectively. The pairwise efficiency game formulation is
utilized to obtain the optimal weights that are used in evaluating the CEM shown in table 21.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
71
It is evident from the CEM that period 2 is the best performer with a mean cross-efficiency
score o f 0.906, and period 7 is the worst performer with a mean score of only 0.571.
Therefore, period 7 is dropped from window 2 analysis, and the new period 11 is added.
Period 2 is used as a benchmark for this new period by implementing the same
worker/machine levels in period 11. However, other factors can now be changed in period
11. For example, the scheduling rule or the sequencing rule can be altered to check its effect
The data utilized in window 2 analysis is depicted in table 22. It is evident that periods
8 and 11 are efficient and all the other eight periods are inefficient. The CEM for window 2 is
shown in table 23. From the CEM, it can be concluded that period 11 is the best overall
performer with a mean cross-efficiency score of 0.965. Thus, the changes made at the
beginning o f this period had a positive impact on the performance. If this does not occur then
the changes can be reversed by implementing the old scheduling/sequencing rule, etc., or make
other modifications and evaluate the performance. This process can be continued for the entire
planning horizon.
There are several techniques that can be used for process monitoring and
improvement. Control charts in statistical process control (SPC) have primarily been used for
process monitoring and control. However, a limitation of these techniques is that they
consider a single attribute of the process each time. The advantage of MWA is that it
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CHAPTER 5
CONCLUSIONS
The NVCFs are envisioned as a solution to meet the constantly changing needs of the
customer at low cost, high quality, small lead times, and high variety. In this dissertation, a
two-phase mathematical programming model is proposed for effective value chain design.
DEA with proposed new extensions is used in the phase 1 of the decision making process to
select supplier, design, manufacturing, and distribution processes with good operating
hi the phase 1 of the decision making process, cross-efficiency analysis in DEA is used
for effectively differentiating between good overall performers and poor performers.
However, the decision maker can also specify input and output weight restrictions in the DEA
model that allow for this discrimination. These weight constraints must be in alignment with
the objectives, policies, and procedures of the NVCF. The inputs and outputs used in phase 1
are not intended to be exhaustive. The decision maker applying this approach may have to
consider other critical I/O measures that are representative of the process being evaluated.
In general, if the initial data is filtered in an optimization procedure, then the solution
to the problem may be sub-optimal This is also true for the proposed analysis, since phase 1
has higher priority over phase 2. In the initial screening process using DEA, some candidates
that are filtered out of the analysis due to relatively poor performance may in fact
72
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
73
have better compatibility characteristics when analyzed in phase 2. To alleviate the severity of
this problem, an astute decision must be made by the decision maker in setting a cutoff point
While the decision making process in this dissertation is performed from the standpoint
o f brokers, it can also be performed from the standpoint of a lead business process. In such a
situation, before becoming involved in the formation of a NVCF, the lead business process
must evaluate its efficiency relative to other similar business processes and implement process
The MWA proposed for continuous process improvement assumes that the change in
outputs is solely dependent on the changes in the input levels. This may not always be true,
because o f the existence of random variation within the system. The MWA changes the
factors). Thus, it may not prove to be very useful for improving a process or system with
large stochastic variation. To make this model useful management must control the
extraneous variation of the process or system under consideration. A method for controlling
such variation is by increasing the time span of each period, and considering the average
chart based on DEA. Such a chart will provide management with a multi-attribute
performance index for a process or system under consideration. It can be used to monitor the
process and identify the effects on process performance by making changes at the functional
levels.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
74
In the formation of the NVCFs, there are other issues to be considered such as
NVCFs workers dividing their loyalty and responsibility between this new venture and their
parent corporation, which may not be feasible in some types of value chains. Also, the
possibility o f losing confidential information through the value chain channels is a key
deterrent to the effectiveness of NVCFs. In spite of all these and other problems, many
companies around the world are becoming involved in such ventures to capture a specific
market opportunity.
Finally, this research illustrates the application of DEA in the area of production and
operations management (POM). Most of the applications o f DEA to date were involved with
organizational units such as bank branches, hospitals, airlines, shops, etc. Researchers and
practitioners in the area of POM are currently realizing the usefulness o f DEA as a tactical and
operational decision tool This application reinforces the effectiveness o f DEA in operations
management.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
APPENDIX A
TABLES
75
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 1. Data for TWA and MWA Comparison
DMU 1 2 3 4 5
1 1.000 0.444 0.708 0.667 0.333
2 1.000 0.750 0.944 0.727 0.500
3 0.794 0.471 1.000 0.578 0.471
4 1.000 0.485 0.773 0.727 0.364
5 0.417 0.278 0.525 0.303 1.000
Mean 0.842 0.486 0.790 0.600 0.534
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
77
Table 4. Supplier Data and Efficiency Scores
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
78
o 00 P CO ix CO p 00 P C M co P C O P P p p
oo COC OP COp rx 00 p p P rx ao P C M h* C M r*. P C Mp
o> 05 P p 05 05 05 05 p P p p P p P P p P P p
o o o o o o P p p O p p o o P P p T P p
oo C Op C Op C M M p co p p C OP p P p
00 C O IX C M rx P C M P p E M -P p C OC O P p
r^ 05 p |x rx P 05 p p oo r- p ao 1^ P p 1^ P
o o o o p o O O o o p o o p p P p P P p
CDo p C D TC O M -CM rx p co p P r- v P fx.
(0 CMC O ID r- P p M- o p oo C O p p h c- CM JX.
COC DC O M - co C O P p P p fx C O M M " M-P M" co P
o o o O o o o o O o p p o o P P O p P p
h- C M h- p p p rx P P p V
U) C D 05 p C Mo C O rx 05 P P p M- P P CO P C OP P
1 00 00 IX ix p p p 05 P P p P fx P P P 00 p P P
o o o o p p o P P P P P P P r- P p p P P
CO CM p ID co p o o o p OO O o l- CO r- p C OC M
M- CMC O Tp P o p o o o r- 00 o co ID p p C O T
r- N* 00 |X p P p r- p o o p fx fx o 00 P p p p
O 9 P p P p p TTp p p Tp p p p p P
CM -M- p p C OC M CO CO n o C M p CO p p p h-
co 05 05 p C O p 05 rx M - c- CO U9 o p CO p p M*P
T h- 00 P p ix. rx |x P P p p p o P p p 00 N- p T
O p P o p o P p O p p p T P p p a P p P
ID C M M - M- p 00 r p p 00 P C O h- P p C O
CM rx 05 p C O T o C M CMp 00 CM
O 00 P ID rx p 0 0 05 P p p P p fx p p r- CD
Table 5. Supplier Cross Efficiency Matrix
T C
o o O o o o p o o o o o o O p p p P P P
CDC MC O 05 >* p C O rx fx C M CO P p f" m- P
- C D M- p p id p C Mp p p P i t P p p p C MP C O
CO P -M - ID M- fx p p p P p P p p M -P
o O P p P p P p o p P o p P p p p P P p
o C M co CM p v p co p P p o P p V P C M
o C DC O p O 05 C Oo p p C M ix P C Mp CD C M
CDC DC O M- M-C Op p p p P M- p p CO P
o O p O O o p o o o O p o O o p p P P P
oo C Mp C DP p p o C M p P T M p P C M ID
h- O C M 05 |x 05 C MC Op p ID p P CMC O N. M-
p C D 1^ C O P P P 05 p p i t P M1 P P P P P p
o o o P o o P o o p p o O P P P P P p P
o 05 p p T p C Op p CO |x P M"P p
o a P p C O 00 C Op C Mp p C OP O P P
oo C DC D M -M p p C D 05 oo p P p M- p ID P P p
o o P O p o o p o o o o O p P P P P p P
00 fx. P fx p C Mp o p C M P C OP P P P p P
C O P 05 C MC M p 05 p p P V P P P P C O |x-
s M - CM CO CO CO 05 co co p V ) co co COC M ID CM T
o O P P P P P o p p p o p P P P P P P
COC D P 05 P o M" ID p co p T C Mr T M- V
00 05 P C M P o T p p P P |x ix "M - h- p P
C D 00 05 P P 05 o 05 a> p P P P p P P P p P 05 P
o o P O o T o o o o O O o P P P p P P P
oo C M N. o C M co T " CO p M - C OP p C OC O M -
T 05 p p M p ID fx IX TCO T p M- p CD P CO
ID C O C OC D 05 P rx p P rx p rx M" p ID P p CD P T
o o P P p P p p P P p o P p P P p P P p
P O o P p C MP p co P fx co M- r- T p
o o ix o C M 00 CM O p CM CM p T p C MP P C M
Tf O) 05 p o |x p 05 00 00 p P p P ix P 00 00 P
o o p T O o o o o o O o O p o p P P p P
r-- fx p C D 05 COp o p CM 00 p p T C MC M co
co C D p 05 p p 00 p 00 00 P 00 fx ID p P P C M
COC O p P 05 fx p p p p p P ID fx P p 1^- P r - T
o O o o o o o p o o O o p P p P P p P
COo p C DC D 00 C
Op V T CM M ID ID P P C O co
CMo C MO o o p p o T TP 00 CD '(P p C M
CM oo o P rx rx ix p p p p p rx P Ix P r*. P P T
o T P o o o o o o o o o o P P p P P p P
r>- CD r o 05 p fx ID ID ID p P r-~ p P CM
05 05 P a COCMCM 00 T P p M - CO 00 ID P
0) P M - p C D p 05 p P 00 P ID M -p p CD T
o o P o O o o o P o O o P P p p o P O
|Mean I
| Std. |
DMU
CDp 00 P o CMCO
M - P p r- CO
Mco
C rx
T" T T T
-
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 6. Supplier False-Positive Indices
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
80
DMU #E OC #P #0 LT EFF
1 14 37.31 11.81 12.18 39.93 0.499
2 13 38.93 9.89 12.68 37.71 0.557
3 16 34.52 11.86 7.25 21.79 0.744
4 14 35.13 9.67 11.59 20.85 0.842
5 15 31.55 10.73 8.22 22.63 0.728
6 19 33.70 8.93 13.59 26.46 0.555
7 17 22.60 10.92 6.90 21.88 0.994
8 18 31.32 10.50 8.51 25.15 0.625
9 12 20.78 6.02 7.07 23.65 1.000
10 14 26.97 6.80 9.70 33.24 0.755
11 10 21.16 7.16 8.73 24.59 1.000
12 20 25.89 9.06 9.16 27.57 0.688
13 19 39.47 9.21 10.29 23.22 0.565
14 18 34.42 11.10 13.28 20.53 0.716
15 18 27.64 8.25 14.06 20.79 0.855
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CO CO o 05 oo in m in o CO CO m o o
m CO CO MT TT in CO 05 r- CO o V CM CO r- m in in
CO CO CO m CO U5 CO CO CO V CO V M- co CO M- t
o o O o o o o o o O o O o o o o o
CO 05 in 00 05 M- CM o 05 r- CM CO o r- 00
in in CD r- CO CO CM CD in m CD * r- o
T- eo CO CO in co MT co 8 eo CO CO co MT 1^- in M-
o o o o o o O o o o o o O o o O o
CO 05 CM l>- 05 o CM o Mt 05 CM in CM CO O T
eo eo o V in m CO 00 M- 00 o O CO m eo m r-.
T M" in 'T M- CO CO CO yf M- in m in o
o o O o o O o o o o o O o o o o o
CO in O CO CO CO r- CO 05 r- in CO CM o o M- 05
C in 05 r>- V o co 05 05 05 CO in CO 00 00 00
TM
CO V m CO in eo CO co CO M" M- to M- o
o o o o o o o o o o o O o o O o
m o 05 o CO o CM CO Mt o in CO MT M1
^ o 00 o 05 in - o 05 T o CO 05
M r - MT
o> o 05 o 05 05 CO 05 N- 05 o CO 05 05 05 05
o * o >- o o o o o O To o O O O o
o r- CO TCM 00 05 oo in CO 00 CD in CO in CO
o o 00 m CM CO m M" M-m CM mt CO CD mt CO in
CO in m in m m m in
Table 8. Designer Cross Efficiency Matrix
in r-- m in in m in m o
o o o o o o o o o o o o o o o o o
co CM CD o CO o r- o CO CO o CO CM o o CO
oo o 05 05 CO o h- CO o m m
05 00 CO
05 05 05
CO f>-
00 co 05 o 05
o
o 05 oo o CO 05 o 05 o
o o O o o o *- o o o o To o
r- 05 CO CO CO CM in o T 05 eo m eo o 00
00 CO lo CO M" 05 oo CM oo OO r- 05 CM CO 05 o co
co in in in MT m CO CO co co in m to m o
o o o o o O o o o o o o o o o o o
in CO in 00 CO 05 00 CO in CO TMT r-> 05
mT oo 05 CO o 05 CO CO 00 oo in co CO CO
-M- CO CO 5 CO m 05 CO CO eo eo CO in r- CO m T
O o o o o o O o o o o o o o o o o
O in mt 05 r- m 00 r-- CO CO h- CO CO
CO CO in CO 00 MT in eo CO CM 05 CO in CM CM in 00
CO CO CO yf eo m CO CO co CO CO CO MT m IO M- o
o o o o o o o o o o o o O o o O o
CO in o 05 co 00 r- o o in N- h- 05 05 05
N. o O T-
m in mo- o r-
O CM s
r*.
05
m
o r-
CO eo M t CO r -
m
m m T-
o o o O o o o o o o o O o o o o O
in 05 CM CO 2 05 oo CO T- CO CO 00
CM CM in "M- CO 00 o CO o O CO 05 i t o CM
M" in in m t M
00 MT eor CO m CO m in r -
o o o o o O o o o O o o o o o o O
in 1 ^ M" TT 05 m 00 o CO 00 O
- o o 05 s o CO 05 o CO E CO eo
co CO tco
-.
M
CO CO MT m m co CO CO co in CD MT
o o o o o O o o o o o o o o O o o
o o mT N- M- m CO o M" eo i n m 05 CO
CM CO in N- CM in co T CO 05 CO in m M ' CO CO C O
M-m CO CO CO CO CO CM CO CO CO eo co CO CO o
o o o O o o o o o o o o o o o o o
05 CO CO 'M- 05 CM o o O M- o CO o o M-
05 eo T Mt C OC OC O oo CO Mf COm m
T*~ CO M- COC OC O eo CMCOC O co C
O CO co CO o
O o o O o o o o O o o o o o o o o
I DMU I
IMean |
1
CO h- 00 05 o CM C in
pis
T O
- MC
C O
M- in T
T T-
1
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
82
Table 9. Designer False-Positive Indices
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
83
Table 10. Manufacturer Data and Efficiency Scores
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
84
Table 11. Manufacturer Cross Efficiency Matrix
DMIJ 1 2 3 4 5 6 7 8 9 10 11 12 13
I
1 0 939 0 194 0 754 0.012 0 805 0 853 0.697 0.003 0.006 0.006 0013 0 413
2 0 877 0 744 0 680 0610 0 046 0 730 0 683 0 502 0 012 0 022 0.022 n n si 0 944
3 0 827 0 213 0.831 0.745 0.013 0.844 0 790 0613 0 004 0006 0.006 0.015 0 455
4 0 874 0 183 0 704 0943 0.011 0 755 0 934 0 777 0003 0 005 0.006 0.012 0 3 9 0
5 0 488 0 537 0 491 0440 0.711 0 613 0 542 0 362 0 190 0 348 0.350 0.659 0.899
fi 0 887 0 178 0 654 0619 0.013 0,862 0 762 0,510 0 003 0 006 nnnfi 0.014 0 442
7 0 718 0 167 0607 0.762 0.011 0,756 0,938 0 627 0.003 0.006 nnnfi 0,013 0.391
8 0 887 0 180 0 698 0930 0 011 0 749 0 920 0 850 0 003 0 005 0.005 0,012 0 384
9 0 188 0 205 0 187 0.168 0.271 0 234 0 207 0138 1.000 0.695 0.467 0 501 O468
m 0 268 0 295 0 269 0.241 0.343 0.337 0 297 0199 0.253 1.000 0.673 0,473 0.673
11 0 299 0 329 0 301 0.269 0 274 0376 0 332 0 222 0.182 0 796 0 750 0 378 0 751
12 0 328 0 361 0 330 0.296 0.478 0.412 0 364 0 244 0.213 0.389 0.392 0.883 0.824
13 0 389 0 249 0391 0.350 0.021 0.488 0432 0 289 0 008 0 014 0.014 0,031 0 976
14 0 469 0 129 0 468 0.422 0.010 0 588 0 520 0.348 0,004 0 006 0.006 0.015 0 459
18 0 522 0 133 0491 0.471 0.011 0 644 0 580 0 388 0 003 0 006 0.006 0.014 0.438
18 0 840 0 174 0 677 0.899 0.011 0,724 0,890 0 822 0 003 0 005 0.005 0 012 Q.371
17 0 208 0 227 0 208 0 186 0.301 0 259 0 229 0153 0 795 0 770 0.518 0 556 0.518
18 0 287 0 316 0 289 0.259 0.259 0.361 0319 0.213 0,179 0 708 0.709 0 357 0 721
18 0 315 0 347 0 317 0.284 0.459 0.396 0350 0.234 0 197 0 361 0.363 0 786 0 791
20 0 382 0 368 0364 0.326 0.030 0455 0 402 0 269 0011 0 021 0.021 0 047 0 910
21 0 429 0 160 0432 0 387 0.013 0.539 0,477 0 319 0 005 0 008 0.008 0 019 0.591
22 0 495 0 115 0420 0 494 0 009 0 553 0.609 0.407 0 003 0 006 0.006 0 0 1 3 0 394
23 0 529 0 123 0 449 0.536 0.010 0.591 0 660 0,441 0 003 0 006 0,006 0 013 0 394
24 0 860 0 178 0693 0921 0.011 0,742 0 912 0 843 0 003 0 005 0.005 0 012 0 381
25 0225 0 247 0.226 0.203 0 327 0 282 0 250 0 167 0 652 0 839 0.564 0 605 O564
28 0 304 0 334 0 306 0.274 0 252 0382 0 338 0 226 0 165 0 651 0,679 0 347 0 764
27 0 331 0 364 0 332 0 298 0438 0.415 0,367 0.245 0,173 0 317 0.318 0 719 0 830
28 0 408 0 167 0410 0.368 0.014 0 512 0 453 0.303 0,005 0 009 0.009 n o 2 n 0,641
29 0 463 0 132 0466 0.418 0.011 0.582 0,514 0344 0 004 0 007 0.007 0 015 0 469
30 0 538 0 123 0455 0 490 0.010 0 598 0,604 0 404 0,003 0 006 0.006 0 013 0411
31 0547 0 126 0464 0.557 0.010 0.609 0 686 0 459 0,003 0 006 0,006 n n i3 0 392
32 0 837 0 174 0674 0.899 0.011 0.723 0 890 0 821 0 003 0 005 nnns 0 0 1 2 0371
33 0241 0 266 0.243 0.218 0.352 0,303 0 268 0179 0 618 0 901 0.606 nfisn OfiOfi
34 0 319 ft 051 0.321 0.287 0 300 0.400 0.354 0,237 0,187 0 661 0 885 n 4 i4 0,801
35 0 345 0 379 0 347 0.311 0.362 0.433 0 383 0,256 0 208 0 731 0.735 0 475 0 866
38 0372 0.409 0.374 0.335 0.332 0.468 0 413 0,276 0 177 0 700 0,730 n4n4 0,901
37 0 503 0 125 0 459 0.453 0.010 0.604 0 559 0 374 0003 0 006 nnnfi n n i4 0 431
38 0 528 0 121 0446 0 546 0.010 0 587 0,672 0,449 0 003 0 006 nnnfi 0 013 0 392
39 0 586 0.135 0 497 0.535 0 011 0.654 0 658 0 440 0 003 0 006 0.006 n m 3 0411
40 0 833 0 174 0671 0.898 0.011 Q.719 0 892 0 817 0,003 0 005 nnns 0 0 1 2 0 372
41 0 249 0 274 0250 0.224 0.352 0.313 0 277 0 185 0 283 0 834 0.625 n486 0 626
42 0 322 0 354 0 324 0.290 0.287 0.404 0 357 0 239 0 177 0 630 0.633 0 396 0 80R
43 0 346 0 380 0 348 0 312 0.414 0.435 0 384 0 257 0 157 0 286 0.288 0 650 0R69
44 0 387 0426 0389 0.349 0 038 0.486 0 429 0 287 0013 0 025 0.025 nnfifi n 939
45 0432 0 260 0.434 0.389 0 021 0.543 0 480 0 321 0 007 0 013 0.013 0 029 0 906
48 0 516 0 120 0 438 0 528 0.010 0.577 0 650 0 435 0,003 0 006 nnnfi 0 013 n 393
47 0811 0 169 0653 0.875 0 010 0.700 0 866 0 795 0 003 0 005 0.005 0 0 1 2 0,361
Maan 0 496 0.249 0 452 0 478 0 147 0.544 0 548 0 404 0 126 0 251 0,219 0 219 0 59R
Std. 0,215 0.128 0 163 0.236 0 184 0.166 0.224 0 215 0.221 0.337 0.287 n274 0.212
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
85
Table 11-Continued
DMU 14 18 16 17 18 19 20 21 22 23 24 25 26
1 0 900 0 908 0 584 0 003 0 005 0.014 0 255 0 6 4 6 0.959 0 906 0.546 0.003 0.006
7 0 8fifi o a o fi 0 421 0.0 1 2 0.021 0 053 0 905 0 642 0.768 0.725 0.393 0.011 0.022
3 0 992 0 933 0 514 0.003 0.006 0015 0 251 0 711 0.889 0 839 0.480 0.003 0 0 0 6
4 0 849 0 843 0 651 0003 0.0 0 5 0 013 0 236 0 609 0 988 0.944 0.608 0003 0.005
5 0 878 0 7fi8 0 304 0 183 0 .3 2 7 0 579 0R62 0 602 0 7 3 2 0.691 0.284 0.178 0 339
6 0 9fi4 0 982 0 427 0.003 0 006 0 015 0 272 0 691 0.908 0.857 0.399 0.003 0 006
7 0 883 0 844 0 576 0003 0.0 0 5 0 013 0 236 0611 0.991 0.948 0.491 0.003 0.005
8 0 837 0 83fi 0 712 0 003 0 005 0.013 0-236 0 600 0.974 0,931 0.666 0.003 0.005
fi 0 397 0 342 0 116 0 .7 3 6 0 .5 1 4 0.504 0,467 0 401 0.340 0.299 0.108 0.505 0 477
m 0 577 0 493 0 167 0260 0.7 4 0 0 432 0 672 0 5 7 7 0,489 0,431 0.156 0.237 0.614
11 0 fi3 8 0 880 0 186 0 .1 8 7 0.7 0 0 0 346 0.750 0 644 0.545 0.481 0.174 0.171 0.685
12 0 7 00 0 fi03 0 904 0 204 0 365 0,775 0 623 0 7 0 7 0.599 0527 0.191 0,199 0.379
13 0 830 0 718 0 949 0 007 0013 0 033 0 576 0 637 0.709 0.625 0 226 0 007 0 013
14 1 000 0 8fi1 0 791 0003 0 006 0 015 0 277 0 710 0.855 0,753 0.272 n n n 3 0 006
15 0 984 0 960 0 395 0 003 0 .0 0 6 0.015 0 270 0 685 0.915 0 840 0.304 0,003 0.006
1fi 0 810 0 810 0 948 0 003 0 .0 0 5 0 012 0.229 0 561 0.942 0 900 0,689 0 003 0.005
17 0 441 0 379 0 128 0 .8 1 6 0.570 0.559 0.516 0 4 4 5 0.377 0.332 0120 0 560 0.473
18 0 fi13 0 828 0 179 0 184 0.793 0,327 0.720 0 618 0.524 0 462 0,167 0168 0.658
1fi 0 fi77 0 879 0 196 0 189 0.338 0 853 0 7 9 0 0 679 0.575 0.506 0.183 0.184 0 351
20 0 773 0 666 0 225 0.011 0.019 0 049 0,909 0 780 0.661 0 582 0.211 0011 0.020
21 0 91fi 0 789 0 267 0 .0 0 4 0.008 0 020 0.361 0 925 0.783 0.690 0.250 0 004 0.008
22 0 889 0 818 0341 0 003 0.005 0 013 0,241 0 616 1.000 0.881 0,319 0 003 0.005
23 0 889 0 880 0 370 0003 0 005 0 013 0.240 0 615 0,998 0.955 0.346 0 003 0 0 0 5
24 0 830 0 830 0756 0.003 0.005 0 013 0.234 0 595 0.966 0.923 0,707 0 003 0 0 0 5
9*5 0 48 0 0 413 0 140 0.628 0.621 0 609 0.564 0 4 8 4 0.410 0 361 0 131 0 610 0 515
2fi 0 fi49 0 889 0 189 0 169 0.672 0.318 0.763 0 6 5 5 0.555 0 489 0.177 0 154 0 697
97 0 70fi 0 608 0206 0.1 6 6 0.297 0 749 0,629 0 712 0.603 0,531 0 192 O 161 0.308
28 0 871 0 780 0 254 0 005 0.008 0 021 0.394 0 878 0744 0.656 0 237 0 005 0 009
9fi 0 989 0 882 0288 0 003 0 006 0 016 0,267 0 734 0.845 0.745 0 269 0 003 0 006
30 0 89fi 0 886 0338 0 003 0.005 0 014 0,253 0 643 0.963 0,874 0 316 0 003 0 0 0 6
31 0 888 0 857 0.384 0 003 0.005 0 013 0,242 0 613 0.995 0.951 0.359 0 003 0.005
32 0 810 0 807 0 944 0 003 0.005 0 012 0.228 0 581 0,942 0 900 0.687 0 003 0 005
33 0 818 0 444 0 150 0.631 0 .6 6 7 0 654 0.606 0 520 0.440 0.388 0140 0 578 0.553
34 0 fi81 0 886 0.198 0 .1 9 2 0.620 0 374 0,600 0 687 0.582 0.512 0.185 0175 0 643
38 0 73fi 0634 0215 0 .2 1 4 0.686 0,417 0.863 0 743 0.629 0 554 0 200 O 196 0 .6 9 4
3fi 0 798 0 684 0.232 0 .1 8 2 0.707 0 355 0.865 0 802 0.679 0,598 0216 0166 Q.696
37 0 940 0 894 0.313 0 003 0.006 0 014 0-265 0 674 0.918 0,809 0,293 0 003 0 006
38 0 888 0 856 0377 0 003 0.005 0 013 0.242 0 614 0 995 0,940 0,352 0 003 0 0 0 5
3fi 0 897 0 898 0 369 0 003 0.005 0 014 0.254 0 643 0,962 0 909 0.345 0 003 0 .0 0 6
40 0 811 0 803 0940 0 003 0.005 0 012 0.227 0 581 0 942 0 902 0,684 0 003 0,005
41 0 837 0 458 0 155 0 791 0.688 0 444 0.625 0 5 3 7 0.454 0 400 0.145 0 265 0 571
42 0 R87 0 591 0.200 0 182 0.591 0 355 0.607 0 693 0,587 0.517 0.187 0 166 0-613
43 0 738 0 636 0215 0 150 0.269 0 678 0.663 0 745 0.631 0 556 0 201 0146 0 279
44 0 89fi 0 711 0 241 0 013 0 023 0 058 0.901 0 833 0,706 0 622 0 225 0 013 0 .0 2 4
45 0 908 0 794 0269 0 007 0.012 0 030 0.550 0 877 0,788 0 694 0.251 0 007 0.012
4fi 0 887 0 849 0364 0 003 0.005 0 013 0 740 0 614 0,996 0 931 0,340 0 003 0 005
47 0 788 0 782 0915 0 003 0.221 0 565
0.005 0 012 0.917 0 876 0.666 0 003 0 005
Mean 0 783 0 724 0.361 0 210 0.495 0 667
0 191 0.221 Q.761 0 697 0.321 0 106 0 707
SM. 0.148 0.164 0.231 0.202 0.293 0 266 0.263 0,113 0.203 0.204 0.178 0 164 0.271
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
86
Table 11-Continued
DMU 77 78 29 30 31 3? 33 34 35 36 37 38 39
1 0 014 0 557 0 757 0871 0 814 0 500 0,003 0 005 0 005 0.006 n 7 7 i 0.785 0,742
2 0 053 0 797 0 751 0 697 0 6 5 2 0.360 0011 0.021 0 021 0.022 nfifift 0.628 0.594
3 0 015 0 614 0 834 0.807 0.754 0 440 0 003 0 006 0.006 0.006 8 773 0.727 0.687
4 0 013 0 577 0 713 0823 0 856 0,557 0 003 0 005 0 005 n n n s 8 723 n 841 0.699
5 0 588 0 759 0 716 0 664 0.621 0.260 0 169 0.3 2 3 0 3?? 0,335 0.636 0.598 n sfifi
6 0 015 0 598 0 810 0824 0 770 0 366 0 003 0 006 0 006 0.006 n 7 9 n Q.742 0.702
7 0 013 0 523 0 714 0824 0 857 0 450 0 003 0 005 0 005 0.005 0,724 0.842 n 7 n n
8 0 013 0 518 0 704 0816 0.849 0 610 0,003 0 005 0 005 0.005 8 717 0.834 n fi9 3
9 0 4 74 0 395 0 346 0.299 0 ?6? 0,099 0 450 0 3 5 9 0 365 0.363 0 2 9 8 0.263 0.234
m 0 403 0 568 0 497 0431 0 877 0143 0 ?39 0 516 0,525 0.522 0.429 n 378 0.336
11 0 377 0 634 0 555 0481 0.421 0,159 0 17? 0 56? 0 512 0.577 8 478 0.422 0,375
17 0 747 0 695 0 609 0.528 0 46? 0175 0 189 0 362 0.343 0.331 0.525 0.463 0.412
13 0 033 0 874 0722 0625 0,548 0 ?07 0 007 0.013 0,013 0,013 0,622 0.548 0,488
1A 0 015 0 608 0831 0 753 0 660 0 749 0 003 0 0 0 6 0 006 0.006 0.749 0.661 0.588
15 0 015 0 591 0802 0 831 0 736 0,278 0 003 0 006 0.006 0.006 0.796 0.736 0.655
1fi 0 017 0 501 0680 0790 0 8?? 0811 0 003 0 005 0,005 0.005 0 693 0.808 0,671
17 0 470 0 437 0383 0.332 0 ?91 0 110 0 499 0 398 0.404 0.402 n 33n 0.291 0.259
18 0 304 0 608 0533 0.462 0.404 0153 0 170 0 530 0 483 0,559 0 459 0.405 0,360
19 0 7 1 7 0668 0.585 0,507 0,444 0168 0175 0 332 0 305 0.294 0.504 0.444 0.395
70 0 049 0 768 0.673 0 583 0.510 0193 0 010 0019 0019 0.018 0 580 0,511 0 455
71 0 020 0 792 0797 0 690 0 605 0 ??9 0 004 0 0 0 8 0 008 0.008 0 687 0 605 0 539
77 0 0 1 3 0 528 0.721 0 606 0 77? 0,292 0,003 0 005 0,005 0,005 0 731 0,773 n fi? 8
73 0 013 0 527 0 720 0.830 0,837 0 316 0003 0.0 0 5 nnns 0.005 0 729 0 828 0 672
24 0 013 0 513 0.697 0.810 0.843 0,647 0 003 0.0 0 5 nnns 0.005 0,710 0 628 0 688
75 0 517 0 476 0.417 0.362 0,317 0120 0 537 0.4 3 3 n 44n 0,438 n 3fin 0 317 0 ?82
7fi 0 298 0644 0 564 0489 0 428 0,162 0.156 0.5 1 6 0,470 0,550 0 486 0 429 0 382
77 0 753 0.701 0 614 0.532 0 466 0,176 0.154 0.2 9 4 0 ?81 0,271 0 529 0 466 0 415
0.008 n fi5 3 0 575 0 512
00
(0
0.757 0 656
31
79 0 018 0 630 0.860 0.653 0 ?47 0 003 0.0 0 6 0.006 0,006 0 741 0 653 0 581
30 0.014 0 554 0.753 0,875 0,766 0,290 0.003 0 005 0.005 0,006 0 767 0 767 0 680
31 0 013 0 529 0.719 0 835 0 871 0 329 0,003 0.0 0 5 0 005 n nns n 7 3 2 0 839 0 694
37 0 017 0 500 0.680 0.788 0.820 0.854 0 003 0.0 0 5 0.005 0.005 0 692 0 805 0 669
33 0 5 5 0 0 512 0.448 0 688 0.340 0129 0,584 0 465 n 473 0,470 0.386 0 340 0 303
3A 0 3 5 7 0 676 0.592 0.513 0.449 0170 0,177 0 .6 1 4 0.560 0,545 0 510 0 450 Q.4QQ
35 0 408 0 731 0.640 0,555 0,486 0,184 0,197 0 613 0 676 0 653 n 552 0 486 0 433
3fi 0 347 0 761 0.691 0 599 0.524 0198 0.168 0.5 8 3 0 575 0.725 n 595 0 525 0.467
37 0 014 0 581 0.790 0 809 0,709 0 268 0 003 0.0 0 6 n n n 6 0.006 0,805 0 710 0 631
38 0.013 0 529 0.719 0,835 0,837 0 322 0 003 0 005 n n n s 0.005 0 732 0 854 0 667
39 0 014 0 555 0.754 0 874 0.817 0316 0,003 0 005 n n n s 0.006 0.768 0 787 0 7 4 4
AO 0 0 1 7 0 497 0 680 0 784 0 816 0 849 0,003 0 005 0,005 0.005 9 889 0 801 0 666
A1 0414 0.528 0.462 0 401 0.351 0133 0,?68 0 480 n 488 0.485 n 399 0,351 0 313
47 0 33 7 0682 0.597 0,518 0 453 0,171 0.168 0 583 0.535 n 524 n 5 i 5 R 454 0 404
43 0 870 0 733 0642 0 556 0.487 0184 0 139 0 266 0 ?65 0,257 0,553 9 488 0 434
4 4 0 059 0 793 0.718 0 622 0 545 0 206 0,012 0 023 0 023 0 0?4 n R i9 n 54R 0 485
45 0 030 0 830 0.783 0 695 0 608 0,230 0.006 0 012 0,012 n m ? n fi9 i n fin 9 0 542
48 0 013 0 526 0 719 0 879 0.8?1 0 312 0 003 0 005 n n n s 0,005 0 728 n 8?R 0 655
4 7 0 017 0 484 0.662 0.763 0 794 0 827 0,002 0 005 n n os n n n s n fi7 i 0 780 0.648
Mean 0 195 0 614 0 668 0 662 0 6 1 9 0 312 0,101 0 180 0,175 n 181 n fi?3 n f ii3 0 535
Stri 0.244 0 117 0.121 0 169 0.187 0.205 0.153 0.232 0.226 0.237 0.138 n 18 ? 0,149
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
87
Table 11-Continued
DMU 40 41 42 43 44 45 46 47
1 0.463 0,004 0 .0 0 5 0.0 1 2 0146 0 278 0 6 8 5 Q.4QQ
2 0.334 0.014 0 .0 1 9 0 045 0 563 0 587 0.549 0.289
3 0.408 0 004 0 .0 0 5 0 013 0 161 0 307 0.635 0.353
4 0,516 0.003 ILQQ&0 011 0138 0 263 0.723 0 4 4 6
5 0 2 4 1 0.215 0 .2 9 3 0.492 0.594 0 559 0.5 2 3 0-208
6 0.339 0.004 0 .0 0 5 0.013 0 157 0 298 0.6 4 9 0.2 9 3
7 0.417 0.003 0 .0 0 5 0.011 0139 0 264 0 725 0.3 6 0
ft0,565 0 003 0 .0 0 5 0 011 0136 0 259 0.712 Q.488
9 0.092 0.418 0 .3 0 7 0.313 0.299 0,263 0.2 3 2 0.079
10 0.132 0 500 0.441 0.335 0.430 0,379 0.3 3 4 0.114
11 0,147 0,360 0 4 9 3 0.268 0 480 0 423 0.372 0.1 2 8
12 0,162 0 240 0 .3 2 8 0 552 0 527 0464 0 4 0 9 O 140
13 0.192 0.008 0 .0 1 2 0.028 0 330 0.550 0.484 0.1 6 6
0 231 0.004 HODS. 0.013 0.163 0 303 0.583 0.2 0 0
14 .
15 0,258 0.004 0 .0 0 5 0.013 0 155 0.295 0.6 5 0 0.2 2 3
16 0752 0.003 0 .0 0 4 0.011 0.132 0 250 0.689 0.6 5 0
17 0.102 0 463 0 3 4 0 0,347 0 331 0 292 0,257 0.0 8 8
18 0,142 0.354 0 .4 7 3 0253 0461 0.406 jL25& 0.1 2 2
19 0.155 0 222 0 .2 9 8 0 530 0 506 0 446 O 392 O 134
20 0,179 0.013 0 .0 1 7 0 042 0 485 0512 0.451 0.1 5 5
21 0212 0005 0 .0 0 7 0.017 0 210 0.376 0.535 0.1 8 3
22 0 270 0 003 O 0 0 5 0 011 0.140 0 266 0.683 0 234
23 0 293 0.003 0 .0 0 5 0.011 0,140 0,266 0.723 0.2 5 3
24 0600 0.003 0 .0 0 5 0 011 0 135 0 257 0.706 0.5 1 8
25 0,111 0.505 0 .3 7 0 0 378 0.361 0 318 0.2 8 0 0.096
26 0,150 0.326 0 .4 6 5 0.246 0 488 0 430 0.379 0.130
27 0 163 0.195 0 .2 6 7 0 557 0.531 0.468 0.4 1 2 0.141
28 0 201 0.006 0 .0 0 8 0.0 1 8 0.220 0.391 0.508 0 .1 7 4
29 0 229 0 004 0 .0 0 6 0 013 0.167 0.309 0.5 7 7 0.1 9 8
30 0,268 0.004 0 .0 0 5 0 012 0,146 0 277 0,677 0.2 3 2
31 0 305 0,003 0 .0 0 5 0 011 0.139 0.264 0.728 0 .2 6 4
32 0 791 0,003 0 .0 0 4 0.011 0 132 0 250 0.6 8 9 0 6ft4
33 O 119 0,542 0 .3 9 8 0.406 0.387 0.341 0.301 0.103
34 0,157 0.369 0 .5 2 5 0.293 0512 0 451 0.397 0.136
35 0,170 0.411 0 .5 2 9 0.3 5 4 0 553 0 488 0.4 2 9 0.1 4 7
36 0.184 0.350 0 .5 3 0 0.302 0,596 0,526 0.4 6 4 0.159
37 0 248 0 004 0 .0 0 5 0.012 0.153 0.291 0.6 2 7 0 215
38 0 299 0.003 0 .0 0 5 0011 0.139 0 264 0 728 0-258
39 0 293 0,004 0 .0 0 5 0 0 1 2 0 146 0 277 0.688 0-253
40 0,801 0.003 0 .0 0 4 0,011 0 132 0 251 0.690 0 693
41 0.123 0.559 0 4 1 0 0 344 0.400 0.352 0.3 1 0 0.106
42 0 159 0,350 0-530 0 280 0.516 0 455 0.401 0.1 3 7
43 Q.171 0.177 0.241 0 582 0.555 0 489 0.431 0.148
44 0.191 0,015 0.021 0 050 0 621 0,547 0482 0.165
45 0 213 0,008 0.011 0026 0 321 0,611 0.538 0.1 8 4
46 0 289 0,003 0.005 0.011 0,139 0.265 0.729 0.250
47 0,778 0.003 0 .0 0 4 0.010 0,128 0 244 0 670 1.000
Mean 0.290 0.143 0 .1 5 8 0 156 0 307 0,364 0,536 0.2 5 7
Std 0.192 0.191 0.203 0.191 0.177 0.11 0.154 0.189
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
88
22 0.761 31.41
23 0 .9 5 5 0.6 9 7 37 02
24 0 .7 0 7 0.321 120.25
25 0.6 1 0 0.1 0 5 480.95
26 0 .6 9 7 0 .2 0 7 236.71
27 0 .7 5 3 0.195 286.15
28 0 .8 6 4 0.6 1 4 40.72
29 0 .8 6 0 0.6 6 8 28 74
30 0.8 7 5 0.6 6 2 32.18
31 0.871 0.619 4 0 71
32 0 854 0.312 173.72
33 0 .5 8 4 0.101 478.22
34 0 .6 1 4 0.1 8 0 241.11
35 0.6 7 6 0.175 286 29
36 0 725 0 181 300.55
37 0805 0.623 29.21
38 0 .8 5 4 0.613 39.31
39 0 .7 4 4 0.535 39.07
40 0.801 0.2 9 0 176.21
41 0 .5 5 9 0.143 290.91
42 0 .5 3 0 0.158 235.44
43 0.5 8 2 0.156 273.08
44 0.621 0.307 102.28
45 0.611 0 .3 6 4 67.86
46 0 729 0.536 36.01
.A 7. 1.000 0257 .289.11.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
89
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
90
1 0.863 1
1 0.857 1
1 0.016 |
| C980
| 0S80
o CO CO CO
0.825
0.870
0.853
0.825
0.849
0.871
CO CO CO CO
CO CO 00 CO
o' o* d d
|
|
in CO T r- CO o T- in CO CM o
CM GO r* CO 05 00 r- o o CO o CM o 05 CM
co CO co CO 00 00 05 05 05 CO 05 05 05 00 O
o' o o o o o O O O o o o o o o
05 o in t- T CO CO M" 00 m 00 in
r* r* m C0 CO CO 05 05 00 m 05 05 8 00
00 co CO GO CO oo CO 00 CO CO CO CO 00 CO o
o' o' o o' o o o o o o o o o o o
r*. CM o CO CO M- CO o h- CO M o
o oo 05 o CO CO 05 I". CO CO o r- r*- 00
05 05 o. 05 05 05 05 05 05 o 05 05 05 05 o
o' o' T _ o' o O o o o T o O o o o
m 00 CO CO o 00 o o CO in CM 05 CM co
m h- h - o 05 r>-
Table 14. Distributor Cross Efficiency Matrix
05 U05) 05 CM r^ - CM
05 $ 05 05 05 05 o 8 05 05 05 8 o
o' o' o ' o' o' O* O* o o d O d o d
CO o CO CO o fo- CO CM o CO 05 CO CO CM
00 05 05 o o CO o o o o o 05 T
CO 00 CO CO co f- 00 CO 00 o
o o* o o' o o o o o o o o d o
co o m CO o CD o eo in f- m 05 CM CO
in m CO CO CO M- CO r CO h - h - CO
s 00 oo oo 00 co CO CO 00 CO oo co co CO CO o
o o o o o o o o o o o o o o o
T < CO o r-'- CO m r*. in 05 CO l- m
o 05 05 o o 05 05 CO 05 05 00 05 05 o
CO O r- CO CO r~- CO CO CO CO CO CO CO CO o
o o o ' o' o o o o o o o o o o o
CO in o CD 00 r- r-. t" . o in CO 05
CO CO eo
lO 05 C
Tf O M- CM T co CO co CO o
05 05 05 05 05 05 05 05 05 05 05 05 05 o
o o' o ' o' O o o o o O d o o o o
CO CM in 00 co 05 CM o CM o M- 00 CM
CO CO 05 CO CO 05 CO CO 05 CO CO 00
co co oo oo 00 00 CO 00 CO CO co CO CO CO o
d o' o o o o o o o o o o o o o
CO o 05 CM in CM co 05 00 in r * CO m CO
M in T CM o o 05 CM o o o
CO 05 C
t-
05 05 05 05 05 05 05 00 05 05 05 05 05 o
o' o' o o O O o o o o O o o o o
CO M- T in CO CO CO o m CO
in in m O CO M- M- O O o
CM 05 C
m O
05 05 05 05 05 05 05 05 05 05 05 05 05 o
o o' o ' o' o o o o o O o O o o o
CD CD 05 ^ m CM in 00 CO 05 o m M" CO
W CM m in M" CO CM o
05 05 05 05 05 05 05 05 05 05 05 05 05 05 o
d d o ' o' o' o' o o o o o o o o o
|
c
M eo CO
I Std.
2 T- CM co M- in CO 00 05 o
T
T C
T " T T o
Q 2
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table IS. Distributor False-Positive Indices
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
92
Table 16. Compatibility Data for all Possible Value Chain Combinations
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
93
Table 16-Continued
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 17. Zero-One Integer Programming Model
subject to:
Xl + X2+X3 + X4+ X5+ X6 + X7+ X8 +X9 + XU)+Xn + Xi2+Xi3 + Xi4 + Xi5 + XI6 + Xn +
X lg + X 19 + X 20 + X21 + X22 + X 23 + X 24 + X25 + X26 + X 27 + X28 + X29 + X30 + X3 1 + X32 + X33 +
X34 + X35 + X 36 + X37 + X38 + X39 + X40 + X4 [ + X42 + X43 + X44 + X45 + X46 + X47 + X48 + X49 +
X50 + X5 1 + X 52 + X 53 + X54 + X55 + X56 + X57 + X58 + X 59 + X60 + X61 + X62 + X63 + X64 + X65 +
X66 + X67 + X68 + X69 + X70 + X7 1 + X72 + X73 + X74 + X75 + X76 + X77 + X78 + X79 + Xso + X 8 1 +
X82 + X 83 + X84 + Xg5 + X86 + X87 + X8 8 + X 89 + X90 = I
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8 5 x i + 4 7 x 2 + 6 5 x 3 + 4 2 x 4 + 6 1 x 5 + 7 2 x 6 + 5 6 x 7 + 7 9 x g + 6 5 x 9 + 5 9 x [0 + 8 2 x M +
89 Xi2 + 7 7 Xu + 9 0 Xu + 3 7 x I5+ 35 x t6+ 3 3 xn + 65 x i8 + 5 2 x , 9 + 7 2 x 20 + 3 6 x 2i +
7 0 X 2 2 + 5 5 X2 3 + 5 7 X 2 4 + 8 0 X 2 5 + 81 xm + 3 2 X 27+ 8 9 x 2 8 + 5 6 x 2 g + 3 4 x 30 + 8 6 X 31 +
4 9 X 3 2 + 3 5 X 3 3 + 71 X 3 4 + 7 6 X 3 5 + 6 2 x 36 + 7 4 X3 7 + 4 2 x 38+ 8 9 X 3 9 + 5 5 X4 0 + 8 2 x 4i +
4 2 x 4 2 + 6 5 X4 3 + 4 2 X 4 4 + 8 5 X 4 5 + 7 9 X 4 6 + 7 4 X4 7 + 3 7 x 4 8 + 5 2 X 4 9 + 6 4 x 5 0 + 4 8 x 5! +
3 6 X 5 2 + 6 3 X 53 + 9 0 X 5 4 + 5 2 X 5 5 + 6 8 X 5 6 + 3 2 X 5 7 + 4 2 Xs8 + 3 7 Xs9 + 5 7 xj + 4 6 X 6 i +
86 x62 + 33 x63 + 44 Xg4 + 58 x65 + 38 X66 + 75 x67 + 84 Xes + 56 x69 + 84 x70 + 55 x7i +
54 X72 + 34 X73 + 60 X74 + 46 X75 + 68 x76 + 58 X77 + 33 x78 + 31 X79 + 76 Xgo+ 39 x8i +
35 X82+ 39 Xg3 + 73 xw+ 63 XK+ 60 Xs6 + 54 Xs7 + 85 X8s+ 31 Xs9 + 39x90 - v3 = 31
x; = 0 or 1 where / = 1, 2, 3 , . . . . , 90.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
96
Table 18. Value Chain Combinations in Ascending Order o f Objective Function Value
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
97
Table 18-Continued
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
98
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
99
DMU 1 2 3 4 5 6 7 8 9 10
1 1.000 0.959 0.862 0.580 0.613 0.797 0.499 0.662 0.816 0.638
2 0.866 1.000 0.801 0.573 0.606 0.700 0.432 0.679 0.842 0.577
3 0.915 0.926 1.000 0.531 0.561 0.744 0.539 0.768 0.818 0.646
4 0.888 0.865 0.766 0.847 0.774 0.858 0.515 0.588 0.864 0.677
5 0.918 0.894 0.791 0.755 0.800 0.862 0.517 0.608 0.793 0.690
6 0.871 0.848 0.751 0.707 0.748 1.000 0.600 0.576 0.743 0.664
7 0.901 0.865 0.899 0.705 0.743 0.991 1.000 0.826 0.736 0.761
8 0.886 0.955 0.948 0.547 0.579 0.706 0.613 1.000 0.849 0.653
9 0.791 0.837 0.732 0.563 0.528 0.658 0.395 0.614 1.000 0.520
10 0.949 0.911 0.906 0.658 0.696 0.866 0.600 0.695 0.775 0.801
Mean 0.898 0.906 0.846 0.647 0.665 0.818 0.571 0.702 0.823 0.663
Std. 0.055 0.054 0.091 0.105 0.098 0.119 0.167 0.132 0.076 0.081
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 23. Cross Efficiency Matrix for Window 2
DMU 1 2 3 4 5 6 8 9 10 11
1 0.951 0.913 0.820 0.551 0.583 0.758 0.630 0.776 0.606 0.979
2 0.801 0.925 0.742 0.530 0.561 0.648 0.629 0.779 0.534 0.993
3 0.883 0.892 0.965 0.511 0.541 0.718 0.740 0.789 0.623 0.995
4 0.888 0.865 0.766 0.847 0.774 0.858 0.588 0.864 0.677 0.928
5 0.896 0.873 0.773 0.737 0.781 0.841 0.593 0.774 0.674 0.937
6 0.871 0.848 0.751 0.706 0.747 1.000 0.576 0.743 0.664 0.910
8 0.813 0.875 0.870 0.501 0.531 0.648 0.918 0.779 0.600 1.000
9 0.791 0.837 0.732 0.563 0.529 0.658 0.614 1.000 0.519 0.956
10 0.928 0.890 0.885 0.643 0.681 0.846 0.679 0.757 0.783 0.955
11 0.770 0.875 0.712 0.501 0.531 0.623 0.618 0.749 0.513 1.000
Mean 0.859 0.879 0.801 0.609 0.626 0.760 0.659 0.801 0.619 0.965
Std. 0.062 0.027 0.082 0.119 0.108 0.124 0.103 0.078 0.084 0.033
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
APPENDIX B
COMPUTER PROGRAMS
101
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
102
C THIS SUBROUTINE WILL DEVELOP A MATRIX FOR THE BASIC CCR
C DEA MODEL
SUBROUTINE USER(LIN)
COMMON/TWO/01(600,14)
COMMON/FOUR/IRINGE
INTEGER DMU
READ(1,*) DMU,NUMINP,NUMOUT
NUMTOT=NUMINP +NUMOUT
50 FORMAT(I2, IX,I2,1X,I2)
DO 20 K = 1,DMU
READ(1,*) (01(K,L), L=l,NUMTOT)
20 CONTINUE
DO 10 J=1,DMU
NUM1 = J
NOISE = 0
CALL QUIET(NOISE)
CALL SOLVE(DMU,NUM 1,NUMINP,NUMOUT)
CALL CLRBAS
10 CONTINUE
RETURN
END
SUBROUTINE SOLVE(DMU,NUMl,NUMINP,NUMOUT)
C MATRIX GENERATOR FOR DEA PROBLEM
COMMON/TWO /0 1(600,14)
DIMENSION NAMEV(8),IRO(650),VAL(650)
DIMENSION KALFNM(36), VALUE(24)
LOGICAL TRUBLE
INTEGER DMU
CHARACTER* 1 NAMEV,NULL,KALFNM
REAL*8 ACCUM_A, ACCUM B, USED TIME
NULL = ''
DATA KALFNM/O',' 1','2',3,'4','5','6','T,'8','9',
+ ,A,,B,,,C,,T),,E ,,,F ,,G',,H,T,
+ ,J,lK','L',,M,,Tsr,'0 ,,T',,Q','R',
+ 1S \'T ,V ','Y ,tWr,''X,'Y,'Z/
C GET INPUT FILE
NUMTOT=NUMINP+NUMOUT
C INITILIAZE ROWS
IF(NUMLGT. 1) GO TO 5
C OBJECTIVE ROW
CALL DEFROW(-1,0.,IRNO,TRUBLE)
C OUTPUT AND INPUT CONSTRAINTS
DO 22 IT = 1,DMU
CALL DEFROW(-1,0.,IRNO,TRUBLE)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
103
22 CONTINUE
CALL DEFROW(0,1.,IRNO,TRUBLE)
5 CONTINUE
C VARIABLES DEFINITION
NONZ = DMU+1
NAMEV( 1)=KALFNM(32)
DO 900 IT = 1, NUMINP
NAMEV(2)=KALFNM(IT+1)
NAMEV(3 )=NULL
NAMEV(4) =NULL
NAMEV(5) = NULL
NAMEV(6) = NULL
NAMEV(7) = NULL
NAMEV(8) = NULL
DO 100 JT = 1, DMU
IF(JT.EQ.NUM 1) GO TO 100
IF(JT. GT.NUM 1) GO TO 99
VAL(JT)= 01(JT,IT)
IRO(JT) = JT+1
GO TO 100
99 CONTINUE
NT=JT-1
VAL(NT)=01(NT+1,IT)
ERO(NT)=NT+l
100 CONTINUE
VAL(DMU)=01(NUM1,IT)
ERO(DMU)=DMU+1
IF(NUMl.EQ.l) GO TO 112
DO 111 JT=1,DMU
CALL INSERT(IRO(JT),IT,VAL(JT), 1)
111 CONTINUE
C CONSTRAINT FOR EQUAL TO ONE FOR SUM OF OUTPUTS
112 CONTINUE
VAL(DMU+1)=01(NUM1,IT)
IRO(DMU+1)=DMU+2
IF(NUMl.EQ.l) GO TO 113
CALL INSERT(IRO(DMU+1),IT,VAL(DMU+1), 1)
GO TO 900
113 CONTINUE
CALL APPCOL(NAMEV,NONZ,VAL,IRO,TRUBLE)
900 CONTINUE
C OUTPUT VARIABLES
NONZ = DMU+1
NAMEV( 1)=KALFNM(31)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
DO 800 IT = NUMINP+1,NUMTOT
I=IT-NUMINP+I
NAMEV(2)=KALFNM(I)
NAMEV(3 )=NULL
NAMEV(4) =NULL
NAMEV(5) = NULL
NAMEV(6) = NULL
NAMEV(7) = NULL
NAMEV(8) = NULL
VAL(1)=01(NUM1,IT)
IRO(l)=l
IF(NUMl.EQ.l) GO TO 107
CALL INSERT(IRO( 1),IT,VAL( 1), 1)
107 CONTINUE
C CONSTRAINT FOR OTHER DMU'S
DO 110 JT = 1, DMU
IF(JT.EQ.NUM1) GO TO 110
IF(JT. GT.NUM 1) GO TO 160
VAL(JT+1)=-01(JT,IT)
IRO(JT+l) = JT+1
GOTO 110
160 NT=JT-1
VAL(NT+1)= -01(NT+1,IT)
IRO(NT+l) = NT+1
110 CONTINUE
VAL(DMU+1) = -01(NUM1,IT)
ERO(DMU+l) = DMU+I
IFCNUM1.EQ.1) GO TO 118
DO 117 JT=2,DMU+1
CALL INSERT(IRO(JT),IT, VAL(JT), 1)
117 CONTINUE
GO TO 800
118 CONTINUE
CALL APPCOL(NAMEV,NONZ,VAL,IRO,TRUBLE)
800 CONTINUE
CALL GO(0,KONDN)
C PRINTING OUT VARIABLE VALUES
8 FORMAT(4(D 16.10,2X))
DO 143 J=l,NUMTOT
CALL REPVAR(J,PRIMAL,DUAL)
VALUE(J) = PRIMAL
143 CONTINUE
WRITE(3,8) (VALUE(J),J= 1,NUMTOT)
CALL REPROW(l,PRIMAL,DUAL)
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
WRITE( 11,*) PRIMAL
C SEND OPTIMAL VALUE TO A FILE, ALONG WITH SEED,MACHINE #, TIME,
C PART#
RETURN
END
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
C TfflS SUBROUTINE WILL DEVELOP A MATRIX FOR THE PAIRWISE
C CROSS EFFICIENCY MODEL
SUBROUTINE USER(LIN)
COMMON/TWO/01(600,14),EFF(600)
COMMON/FOUR/IRINGE
INTEGER DMU
READO,*) DMU,NUMINP,NUMOUT
NUMTOT=NUMINP +NUMOUT
50 FORMAT(I2, IX,12, IX,12)
DO 20 K = I,DMU
READ(1,*) (01(K,L), L=l,NUMTOT)
20 CONTINUE
READ(I ,*) (EFF(K),K= 1,DMU)
DO 10 1=1,DMU
DO 10 J=1,DMU
NUM1 = I
NUM2 =1
NOISE = 0
IF(I.EQ.J) GO TO 10
CALL QUIET(NOISE)
CALL IN1T
CALL SOLVE(DMU,NUM 1,NUM2,NUMINP,NUMOUT)
CALL CLRBAS
10 CONTINUE
RETURN
END
SUBROUTINE SOLVE(DMU,NUM 1,NUM2,NUMINP,NUMOUT)
C MATRIX GENERATOR FOR DEA PROBLEM
COMMON/TWO/01(600,14),EFF(600)
DIMENSION NAMEV(8),IRO(650),VAL(650)
DIMENSION KALFNM(36),VALUE(24)
LOGICAL TRUBLE
INTEGER DMU
CHARACTER* 1 NAMEV,NULL,KALFNM
REAL*8 ACCUM A, ACCUM B, USED TIME
NULL = ''
DATA KALFNM/O',' 1','2','3',4','5','6','7','8','9',
+ 'A','B',,C',T),,,E',,F ,,G,,TF,T,
+ T ^ lA T V tNVO'^'QVR',
+ 'SVT/UTVTWYXVY','Z/
C GET INPUT FILE
NUMTOT=NUMINP+NUMOUT
C INITILIAZE ROWS
C OBJECTIVE ROW
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
107
CALL DEFROW(1,0.,IRNO,TRUBLE)
CALL DEFROW(0,1.,IRNO,TRUBLE)
CALL DEFROW(0,0.,IRNO,TRUBLE)
CALL DEFROW(1,0.,IRNO,TRUBLE)
C VARIABLES DEFINITION
NONZ=3
NAMEV( 1)=KALFNM(32)
DO 900 IT = 1, NUMINP
NAMEV(2)=KALFNM(IT+1)
NAMEV(3)=NULL
NAMEV(4) =NULL
NAMEV(5) = NULL
NAMEV(6) = NULL
NAMEV(7) = NULL
NAMEV(8) = NULL
VAL(1)= 01(NUM2,IT)
IRO(l) =2
C0EFF=-01(NUM1,IT)*EFF(NUM1)
VAL(2)=COEFF
ERO(2)=3
VAL(3)=-OI(NUM2,IT)
IRO(3)=4
CALL APPCOL(NAMEV,NONZ,VAL,IRO,TRUBLE)
900 CONTINUE
C OUTPUT VARIABLES
NONZ=3
NAMEV( 1)=KALFNM( 31)
DO 800 IT = NUMINP+1,NUMTOT
I=IT-NUMINP+1
NAMEV(2)=KALFNM(I)
NAMEV(3)=NULL
NAMEV(4) =NULL
NAMEV(5) = NULL
NAMEV(6) = NULL
NAMEV(7) = NULL
NAMEV(8) = NULL
VAL(1)=01(NUM2,IT)
IR O (l)= l
VAL(2)=01(NUM1,IT)
IRO(2)=3
VAL(3)=Ol(NUM2,IT)
IRO(3)=4
CALL APPCOL(NAMEV,NONZ, VAL,IRO,TRUBLE)
800 CONTINUE
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
108
CALL GO(0,KONDN)
8 FORMAT(4(D 16.10,2X))
DO 143 J=l,NUMTOT
CALL REPVAR(J,PRIMAL,DUAL)
VALUE(J) = PRIMAL
143 CONTINUE
WRITE(3,8) (VALUE(J),J=1,NUMTOT)
C SEND OPTIMAL VALUE TO A FILE, ALONG WITH SEED,MACHINE #, TIME,
PART#
RETURN
END
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
C THIS PROGRAM COMPUTES THE PAIRWISE CROSS EFFICIENCY
C MATRIX
DIMENSION 01(600,20),02(2600,20), 03(600,600), 04(600),EFF(600)
DIMENSION 05(600,600)
INTEGER DMU
READ(1,*) DMU,NUMINP,NUMOUT
NUMTOT=NUMINP +NUMOUT
DO 20 K = I,DMU
READ(1,*) (01(K,L), L=l,NUMTOT)
20 CONTINUE
DO 19 K = I,DMU
READ(1,*) EFF(K)
19 CONTINUE
DO 21 K = 1,DMU*(DMU-1)
READ(2,*) (02(K,L), L=l,NUMTOT)
21 CONTINUE
DO 10 1=1,DMU
DO 10 J=1,DMU
IF(I.EQ.J) GO TO 10
IF(I.LT.J) THEN
JT=J-1
ELSE
JT=J
ENDIF
TOP = 0
T1 = 0
K=(DMU-1)*(I-1)+JT
DO 30 NUM3NP1=1,NUMINP
T1=01( J,NUMINP 1)*02(K,NUMINP I)
TOP =TOP+Tl
30 CONTINUE
BOT = 0
B1 = 0
DO 40 NUMOUTl=l,NUMOUT
B1 = O l(J,NUMINP+NUMOUT 1)*02(K,NUMINP+NUMOUT 1)
BOT = BOT+B1
40 CONTINUE
03(1,J) = BOT/TOP
10 CONTINUE
DO 13 1=1,DMU
DO 13 J=1,DMU
IF(I.EQ.J) THEN
05(I,J)=EFF(I)
GOTO 13
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ENDEF
05(U)=03(I,J)
13 CONTINUE
DO 14 1=1,DMU
WRITE( 11,*) (05(1,J), J=1,DMU)
14 CONTINUE
DO 60 J=1,DMU
AVG1= 0
AVGTOT=0
DO 50 1=1,DMU
AVG1= 05(1, J)
AVGTOT =AVG1+AVGT0T
50 CONTINUE
04(J)=AVGT0T/DMU
60 CONTINUE
DO 70 1=1,DMU
WRITE(11,*) DMU =\ I, 'COLUMN AVERAGE VALUE=',04(I)
70 CONTINUE
DO 65 J=1,DMU
AVG1 = 0
AVGTOT=0
DO 55 1=1,DMU
AVG1= 05(J,I)
AVGTOT =AVG1 +AVGTOT
55 CONTINUE
04(J)=AVGTOT/DMU
65 CONTINUE
DO 75 1=1,DMU
WRITE(11,*) DMU =\ I, 'ROW AVERAGE VALUE=',04(I)
75 CONTINUE
RETURN
END
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
BIBLIOGRAPHY
Athanassopoulos, AD. and Ballantine, J.A. (1995) Ratio and Frontier Analysis for Assessing
Corporate Performance: Evidence from The Grocery Industry in the UK, Journal of the
Operational Research Society. 46(4), pp. 427-440.
Bailes, J. C and Kleinsorge, I. K (1992) Cutting waste with JIT, Management Accounting, 73
(11), pp 28-32.
Barr, R.S. and Seiford L.M. (1994) Benchmarking with Data Envelopment Analysis, resented
at the October, 1994, ORSA/TIMS Joint National Meeting, Detroit, MI.
Bell, H. (1976) Geochemical Reconnaissance Using Heavy Minerals From Small Streams in
Central South California, Geological Survey Bulletin 1404, Washington, D.C.: U. S.
Government Printing Office, pp 1-22.
Bell, R.A and Morey C., (1995) Increasing the Efficiency of Corporate Travel Management
through Macro Benchmarking, Journal of Travel Research, 33(3), pp. 11-20.
Boussofiane, A , Dyson, R.G and Thanassoulis E (1991) Applied data envelopment analysis,
European Journal of Operational Research, 52, pp 1-15.
Busby, J. S and Fan, I. S (1993) The extended manufacturing enterprise : Its nature and its
needs, International Journal o f Technology Management, 8(4), pp 294-308.
Ill
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
112
Chames, A. and Cooper, W. W (1961) Management models and industrial applications of
linear programming, John Wiley & Sons, New York.
Chames, A., Cooper, W.W. and Rhodes, E (1978) Measuring the efficiency of decision
making units, European Journal Of Operational Research, 2(6), pp 429-444.
Chames, A., Clark, T., Cooper W. W and Golany B (1985) A Development Study of Data
Envelopment Analysis in Measuring the Efficiency of Maintenance Units in the U.S. Airforces,
Annals o f Operational Research, 2, pp 95-112.
Chames, A., Cooper, W.W., Golany, B., Seiford, L. and Stutz, J. (1985) Foundations of data
envelopment analysis for Pareto-Koopmans efficient production functions, Journal of
Econometrics, 30, pp 91-107.
Chames, A., Cooper, W.W., Huang, Z. M. and Sun D. B. (1990) Polyhedral cone-ratio DEA
models with an illustrative application to large commercial banks, Journal of Econometrics,
30, pp 91-107.
Chames, A., Cooper W.W, et al, Eds. (1995) Data envelopment analysis : Theory,
methodology and applications, Boston, Khxwer
Christopher, M. (1994) Logistics and supply chain management, Irwin, Inc., New York.
Collier, D.A. and Storbeck, J.E., (1993) A Data Envelopment Approach to Benchmarking in
the Telecommunications Industry, Ohio State Faculty of Management Science Working
Paper, Columbus: Ohio State University.
Davis, T. (Summer 1993) Effective supply chain management, Sloan Management Review, pp
35- 46.
Dresner, M. and Xu, K. (1995) Customer Service, Customer Satisfaction, and Corporate
Performance in the Service Sector, Journal o f Business Logistics, 16(1), pp 23-40.
Dyson, R.G., and Thannassoulis, E. (1988) Reducing weight flexibility in data envelopment
analysis, Journal of Operational Research Society, 39(6), pp 563-576.
Fanizza, G. and Bogyo, T. P. (1976) A Cluster Analysis of Almond Varieties in Apulia. Riv.
Ortoflorofrutt. It. 60, pp 277-281.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
113
Henderson, C. J (1990) Plugging into strategic partnerships : The critical is connection, Sloan
Management Review, 31(3), pp 7-18.
Herald, C., Williams, D. and Fitzgerald, L. (1993) Supply chain methodology, Human
Systems Management, 12, pp 17-23.
Iacocca Institute (1991) 21st Century manufacturing strategy, Lehigh University, Bethlehem,
PA.
Johnston, R. and Lawrence, L. R (1988) Beyond vertical integration - The rise of the value-
adding partnership, Harvard Business Review, 66(4), pp 94-101.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
114
Lee, S. M (1972) Goal programming for decision analysis, Auerbach Publishers, Philadelphia.
Leib, R. C. and Miller, R. A (Spring 1988) JIT and corporate transportation requirements,
Transportation Journal, 27(3), pp 5-10.
Miles, R. E and Rosenberg, H. (Winter 1982) The Human research approach to management:
second-generation issues, Organizational Dynamics, 5, pp 24-38.
Miles, R. E and Snow, C. C (Spring 1984) Fit, failure, and hall o f fame, California
Management Review, 3, pp 15-28.
Miles, R. E and Snow, C. C. (1986) Network organizations: new concepts for new forms,
California Management Review, 3, pp 62-73.
Penrose, L. S. (1953) Distance, Size and Shape, Annals ofEugenics, 18, pp 337-343.
Presley, A., Barnett, B. and Liles, D.H. (1995) A virtual enterprise architecture, Fourth
Annual Agility Forum Conference, 2, pp 3-12.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
115
Rosenthal, S. R. (1992) Effective Product Design and Development, Business One Irvin,
Homewood, Illinois.
Rousseau, J.J and Semple, J. (1993) Two-person ratio efficiency games, Management Science
(In Press).
Sarkis, J. and Talluri, S. (1995) Efficiency valuation and business process improvement
through internal benchmarking. Journal of Valuation and Cost Analysis, 1(1).
Schonberger, R. J (1986) World Class Manufacturing, New York: The Free Press.
Sexton, T. R, Slinkman, R. H and Hogan, A. (1986) Data envelopment analysis: Critique and
extensions, in: Richard HL Slinkman (Ed.), Measuring Efficiency : An Assessment of Data
Envelopment Analysis, Publication no. 32 in the series New Directions of Program
Evaluation, Jossey Bass, San Francisco.
Slowinski, G. (1992) The Human Touch in Successful Strategic Alliances, Mergers and
Acquisitions, 27(1), pp 44-47.
Snow, C. C and Thomas, J. (1992) Building networks: broker roles and behaviors, in Peter
Lorange, et aL, Editors, Strategic Processes: Designing for the 1990s, Basil Blackwell.
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
116
Thompson, R. G., Singleton, F. D., Thrall, R. M. and Smith, B. A (1986) Comparative site
evaluations for locating a high energy physics laboratory in Texas, Interfaces, 16, pp 35-49.
Thompson, R.G., Dharmapala, P.S. and Thrall, R. M. (1995) Linked-cone DEA profit ratios
and technical efficiency with application to Illinois coal mines, International Journal of
Production Economics, 39, pp 99-115.
Talluri, S., Huq, F. and Pinney, W.E. (1995) Application of data envelopment analysis for cell
performance evaluation and process improvement in cellular manufacturing, In Press,
International Journal o f Production Research.
Turner, J. R (1993) Integrated supply chain management: whats wrong with this picture?
Industrial Engineering, 25(12), pp 52-55.
Vakharia, A. (1986) Methods o f Cell Formation in GT: A Framework for Evaluation, Journal
of Operations Management, 6(3), pp 257-72.
Wong, Y.H.B. and Beasley, J. E. (1990) Restricting weight flexibility in DEA, Journal of
Operational Research Society, 41(9), pp 829-835.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.