Strategy in Industrial Networks:

Transcription

Strategy in
Industrial Networks:
EXPERIENCES FROM IKEA
Enrico Baraldi
I
KEA was founded over 60 years ago in Southern Sweden. It has since
grown to become the world’s largest furniture retailer; in 2006, it had
sales of over Euro17 billion, as well as 12,000 product items and
104,000 employees. The company’s focus has consistently been on marketing products at extremely low prices. Its first purchases in the 1950s were
made from producers’ unsold stocks, in order to keep costs low. However, large
sales success soon allowed IKEA to start ordering models of its own design from
local manufacturers. Next, IKEA introduced innovations, such as flat packs,
which reduced production and transport costs, and the “showroom-warehouse”
concept, which reduced retailing costs.
During its expansion in the 1960s, IKEA also laid the groundwork for its
purchasing strategy, relying on long-term relationships with selected suppliers as
external sources for its offerings. Today, its supply network spans the entire
world and has become increasingly complex. However, the use of this network is
still in accordance with the same basic strategy as in the 1960s: to design and
purchase products that entail low production and transportation costs. IKEA
achieves this by carefully taking into account, in its design and purchase strategy, all the activities performed in the network, from raw materials to customer
homes. Remaining faithful to its original external orientation, IKEA performs
only a few of these activities internally, while it intensively uses its relationships
with suppliers to combine its internal and their external resources for the sake of
both efficiency and development. For instance, products are developed in close
interaction with suppliers while taking into consideration the impact of the raw
materials, components, and facilities involved, since all these resources entail
The author would like to thank Jan Gardberg of IKEA and all other managers that kindly provided the
empirical material for this article, and two anonymous reviewers for their insightful comments.
CALIFORNIA MANAGEMENT REVIEW VOL. 50, NO. 4
SUMMER 2008
CMR.BERKELEY.EDU
99
Strategy in Industrial Networks: Experiences from IKEA
FIGURE 1. IKEA and Its Industrial Network
IKEA’s
Boundary
IKEA Trading
Offices (40)
IKEA of
Sweden
IKEA
Distribution
Centers (26)
IKEA Stores
(220)
Suppliers
(1,300)
Logistic Partners
(500)
Sub-suppliers
(10,000)
costs and have an impact on quality, design, and function. In fact, next to low
costs, reasonable quality, appealing designs, and adequate product functionality
are major goals for IKEA. These goals induce the company to promote a constant
product and technical development, which contributes to its image as an innovative and fashion-oriented firm, but which depends heavily on the contribution
of its entire network of suppliers.
To cope with such tasks, IKEA needs advanced skills in marketing, retailing, logistics, purchasing, product development, and technologies. This need of
competence is reflected by IKEA’s complex organization, which consists of over
550 business units specializing in these fields and spread over more than 50
countries. However, the complexity of IKEA’s organization is overshadowed by
that of its industrial network (see Figure 1). This network includes 1,300 direct
suppliers and about 10,000 sub-suppliers,
spread over 60 countries. Over 220 IKEA
Enrico Baraldi is an Associate Professor at
stores are located in 30 countries including
Uppsala STS Center and at the Department of
Business Studies, Uppsala University, Sweden.
Europe, Australia, the U.S., and China.
<[email protected]>
Between IKEA’s stores and suppliers stands a
vital, but less visible part of IKEA’s network:
its wholesale and logistic operations, comprising 26 Distribution Centers spread
over 12 countries. Since IKEA does not own any transport facilities, this network is physically connected via another group of external actors, a few hundred logistic partners.
A pivotal role in this network is played by “IKEA of Sweden,” a leading
business unit that not only manages IKEA’s product range, but also supervises the
entire IKEA universe and develops long-term marketing, logistics, and purchas-
100
UNIVERSITY OF CALIFORNIA, BERKELEY
VOL. 50, NO. 4
SUMMER 2008
CMR.BERKELEY.EDU
ing strategies. In fact, whereas most IKEA units are rather specialized (e.g., local
purchasing for IKEA’s 40 Trading Offices), IKEA of Sweden has both an overall
responsibility and a coordinating role in the development, purchase, distribution, and marketing of each single product.
This article analyzes the experience of IKEA in dealing with its industrial
network and discusses the structural components and dynamic interactions of a
“network strategy,” that is, a strategy that considers and uses the external network for a company’s goals. The case study was built through 70 interviews,
conducted mainly in 1999-2003 with personnel at IKEA and suppliers in Sweden, Poland, and Italy.1
The Pervasiveness of Industrial Networks
Networks are widely publicized and researched phenomena, especially in
high-tech sectors,2 where the likes of Dell, Microsoft, or Genentech pursue their
network strategies through R&D joint ventures, cross-licensing, or strategic alliances.
However, these fashionable terms do not fit the type of networking going on
within the furniture industry or other low-tech industries. However, traditional
sectors also present network-like structures. Examples include the tile industry,3
the apparel industry,4 and the Italian districts5—which specialize, for instance, in
knitwear (Carpi), packaging machines (Bologna), or textiles (Prato). These sectors are composed of many small and medium-sized firms that develop close
links to their suppliers and partners, and often also cooperate intensively with
them on technical issues.
Networks are not only important for small firms that need to interact
with their peers to supplement their limited resources, but networks are fundamental for large companies as well. For instance, multinationals in the steel,6
paper,7 and automotive8 industries interact tightly with their suppliers, sub-suppliers, distributors, and customers to develop new technologies or increase efficiency. There are many examples of large firms from several sectors that relied
strongly on networks for their rapid growth: Apple, Benetton, Toyota, Corning,
and McDonald’s, to name a few.9 Finally, networks of stable relationships are the
norm in several industries, including construction, publishing, textiles, and cultural production.10
It seems that all types of firms, large or small, high-tech or low-tech,
interact closely with other firms and organizations around them. In other words,
inter-firm interactions and networks are everywhere in our economy.11 However,
despite all this interaction, inter-firm networks went practically unnoticed by
mainstream management research until around the 1980s. This neglect of interfirm relationships stems from both the actual behavior of firms—in periods characterized by arm’s-length relations,12 close interactions and networks are more
difficult to discern—and the dominant research paradigms, which either had
different units of analysis or viewed relationships simply as odd exceptions.13
Although Richardson14 recognized the importance of business relationships as early as 1972, it took time for these ideas to enter mainstream strategy
SUMMER 2008
CMR.BERKELEY.EDU
101
literature; and when this finally happened, the impetus came from sociology.
Granovetter stressed in 1985 that economic transactions are embedded in networks of social relations where trust matters,15 and Powell attributed networks a
status equal to markets and hierarchies, viewed as three alternative forms for
organizing economic activity.16 Eventually, Swedberg suggested that all markets
can be viewed as social structures filled with interactions, rather than as pure
price-driven mechanisms.17
In sum, there exists compelling evidence for the diffusion and persistence
of business relationships and networks, in all sectors and for all sizes of firms.
Mainstream strategy research was late in recognizing the importance of
networks, and did not take them into account until the influence of sociology
and widely publicized strategic alliances or joint ventures made it impossible to
neglect this interactive side of business life. However, there is still a risk that
focusing on these special and conspicuous networking episodes may hide the
bulk of networking activities and interactions that go on silently under the surface of daily business activity.18 To avoid such a risk, one can rely on a theoretical perspective that views industrial networks and relationships as essential
phenomena, that is, as the norm rather than the exception: this is the viewpoint
of the “Markets-as-Networks” approach.
“Markets-as-Networks”:
A Network-Based View of Business Management
Are we really sure that inter-firm relationships and networks are merely
exceptions in economic organizing, in a world where firms can only choose
between pure market exchanges or hierarchical control? What happens if we
overstate the argument of Swedberg and start considering instead networks as
the norm, that is, as the most natural and normal form of economic
organizing?19 Following this reasoning, networks came first, while markets and
hierarchies are both human constructions, that is, structures imposed on networks for the sake of transparency (markets) or control (hierarchies).20 Accepting networks as the norm can also help focus on the effects that they produce.
Examples of such “network effects” are unexpected product failures,21 irrational
patterns of electricity consumption,22 and imbalances spread by IT systems.23
However, to enable researchers and managers make sense of networks and their
effects, appropriate analytical tools are necessary.
A set of such tools for analyzing industrial networks as pervasive and
“normal” phenomena is offered by a research tradition known as “IMP” (Industrial Marketing and Purchasing).24 This approach, sometimes referred to as
“Markets-as-Networks,” grew out of extensive empirical studies of industrial
buyer and seller relationships conducted in Europe in the 1960s and 1970s, that
is, well before the modern frenzy over networks.25 The early empirical findings
were then related to sociological theories of exchange26 in order to develop a
series of models of the dyadic interaction between firms. These models stressed
the importance of power/dependence, cooperation, closeness, and expectations
in the daily interactions between buyer and seller.27 It was only a short step from
102
VOL. 50, NO. 4
SUMMER 2008
CMR.BERKELEY.EDU
TABLE 1. The Structural Components and the Dynamic Interactions of a Network Strategy
Structural Components
Dynamic Interactions
1. Defining Relationship Contents:
Contents include exchanged volumes, learning, trust,
commitment, duration, and control (depending on a
relationship’s goals, role and function).
Combining Resources:
Concrete and complex resource combinations created
through interaction processes with external actors:
inter-organizational routines/joint projects
2. Forming the Network Structure:
This structure is based on the number of business
relationships, their roles in the network, hierarchy, and
geographical location.
1. Interacting via Inter-Organizational
Routines:
Repetitive and formal processes that mobilize
relationships and enable resource combinations to
maintain efficiency.
3. Evaluating Goal Matching with the
Network:
Comparing a firm’s goals/resources with those of the
network.Which external resources are needed and
available? What are the goals of other actors?
2. Interacting via Joint Projects:
Ad hoc and informal processes that mobilize
relationships and enable resources combinations to
foster development.
single relationships to networks of relationships: when IMP researchers recognized
that single relationships are related to each other via technical, economic, and
social interdependencies, the way was paved for more complex models stretching to the entire network surrounding a firm.28
The Markets-as-Networks approach has gained currency within the field
of industrial marketing29 and international business,30 but it can also be used to
cast new light on strategic issues of efficiency and development.31 To summarize,
the Markets-as-Networks approach focuses on all types of inter-firm interactions. The starting point is that firms constantly interact with counterparts such
as key suppliers and customers through business relationships that are related in
a network structure. Consequently, firms are embedded within networks that
exist beyond their will. From a strategic point of view, firms need to be aware of
these networks and of how to use them actively. Companies vary widely in their
will and capacity to do this: not all firms can become “strategic centers,” which,
like IKEA, are able to manage a web of partners.32 Nevertheless, all firms are
embedded in a network, which can be both good and bad.33 It is therefore advisable for firms to understand how networks work and how they can be
approached for strategic purposes.
The Structure and Dynamics of a Network Strategy
A network strategy can be understood in terms of structures and dynamics,
according to the idea that a network structure composed of relationships and
external resources needs to emerge before a company can use the network in its
daily interactions with counterparts. Therefore, the analytical frame that we will
apply to discuss IKEA’s network strategy comprises three structural components
and two types of dynamic interactions (see Table 1). The structural components
concern the architecture of the network (e.g., the number of firms involved),
the long-term features of each business relationship (e.g., the goals of the
SUMMER 2008
CMR.BERKELEY.EDU
103
involved actors), and the configuration of external resources (e.g., their distribution and technical connections). These structural elements are relatively stable
and do not change overnight in a network, as stressed by the Markets-as-Networks approach.34 The dynamic interactions concern instead the processes that
go on daily in any business relationship in terms of the activities performed, the
communications among actors, and the concrete combinations and adaptations
of resources across organizational boundaries. These interactions are termed
“dynamic” because the underlying processes change more frequently and can
contribute to changes in the structural components (e.g., to the emergence of
new long-term goals or to the inclusion of new actors in the network structure).
The three structural components are: definition of the content of each
business relationship; formation of the network structure; and evaluation of the
matching of IKEA’s goals and resources with those of the network. The logic of
Component 1 is that IKEA strives to influence such relationship contents as
exchanged volumes, commitments, trust and learning depending on how each
relationship can contribute to achieving IKEA’s goals. However, Component 2
stresses that focusing on a single relationship would not be enough because
IKEA’s goals can be reached only by connecting several relationships into a
broader network structure, including the establishment of new relationships and
the assignment of specific roles to certain counterparts within a hierarchy of
relationships. Finally, Component 3 suggests that a firm needs to evaluate how
its own goals and resources can match those of its counterparts in the network:
unless this match can be achieved by means of dynamic interactions,
adjustments might be necessary in the other two structural components, such as
establishing new relationships or changing their contents.
The three structural components provide the basic network structure
within which dynamic interactions unfold continuously. At a general level, these
daily interaction processes entail combinations of resources by IKEA and other
actors in the attempt to match IKEA’s goals and resources with those in the network. At a specific level, there are two types of dynamic interactions: interacting
via inter-organizational routines for efficiency purposes; and interacting via joint
projects for development purposes. Whereas inter-organizational routines are,
for efficiency’s sake, rigid scripts executed repetitively, joint projects are development processes that can stretch over several years, thus becoming less controllable by IKEA and more uncertain.
IKEA,The Interacting Company:
Handling a Network of Relationships
IKEA’s relationships and network are pivotal in fostering development of
IKEA’s products and technologies, and in sustaining efficiency in its daily operations. IKEA is certainly not the only firm that relies on extensive network interactions for its strategy. In particular, other retailing firms such as Benetton and
Hennes & Mauritz apply a similar approach to complex webs of partners. However, IKEA does it in a special way. A major difference in comparison, for
104
VOL. 50, NO. 4
SUMMER 2008
CMR.BERKELEY.EDU
instance, with Wal-Mart,35 is that IKEA stretches its interactions as far upstream
as possible in the network, all the way to raw material suppliers. IKEA intensively cooperates with these suppliers in order to ensure the quality and the
environmental friendliness of inputs to its direct suppliers downstream. Another
important difference is the extent of IKEA’s cooperation with its partners: this
extends to complex and enduring development projects whereby IKEA’s products and technologies are co-developed with suppliers.36
Finally, IKEA’s interaction mechanisms strongly differ from both hierarchically steered networks, such as the Japanese Keiretsu,37 and looser arm’slength relations mostly based on purchase power, such as those of Wal-Mart.38
The main difference is that instead of solely exploiting the power of being a large
buyer, IKEA takes a long-term approach and strives to build lasting relationships
based on mutuality. This means that it also explicitly considers the interests of
suppliers, who would otherwise lose the motivation to interact with IKEA.
Moreover, IKEA does not strive to unilaterally control these relationships, but
relies on extensive delegation of tasks to its suppliers, and even accepts being
dependent on some of them. Thus, for IKEA, mutual trust and commitment are
more important interaction mechanisms than power.
IKEA’s Background and the Structure of Its Network
IKEA’s concern with providing low-price products characterizes both its
current strategy and its history. The introduction of flat packs in the 1950s
allowed important savings in transportation and production costs. In fact, IKEA’s
customers took over assembly activities, and suppliers only needed to deliver
un-assembled furniture components. Later on, selling costs could be contained
thanks to “showroom-warehouses”: retail stores were redesigned so as to combine a large exhibition area with an adjacent self-service warehouse. IKEA could
afford such low retail prices that its sales rocketed in the 1960s, signaling the
start of its expansion, with several stores in Sweden and abroad. However, a
strong reaction soon followed from Swedish furniture retailers, who tried to
strangle IKEA’s purchase sources by requiring that all Swedish producers stopped
supplying IKEA. IKEA’s countermove was to go looking for suppliers abroad.
Thus, the first agreements with Polish producers were signed in the 1960s, laying the groundwork for many business relationships that still exist today.
Long-lasting relationships with selected key suppliers are still the hallmark of IKEA’s purchasing and product development strategy. As IKEA is
directly involved only in conceiving, distributing, and selling its products, it
needs partners that can concretely develop and produce a total of 12,000 items
that meet its cost, quality, and design goals. Still, extensive knowledge of the
network, from raw materials to customer homes, is pivotal for IKEA to conceive
products that are not only “cool” and functional enough to sell, but that can also
be produced according to set cost and quality goals.
Figure 1 gave a simplified idea of the extension and complexity of IKEA’s
network, encompassing about 10,000 organizations, from sub-suppliers to IKEA
retail stores. Connecting this network geographically requires a very advanced
SUMMER 2008
CMR.BERKELEY.EDU
105
logistic system, including 20,000 transport corridors. However, what counts
more than these physical connections are the organizational connections that
IKEA needs in order to interface with this complex network. In fact, IKEA created specific organizational interfaces to handle relationships with the various
actors: not only the 1,300 direct suppliers and 500 carriers, but also hundreds of
the 10,000 sub-suppliers, especially large firms supplying core materials such as
packaging (e.g., SCA) or coatings (e.g., Akzo-Nobel).
The strategic role of the unit known as IKEA of Sweden has already been
mentioned. From IKEA’s hometown of Älmhult, its 700 employees perform
another fundamental role: interacting from a central position with key logistic
partners and suppliers on such strategic issues as long-term capacity planning and
major technical development projects. Conversely, at a local level, the main interfaces
with suppliers are IKEA’s 40 Trading Offices, employing 3,000 people worldwide.
Even if these local purchasing offices replicate the structure and competence of
the central unit, IKEA of Sweden (with purchase strategists, product category
specialists, technicians, order managers, and logisticians), their interactions with
suppliers mostly concern daily issues (e.g., orders and deliveries) and occasionally development or tendering for new product assignments. IKEA pays its Trading Offices a percentage on the purchases made from the suppliers they
“represent.” Finally, the daily logistic coordination with all suppliers and carriers
is handled by IKEA’s 26 Distribution Centers.
IKEA’s Approach to Business Relationships
IKEA’s approach to supplier relationships depends on the product
involved. Complex products, both in terms of construction (e.g., sofas) and of
production technology (e.g., the “Lack” table, featured below), are assigned to
suppliers with which extensive mutual trust, commitment, and knowledge have
been established through long-term relationships. These strong relationships
entail extensive joint investments in facilities. On the other hand, products
whose technical simplicity means they are easily interchangeable (e.g., rugs) are
usually purchased through shorter-term relations. A similar variation exists in
the relationships with logistic partners: out of over 500 such partners, IKEA has
developed close cooperation with only 50 (e.g., Maersk, Willy Betz, SJ Cargo,
and TNT), which between them account for 80% of IKEA’s transport volumes.
Still, the majority of IKEA’s purchases happen through deep and established
relationships. A common trait is IKEA’s attempt to avoid abusing its power position. The focus is instead on the mutual benefits accruing to both IKEA and its
suppliers. Cooperation with IKEA should ideally bring suppliers advantages such
as stable and long-term orders or technical development projects where IKEA
can “pay” for thousands of tests. In fact, IKEA is well aware that those actors
that no longer have advantages may end their cooperation (as did the coating
supplier Becker-Acroma in the episode reviewed in Appendix A).
Moreover, IKEA does not unilaterally control these relationships but
accepts that it may sometimes be strongly dependent on its suppliers, as in the
case of key logistic partners and those suppliers that daily refill IKEA’s stores
106
VOL. 50, NO. 4
SUMMER 2008
CMR.BERKELEY.EDU
through “Vendor-Managed Inventory.” IKEA does not even unilaterally control
such key sources of its innovativeness as product and technology development
projects. Here, IKEA delegates much responsibility to the most competent partners, either those who have long been in charge of manufacturing a certain
product, or those who have specific technical competences.
Different Types of Relationships in IKEA’s Network
The composition of IKEA’s network varies greatly in terms of the size of
the actors involved. There are many small suppliers (and especially sub-suppliers) that are highly dependent on IKEA, and which IKEA can more directly
influence with its powerful position. However, IKEA exerts its influence not
only by holding down prices, but also by inducing suppliers to upgrade their
technologies in ways that eventually benefit the suppliers themselves. This mutuality is in fact the hallmark of all of IKEA’s business relationships, even of those
where IKEA could exploit a power position due to the overdependence of a
supplier.
On the other hand, there are also larger counterparts with which IKEA
has a much more balanced power relation. These large actors are not easily
influenced by IKEA, and they comply with IKEA’s requests only if they gain
something from a specific cooperation. For instance, Akzo-Nobel, a sub-supplier
of coatings, is a 15,000 employee and €6 billion chemical group. IKEA is certainly an important customer for Akzo-Nobel, but not to the extent that Akzo
will blindly comply with any requests. Instead, Akzo-Nobel chooses to engage in
IKEA’s development efforts when this provides it with specific advantages, such
as learning a new technology. Similarly, the key logistic partner Maersk has
thousands of trucks and employees, alongside hundreds of vessels and local
offices: in this case, it is more IKEA’s transportation routines that need to fit into
the logistic network of this partner rather than vice versa. Relationships with
such large actors directly involve IKEA of Sweden for central negotiations; and
even if IKEA still remains a key account for most suppliers (covering at least 1%
of their sales), there is more balance in power and dependence.
IKEA’s relationships are also very heterogeneous from a geographic point
of view,39 because they are spread over the regions that provide specific
resources or location advantages, such as nearness to IKEA’s major markets
(Germany and Central Europe). Geographical location is one of the key factors
when selecting new suppliers, because it strongly affects costs, competences, and
delivery times. The resulting geographic pattern is as follows: Chinese suppliers
rank first, with nearly 20% of purchase volumes, mainly due to cost reasons;
Polish ones rank second, thanks to a good mix of low costs, technical competence, and nearness to Central Europe; and Swedish suppliers, despite high
costs, still rank third thanks to their advanced technical competence.
Geography is an important factor, but not the only factor, in supplier
selection and hence in constructing the structure of IKEA’s network. During the
selection process, IKEA of Sweden and the local Trading Offices that propose a
supplier also evaluate these other factors, in order of importance: total costs,
SUMMER 2008
CMR.BERKELEY.EDU
107
provided that IKEA’s quality and environment requirements are respected; current and planned production capacity; technical competence; and readiness to
make investments for and with IKEA. An additional rule is that suppliers should
not depend on IKEA for more than 50% of their turnover. The rationale here is
not only the avoidance of over-dependent suppliers that would be hurt too
much if IKEA’s order volumes should decrease, but also reliance on the learning
and development achieved by a supplier thanks to interactions with customers
other than IKEA.
Then, with some suppliers, IKEA develops long-lasting and complexcontent relationships, which entail large volumes and commitments. IKEA often
even purchases machinery for these suppliers and trains their personnel. An
even more restricted group of highly trusted suppliers is then invited to take part
in complex technical development projects with IKEA: these suppliers are those
with greater competences (e.g., in logistics or coating technologies) and those
willing to become more committed to IKEA due to a positive history of interactions or strong expected benefits. IKEA already understood in the 1960s that
using a long-term approach to purchasing actually favors IKEA itself, by allowing lower production costs and purchase prices and faster and improved development for its products. This approach produces even better results if IKEA
follows a philosophy of mutuality and also takes into account the interests of its
suppliers: this is reflected in IKEA’s attempt to balance production volumes
among several suppliers and in IKEA’s investment programs to upgrade a supplier’s competences (e.g., technical or administrative training).
For instance, IKEA applies a ladder model to IT and supplier logistics issues
(see Figure 2): increasing supplier responsibility in deliveries (from simple fulfillment of IKEA’s orders to “Vendor-Managed Inventory”) must correspond both to
increased IT integration with IKEA and to improved logistics capabilities. However, this model does not build only on improved routines and IT, but also on
the development of a stronger business relationship, entailing more trust and
commitment between IKEA and a supplier. This also implies better information
flow and the commitment of concrete resources, ranging from dedicated personnel to building new warehouses at a supplier’s site.
The deepening of a supplier relationship around IT and delivery issues
starts from consistent delivery performances, and proceeds along a supplier’s
improvements on these dimensions:
▪ increased experience of IKEA’s ordering routines or retail sales patterns;
▪ logisticians and order management teams dedicated to handling IKEA’s
orders;
▪ improved communication with IKEA’s units such as retail stores and
IKEA of Sweden;
▪ improved IT competences acquired from daily use of advanced IT tools
such as ERPs;
▪ the willingness to invest in expensive new IT solutions;
▪ a direct knowledge of the IT ordering systems located at IKEA; and
108
VOL. 50, NO. 4
SUMMER 2008
CMR.BERKELEY.EDU
FIGURE 2. IKEA’s Ladder Model for Supplier Interactions and IT/Logistics Capabilities
VMI
(Vendor-Managed
Inventory)
OPDC
(Order-Point
Distribution Center)
Capability Requirements:
(1% of suppliers qualify:
trustworthy and proactive)
Call-Off
• Same as OPDC, plus…
(almost 30% of supplies)
(entry step: all suppliers)
• delivery precision enough to
fulfil only 13 orders per year
within 4 weeks;
• simple IT connection via nononline system ECIS
(Electronic Commerce for
IKEA Suppliers);
• no need to be sole item
supplier in the DC area.
• increased delivery precision
to fulfil daily orders within 12
days;
• high production flexibility expanded warehousing
facilities;
• IKEA-dedicated order
managers team,
• online EDI-ERP connection
with IKEA’s order system
INOS;
• need to be sole item supplier
in the DC area.
• highest delivery precision;
• ability to manage IKEA stores’
inventories;
• forecast retail sales;
• set security stock levels;
• plan/organize transport;
• interact with all IKEA
distribution units and with
IKEA of Sweden for planned
sales campaigns
• deep knowledge of IKEA IT
systems for inventory,
ordering, and transport
• direct online access to all the
above IT systems.
Key:
Call-Off: IKEA emits orders every fourth week and suppliers must deliver within the next 4 weeks.
OPDC: IKEA emits orders daily and suppliers must deliver within 12 days.
VMI: Suppliers are in charge of deciding when and how much to deliver to IKEA.
▪ improved manufacturing flexibility and willingness to make physical
investments (in production or warehousing capacity) in order to cope
with fluctuating order volumes.
Suppliers who achieve the aforementioned capabilities (often thanks to
support and training provided by IKEA) and make the required commitments
can move up along the ladder depicted in Figure 2. However, IKEA is also costdriven and a highly demanding customer that puts pressure on its suppliers.
Therefore, if repeated efforts to improve a supplier competence and efficiency
do not produce good results, IKEA is ready to terminate that relationship. As
a result of IKEA’s efforts to develop (or terminate) supplier relationships, its
SUMMER 2008
CMR.BERKELEY.EDU
109
network is heterogeneous, a feature that IKEA views as a key source of
development.
What Goes on in IKEA’s Network? Two Interaction Processes
The rich structure of relationships and the widespread competences
accessed by IKEA through its network would be useless if IKEA did not have the
ability to combine these resources to achieve efficiency and foster development.
At a general level, this ability relies on IKEA’s organizational structure, on its
strong competence in such areas as product and technology development or
logistics, and on its long-term approach and network-oriented culture. At a
more concrete level, the combination of external resources and competences
relies on two managerial tools: detailed routines performed repetitively by IKEA
and its suppliers in such efficiency-driving processes as order management; and
ad hoc projects that tackle specific product and technical development issues and
often involve up to 20 firms. Inter-organizational routines and projects are
essential mechanisms to promote and handle the numerous interaction
processes unfolding in the network structure reviewed above.
Order Management Routines at IKEA and Its Partners
This process defines the replenishment needs of IKEA stores and communicates them to the supplier in charge of each specific product. Setting exact
order quantities is critical in order to avoid stock-outs in retail stores or extra
inventory costs. Timing is essential too, because suppliers are bound to given
lead-times and cannot react to a delayed order with immediate deliveries. IKEA
achieves efficiency in the order-management process through an advanced procedure that estimates its own replenishment needs and very structured routines
that transfer these needs to each supplier, stating such details as the exact
response time and modes required from them. Nothing is left to chance, and
IKEA considers many variables (such as stock levels, lead-times, and goods-intransit) to define when and how much should be ordered of a certain product.
However, matters are complicated by the fact that IKEA sells 12,000 products, with very different production approaches and lead-times. For instance, a
sofa will be finished after a customer order is placed, because the producer needs
to wait until the customer has chosen the fabric; whereas standard products
such as coffee tables are produced before orders, against sales forecasts. Therefore, IKEA’s orders of sofas and other customized products are steered by endcustomer orders, and so need longer lead-times; whereas orders of standard
products are triggered by automatically preset reorder points at IKEA stores and
Distribution Centers, and have shorter lead-times. To be able to fulfill an order, a
supplier needs to master the details of the ordering routine specific to each of
the products it delivers to IKEA. Manufacturers supplying IKEA with volumes
higher than Euro50 million yearly have large teams of IKEA-dedicated order
managers, fully engaged in handling IKEA’s orders.
To further complicate things, not all IKEA suppliers are equally able to
respect short lead-times and to fulfill orders coming in daily. Therefore, different
110
VOL. 50, NO. 4
SUMMER 2008
CMR.BERKELEY.EDU
suppliers interact with IKEA by means of different ordering routines, depending
on their ability and type of product. For instance, IKEA cannot implement with
all suppliers the ordering routine “OPDC” (Order-Point Distribution Center),
which involves issuing orders every day and requiring fulfillment within 12
days. Instead, many suppliers follow the “Call-Off” routine, which involves issuing orders every fourth week and requiring deliveries within the following 4
weeks. The OPDC modality is clearly more precise for IKEA’s receiving units and
reduces stock levels and goods-in-transit within IKEA. However, OPDC creates
stronger pressures on suppliers not only to learn a new ordering routine, but
also to become more flexible and speculative, or even to increase finished goods
stocks in order to react to shorter lead-times.
Still, OPDC is not IKEA’s most demanding ordering routine for a supplier.
The “VMI” routine (Vendor-Managed Inventory) grants suppliers access to
IKEA’s stock data and assigns them the responsibility to decide when and how
much to deliver. In this way, a supplier is empowered to exploit its knowledge of
IKEA’s ordering patterns. However, this knowledge, essential to forecasting
IKEA’s needs, only comes after having interacted daily with IKEA for some
years. Additionally, with VMI, close interaction and joint planning between a
supplier and IKEA stores become necessary to secure product coverage for special events such as store openings and sales promotions. All order management
modalities are highly interactive routines performed daily by IKEA and its suppliers according to rigid scripts, mostly decided unilaterally by IKEA. At the same
time, IKEA strives to introduce new ordering routines such as OPDC with key
suppliers, according to the aforementioned ladder model that requires direct
supplier involvement and large mutual investments.
The “Printed Veneer” Project for the “Lack” Table
An important type of development projects in IKEA’s network concerns
new technologies. In fact, IKEA searches constantly for technologies that can
reduce costs, improve quality, or allow new designs. Coating technology, for
instance, is pivotal in furniture manufacturing, because it strongly affects both
design and quality, but it is also costly and, if badly handled, can become hazardous to health. IKEA is therefore very concerned with coating technologies
and has promoted hundreds of projects to improve them at its suppliers. One of
these projects is the “printed veneer” project, which addressed the high cost of
the veneers used for IKEA’s best-selling “Lack” table, with more than 2.5 million
units sold yearly.
In 2000, IKEA of Sweden raised some concerns about the high cost of
veneers, which accounted for about 20% of the material costs for the veneered
versions of Lack. These concerns triggered a series of discussions and an evaluation of potential solutions at Swedwood Poland, the manufacturer of Lack tables.
Akzo-Nobel, one of their major coating suppliers, proposed to Swedwood a new
coating technology that would allow substituting real veneers with a printed
pattern. Thus, Swedwood and Akzo-Nobel initiated a large technology development project with the aim of printing veneer on wood. This project was particu-
SUMMER 2008
CMR.BERKELEY.EDU
111
FIGURE 3. The Network of Resources Combined in the “Printed Veneer” Project
IKEA
IKEA of Sweden
HDF
Swedwood
Coating Line
"Lack"
Coatings/Inks
Bürkle
Sorbini
Akzo-Nobel
Product
Business Unit
Equipment
Business
Relationship
larly important not only for cost reduction, but also for product design and aesthetics. Veneers are very visible on Lack tables and so the need to respect the
perceived-quality requirements made this project particularly complex. The aesthetic target was that consumers should not notice any major difference in the
printed pattern compared to real veneers. Akzo-Nobel took a leading role in
perfecting the technology and introducing it into Swedwood’s plants. Leading
the project meant identifying suppliers of the necessary equipment and supervising the many tests required to fine-tune the technical solution. Many technical
and organizational resources had to be combined during this project (see Figure
3, which shows the products, equipment, business units, and relationships that
were involved).
The involved business units were all related by long-term relationships. In
particular, Akzo had been supplying Swedwood for 20 years and closely interacted with IKEA of Sweden to negotiate the prices and conditions applicable to
IKEA’s direct suppliers. Finding a way to print veneer patterns on Lack’s surface,
which is made of HDF (high-density fiberboard), was not a technically easy task.
112
VOL. 50, NO. 4
SUMMER 2008
CMR.BERKELEY.EDU
To start with, Akzo had to develop new coatings and inks. Moreover, many technical resources fell outside Akzo’s competence: new coating lines, resembling
printing presses instead of traditional coaters, were necessary. Therefore, Bürkle
and Sorbini, two coating lines suppliers who had previously supplied Swedwood
and cooperated with Akzo, modified their equipment to suit the new process.
Following Akzo and Swedwood’s specifications, Bürkle and Sorbini adapted two
coating lines that were installed at Swedwood’s plants. However, the most
demanding part of the project was getting all the technical resources to work
together to satisfy IKEA’s requirements. This took over a year of tests, led by
Akzo at Swedwood’s plants. On several occasions, Akzo had to modify its coatings and inks to allow them to work together with HDF and the new coating
lines.
However, by the time the new “print-on-wood” technology was ready,
the problem of high-cost veneers had already been solved by a large supply of
inexpensive and high-quality veneers purchased by IKEA’s Trading Offices. Still,
the new technology was now available, and IKEA decided to apply it to other
products manufactured by Swedwood, namely, a series of shelves that were
officially launched in 2002. In initiating this project, IKEA had justified its high
costs with large cost savings for a specific large-volume product, the Lack table.
When the new technology became less relevant for these tables, IKEA could rely
on the fact that Swedwood manufactured several other IKEA products with
similar veneering problems.
At the same time, efforts continue now to diffuse the “print-on-wood”
technology both to other IKEA products and to other suppliers across IKEA’s
network. Moreover, in 2006 the aforementioned large supply of inexpensive
veneers dried out. Thus, printed veneer was finally applied to its original target,
Lack tables. IKEA and Swedwood are currently discussing the application of the
“print-on-wood” technology not only to substitute veneers, but also to print
directly on IKEA’s furniture any pattern developed by IKEA’s designers. Thus, a
technical development initiated for cost reduction has opened up another way to
sustain IKEA’s image as an innovative and cool furniture company.
Interaction processes such as ordering routines and the printed veneer
project would not be possible without two important premises: IKEA’s internal
structure and competences and its well developed interfaces for interacting with the
network. For instance, IKEA of Sweden employs 700 people specializing in furniture technologies, logistics, and so on. That unit includes 50 “purchase strategists” and 100 technical experts. Moving from IKEA’s center to its periphery, we
encounter 40 local Trading Offices employing 3,000 people who deal daily with
suppliers. Therefore, IKEA’s multifaceted organization, with its many internal
competences and external interfaces, is necessary for IKEA to pursue its network
strategy and interact with the actors in its network.
SUMMER 2008
CMR.BERKELEY.EDU
113
Discussion: Structures and Dynamic Interactions
in a Network Strategy
What sustains IKEA’s network strategy? In other words, what are the key
factors that enable IKEA to interact with its network in order to fulfill its efficiency and development goals? The framework introduced earlier contains both
structural components and dynamic interactions. The idea is that before interacting
with the network and using it dynamically, a network structure needs to
emerge. This is clearly a simplification, because network structure and dynamics
are tightly related. Indeed, a network structure can very well emerge more from
ongoing interactions (inter-organizational routines or ad hoc projects)40 than
from IKEA’s active choices to shape it. Still, as the purpose of this framework is
to simplify matters, structural and dynamic factors are discussed separately.
The Structural Components of a Network Strategy
IKEA strives to shape the structure of its network by selecting suppliers,
at least the 1,800 direct suppliers and logistic partners (see Figure 1). The structure and composition of this network are strategically important because they
create both opportunities and restrictions for the actions that IKEA can undertake.
Therefore, foresight is necessary over the goals to be reached, the resources
needed to accomplish them, and the actors to be mobilized to access external
resources. The actors and resources in IKEA’s network enable IKEA to achieve its
main strategic goal: “providing inexpensive good-quality products to as many
customers as possible.” Moreover, this network structure reflects more specific
efficiency goals, such as low sourcing costs or smooth product flows, and development goals, such as new products or new technologies.
Structural Component 1: Defining Relationship Contents
Because of the importance of specific counterparts for achieving its goals
IKEA needs to define the content of each relationship, in terms of exchange volumes, learning, trust, commitment, and duration. However, the content of one
relationship cannot be defined in isolation from other relationships because it is
related to the role played by that relationship in the overall network: for
instance, compared to an interchangeable sub-supplier, a first-tier unique supplier often receives larger volumes, sustained by the high trust and commitment
emerging from a long-term interaction. Not surprisingly, IKEA aims to
strengthen the latter type of relationships. But among first-tier suppliers there
are even more subtle differences based on the degree of IT integration with
IKEA and on the operational responsibilities assigned to each supplier. To
increasing supplier capabilities and commitment correspond increasing logistic
responsibilities, according to a ladder model (see Figure 2). This increase in supplier responsibility entails an increase in IKEA’s dependence on selected suppliers. Another dimension to categorize relationship content is the initiative and
leadership taken by a partner in development processes. In fact, partners such as
Akzo-Nobel, who are formally second-tier suppliers, often display strong initiative and take a leading role because of their advanced technical competences
114
VOL. 50, NO. 4
SUMMER 2008
CMR.BERKELEY.EDU
and their own network of contacts. This type of relationships entails important
exchanges of technical knowledge with IKEA and is oriented towards the long
term since many development processes are time-consuming.
All the above forms of relationship content are issues that IKEA can
rarely shape unilaterally. Moreover, IKEA’s ability to control and define a relationship’s content is even more limited for its indirect relationships, such as those
between its direct suppliers and their sub-suppliers. Therefore, IKEA must agree
to delegate, for instance, project leadership to a key partner that can select other
actors and define the contents of their own relationships (as Swedwood and
Akzo-Nobel do with their own partners). This delegation is extensive in development projects, where IKEA’s trusted key partners can partly affect the project
goals and have ample freedom to mobilize other actors that have the required
technical capabilities. IKEA’s delegation and a partner’s freedom in defining a
relationship’s content are instead more restricted when it comes to inter-organizational routines (e.g., order management), which for efficiency’s sake need to
be more controllable and uniform.
However, even within IKEA’s direct relationships (e.g., with Swedwood),
IKEA cannot unilaterally decide the relationship content. Content such as
exchanged volumes, technical learning, and trust emerge instead from a twoparty game that requires the capabilities, engagement, and commitment of
IKEA’s counterparts as well. In this game, IKEA is sometimes more powerful,
especially in relation to smaller actors, but in other cases IKEA is so technically
or volume dependent on certain counterparts that it must partly accept their
efforts to shape the relationship content: for instance, exchanged volumes and
technical investments are typically negotiated with key partners such as Swedwood.
Structural Component 2: Forming the Network Structure
Each relationship plays a well-defined role within a partly hierarchical
structure. IKEA’s network includes, for instance, first-tier as well as second-tier
suppliers, some of which play the role of “major” as opposed to “reserve” suppliers. Moreover, IKEA has direct relationships to certain actors (e.g., Swedwood)
and indirect relationships to others (e.g., Sorbini). IKEA strives to shape this
structure firstly by affecting the number of firms that are part of the network.
For instance, between 2002 and 2007, IKEA reduced the number of direct suppliers from 2,400 to 1,300.41 Therefore, the remaining direct suppliers were
given more responsibility. This implies not only larger volumes, but also that
each direct supplier now handles more complex relationships with an increased
number of sub-suppliers. Each relationship also has a specific function for the
entire network because it connects other relationships: for instance, the relationship IKEA-Akzo assigns a leading role to Akzo in setting the technical agenda of
its relationships with Sorbini and Bürkle (see Figure 3).
Another network structure feature essential for a global player is the geographical spread of relationships. Suppliers are selected depending on the placerelated features of their resources (e.g., labor costs, raw materials, and proximity
SUMMER 2008
CMR.BERKELEY.EDU
115
to end markets) and on their role in IKEA’s network (e.g., project-leading partners need to be located near other involved units). The geographical structure of
IKEA’s network is evolving towards a stronger emphasis on Chinese suppliers,
who increased their share of IKEA’s purchases from 14% to 18% between 2003
and 2006. The macro-reasons behind this trend are not only easier communication and lower transportation and production costs, but also the technical
advances made by Chinese suppliers, who now guarantee quality levels comparable with European ones. IKEA’s retail expansion in China further motivates
selecting local suppliers, for both political and transport reasons. However,
despite this geographic trend, suppliers from Sweden, Italy, and Poland still play
a major role when it comes to sophisticated technical developments that require
very specific competences, tight coordination, and co-location (see the Lack
example).
Although IKEA strives to shape the configuration of its network, its control over this structure is never complete: partners’ goals matter too. Furthermore, despite IKEA’s will, some actors take steps either to leave the network or
to change their position within it (see the case of Becker-Acroma in Appendix
A). IKEA’s control and overview of some parts of the network are also limited
because some partners include in a project wholly new actors that IKEA had not
considered. This happened, for instance, during the “printed veneer” project,
whereby Akzo-Nobel involved a new equipment supplier, an Italian cylinder
engraver, in IKEA’s network. In such cases, IKEA agrees to delegate to reliable
partners not only the performance of key tasks, but also the selection into IKEA’s
network of other relevant actors: this clearly affects the structuring of the network but is partly beyond IKEA’s control.
Structural Component 3: Evaluating Goal Matching with the Network
IKEA’s specific goals can be accomplished thanks to the resources of
other firms and accessed through the direct and indirect relationships that shape
a network structure (see Figures 1 and 3). Therefore, it is important to assess
the matching of the goals and resources of the firms involved. A structure that
enables matching includes both relationships that provide similar resources (e.g.,
IKEA’s parallel suppliers) and those that provide complementary resources (e.g.,
the relationships between IKEA’s suppliers and their own sub-suppliers).
However, it is clear that IKEA’s goals and those of its partners do not
always match each other. IKEA certainly seeks firms with which it can share
goals at a general level: for instance, IKEA and Akzo-Nobel share the goal of
developing coating technologies that improve furniture quality or costs. However, goals can sometimes be matched only after intense negotiations and discussions. Moreover, unexpected turns of events can change the initial goal
congruence: this happened for example when IKEA decided not to launch
the “printed on” Lack tables, which had been the main development goal of
Swedwood and Akzo-Nobel. However, at a general level, Akzo-Nobel’s goal of
developing new coating technologies could still be accomplished within its relationship to IKEA. In fact, IKEA is a perfect test customer that can subsidize
116
VOL. 50, NO. 4
SUMMER 2008
CMR.BERKELEY.EDU
Akzo’s own developments: how many furniture retailers could afford to sacrifice
thousands of real products just to perform technical tests?
IKEA’s goals are more easily matched with those of the network in routine interactions, such as order management routines, which are applied daily
without being questioned. However, goal convergence becomes more problematic in situations requiring development efforts and sacrifices from IKEA’s counterparts: for instance, rising conflicts over technical developments for IKEA’s
Billy bookshelf induced the coating supplier Becker-Acroma to withdraw from
an unwanted joint venture with IKEA (see Appendix A). Perfect goal congruence in the network cannot be achieved, simply because the network is not
composed of firms that IKEA can unilaterally control. Despite IKEA’s size and
power, these firms remain independent and hence are relatively free to set their
own goals. More precisely, the goals of IKEA and all firms in the network are
interdependent. This means that achieving one’s goal depends on the goals of
other firms. Håkansson and Ford stress that managing the network is impossible
or even undesirable.42 Instead, all firms need to manage in networks, which
requires carefully taking into account the goals and resources of all the firms
involved. Therefore, to enable “managing in the network,” a key component of
a network strategy is assessing the matching between one’s own goals and those
of the other firms involved.
The Dynamic Interactions of a Network Strategy
Next to evaluating and shaping the structure of the industrial network, it
is necessary to understand what happens within this structure. Thus, the analytical focus shifts to the interaction processes that unfold in the network and that
stand for its dynamics. These processes are particularly important because
achieving strategic goals depends on how the interaction processes unfold, which
now becomes more important than who performs them (i.e., the structural components). In strategic jargon, dynamic interactions are more concerned with
implementation than are structural components, but they are essential to
achieving the specific efficiency and development goals associated with a given
network structure.
While forming the network structure can rely on simply assessing the
matching between internal and external resources, implementing a network
strategy with specific goals requires combining dynamically and more profoundly
the resources in the network. These concrete resource combinations are created
through continuous interaction processes. For instance, through the order management routines, IKEA achieves efficiency when all firms involved combine
daily such resources as 12,000 products, countless components, transport means,
and their specific competences. Resource combinations are particularly dynamic
when the goal is to develop new technologies: during the “printed veneer” project, coating-related resources were combined and recombined in complex ways,
as witnessed by the many tests required to find out how new resources could
work together.
SUMMER 2008
CMR.BERKELEY.EDU
117
Inter-organizational routines and projects drive a network strategy because
they are the concrete mechanisms through which relationships are practically
mobilized and they provide the context in which dynamic resource combinations occur. However, these two types of interaction processes differ in their
functions and applicability to efficiency goals, for which routines are more appropriate, as opposed to development goals, for which projects are a better mechanism.
Dynamic Interactions 1: Interacting via Inter-Organizational Routines
Compared to development projects, inter-organizational routines are
more structured and rigid mechanisms because they clearly specify in advance the
nature and timing of the activities performed by IKEA and its partners. For these
reasons, inter-organizational routines are adequate for daily interactions that
aim to maintain efficiency. Examples of key efficiency goals for IKEA include
reduced lead-times and inventory costs as well as more precise deliveries. To
achieve such goals, order management routines, for instance, clearly specify the
time frames, each actor’s roles, the interaction patterns, and the resources
involved.
The continuous performance of inter-organizational routines also brings
about adaptations in the firms’ activities that can be better linked for mutual
efficiency gains, as stressed by Håkansson and Snehota.43 Moreover, according to
Zollo, Reuer, and Singh, inter-organizational routines improve the performance
of cooperative agreements.44 Still, compared to the open-ended nature and
ample delegation of joint projects, IKEA strives to control more unilaterally the
performance of critical routines such as order management, for the sake of certainty and homogeneity across its network. Therefore, these inter-organizational
routines need to closely follow the script that IKEA requires from its partners.
Dynamic Interactions 2: Interacting via Joint Projects
Joint projects are ad hoc and more flexible mechanisms than routines, and
are therefore employed for interaction processes that aim to foster product and
technology development.45 For instance, the “printed veneer” project had rather
broad goals (i.e., tackling the issue of costly veneers), timeframes, roles, and
interaction patterns, and many unplanned resources could be involved as the
project unfolded. In comparison to routines, projects entail less control and more
uncertainty for IKEA. For instance, the “printed veneer” project was highly
uncertain because it was rather open-ended and technically challenging and had
an open timeframe. Moreover, IKEA was not able to control the selection of new
participants in this project, as the technical leadership was assigned to Akzo.
Nevertheless, no matter how strongly IKEA can control a project, interorganizational projects are important drivers of development for at least three
reasons: they organize IKEA’s interactions with its network by allocating tasks
and responsibilities, they focus attention and efforts on a specific development
problem, and they select and mobilize key actors who can contribute relevant
knowledge to tackle the problem at hand. Similar advantages of projects as tools
118
VOL. 50, NO. 4
SUMMER 2008
CMR.BERKELEY.EDU
to foster innovation have been recognized (although in an intra-organizational
context) by Bowen, Clark, Holloway, and Wheelwright.46 Moreover, as stressed
by Brady and Davies,47 projects facilitate “explorative” learning, that is, they
enable learning by exploring new areas of knowledge. Such exploration is particularly important when IKEA does not know what technology needs to be
developed or who can develop it, as in the case of “printed veneer.” Then, the
flexibility of projects as coordination mechanisms comes in handy, together with
IKEA’s delegation to selected actors, who are in turn free to engage their own
trusted partners. Furthermore, if the new actors remain permanently engaged,
projects assume an even broader renewal potential by affecting the structure and
composition of IKEA’s network.
Granted the importance of the external network for IKEA, a natural question is: What does IKEA need to have itself in order to interact successfully with
this network? IKEA’s competence, organizational structure, and culture are important
prerequisites for IKEA to be able to initiate and take part in the interaction
processes that drive its network strategy. Those internal features strongly affect
how the focal firm interacts with specific partners, via inter-organizational routines and joint projects,48 and are fundamental for any strategic center to successfully interact with a web of partners.49 Firstly, IKEA possesses strong
competences in all areas necessary to interact with and support suppliers: IKEA
of Sweden employs about 100 experts in important areas for its product and
technology development, such as wooden materials, coatings, and surface treatments; while to support interactions with manufacturers and carriers concerning
distribution routines, IKEA of Sweden employs about 100 logistics and order
management experts. These technical and logistic competences are also reproduced on a local level within IKEA’s Trading Offices.
Secondly, IKEA has created both central and local organizational interfaces
to interact with suppliers. The central business unit, IKEA of Sweden, both
negotiates with major suppliers the worldwide prices and discusses long-term
investments and capacity plans. IKEA of Sweden also intervenes during supplier
selections and extensive development projects. Alongside these central interorganizational interfaces, IKEA also interacts locally with single suppliers
through its 40 Trading Offices. These business units are in charge, first and foremost, of daily issues such as ordering and logistic coordination; but they also
intervene in yearly capacity planning and supplier selection meetings, where
they represent their own local suppliers against other Trading Offices. Thirdly,
IKEA’s culture may be inspired by cost-containment, but there is a strongly
rooted understanding that this goal requires mutual long-term commitments with
key external partners. These values inspire IKEA’s criteria for supplier selection
(which stress a supplier’s readiness to make long-term investments for and with
IKEA) and management systems (which reward Trading Offices based on the
volumes that they, together with their local suppliers, supply to IKEA worldwide).
Such mechanisms strengthen the specific Trading Office-supplier relationship
and facilitate the emergence of mutual trust. Thus, IKEA can accept the neces-
SUMMER 2008
CMR.BERKELEY.EDU
119
sity of being dependent on certain highly trusted partners without the need to
unilaterally control the relationship.
However, despite IKEA’s positive experiences, its competences and organizational structure often need to change to fit specific partners and interaction
processes that unfold in the network. Changes in IKEA’s competences may be
necessary to exploit new opportunities and are often driven by joint projects: for
instance, IKEA learned new coating technologies and “VMI” by interacting with
its suppliers and logistic partners. A specific joint project may even lead IKEA to
create special business units (see the joint venture GIAB presented in Appendix
A). On a broader scale, IKEA currently deals with three major organizational
changes stimulated by developments in the network: opening new Trading
Offices in countries offering interesting sourcing opportunities; increasing the
responsibilities and competences of Trading Offices in Asia, as a reaction to the
increased volumes and know-how of Asian suppliers; and creating direct interfaces between retail stores and selected suppliers, in order to improve product
development and deliveries during sales promotions.
Conclusions and Implications from the IKEA Case
Industrial networks and business relationships play key roles for the strategy of IKEA and of most firms. Therefore, firms need a “network strategy,” that
is, they need to consider and use the external network in order to accomplish
their own goals. The framework presented in this article helps to systematize the
factors that can improve interaction with the external network. The structural
components and the dynamic interactions were elaborated on the basis of IKEA’s
network strategy; the key implications of IKEA’s experiences of network interactions can be summarized as follows:
▪ In order for a firm to implement a network strategy and achieve its own
goals, the focal firm’s resources must be combined with those of external
actors. This combination is made concrete through two types of interaction processes: inter-organizational routines—well suited for achieving
efficiency goals—and joint projects, aptly addressing development goals.
▪ These interaction processes are facilitated if the goals and resources of the
various parties match each other. However, perfect goal congruence never
exists in a network of independent firms. Therefore, evaluating the goal
and resource matching with specific counterparts can help in two major
ways: in choosing from the beginning partners with more attuned
resources and goals (e.g., those willing to make long-term commitments);
and in supporting the negotiations necessary to increase the goal congruence with specific partners.
▪ Even if a firm can improve its “network matching” by forming the structure of the network (e.g., by selecting suppliers), no absolute control can
be established over the network. Some counterparts may not accept the
relationship content expected by the focal firm, or may even choose to
interrupt their interactions and leave the network. Therefore, business
120
VOL. 50, NO. 4
SUMMER 2008
CMR.BERKELEY.EDU
relationships with key partners need to be carefully handled to establish
strong outposts in the network. In fact, the limits to controlling the network structure and processes suggest the need to delegate responsibilities
to trusted and competent external actors who are capable, in turn, of
engaging other relevant actors. This is what IKEA does, and it is especially
important in technical development projects that involve very widespread
competences.
▪ The external network cannot compensate for the gaping weaknesses of
unprepared firms: the strategic center cannot be “hollow.” In fact, forming the network by attracting counterparts and continuously interacting
with them requires that a firm is capable and prepared to “meet the network” in three main ways: by possessing extensive and specialized competences (see IKEA’s considerable logistic and technical competences); by
creating appropriate inter-organizational interfaces (see IKEA’s many specialized purchasing units and its product developers, who travel 200 days a
year to meet suppliers); and by promoting a network-oriented culture that
favors a long-term approach and the creation of mutual trust instead of
the abuse of power over partners.
▪ While being prepared to “meet the network” also means being flexible
enough to change internal competences and inter-organizational interfaces to better interact with a changing network, a network-oriented culture is instead more of a stable pillar. In fact, IKEA’s entire network
strategy strongly relies on mutual trust and commitment to selected partners as substitutes for absolute control: its is only if you trust a committed
partner that you can accept dependence on it and delegate essential tasks
such as technical developments. However, trust and commitment do not
appear overnight in a business relationship, but require a long-term
approach. This is why IKEA takes so seriously its supplier development
programs to improve selected partners’ IT and logistics capabilities (see
Figure 2) or its joint technical development projects. These initiatives take
time, but they eventually pay off when these suppliers can be more
trusted for IKEA’s daily replenishments or for technology issues.
APPENDIX A
From Saving the “Billy” Bookshelf to Conflicts with Suppliers
The “Billy” bookshelf is one of IKEA’s best sellers; it sells over one million
units yearly. In the 1980s, Billy was one of IKEA’s most representative products,
strongly associated with its design style and with its image of a family-friendly
company. However, this situation came under threat in 1991, when traces of
formaldehyde were found in Billy packs in Germany. In IKEA’s largest market,
the media started accusing IKEA of “poisoning” customers: a commercial catastrophe and a complete loss of reputation were near, and IKEA had to solve the
problem rapidly in order to restore its goodwill. IKEA of Sweden began a close
SUMMER 2008
CMR.BERKELEY.EDU
121
scrutiny of the whole production process at Sydpoolen, a group of Swedish suppliers that manufactured Billy. The coating process soon became the key suspect,
since hazardous chemicals were used in this production phase, and there was a
risk that these chemicals might remain on the surface of the furniture.
In a state of emergency, IKEA of Sweden mobilized Becker-Acroma, one
of its major coating suppliers. In consultation with Sydpoolen’s technicians, they
were able to relate the formaldehyde emissions to two causes: the use in Sydpoolen’s factories of acid-curing coatings known to emit this gas; and too short a
time between curing and packaging, which did not allow all the formaldehyde to
dissipate before packing took place. Therefore, the first action taken was prohibiting the use of acid-curing coatings at all IKEA’s suppliers. Meanwhile,
Becker and other coating producers started supplying an older technology
known as water-borne lacquers. IKEA, however, saw this solution as only temporary, and started searching for a robust solution. Becker suggested the use of
UV-curing, in which ultra-violet lamps are used to attach coatings to wood panels.
After a few months, IKEA also founded GIAB, a new firm with the specific purpose of making this UV technology more economically viable and introducing it to all IKEA’s suppliers. At the same time, IKEA made it clear to all the
firms involved with this technology that they were required to take shares in
GIAB, which was then transformed into a joint venture. Among these actors
were Becker-Acroma, Sydpoolen, and the sub-supplier of wood panels. GIAB
was equipped with a UV-based coating line that functioned as a full-scale testing
facility around which the above actors gathered to perform tests. The goal was to
find out how coatings needed to be modified and which parameters should be
reproduced in the coating lines of other IKEA suppliers. These resources are
shown in Figure A1.
IKEA’s goal with the joint venture GIAB was then extended to making it a
powerhouse for the development of furniture coating technologies in general.
However, even though cooperation was achieved on the factory floor, the project was ridden with conflicts. Becker-Acroma felt that the coating suppliers had
been used by IKEA as scapegoats for the Billy “scandal,” and that they were now
obliged to do most of the work of finding a solution. Becker-Acroma also saw
the establishment of GIAB as an accusation that these suppliers were incapable
of developing new solutions by themselves. Despite these hard feelings, many
tests were performed, which eventually led to finding the right combination of
UV coatings, particle boards, and coating lines. During the 1990s, this technology was rolled out across all IKEA suppliers, which today use UV curing almost
exclusively. However, the forced cooperation eventually became unacceptable,
especially to larger actors such as Becker-Acroma, who more or less stopped
contributing to GIAB’s development. Therefore, at the end of the 1990s, despite
IKEA’s plans to turn GIAB into a consultant with its own customers outside the
IKEA universe, GIAB lost momentum, firstly as it had fulfilled its task, and secondly as it had never been accepted by coating producers.
122
VOL. 50, NO. 4
SUMMER 2008
CMR.BERKELEY.EDU
FIGURE A1. The Network of Resources Combined during Billy’s Coating Project
IKEA
Wood Panel Producer
Wood Panel
Coating Lines
GIAB
Sydpoolen
Test Facility
Coatings
“Billy”
Becker-Acroma
Notes
1. Next to 70 in-depth interviews, the empirical material was collected via a dozen visits and
site observations made at IKEA and its suppliers’ facilities in Sweden, Poland, and Italy.
These sources were completed by first-hand written sources such as documents, catalogues,
brochures, and company presentations, and by secondary sources such as newspaper clips
and other publications related to IKEA.
2. For examples in the software sector, see B. Iyer, C.H. Lee, and N. Venkatraman, “Managing
in a ‘Small World Ecosystem’: Lessons from the Software Sector,” California Management
Review, 48/3 (Spring 2006): 28-47. For examples in the biotech sector, see W. W. Powell,
K.W. Koput, and L. Smith-Doerr, “Interorganizational Collaboration and the Locus of Innovation: Networks of Learning in Biotechnology,” Administrative Science Quarterly, 41/1 (March
1996): 116-145; W.W. Powell, “Learning From Collaboration: Knowledge and Networks in
the Biotechnology and Pharmaceutical Industries,” California Management Review, 40/3
(Spring 1998): 228-240.
3. See the Sassuolo area in Italy as featured in M. Russo, “Technical Change and the Industrial
District: The Role of Interfirm Relations in the Growth and Transformation of Ceramic Tile
Production in Italy,” Research Policy, 14/6 (December 1985): 329-343; M. Russo, “Complementary Innovations and Generative Relationships: An Ethnographic Study,” Economics of
Innovation & New Technology, 9/6 (December 2000): pp. 517-557.
4. See the study on the garment industry in New York City by B. Uzzi, “Social Structure and
Competition in Interfirm Networks: The Paradox of Embeddedness,” Administrative Science
Quarterly, 42/1 (March 1997): 35-67.
5. See M.H. Lazerson and G. Lorenzoni, “The Firms that Feed Industrial Districts: A Return to
the Italian Source,” Industrial and Corporate Change, 8/2 (1999): 235-265.
6. See B.R. Koka and J.E. Prescott, “Strategic Alliances as Social Capital: A Multidimensional
View,” Strategic Management Journal, 23/9 (September 2002): 795-806; E.H.M. Moors and P.J.
SUMMER 2008
CMR.BERKELEY.EDU
123
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
124
Vergragt, “Technology Choices for Sustainable Industrial Production: Transitions in Metal
Making,” International Journal of Innovation Management, 6/3 (September 2002): 277-299.
T. Wedin, Networks and Demand: The Use of Electricity in an Industrial Process (Uppsala, Sweden:
Department of Business Studies, Uppsala University, 2001).
See I. Stuart, P. Deckert, D. McCutcheon, and R. Kunst, “A Leveraged Learning Network,”
Sloan Management Review, 39/4 (Summer 1998): 81-93; C. Carr and J. Ng “Total Cost Control:
Nissan and Its U.K. Supplier Partnerships,” Management Accounting Research, 6/4 (December
1995): 347-365.
The list is presented on page 148 of G. Lorenzoni and C. Baden-Fuller, “Creating a Strategic
Center to Manage a Web of Partners,” California Management Review, 37/3 (Spring 1995):
146-163.
See W.W. Powell, “Hybrid Organizational Arrangements: New Form or Transitional Development?” California Management Review, 30/1 (Fall 1987): 67-87.
Networks have probably always been there, since the early days of the modern corporations,
even if periodic increases in vertical integration tend to hide inter-firm networks. Even the
Fordistic model emerged in the early 1900s from a highly networked structure in the car
industry. See pages 365-368 in R.N. Langlois and P.L. Robertson, “Explaining Vertical Integration: Lessons from the American Automobile Industry,” Journal of Economic History, 49/2
(June 1989): 361-75.
These types of adversarial postures were for instance typical in the car industry until the
early 1980s, as discussed by L.-E. Gadde and H. Håkansson, Professional Purchasing (London:
Routledge, 1993).
Dominating paradigms within strategic management have been for decades the microeconomics-inspired Industrial Organization [see J.S. Bain, “The Theory of Monopolistic Competition After Thirty Years: The Impact on Industrial Organization,” American Economic Review,
54/3 (May 1964): 28-32; M.E. Porter, Competitive Strategy: Techniques for Analyzing Industries
and Competitors (New York, NY: Free Press, 1980)] and Transaction Cost Economics [see O.E.
Williamson, “The Economics of Organizing: The Transaction Costs Approach,” American
Journal of Sociology, 87/3 (November 1981): 548-577; “Strategizing, Economizing, and Economic Organization,” Strategic Management Journal, 12/8 (Winter 1991): 75-94]. Whereas
Industrial Organization could not consider networks because it used as key unit of analysis
whole sectors, Transaction Costs Economics views inter-firm relationships and networks as
exceptions in economic organizing.
G.B. Richardson, “The Organization of Industry,” Economic Journal, 82/327 (September
1972): 883-896.
M.S. Granovetter, “Economic Action and Social Structure: The Problem of Embeddedness,”
American Journal of Sociology, 91/3 (November 1985): 481-510.
W.W. Powell, “Neither Market Nor Hierarchy: Network Forms of Organization,” in B.M.
Staw and L.L. Cummings, eds., Research in Organizational Behavior, Vol. 12 (Greenwich CT:
JAI Press, 1990), pp. 295-336.
R. Swedberg, “Markets as Social Structures,” in N.J. Smelser and R. Swedberg, eds., The
Handbook of Economic Sociology (Princeton, NJ: Princeton University Press, 1994), pp. 255282.
Even scholars studying strategic alliances are dissatisfied with the research bias towards
applying atomistic views to interactive phenomena such as alliances. For instance, Ranjay
Gulati called for a paradigm shift, from an atomistic to a social network-embedded view of
alliances. See page 295 in R. Gulati, “Alliances and Networks,” Strategic Management Journal,
19/4 (April 1998): 293-317.
This view is presented for instance by H. Håkansson, “Organization Networks,” in A. Sorge
and M. Warner, eds., The IEBM Handbook of Organizational Behaviour (London: Thomson
Business Press, 1997), pp. 232-239; M.J. Piore, “Fragments of a Cognitive Theory of Technological Change and Organizational Structure,” in N. Nohria and R.G. Eccles, eds., Networks
and Organizations: Structure, Form and Action (Boston, MA: Harvard Business School Press,
1992), pp. 430-444.
Markets and hierarchies, the norms in many established theories (e.g., marketing, strategy),
would then become just extreme cases whereby the informal network interactions are constrained either inside a strict hierarchy (based on command and bureaucratization) or by a
perfect market (composed of autonomous firms that coordinate solely via price signals).
See H. Håkansson, “Product Development in Networks,” in H. Håkansson, ed., Industrial
Technological Development—A Network Approach (London: Croom Helm, 1987), pp. 84-115.
VOL. 50, NO. 4
SUMMER 2008
CMR.BERKELEY.EDU
22. See Wedin, op. cit.
23. See E. Baraldi, When Information Technology Faces Resource Interaction: Using IT Tools to Handle
Products at IKEA and Edsbyn (Uppsala, Sweden: Department of Business Studies, Uppsala
University, 2003).
24. A comprehensive literature overview over the IMP tradition is presented on pages 25-30 of
H. Håkansson and A. Waluzsewski, Managing Technological Development: IKEA, the Environment
and Technology (London: Routledge, 2002).
25. These studies included both a large-scale questionnaire studying hundreds of business relationships in Europe in the late 1970s and numerous extensive, deep case studies. The latter
research approach still characterizes current IMP research: since the 1970s, hundreds of case
studies have traced the industrial network around a focal firm or a technical solution by
means of snowballing data collection techniques reaching up to 100 interviews.
26. See G.C. Homans, Social Behavior: Its Elementary Forms (New York, NY: Harcourt, Brace &
World, 1961); S. Macaulay, “Non-Contractual Relations in Business: A Preliminary Study,”
American Sociological Review, 28/1 (February 1963): 55-67; P.M. Blau, “The Hierarchy of
Authority in Organizations,” American Journal of Sociology, 73/4 (January 1968): 453-467.
27. See, for instance, the model on page 24 of H. Håkansson, ed., International Marketing and
Purchasing of Industrial Goods: An Interactive Approach (Chichester: Wiley, 1982).
28. An example of these network-level models is the so-called “ARA” model, which decomposes
the substance of industrial networks into three layers: inter-firm Activity patterns, Resource
constellations, and webs of Actors. This Activity-Resource-Actor model is elaborated in
H. Håkansson and I. Snehota, eds., Developing Relationships in Business Networks (London:
Routledge, 1995).
29. See J.C. Anderson, H. Håkansson and J. Johanson, “Dyadic Business Relationships Within
a Business Network Context,” Journal of Marketing, 58/4 (October 1994): 1-15; L. Hallén,
J. Johanson, and N. Seyed-Mohamed, “Interfirm Adaptation in Business Relationships,”
Journal of Marketing, 55/2 (April 1991): 29-37.
30. See U. Andersson and M. Forsgren, “Subsidiary Embeddedness and Control in the Multinational Corporation,” International Business Review, 5/5 (October 1996): 487-508; and U.
Andersson, M. Forsgren and U. Holm, “The Strategic Impact of External Networks: Subsidiary Performance and Competence Development in the Multinational Corporation,”
Strategic Management Journal, 23/11 (November 2002): 979-996.
31. For a Markets-as-Networks view of business strategy, see H. Håkansson and I. Snehota, “No
Business is an Island: The Network Concept of Business Strategy,” Scandinavian Journal of
Management, 5/3 (1989): 187-200.
32. The concept of “strategic center” is used by Lorenzoni and Baden-Fuller, op. cit.
33. There is, in fact, also a “dark side” of networks, stressed for instance by H. Håkansson and I.
Snehota, “The Burden of Relationships or Who’s Next?” in D. Ford, ed., Understanding Business Markets (London: Thompson Learning, 2002), pp. 88-94. Networks do not only produce
positive effects for development and efficiency, but they are also ridden with conservative
forces that embed (in the sense of “constraining”) the firm. Too close business relationships
can be harmful for novelty, a problem defined as “over-embeddness” by Uzzi [op. cit., pp.
60-63]. However, the dark side of networks derives not only from social aspects: networks
can in fact generate technical lock-ins and economic overdependence for the involved firms.
34. In fact, business relationships require time to be established, but then display a strong continuity and can stretch over decades (as is the case for several of IKEA’s relationships). This
continuity includes institutionalization and dependencies that make relationships hard to
change or terminate. These combined effects create stability in a network, where the
involved actors, their goals, and their resources change only slowly. For the structural characteristics of business relationships and networks, see H. Håkansson and I. Snehota, eds.,
Developing Relationships in Business Networks (London: Routledge, 1995).
35. For a description of Wal-Mart’s supplier relationships, see M. Petrovic and G.G. Hamilton,
“Making Global Markets: Wal-Mart and Its Suppliers,” in N. Lichtenstein, ed., Wal-Mart: The
Face of Twenty-First-Century Capitalism (New York, NY: The New Press, 2006), pp. 122-138.
36. IKEA’s approach can be contrasted with Charles Fishman’s rendering of Wal-Mart: “WalMart has the power to squeeze profit-killing concessions from suppliers...Wal-Mart price
pressure can leave so little profit that there is little left for innovation.” The quote is found in
C. Fishman, The Wal-Mart Effect (New York, NY: Penguin, 2006), p. 89.
37. For the basic features of Keiretsu, see K. Miyashita and D. Russell, Keiretsu: Inside the Hidden
Japanese Conglomerates (New York, NY: McGraw-Hill, 1994); H. Kim, R.E. Hoskisson and
SUMMER 2008
CMR.BERKELEY.EDU
125
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
126
W.P. Wan, “Power Dependence, Diversification Strategy, and Performance in Keiretsu Member Firms,” Strategic Management Journal, 25/7 (July 2004): 613-636.
See Fishman, op. cit.; C. Fishman, “The Wal-Mart Effect and a Decent Society: Who Knew
Shopping Was So Important?” Academy of Management Perspectives, 20/3 (August 2006): 6-25,
at pp. 13-14.
For a comprehensive analysis of how IKEA utilizes geographical factors across several places,
see E. Baraldi, “The Places of IKEA: Using Space in Handling Resource Networks,” in
E. Baraldi, H. Fors, and A. Houltz, eds., Taking Place: The Spatial Contexts of Science, Technology,
and Business (Sagamore Beach, MA: Science History Publications, 2006), pp. 297-320.
Development projects, regarded here as dynamic interactions, entail also strong structuring
elements: even in informal projects, a specific organization—”a network in the network”—
is typically built. This can alter the content of existing relationships between IKEA and its
partners, as it partly did in the Billy episode among IKEA and Becker-Acroma (see Appendix
A). Moreover, IKEA’s technical projects are often an occasion for new actors to be included
in the structure of IKEA’s network: the “printed veneer” project, for instance, involved for
the first time an Italian firm that is now a stable supplier of engraved cylinders.
Meeting with Anders Brorström, Deputy Procurement Director, IKEA Russia, Moscow,
February 28, 2007.
See H. Håkansson and D. Ford, “How Should Companies Interact in Business Networks?”
Journal of Business Research, 55/2 (February 2002): 133-139. On page 138, the authors stress
that achieving total control over a network, however unlikely it is, is undesirable because it
would transfer all burden and source of innovation and wisdom to the controlling company
itself.
On the concept of “activity links,” one of the key dimensions in the IMP network “ARAmodel,” see H. Håkansson and I. Snehota, eds., Developing Relationships in Business Networks
(London: Routledge, 1995), pp. 28-30.
See M. Zollo, J.J. Reuer, and H. Singh, “Interorganizational Routines and Performance in
Strategic Alliances,” Organization Science, 13/6 (November/December 2002): 701-713.
These advantages of both intra- and inter-organizational projects as coordination mechanisms are recognized, for instance, by J. Sydow, L. Lindkvist, and R. DeFilippi, “ProjectBased Organizations, Embeddedness, and Repositories of Knowledge: Editorial,” Organization
Studies, 25/9 (November 2004): 1475-1489.
H.K. Bowen, K.B. Clark, C.A. Holloway, and S.C. Wheelwright, “Development Projects: The
Engine of Renewal,” Harvard Business Review, 72/5 (September/October 1994): 110-120.
See T. Brady and A. Davies, “Building Project Capabilities: From Explorative to Exploitative
Learning,” Organization Studies, 25/9 (November 2004): 1601-1621, at 1605-1607.
A routine such as OPDC strongly depends on IKEA’s competences and organization and is
impossible to perform without adequate external interfaces. The effect of IKEA’s internal
structure is present, although less evident, on development projects, whose success certainly
depends on IKEA’s competence and organizational structure, but where much responsibility
and control is typically delegated to actors farther away from IKEA.
See Lorenzoni and Baden-Fuller, op. cit.
VOL. 50, NO. 4
SUMMER 2008
CMR.BERKELEY.EDU
MANAGEMENT SCIENCE
informs
Vol. 56, No. 3, March 2010, pp. 468–484
issn 0025-1909 eissn 1526-5501 10 5603 0468
®
doi 10.1287/mnsc.1090.1117
© 2010 INFORMS
The Impact of Misalignment of Organizational
Structure and Product Architecture on Quality in
Complex Product Development
Bilal Gokpinar
Department of Management Science and Innovation, University College London,
London WC1E 6BT, United Kingdom, [email protected]
Wallace J. Hopp
Stephen M. Ross School of Business, University of Michigan, Ann Arbor, Michigan 48109, [email protected]
Seyed M. R. Iravani
Department of Industrial Engineering and Management Sciences, Northwestern University,
Evanston, Illinois 60208, [email protected]
P
roduct architecture and organizational communication play significant roles in complex product development efforts. By using networks to characterize both product structure and communication patterns, we
examine the impact of mismatches between these on new product development (NPD) performance. Specifically, we study the vehicle development process of a major auto company and use vehicle quality (warranty
repairs) as our NPD performance metric. Our empirical results indicate that centrality in a product architecture
network is related to quality according to an inverted-U relationship, which suggests that vehicle subsystems
of intermediate complexity exhibit abnormally high levels of quality problems. To identify specific subsystems
in danger of excessive quality problems, we characterize mismatches between product architecture and organizational structure by defining a new metric, called coordination deficit, and show that it is positively associated
with quality problems. These results deepen our understanding of the impact of organizational structure and
product architecture on the NPD process and provide tools with which managers can diagnose and improve
their NPD systems.
Key words: new product development; product architecture; organizational structure; complex networks
History: Received June 23, 2008; accepted August 30, 2009, by Christoph Loch, R&D and product
development. Published online in Articles in Advance January 12, 2010.
1.
Introduction and Literature Review
that make up the product, and (2) how to ensure that
people communicate/collaborate effectively in the
performance of design tasks. As evidence that these
problems are universal and difficult, a recent joint
study by BusinessWeek and the Boston Consulting
Group, reported that 1,000 senior managers around
the globe cited a lack of coordination as the secondbiggest barrier to innovation (McGregor 2006).
From an operations management standpoint, we
can view the new product development (NPD) process as a network of engineers designing a network
of parts. Consequently, in this paper, we study the
problem of coordinating parts and people in an NPD
process by means of network analysis. As such, this
paper is part of a growing literature that makes use
of networks to represent both product architecture
(Krishnan and Ulrich 2001, Henderson and Clark
1990, Ulrich 1995) and organizational structure (Clark
and Fujimoto 1991, Brown and Eisenhardt 1995). In
the work closest to our own, Sosa et al. (2004) adopted
Product innovation is central to business creation and
growth. Firms that are able to bring a steady stream
of timely and well-executed products to market are
likely to enjoy long-term financial success. Designing
products and bringing them to market, however, is
not a straightforward process. Product development
efforts usually involve many design iterations. The
ability of the organization to manage these inevitable
design changes has a major impact on product quality
and firm competitiveness.
The development process is particularly challenging for complex products such as automobiles or
airplanes, which involve thousands of engineers
spending years designing, testing, and integrating
hundreds of thousands of parts. Consequently, a key
challenge in these product development processes
is matching the organization to the product being
developed. This involves two fundamental problems:
(1) how to assign people to the parts and subsystems
468
Gokpinar, Hopp, and Iravani: Impact of Misalignment of Organizational Structure and Product Architecture
Management Science 56(3), pp. 468–484, © 2010 INFORMS
a combined perspective in a study of the alignment
of design interfaces and communication patterns. In
a subsequent paper, they identified factors that make
some teams better than others at aligning their crossteam interactions with design interfaces (Sosa et al.
2007). Although the insights from these studies are
interesting and potentially useful in improving NPD
processes, they are premised on a basic assumption,
namely that misalignment of the design organization
and the product architecture is detrimental to performance. But, because none of these studies actually
measured performance, they could not corroborate
this assumption.
In this paper, we build on this stream of research
by (a) defining a new metric, called coordination deficit,
which quantifies mismatches between product architecture and organizational structure, and (b) empirically investigating the effect of coordination deficit on
product quality.
In a broader sense, our work builds upon and
integrates two streams of research: (i) operations of
complex product development and (ii) social network
analysis of organizational performance. Although
these areas are very broad and have been studied
from a variety of perspectives, they have been studied
together under two major research headings: knowledge networks (see Nonaka and Takeuchi 1995, Contractor and Monge 2002) and modularity (see Ulrich
1995, Baldwin and Clark 2000).
Researchers have investigated various aspects of
knowledge networks in the context of product development and have provided critical insights into why
some business units are able to make effective use of
knowledge from other parts of the company, while
other units find knowledge to be a barrier to innovation (Hansen 2002, Carlile 2002). In this paper,
we construct a very specific knowledge network
that characterizes collaboration and communication
between design engineers and identify structural features of this network that are correlated with quality
problems in the final product.
Modularity refers to methods for reducing the number of interactions and interfaces among parts and
components in product design (Ulrich 1995, Baldwin
and Clark 2000). Organizational implications of modularity, as well as the organizational factors that support the use of modularity, have been studied by
several researchers (see Sanchez and Mahoney 1996,
Schilling 2002, Ethiraj and Levinthal 2004, Fleming
and Sorenson 2004). Unlike other network studies
of modularity, which represent interfaces as either
present or not present, we make use of engineering
data to characterize the strength of interfaces between
components. This gives us a more detailed representation of the product architecture, which we compare
469
to the organizational structure to quantitatively measure the degree of misalignment.
This paper also contributes to the literature on the
use of social network tools in empirical studies of
organizations (see, e.g., Wasserman and Faust 1994).
A distinctive feature of our study is that we make use
of archival data, rather than surveys, to construct a
social network. Because such data is readily available
in NPD environments, this approach may ultimately
make network analysis more practical as a management tool.
Finally, from a practical perspective, our work can
help managers to systematically identify and quantify potential problem areas that can be addressed to
improve the quality of the resulting products. Our
metric of organizational misalignment (i.e., coordination deficit) can be computed using standard data
from an engineering change order system. As such, it
provides a way to highlight opportunities for improving coordination among design engineers without collecting additional data. This should be particularly
valuable in environments where evolution of product
architectures changes coordination needs over time
and makes static organizational policies ineffective.
The remainder of this paper is organized as follows. In §2, we provide the theoretical background
and frame our hypotheses. In §3, we present a detailed
description of the data and the system in which we test
the hypotheses. In §4, we describe the model development, and in §5, we present the analysis and results.
We discuss our results in §6 and conclude in §7.
2.
Theory and Hypotheses
Ulrich (1995, p. 420) defined product architecture
as “(i) the arrangement of functional elements, (ii) the
mapping from functional elements to physical components, and (iii) the specification of the interfaces
among interacting physical components.” All three of
these dimensions may influence ultimate product performance at either the local (component) level (Ulrich
1995, Baldwin and Clark 2000, Mihm et al. 2003) or
the global (product) level (Clark and Fujimoto 1990).
An intermediate level between the component and
product levels is the subsystem level, which is widely
used by firms to describe product architectures for
management purposes. From a network perspective,
product architecture can be characterized by representing subsystems as nodes and interfaces between
subsystems as links. Network metrics can then be
used to describe the nature and position of product subsystems. For example, a subsystem with many
(physical and functional) interfaces will have high
centrality in the product architecture network.
Given this interpretation, the centrality of a subsystem can serve as a proxy for complexity, because
470
more interfaces imply more design issues. So, if
management were to devote equal resources and
attention to all subsystems, we would expect highly
central subsystems to exhibit worse performance
(e.g., more quality problems). But management does
not usually do this. Previous research on product
development has shown that architectural/technical
interdependence is associated with organizational
communication (Henderson and Clark 1990, Brown
and Eisenhardt 1995, Adler et al. 1995). Highly central
subsystems, which are heavily connected to other subsystems, logically receive intense attention from the
organization, which offsets the potential quality problems resulting from their complexity.
If management were to perfectly correlate resources
and attention to the complexity of each subsystem, we
would not expect to see any correlation at all between
centrality and performance (e.g., quality). But we do
not think this is realistic either. Complexity is not
directly observable. Therefore, mismatches between
organizational attention and subsystem interconnectivity, of the type observed by Sosa et al. (2004), are
likely to occur. We conjecture that they are most likely
to occur for subsystems of intermediate centrality. The
reason is that highly central subsystems are obviously
highly complex, and hence receive substantial organizational attention. Indeed, they may receive even
more attention than they need because they present
such manifest design challenges. At the other end
of the scale, low centrality subsystems, which have
few interfaces, require relatively little coordination
effort and so are unlikely to be underattended. Even
the minimal amount of organizational coordination
built into standard design practices is likely to be
enough for these subsystems. But intermediate centrality subsystems are neither complex enough to be
obvious nor simple enough to be easy. These are the
subsystems where a delicate matching of resources
and attention to the design complexity is most difficult. Therefore, this is where we expect to find the
most mismatches, as we conjecture in the following
hypothesis:
Hypothesis 1. The centrality of a product subsystem
in the product architecture has an inverted-U association
with the quality problems observed in that subsystem.
The earlier discussion provides a very general
sense of where a lack of organizational coordination
might lead to excessive quality problems. But because
centrality only characterizes subsystems in a coarse
manner, it would not provide any explicit managerial guidance on which subsystems are most prone
to quality problems. Henderson and Clark (1990)
established a relationship between product architecture and design organization, and pointed out the
importance of matching team interfaces to technical
interfaces. Sosa et al. (2004) observed that product
development teams tend to ignore certain types of
technical interfaces. In a more recent study, Sosa et al.
(2007) presented anecdotal evidence from industry
where the shortage of organizational attention to technical interfaces resulted in poor performance. Based
on these findings and the intuition of NPD managers in our client firm, we conjecture the following
hypothesis:
Hypothesis 2. Mismatches between product architecture and organizational coordination in subsystems are
positively associated with the quality problems observed in
these subsystems.
We note, however, that neither Henderson and
Clark (1990) nor Sosa et al. (2004) actually measured
NPD process performance to test the above hypothesis. To do this, we will first establish a quantitative measure of the degree of mismatch between the
organization and the product and then correlate this
metric with empirically observed performance (i.e.,
warranty claims).
3.
Overview of the Vehicle
Development Process
Our empirical analyses are based on a detailed study
of the new vehicle development process of a large
U.S. auto manufacturer. Because it involves many
interdependent tasks over an extended period, automotive design is a prototypical example of complex
product development. To create a useful model, one
of the authors spent two summers (about six months)
on site for data collection and analysis. This allowed
us to gain a good understanding of the product development process through observation of common practices and obstacles. We also collected an extensive
data set from the engineering change order (ECO) system, which is used by the firm to manage and document the design process.
3.1. Vehicle Development Process
Our main unit of reference regarding the vehicle
development process is a vehicle program. For a large
company, such as the one we studied, there are typically multiple models with different brand names
within the same vehicle program. A model refers
to the end product that is sold to the customers in
dealerships (e.g., Chevrolet Malibu, Toyota Camry,
etc.). Although models under the same vehicle program may be sold under different brand names, their
underlying architectural structure and product development effort is similar. Because vehicles are built off
of platforms, there is a good deal of component commonality across models. Some of these components
are entirely new to the program, whereas some are
carried over from previous programs.
It takes two to three years to complete the entire
product development process for a program. A
program provides a platform for several vehicles,
which are typically launched in a staggered fashion
to smooth demands on engineering and marketing
resources. Once in the market, vehicles are usually
given major refreshes (redesigns) every five to six
years, with a minor update at about the midpoint of
a model life cycle. Most models are eventually retired
after being in the market for several design cycles.
For practical reasons of data availability, we followed our client in dividing an automobile into 243
architectural subsystems, which contain roughly 150,000
parts that interact with each other. Consequently, we
will construct the product architecture network by
describing interfaces between subsystems.
3.2. Organization
The primary actors in a vehicle development organization are design engineers. Although other types
of engineers (e.g., materials engineers, quality-control
engineers, and testing engineers) are employed by the
organization, their direct involvement in the design
process is limited. Therefore, we focus our attention
exclusively on the design engineers. In the system
we studied, there were about 10,000 engineers who
participated in the engineering design work. These
engineers are responsible for creating the parts, making sure that they meet design specifications and
coordinating interfaces with other parts. Design engineers typically work in groups of 5–10 people, led
by a manager who is responsible for supervising the
design of parts and components, as well as coordinating efforts within and beyond the group. Within
the product development system, design engineers
coordinate with each other through distribution lists.
Whenever there is activity related to a part, designated engineers who are directly involved (e.g., an
engineer whose parts share a direct physical interface with a modified part) or indirectly involved (e.g.,
an engineer whose part shares an indirect functional
interface) are notified via the distribution list. Individual engineers are placed on these distribution lists
as a result of both management policy and requests
by engineers. As such, distribution lists capture both
formal connections inherent in the organization chart
and informal connections that emerge from the experience of engineers.
3.3.
Information System and
Engineering Change Orders
Because of significant advances in computer data storage and processing capacities, firms are able to accumulate and track large amounts of data about the
product development process. In our study, we made
use of an ECO system like that used in most product development processes as a tool to control and
document the product development process. (For a
471
detailed overview of ECO systems, see Loch and
Terwiesch 1999.) An ECO is filed by a design engineer every time a new part is released or an existing
part is changed in any way. Although the details vary
from one company to another, the basic features of
most ECO systems are similar.
In our client’s vehicle design process, the ECO
database contained approximately 100,000 separate
ECOs for one model year. A typical ECO contains the
identity of the engineer who initiated it, a unique reason code that explains why the ECO was issued, the
identities of other engineers to be notified as part of
a distribution list about activity related to the ECO,
part numbers associated with it, and the targeted and
actual dates of completion. Figure 1 shows a simplified version of our client’s ECO process. Note that
there are several different situations for which ECOs
are created, including when a part is initially released,
when there is a design problem that must be corrected,
and when there is an exogenous change (e.g., because
of a government regulation, styling change, or supplier request). Each ECO is notated with a reason code,
which describes the specific motivation for it.
For purposes of analysis, we have grouped ECOs
into three mutually exclusive sets according to their
reason codes: (1) new release ECOs, which are filed for
all parts of a new model (note that some of these parts
are new, whereas others are existing parts from a previous model that have been renumbered for the new
model); (2) problematic ECOs whose reason codes were
identified by several design engineers with whom we
consulted as indicating problems in the design process; and (3) other ECOs, which include all ECOs not
contained in the above categories (e.g., ECOs due to
a cost reduction initiative or a change in government
regulations). The role of each of these ECO types in
the design process are illustrated schematically in Figure 2. We use this classification to create variables in
the empirical model in the next section.
There have been several studies (see Clark and
Fujimoto 1991, Huang and Mak 1999, Terwiesch and
Loch 1999, Loch and Terwiesch 1999) of ECOs in
the design process. These examined the broad significance of ECO generation without specifically capturing product architecture information or organizational
structure. Because ECOs are filed when an individual part fails to meet specifications, two or more parts
have interface problems, or product changes are made
that affect part designs, the ECO database contains
a great deal of information. To our knowledge, this
study is the first attempt to use the ECO system to
capture product and organization interactions.
Previous studies (e.g., Sosa et al. 2004, 2007) have
relied on surveys to construct networks for both product architectures and organizational structures. This is
(a) time consuming, which may limit use in practice,
472
Figure 1
A Schematic of ECO Flows in the Vehicle Design Process
Design engineer/group manager
Sample reasons for issuing an ECO
• New release
• Performance test failure
• Cost reduction
• Government regulation, etc.
Discuss, draft and
pre-approve ECO
Responsible engineer
• Identify engineers who need to be notified in the
distribution list
• List related part numbers
• Specify dates of completion
Responsible engineer
Other engineers on the distribution list
Incorporate necessary changes to
owned parts and sign off
Perform necessary changes
and modifications
Group manager
Approve changes and
sign off
Not resolved
Change approval: cross system,
cross functional board
Resolved
and (b) subject to people’s memories (e.g., a vehicle
program lasts several years, so people must remember
with whom they collaborated years ago to construct
a relevant organizational structure network through a
survey). Because the ECO system contains information about both parts and the engineers who work on
them, we can use it instead of surveys to construct the
Figure 2
product architecture and organizational coordination
networks.
4.
Model Development
In this section, we describe how we created the product architecture and organizational coordination networks
from the ECO data described above. These networks
Modification of a Part Through the ECO System
Problematic
ECO
Design/Integration
problems
Part
release
New release
ECO
Pass
Design
Test
Other problems
Other
ECO
Launch
Figure 3
473
Product Architecture Network: A Network of Vehicle Subsystems
are the basis for the key independent variables in our
empirical study of vehicle quality. So, once we have
described the networks, we discuss the construction
of the dependent variable, independent variables, and
control variables in our regression model.
4.1. Creating the Product Architecture Network
We constructed the product architecture network by
defining vehicle subsystems as nodes. We defined
links between these nodes by looking only at new
release ECOs. Note that these new release ECOs are
not a result of a problem or later changes, but purely
a result of initiating all parts of a new vehicle program. As such, they provide an unbiased summary of
the linkages between parts. For example, when a part
in the steering wheel subsystem is newly released, all
parts related to it, which may be in the steering wheel,
electrical traction, or other subsystems, will be automatically listed on the new release ECO for that part.
The logic behind this construction is straightforward: When a part is initiated by issuing a new
release ECO, all parts that share some sort of physical
or functional interface with that part are also listed
in the ECO. Therefore, if we look at all such ECOs
and count how many times two subsystems appear
in the same ECO, we can get a proxy for the strength
of the architectural interaction (number of interfaces)
between the two subsystems. Specifically, we use the
number of new release ECOs that include parts both
from subsystems i and j as the weight for the link
between nodes i and j in the product architecture
network.
This network reveals that the various subsystems
differ substantially in terms of their connections to
other subsystems. For example, in a car, the wiring
harness subsystem has physical connections to almost
every other subsystem, while the air cleaner subsystem has only a limited number of physical connections with the rest of the vehicle. Figure 3 depicts a
visual representation of the product architecture network, which shows that the network is too large and
complex to analyze visually. Clearly, we need quantitative metrics to characterize the product architecture
in a useful manner.
4.2.
Creating the Organizational
Coordination Network
Many organizational studies (see Ibarra 1993, Krackhardt and Hanson 1993, Burt 2004) have studied communication and advice networks of individuals by
using empirical data sets that are usually collected
through surveys or questionnaires. Our study differs
from these by making use of formal institutional connections, rather than informal social ones. One benefit
of this approach is that it permits organizational analysis with data already being recorded, and so does not
subject the organization to the burden of a detailed
survey. A second benefit is that it focuses on links over
which management has a great deal of influence (i.e.,
who is listed on which distribution list). Hence, any
levers indicated by this analysis can be translated into
concrete management policies.
To construct the organizational coordination network, we again used vehicle subsystems as nodes and
proceeded in two steps. In the first step, we used only
the new release ECOs to determine which engineers
are associated with which subsystems. We did this
because our client indicated that only key engineers
474
involved in the design of the parts (and hence subsystems) are listed in the new part release ECOs. (Note
that an engineer may be associated with more than
one subsystem, while a subsystem always involves
more than one engineer.) In the second step, we used
all ECOs to characterize communication between subsystems, to capture the full range of communication
over the duration of the project. Specifically, we used
the number of distribution lists that include engineers
from both subsystem i and subsystem j as the weight
of the link between nodes i and j in the organizational
coordination network. Note that each distribution list
corresponds to an issue in the product development
work, so we count the number of distribution lists
rather than the number of people in establishing the
links between subsystems.
4.3. Scope of the Model
We tested our hypotheses by developing a regression
model. We examined 13 vehicle programs, with 243
subsystems in each, giving us a total of n = 243 ×
13 = 3159 observations in the model. Each of the 13
programs corresponded to a 2005 model-year vehicle
designed in the United States and sold solely to U.S.
customers. Note that these programs correspond to
platforms from which many models are introduced.
For example, our client launched 32 distinct models
in the 2005 model year.
As the dependent variable in the model, we used
warranty claims data aggregated from roughly 17,000
unique problem codes up to the subsystem level. We
followed our client in using IPTV (incidents per thousand vehicle) as a measure of quality. We used the
number of warranty incidents (IPTV) reported during
the first 12 months after the vehicle launch. Note that
we observed warranty data during the first year of the
vehicle use (i.e., in calendar years 2005 and 2006), but
the ECOs that describe the product and organization
networks for these programs were initiated during
calendar years 2002–2005. Therefore, collecting design
and quality data for one model year requires examining over four years of data within the company.
We conducted a similar study by examining the
vehicles that were launched in the 2006 model year
to check the robustness of the model. As before, we
focused on 13 vehicle programs, which correspond
to 26 distinct models. Because the procedure and the
results are very similar to those for the 2005 model
year, we present them in the online appendix (provided in the e-companion).1
1
An electronic companion to this paper is available as part of the online version that can be found at http://mansci.journal.informs.org/.
Figure 4
Calculating Degree Centrality in the Product Architecture
Network
Centrality score = 66
Electrical
traction
4
Door
trim
Centrality
score = 17
4.4.
Centrality score = 44
19
43
13
Steering
wheel
25
Wiring
harness
Centrality
score = 9
9
Battery
Centrality score:
13 + 43 + 25 + 9 = 90
Independent Variables
4.4.1. Centrality in the Product Architecture Network. After creating the product architecture network
as outlined in the previous section, we calculated
the centrality scores of the nodes (subsystems) using
UCINET 6,2 Borgatti et al. (2002). We use degree centrality, which is computed as the sum of the weights
of the links emanating from a node to characterize the
level of connectivity of a subsystem. Subsystems with
higher degree centrality have more interfaces and are
therefore, presumably, more complex. Figure 4 illustrates this by showing partial centrality scores for a
portion of the product architecture network.
To look for the U-shaped relationship conjectured
in Hypothesis 1, we also included the square of the
degree centrality as an independent variable. A positive coefficient for the linear variable and a negative
coefficient for the squared variable would suggest an
inverted U-shaped relationship between degree centrality and warranty claims.
4.4.2. Coordination Deficit. Hypothesis 2 conjectures that misalignment between the product architecture and organizational structure is associated with
quality problems. The product architecture and organizational coordination networks defined above provide a means for quantifying misalignment. But there
are many ways to specify and measure mismatches
between the two networks. As long as (a) the metric is computed from the data contained in the
networks, and (b) the metric monotonically increases
in the extent to which the two networks are misaligned, then it can be considered as a possible metric. Below, we discuss one such metric that we feel
fits the NPD process, along with three other plausible
metrics.
To develop a misalignment metric, we first posit
that the interfaces between two subsystems in the
2
UCINET is a social network analysis software package that graphically displays networks and computes most standard network
metrics.
475
product architecture network imply a certain number
of design issues that must be resolved. This number
may be uncertain, but we assume that it is proportional in expectation to the number of interfaces
indicated by the product architecture network. We
further assume that each issue requires some number of communications to resolve, which again may
be uncertain. The product of these two numbers is
the number of communications required to successfully coordinate the two subsystems. If the number of communications falls short of this limit, then
unresolved issues may result in design flaws that
lead to warranty claims. Because each unresolved
issue represents an additional flaw, we conjecture
that the expected number of warranty claims is linearly related to the difference between the actual
and required number of communications. However,
if the number of actual communications exceeds
the required number, then no additional benefit is
gained, because communications about the interfaces
between subsystems i and j will not impact design
issues involving interfaces between other pairs of
subsystems.
We quantify the above reasoning into a metric
that we call the coordination deficit metric. To do this,
we let W A and W C represent the product architecture and coordination networks, respectively, where
W A = WijA , and WijA represents the weight of the link
between nodes i and j in the product architecture network; and W C = WijC , and WijC represents the weight
of the link between nodes i and j in the organizational coordination network. Because these weights
may have different magnitudes, we normalize them
by dividing by the total weight of the links in each
network. This yields Aij = WijA / i j WijA /2 and Cij =
WijC / i j WijC /2, where Aij and Cij represent the proportion of total links that are from subsystem i to
subsystem j in the product architecture and organizational coordination networks, respectively. With these,
Figure 5
we define i as the coordination deficit for node (subsystem) i as
i = maxAij − Cij 0
(1)
j
Note that this metric includes only links where
Aij − Cij is positive (i.e., the connection between nodes
i and j is stronger in the product architecture network than in the organizational coordination network) to capture undercoverage of subsystem linkages. Because problems from lack of coordination
cannot be reduced below zero, we would not expect
excess coverage along one link to offset inadequate
coverage along another. Hence, we omit links where
Aij − Cij is negative.
Figure 5 illustrates calculation of the coordination
deficit metric for a subset of the nodes in the vehicle development system. In this example, the wiring
harness subsystem has four links in the product architecture network to the door trim, electrical traction,
steering wheel, and battery subsystems, with weights
of 13, 43, 25, and 9, respectively (see Figure 5).
Because the total weight of all the links in the network
is 4 + 19 + 13 + 43 + 25 + 9 = 113, these links represent
the following fractions of the total: 0.115, 0.380, 0.221,
0.079. In the organizational coordination network, the
wiring harness subsystem has four links to the same
nodes as in the product architecture network, with
weights of 2, 27, 33, and 7. These represent the following fractions of the total: 0.024, 0.325, 0.397, and 0.084.
For each link, we compute the difference between the
fraction of weight in the product architecture network
and the fraction of weight in the organization network
(inserting a zero if this difference is negative). This
yields a coordination deficit for the wiring harness
subsystem of 0115 − 0024 + 0380 − 0325 + 0 + 0 =
0146. Once we have computed coordination deficit in
this manner for all subsystems (nodes), we can use it
as an independent variable in our model.
An Example of Calculating Coordination Deficit for a Subset of Nodes
Product architecture network
Electrical
traction
19
4
43
25
13
0.380
0.036
Electrical
traction
9
Organizational coordination network
Door
trim
0.115
2
0.221
0.134
Wiring
harness
Steering
wheel
Steering
wheel
0.079
0.397
Battery
0.036
0.325
11
27
0.169
33
3
7
Door
trim
0.024
Wiring
harness
0.084
Battery
476
Although our coordination deficit metric is reasonable, it is not the only way to measure mismatches
between the product architecture and organizational
coordination networks. To see if another measure
might work better, we considered three alternatives
that also satisfy the two criteria we defined above for
a metric to measure misalignment:
1. The ratio metric is computed as the ratio of the
percentage of links in the two networks. That is, we
first calculate the percentage of the entire network
flow at each link for both architectural and coordination networks as we did for the coordination deficit
metric (i.e., calculating the Aij and Cij ). However,
unlike the coordination deficit metric, which calculates the difference between the flow at links in two
networks, this metric calculates the ratio between the
flow at links in two networks. After calculating the
ratios, it proceeds similar to the coordination deficit
metric, and aggregates these ratio values at each node.
More formally, the ratio metric for node (subsystem)
i is given by
Ri =
j
max
Aij
Cij
0 (2)
Although this metric is monotonic in the degree
of mismatch between the product architecture and
organizational coordination networks, it implies that
reducing the number of mismatches will affect quality
(warranty claims) in a nonlinear fashion.
2. The node difference metric is computed as the difference between the centrality score of the subsystem
(node) in the product architecture network and that in
organizational coordination network. That is, if we let
Ai = j Aij be the centrality of node i in the product
architecture network, and Ci = j Cij be the centrality
of node i in the organizational coordination network,
the node difference metric for node (subsystem) i is
given by
Di = maxAi − Ci 0
(3)
As such, this metric considers node differences
between the two networks, rather than link differences.
3. The local deficit metric is obtained by calculating the percentage of flow along each link emanating
from a node. After calculating these flow percentages at each node for both networks, it proceeds in a
fashion similar to the coordination deficit metric and
calculates the aggregated deficit scores. Formally, Aij
and Cij are now calculated as Aij = WijA / j WijA and
Cij = WijC / j WijC . We then aggregate these for node
(subsystem) i as
Li =
j
maxAij − Cij 0
(4)
Because it normalizes flows at each link by the total
flow from that node, rather than total network flow,
the local deficit metric is not sensitive to the total
amount of coordination effort associated with a subsystem. For example, 1 unit of flow between nodes i
and j out of a total flow of 10 units from node i is
regarded as equivalent to 10 units of flow between
nodes i and j out of a total flow of 100 units from
node i.
We examined both the original coordination deficit
metric and these three alternate metrics in our regression analysis, as we discuss in §5.
4.5.
Control Variables
4.5.1. Previous Year’s Warranty Claims. Although we control for all relevant factors for which
we could obtain data, there may still be unobserved
factors, such as subsystem characteristics or engineer
capabilities, that could bias the results. According to
(Wooldridge 2002), “Omitted variables bias can be
eliminated, or at least mitigated, if a proxy variable
is available for the unobserved variable” (p. 63), and
“often the outcome of the dependent variable from an
earlier time period can be a useful proxy variable” (p.
66). Nerkar and Paruchuri (2005) and Heckman and
Borjas (1980) used this approach by introducing previous performance as an independent variable to predict
current performance. To control for unobserved factors, we used warranty claims in the previous year as
an independent variable.
4.5.2. Fraction of Problematic ECOs. We included
the fraction of problematic ECOs as a measure of internal quality problems. The rationale is that the rate of
internal quality problems could be a signal of external warranty issues. Because problematic ECOs are a
result of design related mistakes, a high percentage of
problematic ECOs is a reasonable proxy for the rate of
internal quality problems.
4.5.3. Fraction of New Parts. Following Clark
et al. (1987), who adjusted for the fraction of new parts
in a vehicle to compare the productivity of different
auto makers, we include the fraction of new parts (relative to the previous model year) in a subsystem as a
control variable. We would expect to experience more
quality problems with new parts than old ones.
4.5.4. Other Controls. We also controlled for the
following additional factors:
• Number of parts: This is the total number of parts
in a subsystem, which may be a proxy for the subsystem size or complexity.
• Number of engineers: This represents the total
number of engineers that appear in the distribution
lists associated with a subsystem, which is another
potential proxy for the complexity of that subsystem.
477
• Number of ECOs: This is the total number of ECOs
that are generated in a subsystem, which may be yet
another indicator of the complexity of a subsystem.
• Average ECO tardiness: This variable is calculated
using the targeted completion dates and actual completion dates of ECOs. Specifically, it calculates the
tardiness for each ECO and then averages it across all
ECOs in a subsystem. Tardiness could indicate trouble
in the design process (bad for quality) or additional
time spent resolving problems (good for quality) and
so does not have an obvious expected relationship
with warranty claims.
5.
that varies randomly over programs for each subsystem. While random-effects models make use of the
(seldom met) assumption that individual effects are
uncorrelated with the regressors, and model individual constant terms as randomly distributed across
cross-sectional units, fixed-effects models impose the
most powerful control on unobserved heterogeneity by only examining within subsystem variation
(Greene 2008).
A random-effects model is more appealing to us
than a fixed-effects model for two reasons: (i) Fixedeffects models can produce biased estimates for panels over short time periods (Greene 2008, Hsiao 1986).
Because we only have 13 programs (similar to having
13 time units) and a large number of cross-sections
(N = 243), a fixed-effects model may not be appropriate. (ii) Fixed-effects models provide poor estimates
of the effects of the variables that vary only slightly
over time (i.e., over the 13 programs) (Kraatz and
Zajac 2001). In our panel data, some of the key variables such as number of design engineers and fraction
of new parts change only slightly across programs.
Random-effects models do not share these limitations.
They allow us to examine both within and between
subsystem variance in independent and dependent
variables.
Nevertheless, we fitted both the random-effects
model and fixed-effects model, and conducted a
Hausman test to determine which specification is
more appropriate (Hausman 1978). In this test, under
the null hypotheses, the two estimates do not differ significantly, and, therefore, the more efficient and
consistent random-effects model is preferable. The
Hausman test resulted in a test statistic of 2 = 1174,
which is well below the critical value of 15.51 from
the chi-squared table. Therefore, the null hypothesis
of the “no statistical differences” is not rejected, which
implies that the random-effects model is the appropriate specification for our data.
Analysis and Results
Table 1 shows descriptive statistics and bivariate
correlations between the variables in our models.
Warranty incidents for the 2004 and 2005 model years
are highly correlated as expected. Furthermore, we
note that both centrality of a subsystem and coordination deficit have positive correlations with 2005 warranty claims.
Our study examines a total of 243 subsystems
across 13 vehicle programs. Because we have all subsystems present in all programs, we have the repeated
observations for each of the 243 subsystems. This
panel structure of our data set (i.e., a cross-section
of 243 subsystems observed 13 times) allows us to
explore both within and between subsystem variation. By using panel data methods, we can control
for the unobserved subsystem characteristics, which
could pose a major problem for the ordinary least
squares estimates (Petersen and Koput 1991).
A fixed-effects model could address the problem
of unobserved heterogeneity by including an error
term that is assumed to be constant over vehicle programs for each subsystem, whereas a random-effects
model could address this by inserting an error term
Table 1
Descriptive Statistics
Variable description
(1) Warranty incidents
for 2005
(2) Warranty incidents
for 2004
(3) Number of parts
(4) Number of design
engineers
(5) Number of ECOs
(6) Fraction of new parts
(7) Fraction of
problematic ECOs
(8) Average ECO tardiness
(9) Centrality of a subsystema
(10) Centrality squared
of a subsystema
(11) Coordination deficit
a
Normalized scores.
Mean
SD
Min
Max
4115
2398
0086
11830
1
(1)
4372
2468
0161
13652
0728
2302
1045
13489
28905
164
17
275
221
2322
0332
0284
16861
0158
0053
132
0076
0109
294
0714
0570
2368
0219
0078
5058
0173
0148
0000
0033
0001
0081
0047
0001
(2)
(4)
1
0202
1
0317
0034
0067
0389
0108
0074
(5)
(6)
(7)
1
0065
0087
1
0314
1
(8)
(9)
0021
0143
0117
82000 −0112 −0094
0092 −0146
0105
0103 −0116
1
0572
0365
0211
0285
0085
0173
0007
0079 −0084
1
0760 −0288 −0192 −0109 −0022 −0111 −0008 −0071
0069 −0173
0244
0269
(10)
(11)
1
0033
0014
−0074 −0085
0059
0165
0133
(3)
0187
0045 −0089
0025
0093
0124 −0118
1
0461 −0297
1
478
Table 2
Models of Warranty Incidents for Product Subsystems
Estimation method:
Variable
Warranty incidents for 2004
Number of parts
Number of design engineers
Number of ECOs
Fraction of new parts
Fraction of problematic ECOs
Average ECO tardiness
Model 1
Random-effects
(controls)
Model 2a
Random-effects
(architecture)
Model 2b
Fixed-effects
(architecture)
Model 3a
Random-effects
(deficit)
Model 3b
Fixed-effects
(deficit)
06943∗∗∗
0077
00038
0013
−00098
0006
00061
0004
3752∗∗
1440
1016∗∗
5041
−0153∗∗
0059
06881∗∗∗
0077
00053
0010
−00087∗
0005
00066
0004
08319∗∗∗
0126
00084
0019
07024∗∗∗
0079
00049
0014
08608∗∗∗
0131
00077
0017
−00125
0008
00039
0006
−00091∗
0005
00064
0004
−00098
0007
00043
0006
3744∗∗
1472
9653∗∗
4890
3108
2140
6741∗∗
3127
3719∗∗
1465
9462∗∗
4851
3325
2144
6722∗∗
3125
Centrality of a subsystem
Centrality squared of a subsystem
−0147∗∗
0064
−0219∗
0115
378∗∗∗
0942
−607∗∗∗
1186
292∗∗∗
0874
−513∗∗∗
1143
Coordination deficit
R-squared (%)
Adjusted R-squared (%)
N
∗
71.00
70.93
3,159
72.55
72.46
3,159
29.53
29.35
3,159
−0116∗∗
0051
−0236∗
0121
26975∗∗∗
0997
23494∗∗
0922
73.95
73.88
3,159
30.76
30.61
3,159
p < 01; ∗∗ p < 005; ∗∗∗ p < 001.
Table 2 presents the results of our panel model.
Model 1 consists of only the control variables. As we
would expect, this shows that warranty incidents in
2004 is significant (p < 001) as a predictor of warranty
incidents in 2005. This confirms that it is an effective
proxy variable. Both the fraction of new parts and
the fraction of problematic ECOs also have significant positive coefficients (p < 005), which indicates a
positive association between these variables and warranty incidents. Average ECO tardiness is significant
(p < 005) with a negative coefficient, which suggests
that ECOs that take longer to resolve tend to result
in fewer quality problems in the field. Finally, the
number of parts and the number of ECOs in a subsystem are not significant. This agrees with our onsite observations that (i) subsystems with more parts
are not necessarily more complex, because some of
the simplest subsystems involve many tiny parts, and
(ii) total number of ECOs itself is not a good quality indicator because many ECOs are not problem
related. Note that Model 1 explains almost 71% of the
variation in 2005 warranty incidents.
5.1. Inverted-U relationship
Model 2a adds the linear and quadratic terms for
subsystem centrality to investigate Hypothesis 1. We
note that both terms are significant, but that the coefficient is positive for the linear term and negative
for the quadratic term. Although this is consistent
with the conjectured inverted U-shaped relationship,
it is not sufficient to demonstrate it. We must also
show an appropriate distribution of the independent
variable around the maximum. Without this, the coefficients might indicate a monotonic concave relationship instead. The Box-Whisker plot in Figure 6, which
provides a simple visualization of the data by dividing the sample into deciles and box-plotting each subsample, also supports the inverted-U relationship.
To further check the inverted-U relationship, we
performed two calculations (see the online appendix
for details): (i) First, we calculated the location of
the inflection point (i.e., the maximum), which corresponds to (378 − 2 × 607 × x = 0) or x = 0311.
This value is about half a standard deviation above
the mean, which supports the inverted-U relationship.
(ii) We divided the data into deciles, conducted separate regressions within each subsample, and observed
the pattern of the coefficient of subsystem centrality.
The estimated coefficients in these separate regressions confirm the inverted-U relationship with the
maximum in the seventh decile. We also checked for
outliers, because a significant curvilinear relationship
between subsystem centrality and warranty incidents
might be attributed to a few outliers in subsystem
centrality. We did not detect any influential outliers
using Cook’s distance (Cook and Weisberg 1982).
479
Figure 6
Box-Whisker Plot for Subsystem Centrality and Warranty Incidents
14
12
q1
Min
Median
Max
q3
Warranty incidents
10
8
6
4
2
0
Decile 1
Decile 2
Decile 3
Decile 4
Decile 5
Decile 6
Decile 7
Decile 8
Decile 9
Decile 10
Centrality deciles
Note that the behavior of the control variables is
quite similar in Models 1 and 2a. Also note that
with the addition of subsystem centrality variables
(both the linear and quadratic term), the adjusted
R-squared improves from 70.93% to 72.46% despite
the loss of two degrees of freedom. Hence, we conclude that Model 2a supports Hypothesis 1 and indicates an inverted-U relationship between subsystem
centrality and warranty incidents.
5.2. Coordination Deficit
Model 3a replaces the two subsystem centrality variables with our coordination deficit metric as a predictor and shows coordination deficit metric to be highly
significant (p < 001) with a positive coefficient. Note
that instead of including both subsystem centrality
and coordination deficit in the model simultaneously,
the coordination deficit variable replaces the subsystem centrality variable.
The reason for this is that it is our theory that
both variables are proxies for the same effect, namely
that mismatches between the organizational coordination network and the product architecture network
increase the likelihood of warranty claims. However, subsystem centrality captures this effect only
very roughly, by suggesting through the observed
inverted-U relationship that intermediate centrality
subsystems tend to exhibit higher levels of warranty
incidents. In contrast, coordination deficit measures
the mismatches much more directly by incorporating information about the organizational coordination
network, as well as the product architecture network.
That the two variables overlap in their predictive role
is supported by the fact that they are correlated (correlation coefficient = 0.461). That they are not identical
is supported by the fact that a regression including
both subsystem centrality and coordination deficit has
both variables significant (at the 5% level).
We observe that although the majority of the
explanatory power of the model comes from the previous year’s warranty claims, adding the coordination deficit metric to the original control variables
causes adjusted R-squared to improve from 70.93% in
Model 1 to 73.88% in Model 3a. Hence, Model 3a supports Hypothesis 2 by suggesting that coordination
deficit and warranty incidents are positively associated. The fact that R-squared is higher in Model 3a
than in Model 2a suggests that coordination deficit
is a better predictor of warranty claims than is subsystem centrality. This makes sense because coordination deficit contains much more information about
the system than does subsystem centrality. Finally,
note that coefficients of the variables are quite stable
across models. This supports our earlier observation
that multicollinearity is not a problem. But, it also
suggests that the magnitudes of the coefficients are a
good gauge of the effects.
In addition to the three models presented in Table 2,
we ran versions of Model 3a using the ratio, node difference and local deficit metrics presented in §4.4.2.
However, none of these metrics were significant.
Given the logical flaws in these metrics, this is not
surprising. The ratio metric assumed a nonlinear
relationship between reduction in mismatches and
warranty claims, which is hard to defend considering the observation in §4.4.2 that architectural interfaces generate issues requiring communications to
resolve, implying a linear relationship between warranty claims and mismatches. The node difference
metric is a coarser measure of mismatches than the
coordination deficit metric because there are many
fewer nodes than links, which explains why it did not
work as well in predicting quality problems. Finally,
the local deficit metric used the normalization of
flows at each node, and did not consider the flows
in the other parts of the network. Hence, this metric cannot detect some kinds of misalignment that are
detected by the coordination deficit metric, and, consequently, it was less effective.
5.3.
Checking for Endogeneity and the
Robustness of Results
Because some of the covariates in our models
include measures that are potentially endogenous
(e.g., assignment of distribution lists, etc.), it is important to examine potential endogeneity issues and their
impact on the results. Although the Hausman test
suggested that a random-effects model is appropriate for our data, we also examined the results of
the fixed-effects model, which specifically controls for
the unobserved heterogeneity. We present the results
of the fixed-effects regressions in Models 2b and 3b.
Note that the coefficients and significance levels of
the variables in these fixed-effects models are different than those in the earlier random-effects models
due to different model assumptions. Also, because the
fixed-effects models only exploit within subsystem
variation, the R2 values are significantly lower than
the random-effects models. The main conclusion from
Models 2b and 3b is that our main variables of interest, product network centrality and coordination deficit,
are significant at the 0.01 and 0.05 levels, respectively.
This provides further support for our earlier results.
One significant variable that may be effected by
such endogeneity problem is coordination deficit. To
examine the endogeneity of coordination deficit, we
created a two-stage least squares (2SLS) randomeffects model (Baltagi 2001). In this procedure, in
the first stage we regress all exogenous variables on
the suspected endogenous variable (i.e., coordination
deficit) and get the fitted values. Here we use number of new release ECOs, number of new release parts,
and number of engineers working on new release parts
as the instruments. We expect these variables to have
a direct effect on coordination deficit, but no effect
on warranty claims other than their indirect effect
through coordination deficit. We then calculate the
second stage model using the fitted values created in
the first stage. The results of the 2SLS random-effects
model showed that all pairs of coefficients are within
Figure 7
Organizational Coordination vs. Product Architecture
Centrality
0.40
Organizational coordination
centrality
480
y = 1.114x 2 – 0.249x + 0.072
0.35
R 2 = 0.471
0.30
0.25
0.20
0.15
0.10
0.05
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Product architecture centrality
a 90% confidence interval of the original randomeffects model, suggesting that endogeneity does not
pose a threat for our analysis.
Results of Model 2a in Table 2 support the proposed
inverted-U relationship between product architecture
centrality and warranty claims. As mentioned in §2,
one potential explanation of this relationship could be
as follows: It may be hard for organizations to gauge
and provide the right amount of attention to intermediate central subsystems, but it is probably easier to
identify highly central subsystems and provide sufficient resources accordingly. If this is the case, then
we should observe fewer warranty claims on highly
central subsystems than on subsystems with intermediate centrality. To explore this further, we investigated the relationship between product architecture
centrality and organizational coordination centrality.
As we see in Figure 7, there is very little difference
between the organizational coordination centrality for
subsystems with product architecture centrality below
0.25, suggesting that the firm does not distinguish
between the complexity of these subsystems when
making decisions that affect coordination activities.
However, for subsystems with product architecture
centralities above 0.25, organizational coordination
centrality increases very rapidly (i.e., consistent with
an increasing convex function), suggesting that these
high centrality subsystems receive extensive coordination attention. These observations support our reasoning behind Hypothesis 1 that intermediate centrality
subsystems may not be receiving coordination effort
commensurate with their complexity.
From the results of the models in Table 2, it is clear
that warranty claims in the previous year has a large
impact on the warranty claims this year. To check
the robustness of coefficients of the other variables
in the models, we removed this lagged variable from
the models and and reran the statistical analysis. We
did not observe a significant change in the direction
or impact of the coefficients, but as expected, the overall explanatory power of the models were reduced
(e.g., adjusted R-squared of Model 2 was reduced to
around 11% from 69%) by excluding of the lagged
variable.
5.4.
Economic Significance of
Coordination Deficit
In addition to the statistical analyses, we conducted
an analysis of economic significance to get a sense
of the magnitude of the association between coordination deficit and warranty claims. Following the
convention used in other studies (see Nerkar and
Paruchuri 2005, Song et al. 2003), we computed the
percentage of change in the dependent variable (i.e.,
warranty claims) associated with a one standard deviation change in the independent variable (i.e., coordination deficit), evaluated at the mean of the data.
Reducing coordination deficit by one standard deviation (which is equal to 0047) from the mean in our
model predicts a 26975 × 0047 = 0127 unit reduction
in warranty claims, which represents a 0127 ÷ 4115 =
308% reduction. For our client, this would translate
into millions of dollars annually in direct savings,
plus an important reputational benefit (i.e., because
Consumer Reports and other rating services consider
warranty claims in their evaluation and recommendation of vehicles).
Although a 308% reduction in warranty claims
is economically important, the percentage of quality issues that are related to coordination deficit may
actually be substantially larger than this. The reason
is that in our model, the variable representing warranty claims from the previous year may also contain
claims that are associated with coordination deficit
(i.e., design flaws that were introduced in previous
years and carried over to this year’s model through
part reuse). So, if the firm were to reduce coordination deficit by one standard deviation in each design
cycle, some of the quality improvements would carry
over to future vehicles (through components based
on previous designs). Although one cannot calculate
this carryover effect of the reduction in coordination
deficit with precision, we can get an approximate figure by using our models.
One way to estimate the potential magnitude of an
ongoing reduction in coordination deficit is to make
use of Model 3 as follows: First, we note that the
regression equation we have is in the following form:
warranty claims in year t + 1
= 0 + 1 warranty claims in year t + · · · where t = 2004, t + 1 = 2005, and 1 = 070. Then, we
suppose that 1 remains constant over time and that
we reduce coordination deficit by x% in each design
481
cycle. (Recall that redesigns occur every five to six
years.) If we let
yn = total percentage of reduction (both direct and
indirect) in warranty claims in the nth redesign
and we assume that warranty claims are constant
between redesigns, then the total reduction in a
given cycle will be equal to the direct reduction plus
carryover, which is
y1 = x
y2 = x + 070x
y3 = x + 070x + 0702 x
(5)
···
2
yn = x + 070x + 070 x + 0703 x + · · · + 070n−1 x
As n → , this geometric series converges to x +
x070/1 − 070 = 333x. So, if coordination deficit is
reduced by one standard deviation (i.e., x = 308%),
and if there is no redesign at all, then the percentage
of warranty reduction in the nth year converges to
333 × 308% = 1026% when n is large. If, instead, we
assume that because of technology change and model
retirement, the dependence on old designs extends
back only five design cycles, then an ongoing one
standard deviation reduction in coordination deficit
ultimately results in a y5 × x (=294 × 308% = 906%)
reduction in warranty claims.
To get a sense of the total number of warranty
claims that are associated with mismatches between
the organizational coordination and product architecture networks (as opposed to the improvement predicted by our model for a realistically achievable one
standard deviation reduction in coordination deficit),
we consider the predicted impact eliminating coordination deficit entirely. According to Model 3, this
would result in a 53% direct reduction in warranty
claims, which would yield a reduction of 333 ×
53% = 1765% in the limit, and a reduction of 294 ×
53% = 1558% if the carryover effect is limited to five
design cycles.
Although these calculations give us a general sense
of the impact of ongoing reduction in coordination
deficit, we should be cautious in interpreting the individual coefficients and carryover effects. In the above
calculations, warranty claims from the previous year is
a proxy variable, which may include many causal
effects, of which coordination deficit is only one.
Using our regression model to estimate the amount of
this variable that is attributable to coordination deficit
is reasonable, but far from precise. Moreover, because
we would expect design flaws from prior years to get
corrected over time, our estimate is probably an upper
bound on the overall economic impact of coordination
deficit reductions on warranty claims.
482
According to a 2006 J.D. Power and Associates
quality survey, an average of 52 of the 124 (i.e.,
42%) of quality problems observed in automobiles
were design defects (Jensen 2006). If that is true,
then our analysis suggests that almost half of these
design related warranty claims are due to organizational coordination problems. The rest, presumably,
are associated with individual errors.
Along with the cost savings, these numbers also
indicate a major potential reputational benefit. For
example, in a recent J.D. Power and Associates initial quality survey, Toyota observed 104 complaints
per 100 vehicles (fourth in ranking), while Honda
observed 110 (seventh in ranking), Ford observed 112
(eighth in ranking), and Chevrolet observed 113 problems (tenth in ranking) in the first 90 days of ownership (Bennett and Boudette 2008). Although these
complaints are not the same as our warranty incidents, they are certainly related. Because these numbers are very close for the brands, a 10% reduction
could move a brand from tenth place to third place.
So, relatively small improvements in warranty incidents could make a significant improvement in a
firm’s quality rankings and hence its reputation.
6.
Discussion
Our analysis shows that warranty claims in the previous year have significant power for predicting warranty claims this year. Indeed, when we use only the
previous year’s warranty claims in a simple regression, it explains about 55% of the variation in this
year’s warranty claims. Though intuitive, this result is
not of great managerial use, because it merely implies
that trouble spots in a vehicle tend to persist over
time. In this sense, using last year’s warranty claim
data to predict this year’s warranty claims is a bit like
using yesterday’s weather to predict today’s weather.
There is a substantial correlation, but the model is
obvious. Only by going beyond this level of prediction can we derive useful forecasts.
We also observed a positive correlation between the
fraction of problematic change orders and the number
of warranty claims. Subsystems for which we observe
a high percentage of problems during design are the
very subsystems that result in a higher number of
warranty incidents. The implication is that engineers
fix some of the design problems by issuing and resolving ECOs, but not all of them. Because some design
problems reach the marketplace and lead to warranty
claims, management efforts to reduce the problematic
change orders will both speed the vehicle development process and improve vehicle quality.
Another factor shown by our analysis to be correlated with warranty claim incidents is the percentage of new parts in a subsystem. This is intuitive
given the learning involved in the design of a new
part. From a management perspective, this implies
that design organizations should devote extra attention and resources to subsystems with higher fractions of new parts. Although our client clearly knew
this already, the fact that warranty claims are still
positively correlated with the fraction of new content suggests that current levels of attention and/or
resources may not be enough.
A somewhat counterintuitive implication of our
results is that tardiness of engineering change orders
and quality problems are negatively correlated.
Although one might expect tardiness to compromise
quality (e.g., by causing haste or chaos in the design
process), we observed that subsystems with more
ECO tardiness tend to have fewer quality problems
in the field. This may be due to a simple time versus
quality trade-off; more time on a component results in
a lower probability of a problem, even at the expense
of missing due dates. Of course, while missing a due
date in order to spend more time on a given component may improve that component, it may also
be detrimental to other components or the vehicle
launch as a whole. So, although this result may suggest that management should be careful about compressing design times too much, it certainly should
not be taken as support for missing due dates established by the ECO system.
The inverted-U association we observed between
subsystem centrality and warranty incidents suggests
that subsystems of intermediate centrality are more
prone to quality problems. We conjecture that this is
because intermediate centrality parts are more difficult to evaluate with regard to their complexity than
are high centrality parts (which are obviously complex) or low centrality parts (which are obviously
simple). As such, it is more difficult to determine
the appropriate amount of resources and coordination
effort for intermediate centrality parts than for either
high or low centrality parts. Though intriguing from
a research perspective, this result does not identify
specific subsystems in need of greater attention and
hence is of limited managerial use.
Our most important contributions are (1) introducing the coordination deficit metric for quantifying
mismatches between the product architecture and the
organizational structure, and (2) showing that this
metric is positively correlated with warranty claim
incidents. This result is significant to the literature
on network analysis of product development systems because (a) it is the first effort to formally measure misalignment between an organization and its
assigned work, and (b) it provides support for the
common conjecture that misalignment of the design
organization with the product architecture is detrimental to performance.
From a management perspective, this work suggests some potentially appealing insights. First, our
analysis highlights a means for mining ECO system
data to monitor the alignment of the organization with
the product being designed. Our coordination deficit
metric provides a simple quantitative measure of the
degree of mismatch and points out specific pairs of
subsystems where the level of formal coordination is
less than the extent of connectivity in the product
architecture. Such pairs of subsystems may be candidates for additional coordination attention. Because
the coordination deficit metric also identifies pairs of
subsystems where the level of coordination activity
exceeds the amount of connectivity in the product
architecture, it may also suggest places where coordination efforts can be reduced with minimal impact
on performance. This suggests that it is possible to
improve the match between organizational coordination and the product architecture without increasing
the total amount of coordination activity.
Although the statistical correlations we have identified in this study only suggest, rather than prove,
causality, the existence of a positive association
between coordination deficit and quality problems is
of managerial interest. Because of the large cost of
design quality problems (e.g., recalls), managers of
product development organizations must be sensitive
to any factor that may have an impact on design quality problems. No statistical study (e.g., of the type
used in Six Sigma programs) can ever provide proof
of causality, so managers can only pursue improvements by addressing factors shown to be associated
with quality problems. Our paper introduces and
quantifies coordination deficit as one such factor. Furthermore, because researchers and practitioners have
been arguing (indeed assuming) that alignment of the
organization with the product is desirable, our findings are consistent with current management theory.
Our results support this theory and provide concrete
guidance on how to act upon it in practice.
7.
Conclusions
In this paper, we have presented an empirical model
that characterizes the misalignment of the product architecture and organizational interactions and
have investigated the impact of this misalignment on
quality (measured by warranty claims) in a vehicle
development process. Our results suggest that organizational factors and product architecture have a
significant impact on quality.
Our analysis made use of data from an ECO
system like that used in most product development
processes. These data enabled us to specify both
networks. As such, our study is the first, of which
we are aware, that bases a network analysis of the
product development process entirely on standard
483
data from a firm’s information system. Because we
do not rely on cumbersome and time consuming surveys, our methodology is more likely to find use in
practice than survey based methods.
Our work, along with the other studies that have
made use of emerging tools of complex networks to
characterize both product architecture (a network of
components) and organizational structure (a network
of people), highlight the potential importance of such
network tools to the science and practice of NPD.
Our results suggest that misalignment of the design
organization with the product architecture negatively
affects product quality and uses network tools to
highlight the specific areas of misalignment. Sosa
et al. (2004, 2007) suggest that such misalignments are
influenced by various features of the organizational
structure and use network tools to characterize these
features. Because of the power and flexibility of these
network tools, they are already becoming a standard
part of the NPD research tool kit. We expect them to
become similarly prevalent as practical management
tools in the future.
To further the science and practice of NPD processes, this work could be extended in several directions. First, our research exclusively relied on archival
(e.g., ECO, warranty) data. Although this is of substantial practical use, because it captures formal connections, it leaves out informal connections, such
as communication outside the channels indicated by
the distribution lists. Hence, a complementary study
could make use of surveys or e-mail and phone
records to characterize informal communication for
use as an additional predictor of quality performance.
A second dimension along which our model could
be refined is the granularity of the product data. We
have performed our analysis at the subsystem level.
This was largely because our client only had warranty
claims data that could be appropriately aggregated at
this level. However, we could obtain warranty claims
at the part level, we could perform a much more
detailed analysis of the impact of coordination deficit
on product quality. Our expectation is that this would
facilitate more precise matching of the organizational
structure to product architecture. It would also enable
a more accurate prediction of potential quality trouble
spots.
Finally, we note that the ultimate managerial purpose of this type of analysis is to better adapt the
design organization to the products being developed.
Our results provide an approach for identifying gaps
between organizational structure and product architecture. However, we have only analyzed vehicle
programs for one model year. To get a deeper understanding of how vehicle architectures evolve over
time and where the organizational coordination practices lag behind product changes, it would be useful
484
to perform a longitudinal study over multiple model
years. Although getting data extending back across
multiple design cycles would be a huge challenge,
such an analysis would represent an important step
in the use of complex network methods to further the
science of product development.
8.
Electronic Companion
An electronic companion to this paper is available as
part of the online version that can be found at http://
mansci.journal.informs.org/.
Acknowledgments
The authors gratefully acknowledge the support of this
research by the National Science Foundation under
Grants DMI-0423048 and DMI-024377. The authors thank
Christoph Loch (the department editor), the associate editor, and the reviewers for their excellent feedback and wise
suggestions, which have improved this paper substantially.
References
Adler, P. S., A. Mandelbaum, V. Nguyen, E. Schwerer. 1995. From
project to process management: An empirically-based framework for analyzing product development time. Management
Sci. 41(3) 458–484.
Baldwin, C. Y., K. B. Clark. 2000. Design Rules: The Power of Modularity. MIT Press, Cambridge, MA.
Baltagi, B. 2001. Econometric Analysis of Panel Data, 2nd ed. John
Wiley & Sons, New York.
Bennett, J., N. E. Boudette. 2008. Ford and GM show gains in
quality survey. Wall Street Journal (June 5), http://online.wsj.
com/article/SB121259964651445519.html.
Borgatti, S. P., M. G. Everett, L. C. Freeman. 2002. UCINET for Windows: Software for Social Network Analysis. Analytic Technologies, Harvard, MA.
Brown, S. L., K. M. Eisenhardt. 1995. Product development: Past
research, present findings, and future directions. Acad. Management Rev. 20(2) 343–378.
Burt, R. S. 2004. Structural holes and good ideas. Amer. J. Sociol.
110(2) 349–399.
Carlile, P. R. 2002. A pragmatic view of knowledge and boundaries: Boundary objects in new product development. Organ.
Sci. 13(4) 442–455.
Clark, K. B., T. Fujimoto. 1990. The power of product integrity. Harvard Bus. Rev. 68(November–December) 107–118.
Clark, K. B., T. Fujimoto. 1991. Product Development Performance:
Strategy, Organization, and Management in the World Auto Industry. Harvard Business School Press, Boston.
Clark, K. B., W. B. Chew, T. Fujimoto, J. Meyer, F. M. Scherer. 1987.
Product development in the world auto industry. Brookings
Papers Econom. Activity 1987(3) 729–781.
Contractor, N. S., P. R. Monge. 2002. Managing knowledge networks. Management Comm. Quart. 16(2) 249–258.
Cook, R. D., S. Weisberg. 1982. Residuals and Influence in Regression.
Chapman and Hall, London.
Ethiraj, S. K., D. Levinthal. 2004. Modularity and innovation in
complex systems. Management Sci. 50(2) 159–173.
Fleming, L., O. Sorenson. 2004. Science as a map in technological
search. Strategic Management J. 25 909–928.
Greene, W. H. 2008. Econometric Analysis, 6th ed. Prentice Hall,
Upper Saddle River, NJ.
Hansen, M. T. 2002. Knowledge networks: Explaining effective
knowledge sharing in multiunit companies. Organ. Sci. 13(3)
232–248.
Hausman, J. A. 1978. Specification tests in econometrics. Econometrica 46(6) 1251–1271.
Heckman, J. J., G. J. Borjas. 1980. Does unemployment cause future
unemployment? Definitions, questions and answers from a
continuous-time model of heterogeneity and state dependence.
Economica 47(187) 247–283.
Henderson, R. M., K. B. Clark. 1990. Architectural innovation: The
reconfiguration of existing product technologies and the failure
of established firms. Admin. Sci. Quart. 35(1) 9–30.
Hsiao, C. 1986. Analysis of Panel Data. Cambridge University Press,
Cambridge, UK.
Huang, G. Q., K. L. Mak. 1999. Current practices of engineering
change management in UK manufacturing industries. Internat.
J. Oper. Production Management 19(1) 21–37.
Ibarra, H. 1993. Network centrality, power, and innovation involvement: Determinants of technical and administrative roles. Acad.
Management J. 36(3) 471–501.
Jensen, C. 2006. Design quirks erode ’06 quality ratings. New
York Times (June 11), http://www.nytimes.com/2006/06/11/
automobiles/11QUALITY.html.
Kraatz, M. S., E. J. Zajac. 2001. How organizational resources affect
strategic change and performance in turbulent environments:
Theory and evidence. Organ. Sci. 12(5) 632–657.
Krackhardt, D., J. R. Hanson. 1993. Informal networks: The company behind the chart. Harvard Bus. Rev. 71(4) 104–111.
Krishnan, V., K. T. Ulrich. 2001. Product development decisions: A
review of the literature. Management Sci. 47(1) 1–21.
Loch, C. H., C. Terwiesch. 1999. Accelerating the process of engineering change orders: Capacity and congestion effects. J. Product Innovation Management 16(2) 145–159.
McGregor, J. 2006. The world’s most innovative companies.
Bus. Week (April 24), http://www.businessweek.com/print/
magazine/content/06_17/b3981401.htm.
Mihm, J., C. Loch, A. Huchzermier. 2003. Problem-solving oscillations in complex engineering projects. Management Sci. 49(6)
733–750.
Nerkar, A., S. Paruchuri. 2005. Evolution of R&D capabilities: The
role of knowledge networks within a firm. Management Sci.
51(5) 771–785.
Nonaka, I., H. Takeuchi. 1995. The Knowledge-Creating Company:
How Japanese Companies Create the Dynamics of Innovation.
Oxford University Press, New York.
Petersen, T., K. W. Koput. 1991. Density dependence in organizational mortality: Legitimacy or unobserved heterogeneity?
Amer. Sociol. Rev. 56(3) 399–409.
Sanchez, R., J. T. Mahoney. 1996. Modularity, flexibility, and knowledge management in product and organization design. Strategic Management J. 17 63–76.
Schilling, M. A. 2002. Toward a general modular systems theory
and its application to interfirm product modularity. Acad. Management Rev. 25(2) 312–334.
Song, J., P. Almeida, G. Wu. 2003. Learning-by-hiring: When is
mobility more likely to facilitate interfirm knowledge transfer?
Management Sci. 49(4) 351–365.
Sosa, M. E., S. D. Eppinger, C. M. Rowles. 2004. The misalignment
of product architecture and organizational structure in complex
product development. Management Sci. 50(12) 1674–1689.
Sosa, M. E., M. Gargiulo, C. Rowles. 2007. Component connectivity,
team network structure and the attention to technical interfaces
in complex product development. Working paper, INSEAD,
France.
Terwiesch, C., C. H. Loch. 1999. Managing the process of engineering change orders: The case of the climate control system
in automobile development. J. Product Innovation Management
16(2) 160–172.
Ulrich, K. 1995. The role of product architecture in the manufacturing firm. Res. Policy 24(3) 419–441.
Wasserman, S., K. Faust. 1994. Social Network Analysis. Cambridge
University Press, London.
Wooldridge, J. M. 2002. Econometric Analysis of Cross Section and
Panel Data. MIT Press, Cambridge, MA.
Journal of Operations Management 29 (2011) 194–211
Contents lists available at ScienceDirect
Journal of Operations Management
journal homepage: www.elsevier.com/locate/jom
Structural investigation of supply networks: A social network analysis approach
Yusoon Kim a,∗ , Thomas Y. Choi b , Tingting Yan b , Kevin Dooley b
a
b
Department of Management, Marketing, and Logistics, College of Business Administration, Georgia Southern University, Statesboro, GA 30460, United States
Department of Supply Chain Management, W. P. Carey School of Business, Arizona State University, Tempe, AZ 85287, United States
a r t i c l e
i n f o
Article history:
Received 26 October 2009
Received in revised form 5 November 2010
Accepted 10 November 2010
Available online 18 November 2010
Keywords:
Supply networks
Supply chain management
Second-tier suppliers
Social network analysis
Network structure
Structural analysis
Network indices
a b s t r a c t
A system of interconnected buyers and suppliers is better modeled as a network than as a linear chain. In
this paper we demonstrate how to use social network analysis to investigate the structural characteristics
of supply networks. Our theoretical framework relates key social network analysis metrics to supply
network constructs. We apply this framework to the three automotive supply networks reported in Choi
and Hong (2002). Each of the supply networks is analyzed in terms of both materials flow and contractual
relationships. We compare the social network analysis results with the case-based interpretations in Choi
and Hong (2002) and conclude that our framework can both supplement and complement case-based
analysis of supply networks.
© 2010 Elsevier B.V. All rights reserved.
1. Introduction
Supply chain management has focused on linear relationships of
buyers and suppliers (Cox et al., 2006; Zhu and Sarkis, 2004). While
a linear perspective may be useful for planning certain mechanical
aspects of transactions between buyers and suppliers, it fails to
capture the complexity needed to understand a firm’s strategy or
behavior, as both depend on a larger supply network that the firm
is embedded in (Choi and Kim, 2008). A firm’s “supply network”
consists of ties to its immediate suppliers and customers, and ties
between them and their immediate suppliers and customers, and
so on (Cooper et al., 1997; Croxton et al., 2001). In the past decade
there has been increased discussion of the benefits of adopting a
network perspective in supply chain management research (Choi
et al., 2001; Lazzarini et al., 2001; Lee, 2004; Wilding, 1998).
From a supply network perspective, the relative position of individual firms with respect to one another influences both strategy
and behavior (Borgatti and Li, 2009). In this context, it becomes
imperative to study each firm’s role and importance as derived
from its embedded position in the broader relationship structure
(Borgatti and Li, 2009; DiMaggio and Louch, 1998). For example,
Burkhardt and Brass (1990) and Ibarra (1993) claim that power and
influence derive from a firm’s structural position in its surrounding network. Others have linked network position to such issues as
∗ Corresponding author. Tel.: +1 912 478 2465; fax: +1 912 478 2553.
E-mail address: [email protected] (Y. Kim).
0272-6963/$ – see front matter © 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.jom.2010.11.001
innovation adoption (e.g., Burt, 1980; Ibarra, 1993), brokering (e.g.,
Pollock et al., 2004; Zaheer and Bell, 2005), and creating alliances
(e.g., Gulati, 1999).
To date, there have been few studies of real-life supply networks,
due to the difficulties in obtaining data. The studies of real networks that have been done have relied on qualitative methods to
derive theoretical and practical insights (e.g., Harland et al., 2001;
Jarillo and Stevenson, 1991). While qualitative interpretations have
their merits, their validity is threatened by a researcher’s bounded
rationality, which includes the difficulty to conceptualize complex
phenomena such as networks. Thus in this paper we propose to
analyze the structural characteristics of supply networks using a
formal, quantitative modeling approach—social network analysis
(Borgatti and Li, 2009; Grover and Malhotra, 2003; Harland et al.,
1999). We will show how social network analysis can both supplement and compliment more traditional, qualitative interpretation
methods when analyzing cases involving supply networks.
Social network analysis (SNA) has recently gained acceptance
among scholars for its potential to integrate the operations and
supply management field with other branches of management
science (Autry and Griffis, 2008; Borgatti and Li, 2009; Carter
et al., 2007). According to Borgatti and Li (2009), SNA concepts
are particularly suitable for studying how patterns of inter-firm
relationships in a supply network translate to competitive advantages through management of materials movement and diffusion of
information.
To date, SNA has not been applied in an empirical study of real
supply networks; in fact there is a general paucity of SNA appli-
Y. Kim et al. / Journal of Operations Management 29 (2011) 194–211
cations in operations and supply management, with only a few
exceptions (e.g., Carter et al., 2007; Choi and Liker, 1995). This is
largely because there is lack of conceptual clarification as to how
the key SNA metrics (e.g. centrality) can be theoretically interpreted
in the context of supply networks. Therefore in this study we link
different SNA metrics at the node- or firm-level to specific roles
in a supply network. We consider supply networks based on both
materials flow and contractual relationships. The metrics yield six
supply network related constructs: supply load, demand load, operational criticality, influential scope, informational independence,
and relational mediation. Different network-level SNA metrics are
also linked to their implications for supply network performance.
We apply our framework to real supply network data derived
from three published case studies of automotive supply networks
(Choi and Hong, 2002). In that study the authors created empirically
three complete network maps of the center console assembly for
Honda Accord, Acura CL/TL, and DaimlerChrysler Grand Cherokee.
In the present paper, we convert the network data from Choi and
Hong (2002) into matrix forms and analyze them using the software
UCINET 6 (Borgatti et al., 2002). These quantitative results are then
interpreted using our theoretical framework. Finally, we discuss
our quantitative SNA results comparing to the qualitative findings
of Choi and Hong (2002) and consider the implications.
2. Literature review
2.1. Supply networks
Supply networks consist of inter-connected firms that engage
in procurement, use, and transformation of raw materials to provide goods and services (Lamming et al., 2000; Harland et al., 2001).
The relatively recent incorporation of the term “network” into supply chain management research represents a pressing need to view
supply chains as a network for firms to gain improved performance,
operational efficiencies, and ultimately sustainable competitiveness (Corbett et al., 1999; Dyer and Nobeoka, 2000; Kotabe et al.,
2002). Therefore, it is increasingly important to analyze the network structure of supply relationships.
In the operations and supply management field, a complex system perspective has been used as a theoretical lens for describing
supply networks. Wilding (1998) studied dynamic events in supply
networks through what he referred to as “supply chain complexity
triangle” (p. 599). Choi et al. (2001) conceptualized supply networks
as a complex adaptive system (CAS). Surana et al. (2005) proposed
how various complex systems concepts can be harnessed to model
supply networks. Pathak et al. (2007) discussed the usefulness of
CAS principles in identifying complex phenomena in supply networks. Others have examined supply networks from a strategic
management perspective. Greve (2009), using supply networks in
the maritime shipping industry, studied whether technology adoption is more rapid in centrally located network positions. Mills et al.
(2004) suggested different strategic approaches to managing supply networks depending on whether a firm is facing upstream or
downstream and whether it is seeking its long-term or short-term
position in the supply network.
Methodologically, simulation models have been used to study
hypothetical supply networks (Kim, 2009; North and Macal, 2007;
Pathak et al., 2007). Others have studied real-world supply networks using the case study approach (Jarillo and Stevenson, 1991;
Nishiguchi, 1994; Choi and Hong, 2002). Scholars in the industrial
marketing have developed descriptive models of supply networks
(Ford, 1990; Håkansson, 1982, 1987; Håkansson and Snehota,
1995). Descriptive case studies in this genre illustrate how companies such as Benetton, Toyota, or Nissan attained competitive
advantage through their supply networks (Jarillo and Stevenson,
195
1991; Nishiguchi, 1994). Other studies focused on developing
taxonomies of supply networks (Harland et al., 2001; Lamming
et al., 2000; Samaddar et al., 2006).
More recently, Borgatti and Li (2009) have highlighted the
salience of SNA to study supply networks. In fact, there have been
a few studies in the operations and supply management field that
used or promoted the use of SNA. Choi and Liker (1995) used SNA to
investigate the implementation of continuous improvement activities in automotive supplier firms. Carter et al. (2007) provided
an example of the application of SNA in a logistics context. Autry
and Griffis (2008) applied the concept of social capital, framed as
part of social network theory, to supply chain context. However,
still lacking in such studies is a theoretical framework that relates
social network theory to supply network dynamics and the comprehensive application of SNA to studying supply networks. In the
following section, we provide a brief overview of SNA, focusing on
the key metrics useful for investigating and explaining phenomena
within supply networks.
2.2. Social network analysis (SNA)
A network is made up of nodes and ties that connect these nodes.
In a social network, the nodes (i.e., persons or firms) have agency in
that they have an ability to make choices. With its computational
foundation in graph theory (Cook et al., 1998; Kircherr, 1992; Li
and Vitányi, 1991), SNA analyzes the patterns of ties in a network.
Naturally, SNA has been used to study community or friendship
structure (Kumar et al., 2006; Wallman, 1984) and communication
patterns (Koehly et al., 2003; Zack and McKenney, 1995). It has been
adopted to explore the spreading of diseases (e.g., Klovdahl, 1985)
and diffusion of innovation (e.g., Abrahamson and Rosenkopf, 1997;
Valente, 1996). In organization studies and strategic management,
scholars have used it to investigate corporate interlocking directorships (Robins and Alexander, 2004; Scott, 1986) and network
effects on individual firms’ performance (e.g., Ahuja et al., 2009;
Burkhardt and Brass, 1990; Gulati, 1999; Jensen, 2003; Rowley
et al., 2005; Stam and Elfring, 2008; Uzzi, 1997).
Operations and supply management scholars have also noted
the methodological potential of SNA. For instance, Choi et al. (2001)
stated that one could approach the study of supply networks from
the social network perspective. Ellram et al. (2006) acknowledged
social network theory as a useful tool to study influence in supply chains. Carter et al. (2007) identified SNA as a key research
method to advance the fields of logistics and supply chain management. More recently, Borgatti and Li (2009) and Ketchen and
Hult (2007) echoed such sentiments. They have also recognized the
difficulty of collecting network-level data in supply networks but
argued its imperativeness for operations and supply management
to be integrated with other management disciplines.
According to Borgatti and Li (2009), a more systematic adoption
of SNA will be instrumental in exploring behavioral mechanisms of
entire supply networks. A SNA approach allows us to better understand the operations of supply networks, both at the individual firm
level and network level—how important the individual firms are,
given their positions in the network and how the network structure
affects the individual firms and performance of the whole network.
Social network scholars (Everett and Borgatti, 1999; Freeman, 1977,
1979; Krackhardt, 1990; Marsden, 2002) have developed a range of
network metrics at the node- or network-level to characterize the
dynamics inside a social network.
2.3. Key network metrics
Network metrics can be calculated at two levels—the node level
and network level. Node-level metrics measure how an individual node is embedded in a network from that individual node’s
196
perspective. In this study, we focus on three types of node-level
metrics—degree, closeness, and betweenness centrality. Networklevel metrics compute how the overall network ties are organized
from the perspective of an observer that has the bird’s eye view of
the network. The network-level metrics we consider are network
density, centralization, and complexity.
2.3.1. Node-level metrics
Identifying the key actors in a social network is one of the primary uses of SNA (Tichy et al., 1979; Wasserman and Faust, 1994).
The concept of centrality is fundamental to node-level network
metrics (Borgatti and Everett, 2006; Borgatti and Li, 2009). Centrality reflects the relative importance of individual nodes in a
network. A node’s central position in a social network has a significant impact on its and others’ behaviors and well-beings (Mizruchi,
1994). Centrality has been associated with social status (Bonacich,
1972; Freeman, 1979), power (Coleman, 1973), and prestige (Burt,
1982).
There are different types of centrality metrics and they identify
nodes that are important, in different aspects. Most prominent are
degree centrality, closeness centrality, and betweenness centrality
(Everett and Borgatti, 1999; Krackhardt, 1990; Marsden, 2002). Of
these, the most straightforward is degree centrality. This concept
builds on an observation that the more links a node has the more
central it is—when a node is connected to a large number of other
nodes, the node has high degree centrality. Due to its greater connectedness with other nodes, a node with high degree centrality
would necessarily be more visible in the network (Freeman, 1979;
Marsden, 2002).
Another centrality concept is closeness centrality. As the term
suggests, this metric focuses on how close a node is to all the other
nodes in the network beyond ones that it is directly connected to.
A node is central if it can quickly reach all the others, and that is
why closeness centrality includes indirect ties. This centrality is
usually associated with node’s autonomy or independence in social
networks (Freeman, 1979; Marsden, 2002)—a node with high closeness centrality has more freedom from others’ influence and higher
capacity for independent actions. Such nodes become less reliant
on other nodes.
Betweenness centrality measures how often a node lies on the
shortest path between all combinations of pairs of other nodes. The
more a given node connects nodes that would otherwise be disconnected, the more central that node is—other nodes are dependent
on this node to reach out to the rest of the network. This metric
focuses on the role of a node as an intermediary and posits that this
dependence of others makes the node central in the network. As
such, the betweenness centrality usually denotes a node’s potential control or influence in the network (Marsden, 2002). A node
with high betweenness centrality has a great capacity to facilitate
or constrain interactions between other nodes (Freeman, 1979).
2.3.2. Network-level metrics
SNA also yields metrics concerning the structure of the overall
network, such as network density, network centralization, and network complexity. Network density refers to the number of total ties
in a network relative to the number of potential ties. It is a measure
of the overall connectedness of a network (Scott, 2000)—a network
in which all nodes are connected with all other nodes would give
us a network density of one.
Network centralization captures the extent to which the overall
connectedness is organized around particular nodes in a network
(Provan and Milward, 1995). Conceptually, network centralization can be viewed as an extension of the node-level centrality
(Freeman, 1979)—if a network had such a highly centralized structure that all connections go through few central nodes, then that
network would be high on network centralization. The network
with highest possible centralization is one with a star structure,
wherein a single node at the center is connected to all other nodes
and these other nodes are not connected to each other. Likewise, the
lowest centralization occurs when all nodes have the same number
of connections to others.
Network centralization and network density are complementary. Whereas centralization is concerned with the distribution of
power or control across the network, density reflects network cohesiveness. A network that has every node connected with everyone
else would have a highest possible density (i.e., density of one).
This network would be a highly cohesive network but would have
a diffuse and distributed control structure.
Network complexity is defined as “the number of dependency
relations within a network” (Frenken, 2000, p. 260) and thus would
depend on both the number of nodes in the network and the degree
to which they are interlinked (Frenken, 2000; Kauffman, 1993). In
the context of a supply network, complexity relates to the collective operational burden born by the members in the network (Choi
and Krause, 2006). For instance, a large number of units in a system
is likely to entail high coordination cost (Kim et al., 2006; Provan,
1983). Further, if these units are highly interdependent, then the
collective operational burden would be high and thus more complex at the system level.
Network complexity is related to network density and network
centralization. First, more complex networks require higher operational burden (Lokam, 2003; Pudlák and Rödl, 1992, 1994). Second,
network density is conceptually linked with network complexity
because a denser network requires more effort to build and maintain (Marczyk, 2006). Finally, network centralization is associated
with network complexity because the highest coordination costs
require when every node is connected to all other nodes (i.e., a
network with the least centralization) (Pudlák et al., 1988).
3. Conceptual framework for analyzing supply networks
3.1. Two types of supply network
There are a number of different ways in which ties can be established between firms in the supply network. For example, a tie
might be established between two firms if they were collaborating
on a new product development or if they had overlapping board
membership or belonged to the same trade organization. In this
paper we focus on two types of ties that Choi and Hong (2002)
collected data for in their study.
Firms can be linked because of the delivery and receipt of materials, or they can be linked through a contractual relationship (Choi
and Hong, 2002). In a tree-like structure of materials flow (Berry
et al., 1994; Chopra and Sodhi, 2004; Hwarng et al., 2005), the
network describes which supplier delivers to which customer. The
other type of network is based on contractual relationships. Often,
when a buying company wants to control the bill of materials, it
engages in directed sourcing, wherein it establishes a contract with
a second- or third-tier supplier and directs the top-tier supplier
to receive materials from them (Choi and Krause, 2006; Chopra
and Sodhi, 2004; Park and Hartley, 2002). In this context, materials flow occurs between two firms who do not have a contract and
vice versa. These two types of supply networks, although based on
the same set of nodes, can have different network structures and,
therefore, different logics and implications (Borgatti and Li, 2009).
3.2. Supply network constructs
3.2.1. Firm-level constructs
We now consider the key node-level SNA metrics and discuss
how they can be used to interpret different roles in supply net-
197
Table 1
Node-level centrality metrics and their implications for supply networks.
Network type
Centrality
metrics
Supply network
constructs
Conceptual definitions
Implication for central nodes
Description
Key capabilities
Materials flow
Indegree
centrality
Supply load
Outdegree
centrality
Demand load
The degree of difficulty faced by a
Integrator
firm in managing incoming
material flows from the upstream
firms
The degree of difficulty faced by a
Allocator
firm in dealing with demands from
the downstream firms
To put together or transform
different parts into a value-added
product and ensure it functions
well
To distribute limited resources
across multiple customers,
focusing on scale economies
Betweenness
centrality
Operational
criticality
The extent to which a firm impacts Pivot
the final assembler’s operational
performance in terms of product
quality, coordination cost and
overall lead-time.
To facilitate or control the flows of
supply across the whole network
System integration
Design/development
Architectural
innovation
Process/manufacturing
Quality management
Component innovation
Out-bound logistics
Risk management
In- and out-bound
logistics
Cross-functional integ.
Degree
centrality
Influential scope
Closeness
centrality
Informational
independence
Betweenness
centrality
Relational
mediation
The extent to which a firm has an
impact on operational decisions or
strategic behavior of other firms in
the supply network
The extent to which a firm has
freedom from the controlling
actions of others in terms of
accessing information in the
supply network
The extent to which a firm can
intervene or has control over
interactions among other firms in
the supply network
Rolea
Contractual
relationship
a
Coordinator To reconcile differences of network
members and align their opinions
with the greater supply network
goals
Navigator
To explore, access, and collect
various information with greater
autonomy in the supply network
Broker
To mediate dealings between
network members and turn them
into its own advantage
Contract management
SRM/CRM
Information acquisition
Strategic alignment
with OEM
Information processing
Strategic alignment
with OEM
Network role given high centrality.
works. Table 1 offers an overview of key centrality metrics, the
corresponding supply network constructs, and their implications
for network roles in the context of modeling supply networks. We
propose this new framework for the interpretation of the SNA metrics in the supply network context.
To illustrate these constructs, we first discuss the calculation
of key SNA metrics and the essential properties of each. Then, we
integrate each key SNA metric separately with the two types of
supply networks (i.e., materials flow and contractual relation). We
should note that the supply network based on materials flow is
directional, whereas the supply network based on contractual relationship is non-directional as legal obligations are mutually agreed
and enacted.
3.2.1.1. Degree centrality in supply network. Degree centrality is
measured by the number of direct ties to a node. Degree centrality
CD (ni ) for node i(ni ) in a non-directional network is defined as:
CD (ni ) =
xij =
j
xji
j
where xij is the binary variable equal to 1 if there is a link between
ni and nj but equal to 0 otherwise (Freeman, 1979; Glanzer and
Glaser, 1959; Nieminen, 1973; Proctor and Loomis, 1951; Shaw,
1954). To account for the impact of network size g, degree centrality
is normalized as the proportion of nodes directly adjacent to ni :
CD (ni ) =
CD (ni )
.
g−1
For comparison purposes, in this study, we convert normalized
degree centrality to a 0–100 scale by multiplying by 100.
A high degree centrality points to “where the action is” in a
network (Wasserman and Faust, 1994, p. 179). Freeman (1979)
describes it as reflecting the amount of relational activities, and
such activities make the nodes with high degree more visible. For
instance, in a non-directional contractual relationship network, the
degree centrality refers to the extent to which the firm influences
other firms on their operations or decisions as the firm has more
direct contacts with others (Cachon, 2003; Cachon and Lariviere,
2005; Ferguson et al., 2005). In contrast, nodes with low degree
centrality are considered peripheral in the same network. If a node
is completely isolated (i.e., zero degree), then removing this node
from the network has virtually no effect on the network. Therefore, a firm who has more contractual ties in the network garners a
broad range of influence on others, and at the same time such a firm
would often be required to reconcile conflicting schedules or interests between others. For the final assembler, for instance, it would
make sense to align with suppliers with high degree centrality.
In a directional network of materials flow, the focus is either
on the flow initiated (out-degree) or flow received (in-degree). For
instance, out-degree centrality of a node is defined as:
CD (ni ) =
xi+
.
g−1
In-degree centrality and out-degree centrality indicate the size
of the adjacent upstream tier and downstream tier, respectively.
A high in-degree or out-degree can capture transactional intensity
or related risks for a firm (Powell et al., 1996). In a materials flow
network, in-degree centrality for a firm can reflect the degree of
difficulty faced by the firm when managing the incoming material
flows. In other words, this metric measures the firm’s operational
load coming from the upstream suppliers. A firm with high indegree centrality may serve the role of an integrator, as they are
tasked with organizing and incorporating a range of parts from various suppliers to maintain the overall integrity of the product or
service (Parker and Anderson, 2002; Violino and Caldwell, 1998).
Such members in a supply network are instrumental and vital in
carrying out the architectural or technical changes in the current
product (Henderson and Clark, 1990; Iansiti, 2000).
Out-degree centrality relates to the firm’s level of difficulty
in managing the needs of customers. The more direct customers
there are in downstream, the more challenging it is for the firm
to ensure on-time delivery, cost-effective inventory, and order
198
management for their customers. The number of direct customers is thus positively associated with the operational load
related to demand integration and resource allocation (Frohlich and
Westbrook, 2002). In a materials flow supply network, a firm with
high out-degree centrality tends to be a common supplier to multiple downstream firms. Such supplier can economize and capitalize
on its own internal resources as it aggregates demands from a range
of customers (Nobeoka, 1996). Further, this firm is more likely than
others to gain access to proprietary assets or information of its
customer firms. This firm is in the best position to allocate or channel production or technical information to others in the network
(Cassiman and Veugelers, 2002).
3.2.1.2. Closeness centrality in supply network. The calculation of
closeness centrality is based on geodesic distance d(ni , nj ) —the
minimal length of a path between two nodes ni and nj (Hakimi,
1965; Sabidussi, 1966). In this study, closeness centrality is considered only in contractual relationship networks, as shown in Table 1.
In a directed network (e.g., materials flow), the geodesic(s) from ni
to nj may not be the same as the one(s) from nj to ni , or there can be
two geodesics between two non-adjacent nodes. In the case of supply networks, this does not make physical sense. Therefore, typical
node closeness is defined as:
⎡
⎤−1
g
CC (ni ) = ⎣
d(ni , nj )⎦
j=1
where
g
j=1
d(ni , nj ) is the total distance between ni and all other
nodes. At a maximum, the index equals (g − 1)−1 , which happens
when the node is adjacent to all other nodes. When all the other
nodes are not reachable from the node in question, the index
reaches its minimum value of zero. The index can be normalized
by multiplying CC (ni ) by g − 1. The value then ranges between 0
and 1 regardless of network size (Beauchamp, 1965). In this study,
the normalized index is converted to a 0–100 scale.
Nodes with high closeness need not much rely on others for
relaying information or initiating communications (Bavelas, 1950;
Beauchamp, 1965; Leavitt, 1951). This metric, in a supply network
context, thus can represent the extent to which a firm can act
autonomously and navigate freely across the network to access
resources in a timely manner. Such a firm has comparatively shorter
supply chains, both upstream and downstream. Shorter chains
translate into less distortion of information and better ability to
access reliable information (e.g., demand forecasts, supply disruption) in a timelier manner (Lee et al., 1997; Chen et al., 2000).
Such accessibility to high-quality information increases the firm’s
capability to match supply and demand (Cachon and Fisher, 2000),
resulting in less inventory and lower operational costs (Lee et al.,
2000).
3.2.1.3. Betweenness centrality in supply network. Betweenness
centrality appears under both types of networks. A firm can lie
between a pair of non-adjacent firms either along their materials flow or contractual relationship. The intermediary will have
different effects on the firms it links, whether directionally or
non-directionally. Measuring betweenness centrality begins with
an assumption that a connection between two nodes, nj and nk ,
follows their geodesics. Therefore, betweenness centrality can be
expressed as (Freeman, 1977):
CB (ni ) =
gjk (ni )
j<k
gjk
where gjk is the total number of geodesics linking the two nodes,
and gjk (ni ) is the number of those geodesics that contain ni . The
ni ’s betweenness is then simply the sum of the probabilities that
the node lies between other nodes. The betweenness reaches the
maximum when ni falls on all geodesics and has a minimum of zero
when ni falls on no geodesics. We normalize it to a value between
0 and 100:
CB (ni ) =
CB (ni )
× 100.
[(g − 1)(g − 2)/2]
The betweenness can be viewed as indicating how much “gatekeeping” ni does for the other nodes (Borgatti and Everett, 2006;
Freeman, 1980; Spencer, 2003). Gatekeeping occurs because a node
on geodesic can control the flows of materials or communication
(Marsden, 2002). When applying to materials flow networks, firms
with high betweenness act as a hub or pivot that transmits materials along the supply chains, and betweenness centrality relates to
the extent to which a firm potentially affects the downstream firms’
daily operations (e.g., lead time) and eventually the performance
(e.g., final product quality) of the whole network. For instance, if a
firm with high betweenness transmits materials to a wrong place
or does not respond to changes in demand in a timely manner,
it can easily lead to supply disruptions (Chopra and Sodhi, 2004).
Similarly, the effects of poor-quality outputs from these firms can
easily infect the broader supply network, interfering with normal
product flows (Kleindorfer and Saad, 2005). Therefore, operational
hiccups caused by such firms can surely hamper the functioning of
the entire supply network (Hendricks and Singhal, 2005). Considering the significance of negative impacts such members can have,
it would be prudent of the final assembler to ensure high or, at least,
consistent operational performance of these firms (Hendricks and
Singhal, 2003).
In a contractual relationship network, the metric can denote the
extent to which a firm can affect the interactions among others in
the same supply network. A firm with high betweenness centrality
mediates many pathways and thus can either facilitate or interfere
with the network communications. The social network literature
suggests that a node linking dense regions of relationships enjoys
the benefits of non-redundant information to increase its control
over others (Burt, 1992, 1998). Supply network research also postulates that a buyer can enjoy the increased sourcing leverage when
it lies between two disconnected, competiting suppliers (Choi and
Wu, 2009; Wu and Choi, 2005). For instance, the buyer can play
two rival suppliers off each other to drive down the purchasing
price.
3.2.2. Network-level constructs
We now discuss the key network-level metrics. Table 2 summarizes the theoretical interpretation of the metrics and their
implications for network performance in the context of supply networks.
3.2.2.1. Supply network centralization. Recall that CD (ni ) is nodelevel degree centrality, and CD (n∗i ) is its maximum value in the
network. Then, a general definition for network centralization is
(Freeman, 1979):
CD =
g
[C (n∗ ) − CD (ni )]
i=1 D
.
g
∗
max
i=1
[CD (n ) − CD (ni )]
Given g nodes in the network, the denominator reduces to
(g − 1)(g − 2). The value of CD reaches the maximum value of 1 when
one node is connected with all other g − 1 nodes, and the others
interact only with this node. Its minimum value of 0 occurs when all
degree centrality values are equal. In supply networks, centralization can refer to how much power or control the core firms exercise
over other network members (Choi and Hong, 2002). In this study,
besides centralization based on degree, two other centralization
199
Table 2
Network-level metrics and their implications for supply networks.
Implication of overall network structurea
Network type
Network-level
metrics
Conceptual definition in supply
networks
Characteristics
Performance implications
Materials flow
Centralization
The extent to which particular
focal firms control and manage
the movement of materials in a
supply network
Operational authority (e.g., power to make
decisions on materials flow) concentrated
in few central firms
Centralized decision implementation
process
Complexity
The amount of collective
operational burden born by the
member firms in a supply
network
More firms engaged in the delivering and
receiving of materials
More steps required to move the materials
along
High level of controllability in production
planning
Low level of operational effectiveness at
the network-level (i.e., more time taken to
reach a decision and take actions on issues
at a local level)
Low level of operational efficiency at the
network level (i.e., longer lead time from
the most upstream to the final assembler
or more parts for the same product
function)
Centralization
The extent to which particular
focal firms exercise bargaining
power or relationship
management control over
other firms in a supply network
Lack of interactions between central and
peripheral firms in a supply network
Decoupled relationships between firms at
different tiers
The amount of load on the
supply network as a whole that
requires relationship
coordination
More firms involved in transferring
information
Active interactions at a local level
Slow relaying communications from
downstream to the final assembler
Contractual
relationship
Complexity
a
High level of controllability in product
design, product quality, and/or cost
management
Low level of responsiveness to or more
time for resolution on issues occurring at a
local level
Low level of robustness or high degree of
vulnerability to supply disruptions (i.e.,
more time to channel information and a
higher likelihood of information distortion
across a supply network)
Implications given high metric score.
indices are also used—ones based on closeness and betweenness
centrality.
Further, there are other proxy measures of centralization used
in this study. They are multiple indices of density that involve the
core and periphery sub-groups in a network (see Table 6). When
a network is partitioned into two clusters, a core cluster appears
among nodes that are densely connected together and a periphery
is formed among nodes that are more connected to core members than to each other (Borgatti and Everett, 2000; Luce and Perry,
1949). For instance, in Fig. 1, there are 19 firms in the core group
(see Table 6) around Honda and CVT who appear at the center. The
rest appears in the periphery.
3.2.2.2. Supply network complexity. Supply network complexity
refers to the load on the network as a whole that requires coordination (Choi and Hong, 2002). While the general state of the literature
regarding the property of complexity at the network level is still
emerging (Butts, 2001a,b; Everett, 1985; Freeman, 1983), we adopt
the idea put forth by Kauffman (1993) and Frenken (2000). They
propose that network complexity can be indicated by the number
of nodes and degree of interdependency among nodes in a given
network. Therefore, we use two types of SNA output metrics—sizetype and density-type—to represent the number of supply network
members and the level of connectedness among them, respectively.
The size-type outputs are shown in network size and core
size, and the density-types include network density, core density,
periphery density, core-to-periphery (CTP) density, and peripheryto-core (PTC) density. Network size relates to the average path
length among nodes in the network (Ebel et al., 2002). More firms
in a network translate into more steps and more time needed to
complete the same task, whereby creating a higher likelihood of
the supply being interrupted en route and higher collective burden
born at the system level (Frenken, 2000). Likewise, between the
two networks of identical size, more links imply a higher probability that the functioning of the individual nodes in the network
is likely to be impeded by others, leading to a greater coordination
load on the whole network (Choi and Krause, 2006). For instance,
if an OEM has two top-tier suppliers, the firm would necessarily
incur a greater amount of coordination load, compared to a situation where there is only one top-tier firm. Therefore, a complex
supply networks would be associated with large network size, large
core size, high network density, high core density, high periphery
density, high CTP density, and high PTC density. Note that in the
contractual relation supply networks, the PTC and CTP densities
are identical, since every link in the network is non-directional and
the adjacency matrix representing this network is symmetric.
4. Research methodology
4.1. Data source
Choi and Hong (2002) (hereafter, denoted as C&H) reported
three supply networks from raw materials suppliers to a final
assembler involved in the production of an automobile center console assembly. The three product lines represented were Honda
Accord, Acura CL/TL, and DaimlerChrysler (DCX) Grand Cherokee.
Using an inductive case study approach, the authors derived propositions regarding the behavioral characteristics of supply networks.
Table 3 provides a review of this particular work.
In our analysis, we include all the firms in the supply network
as identified in C&H—they are direct suppliers and parts brokers,
stretching from raw materials suppliers to the final assembler.
As indicated before, each supply network contains two different
types of network information—one pertaining to materials flow
and another based on contractual relationships. These two different
types of network data yield a total of six supply networks—three
based on directional materials flow and three based on nondirectional contractual relationships.
4.2. Data analysis
The network information from C&H is converted into a binary
adjacency matrix (Wasserman and Faust, 1994) that has firms representing both the rows and columns of the matrix. For instance,
cell (i,j) would equal “1” if the firms i and j were linked either by
materials flow or contractual relationship, and would be “0” otherwise. Supply networks may yield adjacency matrices that are
200
Fig. 1. Materials flow network for Accord.
symmetric (i.e., non-directional) or asymmetric (i.e., directional),
depending on the nature of the linkages. As noted earlier, a materials flow network is directional and thus asymmetric, while a
contractual relationship network is non-directional and thus symmetric. Once generated, the adjacency matrices are imported into
UCINET 6 and are used as inputs for network analysis (Borgatti et al.,
2002).
UCINET is a comprehensive software package for the analysis of
social network data. It has been one of the most widely accepted
SNA tools for conducting the structural analysis of interorganizational networks (e.g., Gulati, 1995, 1999; Human and Provan,
1997; Rowley et al., 2005; Ahuja et al., 2009). The program contains
dozens of network analytic methods such as centrality measures,
subgroup identification, role analysis, elementary graph analysis,
and permutation-based statistical analysis. While performing SNA,
UCINET can create network visualizations. A visualization of each
of the six supply networks, also known as a sociogram, is shown in
Figs. 1–6.
5. Results
5.1. Node-level results
Tables 4 and 5 list key firms in the two types of supply networks. We identify key firms based on their centrality values.
Tables 4 and 5 build on Table 1. The supply network constructs
shown on the top row come from Table 1, and centrality computations are conducted on the corresponding centrality metrics shown
in Table 1.
As indicated below each table, there is a cut-off point for each
supply network construct (e.g., 10 for in-degree, 6 for out-degree).
The cut-off point is determined based on one rule: when there is
a noticeable drop-off in the score, the previous score constitutes
the threshold. In all cases except one, there are multiple key firms.
The exception is out-degree centrality for the materials flow type
of DCX’s supply network. Every node in the network, except for
the OEM, has only one customer, showing the same value on outdegree centrality; consequently, there was no threshold value. In
Tables 4 and 5, the number shown in parenthesis next to a firm
name represents the centrality score.
5.2. Network-level results
Tables 6–8 show SNA results at the network level. Tables 6 and 7
focus on centralization metrics, respectively, for directional materials flow and non-directional contractual relationships. Table 8
summarizes all complexity metrics for both types of networks.
In Table 6, various network-level indicators are shown across
three different supply networks. Beginning with network size
and density, individual node-level centrality scores are averaged
for each supply network. Then, the three network centralization
scores are listed. Up to this point, all values reflect network-level
attributes. Below the network-level values, Table 6 lists values at
the group level. It first shows the size of core group and its density
(see Section 3.2.2.1 on supply network centralization for a discussion on core and peripheral groups). It then moves on to listing
other group-level measures.
Addressing contractual relationship supply networks, Table 7 is
constructed much the same way. Since this supply network type
is non-directional, network measures shown on the left-side column are slightly different from those of Table 6, as discussed under
Table 1. Also note that most of the network-level metrics are in normalized form, which allows us to compare them across the three
different supply networks.
201
Table 3
Summary of case data from Choi and Hong (2002).
Network
measures
Product type
Honda Accord
Acura CL/TL
DaimlerChrysler (DCX) Grand Cherokee
Centralization
Two firms, CVT and JFC, are top-tier suppliers
to Honda
Several second- or third-tier suppliers (e.g.,
Emhart, Garden State, and Miliken) directly
selected by Honda
Some third-tier suppliers directly selected by
CVT, based on Honda’s core supplier list
Honda’s penchant for centralized control when
it comes to the product design and supplier
selection
One top-tier supplier, Intek, a complete
integrator of this supply network
Honda engaging in directed sourcing at the
second, third, and even fourth tiers
Intek likewise engages in directed sourcing by
selecting its own suppliers and even their
supplier’s suppliers, based on Honda’s core
supplier list
Directed sourcing generally for high-priced or
strategic items
Honda’s centralized control of the product
design activities
Textron as the sole top-tier supplier that
integrates parts and subassemblies
Textron-Farmington and Leon Plastics appear
as two key second-tier suppliers
Textron assumes the leading role in designing
console
Directed sourcing occurs only on a limited
basis
Complexity
All together, 50 network entities: 2 first-tier,
21 second-tier, 18 third-tier, 7 fourth-tier, and
2 fifth-tier suppliers
Majority of the suppliers at the second-tier
level
Four different nature of businesses in the
network—manufacturing companies, raw
materials suppliers (e.g., GE Plastics),
distribution centers (e.g., Iwata Bolt), and
trading houses (e.g., Honda Trading)
Reciprocal relationship between CVT and JFC,
two top-tier suppliers, contributing to either
reduction or increase of network complexity
depending on the relational nature
76 entities in the network: 1 first-tier, 20
second-tier, 28 third-tier, 17 fourth-tier, 9
fifth-tier, and 1 sixth-tier suppliers
The coupling between Honda and Intek based
on their shared history may reduce the level of
complexity
The decoupling between Intek and JFC, a
second-tier supplier of the critical
subassembly, may further the complexity
Honda’s effort to centralize second-tier
suppliers may increase complexity of the
network as a whole
41 entities: 2 first-tier, 10 second-tier, 22
third-tier, and 7 fourth-tier suppliers
At the top-tier level, Textron is engaged in
assembly work and also acts as a conduit for a
part from Leon as it ships the front console mat
directly to the DCX plant with Textron’s label
No reciprocal relations among suppliers
As per DCX’s recommendation, Textron has
consolidated the second-tier suppliers, leading
to reduced number of second-tier suppliers
and subsequently reduced complexity
Fig. 2. Materials flow network for Acura.
202
Fig. 3. Materials flow network for Grand Cherokee.
Fig. 4. Contractual relationship network for Accord.
Fig. 5. Contractual relationship network for Acura.
Fig. 6. Contractual relationship network for Grand Cherokee.
203
204
Table 4
List of key firms based on materials flow network.
Supply loada
d
Accord
CVT (59 ), JFC (15), HFI (11)
Acura
Intek (58), Arkay (21), Select Ind. (12)
DCX
Textron (65), Leon Plastics (31)
a
b
c
d
Demand loadb
Operational criticalityc
CVT (15), C&C (7.4), JFC (7.4), GE (7.4), Yamamoru (7.4),
Industry Products (7.4)
Iwata Bolt (9.1), Tobutsu (6.1), Arkay (6.1), Twist (6.1),
Milliken (6.1), Garden State (6.1), Select Ind. (6.1)
None
CVT (13), Emhart (2), Yamamoru (1.7), Fitzerald (1.7),
JFC (1.3)
Intek (3), Arkay (1.7)
Textron (3.8), Leon Plastics (2.5)
Firms with in-degree > 10.
Firms with out-degree > 6.
Firms with betweenness > 1.0.
Centrality score.
Table 5
List of key firms based on contractual relationship network.
Accord
Acura
DCX
a
b
c
Influential scopea
Informational independenceb
Relational mediationc
CVT (52), Honda (30),
Yamamoru (15)
Intek (45), Honda (36), Arkay
(18), Select Ind. (15)
Textron (62), Leon Plastics (35)
CVT (57), Honda (53), Yamamoru (40)
CVT (79), Honda (64), Emhart (21) Yamamoru
(15), Fitzerald (14)
Intek (77), Honda (63), Select Ind. (14), Iwata
Bolt (12), Arkay (10)
Textron (88), Leon Plastics (53) Daimler (15)
Intek (62), Honda (56), Arkay (44), Select Ind.
(43), Tobutsu (41), HFI (40)
Textron (72), Leon Plastics (58) Daimler (46)
Firms with degree > 15.
Firms with closeness > 40.
Firms with betweenness > 10.
Table 6
Network-level results for materials flowa supply networks.
Network measures
6. Interpretation of results
Product type
Accord
Acura
DCX
Network size (firms)
Network density
Average in-degree
Average out-degree
Average betweenness
Centralization (in-degree)
Centralization (out-degree)
Centralization (betweenness)
28
0.046
4.630
4.630
0.809
0.567
0.106
0.128
34
0.037
3.654
3.654
0.231
0.556
0.056
0.029
27
0.037
3.704
3.704
0.234
0.641
0.001
0.038
Core group size (firms)
Core density
Core to periphery (CTP) density
Periphery to core (PTC) density
Periphery density
19
0.067
0.006
0.064
0.000
23
0.059
0.000
0.043
0.000
4
0.250
0.000
0.250
0.000
a
Represented by asymmetric matrix.
Table 8 re-organizes some information from Tables 6 and 7. It
lists values for the select indicators of network complexity—they
represent the degree of interdependency among firms. Network
size is listed as the first indicator. Network density is then listed in
both materials flow and contractual relationships networks. Then,
group-level indicators are listed in both types of supply networks.
Table 7
Network-level results for contractual relationshipa supply networks.
Network measures
Product type
Accord
Acura
Network size (firms)
Network density
Average degree
Average closeness
Average betweenness
Centralization (degree)
Centralization (closeness)
Centralization (betweenness)
28
0.074
7.407
35.716
7.407
0.479
0.459
0.748
34
0.066
6.595
37.747
5.375
0.413
0.513
0.738
27
0.074
7.407
41.959
5.778
0.585
0.641
0.854
Core group size (firms)
Core density
CTP or PTC density
Periphery density
17
0.125
0.048
0.036
6
0.467
0.179
0.000
3
0.667
0.333
0.000
a
Represented by symmetric matrix.
DCX
In this section, we recapitulate the SNA results shown in
Tables 4–8 with reference to the supply network constructs developed in this study (see Tables 1 and 2). We provide network
dynamics implications of the node-level results first and then those
of the network-level results. A summary of the SNA results at the
node- and network-level is shown, respectively, in Tables 9 and 10.
6.1. Node-level implications
6.1.1. Key firms in the materials flow supply networks
Table 4 compares groups of firms across supply load, demand
load, and operational criticality (see Table 1 for definitions). CVT, a
first-tier supplier in Accord supply network, appears highly central, showing the highest scores on all three columns. In other
words, CVT assumes the most operational burden on both the
supply side and demand side. This firm is tasked with integrating multiple parts into a product, which also means the firm can
make the most of its resources by pooling customer demands and
the related risks. CVT is also the pivotal player in the movement
of materials. Without this firm, the entire supply chain would be
disrupted. In contrast, we observe that another top-tier supplier
of Accord, JFC, is not as central. Its centrality scores are markedly
lower than those of CVT, and there are other second- (i.e., C&C,
Emhart, and Yamamoru) and third-tier suppliers (i.e., Fitzerald)
who appear more central than JFC. This is because most suppliers supplying to JFC also serve CVT but not the other way around
(see Fig. 1).
Intek, the top-tier supplier for Acura, appears most central under
both supply load and operational criticality. The bulk of network
resources flow into and through this firm. However, unlike CVT in
Accord network, Intek does not appear central under demand load.
This is because the firm primarily receives materials (see Fig. 2). In
fact, Iwata Bolt, a second-tier supplier for Acura, comes first under
demand load. This simply means that this firm delivers to a relatively large number of buying firms, which implies that this supplier
has leverage in allocating its internal resources across multiple customers. Another noteworthy finding is that Arkay, a second-tier
supplier, is the only firm that ranks high on all the three centrality
metrics. Without conducting SNA, Arkay’s central role in the Acura
supply network may very well be overlooked.
205
Table 8
Key indicators for network complexity.
Network size
(firms)
Accord
Acura
DCX
28
34
27
Materials flow network
Contractual relationship network
Network
density
Core size
Core
density
CTP density PTC density Network
density
Core size
Core
density
Periphery
density
PTC density
0.046
0.037
0.037
19
23
4
0.067
0.059
0.250
0.006
0.000
0.000
17
6
3
0.125
0.467
0.667
0.036
0.000
0.000
0.048
0.179
0.333
There are a comparatively less number of central firms in DCX’s
supply network. The implication is that the structure of the DCX
network is simpler (see Fig. 3) than those of Honda and Acura. For
one, there are no firms listed under demand load. This is because
every supplier in this network has only one customer, including
Textron and Leon, a top-tier and a key second-tier supplier, respectively. These two suppliers appear under both supply load and operational criticality. Both firms engage in value-adding activities by
integrating parts and facilitating their flows. The supply streams in
this supply network take place primarily through Textron or Leon.
6.1.2. Key firms in contractual relationship supply networks
In Table 5, CVT is again prominent on all centrality metrics in
Accord supply network. This firm appears as most influential on
the operation of the contractual relationship supply network, just
as it does in the materials flow network. Nonetheless, there are
a few notable differences. First, Honda does not appear at all in
Table 4, but in this network based on contractual relationships,
Honda emerges quite visibly (second to CVT) on all three columns.
This is because Honda maintains a contractual relationship with
many of its second- and third-tier suppliers (see Fig. 4). Second,
JFC, a top-tier supplier who appears in all three centrality metrics
in Table 4, is gone in Table 5. In other words, when it comes to
managing contracts, Honda emerges as central and JFC disappears.
Clearly, JFC is more isolated in the contractual relationship network.
For Acura supply network, Intek appears yet again as most central, while Honda emerges as central also. Thus, Intek looks like
most influential in the contractual relation network and none could
bypass Intek to connect with Honda. The network position allows
Intek to take control of information and communication flows. One
supplier for Acura that appears in Table 5 but did not in Table 4 is
HFI. HFI is a lone third-tier supplier that SNA picked up as being a
key firm under Informational Independence. This is largely because
0.064
0.043
0.250
0.074
0.066
0.074
HFI does business with other central firms such as Intek and Arkay,
and this is how it stays in the loop (see Fig. 5).
Unlike Accord and Acura, the list of firms that appear in Table 5
for DCX shows little change. There were two firms (Textron and
Leon) in Table 4 and the same firms appear again in Table 5. The
only exception is Daimler. Compared to the materials flow network,
the OEM is more prominent in the contractual relationship network
(Table 5), and this is due to its direct links with two third-tier suppliers, Irwin and E.R. Wagner (see Fig. 6). Daimler thus has leverage
over the relationships between these two suppliers and Textron,
the top-tier supplier.
6.2. Network-level implications
We now turn to discussing the dynamics at the network level.
Characterization we make below pertains to the whole supply networks based on Tables 6 and 7.
6.2.1. Characteristics of the materials flow supply networks
In Table 6, Accord’s supply network shows a comparatively
high density compared to the other two networks of Acura and
DCX. Accord’s supply network also features relatively high average scores on the key centrality metrics. Particularly, on average
betweenness, Accord’s lead is substantial. It implies that firms
in this supply network are more engaged in both delivering and
receiving materials than firms in other supply networks. It also
means that there are more steps required to move the materials
along. From an operational standpoint, it might indicate that this
network provides less efficiency (e.g., longer lead time, more parts
used for the same function) as it imposes more managerial attention on the firms in a central position. Looking at centralization
scores for Accord, indegree score stands out, suggesting the inflow
of materials is concentrated in a small group of firms in the supply network. We also note a rather large discrepancy in the scores
Table 9
Node-level overview.
Accord CVT, a 1st-tier suppliers, is most central, and assumes the most operational
burden on both supply and demand sides
JFC, another 1st-tier supplier, is not as much central as CVT
Honda appears not central in this network
HFI and C&C, two 2nd-tier suppliers, need to handle high degrees of supply
load and demand load, respectively
Two 2nd-tier suppliers (Emhart and Yamamoru), and one 3rd-tier (Fitzerald)
are also central as a go-between along the materials flow
CVT is most central under all three measures—operational flexibility,
managerial independence, and relational control
Honda emerges as the close second to CVT on all centrality metrics
JFC is extinct and becomes isolated in this network
Yamamoru, a 2nd-tier suppliers, emerges as central under managerial
independence
Emhart, a 2nd-tier supplier, is central under relational control
Acura
Intek, the sole 1st-tier supplier, is most central under both supply load and
operational criticality
Arkay, a 2nd-tier supplier, is central on every centrality metric
Iwata Bolt, a 2nd-tier supplier, is most central under
demand load
Honda is virtually out of sight in this network
Intek is again most central on all three
centrality
Honda is the close second to Intek on every centrality metric
HFI, a 3rd-tier supplier, emerges as key under managerial independence due to
its ties with other key suppliers
Two 2nd-tier suppliers, Arkay and Select Industries, rank consistently high on
all three centrality metrics
DCX
Textron, the sole 1st-tier supplier, and Leon, a 2nd-tier supplier, are most
central under both supply load and operational criticality
No central firm under demand load
Daimler is rather central only under supply load
Little change from materials flow
network
Textron and Leon are two most central on every centrality metrics
Daimler comes next but by a large margin on all three metrics
No other firms, than the three firms, appear as central in this network
206
Table 10
Network-level overview.
Accord Comparatively high overall density
Highest average score on all three centrality metrics
Relatively high centralization across all three types, and substantial lead on
average betweenness score
Much higher indegree centralization than the other two types
No connectivity among peripheral firms
Little reciprocity between the core and peripheral firms (much higher PTC
density than CTP density)
Peripheral firms engage solely in supplying to core firms
Relatively high overall density
Highest average betweenness but lowest average on closeness
Largest core group with low
density
Relatively high periphery density
Comparatively low PTC density
There are more interactions overall among more
members
Rather complex at the network level
Relatively less centralized
Acura
Comparatively large overall membership but with low density
Relatively low average scores on all the three centrality metrics
Comparatively low centralization indices
Very large core group with very low density
Virtually no materials flows among peripheral firms
No reciprocity between the core and the periphery firms
Network activities concentrated around the core group
Comparatively less complex at the network level
Largest overall membership but with lowest overall density
Lowest average betweenness score
More tightly coupled core group
No interactions among peripheral firms
Network activities mostly concentrated around the core group
Relatively high PTC density
Comparatively more centralized around the smaller core group
Comparatively less complex at the network level
DCX
Smallest membership with relatively low density
Comparatively high indegree centralization, but quite low outdegree
centralization
Smallest and tightly knit core group
No materials flows among peripheral firms
Largest discrepancy between PTC and CPT density among three SNs
Peripheral firms engage exclusively in supplying to the core firms
Relatively more centralized and least complex at the network level
Highest average scores on closeness centrality
Highest centralization
indices
Smallest core group with very high density
No interactions among peripheral firms
By far higher PTC density
Majority of network activities centers around the core group
Peripheral firms engage only in supplying to the core firms
Most centralized and least complex at the network level
of CTP density and PTC density, which signifies little reciprocity
between the core and peripheral firms. Further, the far-off firms do
not interact at all, as demonstrated in the periphery density of 0,
and this is true for all product types. In other words, the peripheral
firms engage solely in supplying to the core firms.
Acura’s supply network has comparatively large membership
but low overall density. The three average centrality scores are relatively low. Acura’s supply network, compared to Accord’s, has less
number of links and the overall steps required to get things done
are not as many, which may indicate higher operational efficiency.
Further, based on centralization scores, Acura’s supply network
appears as less centralized than Accord’s. At the local level, the core
group has very large membership but with relatively low density.
Since there are more firms in the core group, it suggests that the
power in the network is more spread out; the more flat structure of
operational authority again may be an indication that this network
works more efficiently (e.g., less time expended to make a decision
on issues at a local level).
DCX’s supply network has the smallest membership, and the
overall density is also relatively low. The centralization index based
on in-degree is comparatively high, as is the case with Accord and
Acura. However, notable difference occurs with out-degree centralization. It is quite low, indicating that most of the materials flow
out to few common dominant firms, and this observation is also
supported by the small size of the core group. There is also a huge
discrepancy between CTP density and PTC density, which simply
means that the majority of materials flow links in the network is
concentrated on a small number of firms. As expected, these firms in
the core group are tightly knit, as evidenced by a high core density.
Such simple structure can provide high operational efficiency at the
network-level (e.g., shorter lead time from upstream suppliers to
the final assembler); however, if multiple issues were to happen
simultaneously they could overwhelm the few central players and
could require much more time for resolution.
6.2.2. Characteristics of contractual relationship supply networks
The density for Accord is much higher in Table 7 than it was in
Table 6. This is because contracts can jump across several tiers. As
expected, the same thing happens for Acura and DCX as well. In
terms of centrality metrics, Accord’s supply network shows relatively low average closeness but high average betweenness scores.
Such a structure may be less responsive or more susceptible to
supply disruptions. It would possibly take more time channeling
information and there is a higher chance that information becomes
distorted on its way along the chains as more firms get involved in
transferring it. Therefore, such structure is likely to be less robust
or less effective when it comes to coping with supply disruptions.
By the same token, the structure would provide greater complexity
at the network level for Accord, as also evidenced by Accord’s relatively large core group size (see Table 7). Further, it has relatively
high periphery density, which further indicates that the network is
complex because there are more interactions going on even among
peripheral members. Still, more contacts among members at the
local level might facilitate identifying, if any, supply issues occurring locally.
Acura’s supply network shows relatively low overall density but
with large membership, which correspond with less number of
contractual links overall. Regarding average betweenness, this supply network shows the lowest score, indicating that this network
needs a smaller number of channels to get things done. Comparatively, therefore, this supply network appears as more efficient, for
instance, in managing such issues as supply disruptions because
communications at the network level can be comparatively faster
and more organized than those of Accord’s, which is also supported
by Acura’s comparatively more tightly knit core group and zero
periphery density.
Interestingly, DCX’s supply network shows the highest average
closeness score. In other words, the firms in the DCX’s network
are more readily reachable from each other, indicating that information can travel faster across the network. To put it differently,
the network structure is more conducive to the centralized control by dominant actors. As might be expected, this supply network
features the highest centralization indices among all three supply
networks. There is additional evidence for DCX’s high centralization
at the network level—the majority of the activities in the supply
network seem to center around a very small group of firms (i.e.,
three core firms) that are highly interwoven together (i.e., the highest core density of 0.667). Further, the firms in the periphery, with
no interactions among them, focus on catering to the core firms’
needs, evidenced by high PTC density. Because network information tends to spread relatively fast and converge at a small group
of dominant actors, the network as a whole would be comparatively more effective and robust when it comes to dealing with
supply disruptions. Particularly, active interactions between core
and periphery firms would further enhance such capability of the
supply network.
7. Discussion
7.1. Comparisons between SNA results and C&H study
7.1.1. Overlapping and divergent results
One of the main findings of C&H was the three OEMs’ varying
degrees of centralized control over their supply networks. The SNA
results confirm this. In particular, the final assemblers’ practice of
directed sourcing is captured in the contractual relationship network structure. For instance, the high values in Honda’s various
centralities and overall density in the contractual relationship network, compared to those in the materials flow network, is clearly
attributable to the added links that represent Honda’s directed
sourcing practice involving its second- and third-tier suppliers.
Another finding shared by both studies is the relational salience
of those tertiary-level suppliers in the network that are sourced
directly by OEMs. All of such suppliers (e.g., Emhart for Accord and
Iwata Bolt for Acura) emerge as visible in the contractual relationship network, through their exhibiting high scores on the various
centrality metrics or becoming a member of the core group in their
respective supply networks.
Divergent results between the two studies relate largely to
network-level properties such as network centralization and complexity. First, C&H describe Honda’s two supply networks as more
centralized than DCX’s. However, SNA suggests the opposite (see
Tables 6 and 7). In evaluating network centralization, C&H actually take the perspective of the final assemblers (i.e., Honda and
DCX). They present the argument that Honda is more centralized
compared to DCX because it has more direct ties with its suppliers
(i.e., top-tier as well as second- and third-tier suppliers)—Honda
has more centralized control of its supply networks. However,
SNA, in contrast, looks at how central all firms are in the supply
network, not just the final assembler. SNA evaluates the relative
node-level centrality scores of all the network members to arrive
at the indicators of network centralization. The two studies also
diverge when considering which supply network is most complex.
C&H suggest that Acura’s network is most complex. This judgment
is based on the network-level physical attributes (e.g., total number of entities, average geographical distance between companies)
and qualitative evidence regarding the lack of shared history and
the perceived level of decoupling among members. Contrarily, SNA
points to Accord’s network as being most complex. This is because
SNA focuses on how individual firms and their relationships are
connected to one another at the network level. For instance, SNA
considers various aspects of interdependence among members in
the network, such as network density, core density, periphery density, and PTC density.
The two studies, as such, draw different conclusions on some
aspects of supply network properties. Nonetheless, we want to
caution that this does not mean one is more accurate; rather, we
want to say that they just focus on different aspects of the same
phenomenon—the case approach focuses on contextual information, whereas SNA operates on numerical breakdown of data on
relative positions of members.
207
7.1.2. What C&H offer but SNA does not
C&H’s qualitative approach offers a contextually rich picture of
network dynamics. For instance, they make statements about the
network structure by drawing on such observations as Honda’s
strong penchant toward centralized policy with respect to supplier selection and product design and DCX’s practice of delegating
authority to the first-tier supplier as to who will be second-tier
suppliers and how to design the console. Further, the case method
can provide more detailed accounts of how the supply networks
operate and behave. For instance, in the Honda’s supply networks,
the second-tier suppliers selected directly by Honda tend to be
less cooperative with the top-tier supplier, which contributes to
furthering complexity at the network level; in the DCX’s network, Daimler commissions the top-tier supplier to consolidate
the second-tier suppliers to reduce operational complexity. Such
findings are context-specific and would be very difficult for SNA to
capture.
Also, C&H offer some propositions representing the overarching principles of the supply networks, derived from the qualitative
data. For instance, the study observes, “Formalized rules, norms,
and policies lead to the varying degrees of centralization in the
supply network . . .” (p. 488); “The cost consideration represents
the most salient force that shapes the emergence of the supplynetwork structure” (p.488); and “A centralized approach to supply
network involves a common list of core suppliers and the design
activities are tightly controlled by the final assembler” (p. 489).
Only from case-based qualitative studies could such propositions
be compiled. SNA would be unable to capture such contextually
rich information.
7.1.3. What SNA offers but C&H do not
Simply, SNA offers many quantitative metrics that qualitative
approaches cannot. By analyzing the structural characteristics of
supply networks, SNA brings us new intriguing results that would
likely be overlooked by qualitative methods. First, by producing
various network metrics, from node- to group- and to networklevel, SNA facilitated a comprehensive analysis of supply networks.
For instance, SNA evaluated differing roles of the individual nodes
and their relative importance with respect to others in the same
network (see Tables 4 and 5).
Second, SNA allowed for a comparative analysis of two different
network structures—materials flow and contractual relationship.
Between the two different network structures, we have observed
some divergent results even on the same network metrics (e.g.,
density, betweenness centrality). Those discrepancies, as noted
earlier, come from the fact that the two structures are constructed
based on different types of relational connection. Thus, it is not
proper to say that one type of link is a more accurate depiction of a
given network than the other; but rather the two different types of
network information should be considered jointly to fully understand a supply network. Further, SNA enabled a group-level analysis
by partitioning each supply network into two structurally distinct
clusters—core and periphery sub-groups. The core-periphery analysis in fact facilitated assessing network-level properties across
different supply networks (i.e., network centralization and network
complexity).
7.2. Academic contributions
Our goal in this paper has been to introduce SNA as a means
to analyze the structure of supply networks and draw theoretical
conclusions from such analysis. Our framework translates key SNA
metrics into the context of supply networks, and discusses how
roles of individual supply network members vary depending on
their relative structural position in the network. Subsequently, we
suggested a guideline as to how to identify central nodes and eval-
208
uate them differently. Central firms require possessing a particular
set of capabilities corresponding to the roles they assume in the
network (see Table 1). For instance, firms with high in-degree centrality should focus on developing a capability in system integration
or product architectural innovation (Parker and Anderson, 2002;
Violino and Caldwell, 1998); firms with high betweenness centrality may be in a better position to engage in supply risk management.
Thus, it would be prudent for a buying company (e.g., OEMs), when
selecting or developing a supplier, to consider these issues. We
hope that the theoretical framework of this study would be instrumental in facilitating future supply network research adopting SNA
approach.
The paper’s methodological contribution is two-fold. First, this
study demonstrates the value of SNA in studying supply networks.
SNA considers all member firms in a given supply network to
determine which firms are most important, in what aspect, to
the operation of the whole network. Capitalizing on computating power, SNA can generate various analytic outputs reflecting
either individual- or group-level behavioral dynamics, which in
fact facilitate gaining a more comprehensive and systematic view of
network dynamics. Second, applying the widely accepted networklevel analytical concepts (i.e., network density, centralization, and
core-periphery), SNA can complement qualitative methods in capturing the structural intricacy of the whole network in a more
objective way. As has been demonstrated, SNA has considerable
potential for enhancing our studies of supply networks (Borgatti
and Li, 2009; Carter et al., 2007) and can effectively complement
qualitative methods.
7.3. Managerial contributions
Based on C&H’s data, our study brings to the fore the salience of
two types of supply networks—materials flow and contractual relationship. We propose that managers consider these two types for
any given supply network, as we have demonstrated how the two
networks organize and behave differently. For instance, in Acura’s
supply networks, the size of the core group becomes much larger
when based on materials flow than the contractual relationship (see
Tables 6 and 7 for comparison). Also, managers should note that
there can be different sets of key firms between the two types of
supply network (see Tables 4 and 5). One firm that does not appear
as central in one type (e.g., HFI in the Acura network) may be a
key player in another. Depending on which type of link to focus on,
individual suppliers’ position of importance and the strategic roles
will vary. For instance, the key firms in a materials flow network
can have a considerable effect on the operational quality of overall
supply network, affecting lead time, product quality, OEM’s inventory level, or stockout costs (Bourland et al., 1996). Key suppliers
in a contractual relation network could facilitate the timely indentification or resolution of those system-level operational problems
and other supply disruption risks (Lee, 2002).
Further, it may be prudent for a manufacturing firm to identify
central second- or third-tier suppliers using SNA. Some of these
suppliers become a key player by being linked to more visible other
key firms in the supply network. In other words, some tertiarylevel suppliers emerge as important because they are vital to other
more prominent suppliers in supply networks. We anticipate these
second- and third-tier suppliers that previously went unnoticed
will play a more significant role in future. As the issue of supply
chain scalability takes the center stage for safety and sustainability,
large final assemblers are moving toward identifying and managing
key tertiary-level suppliers. Collecting complete supply network
data and applying SNA, as we have done in this paper, may serve
as a useful approach.
In general, having a pictorial rendition of a supply network will
be useful to managers. SNA can help generate network sociograms
(see Figs. 1–6). As a visual embodiment of relationship patterns in
supply networks, these sociograms can be instrumental in attaining
a realistic picture of networking patterns and the dynamics. Just as
all graphs, network drawings can help save search efforts, facilitate
recognition, and provide interesting new perspectives and insights
into supply networks. Also, SNA provides a methodological frame
for collecting and organizing data, which will be useful for planning and monitoring changes in the operation of supply networks.
The position of a node in the network affects the opportunities and
constraints of that node and of others (Gulati et al., 2000; Rowley,
1997).
7.4. Limitations and future directions
Our study represents a very first step in theorizing and empirically investigating supply networks using SNA concepts. We
acknowledge that our study is limited in ways that suggest opportunities for future research. First, our analysis is confined to a
specific automobile module (i.e., center console assembly). Any one
supplier in the supply network might be involved in several overlapping supply networks across different product lines. A supplier’s
role based on one supply network will look quite different from that
derived by considering the multiple supply networks together it is
a member of. Therefore, the central roles a supplier plays in our
analysis should be qualified to the single product line. It would not
be reasonable to consider the results of our analysis as a general
statement regarding that supplier.
In a similar vein, supply networks are considered basically
“egocentric”—centered around a focal actor (Håkansson and Ford,
2002; Mizruchi and Marquis, 2006). The three supply networks
studied here were also mapped based on information obtained
from the final assemblers. Therefore, any possible effect each
supplier’s extended network can have on the firm’s strategic
importance to the OEM could not be captured in our analysis. For
instance, one second tier supplier to Honda may have a tie to other
OEMs. If such extended ties were also counted, certain centrality
metrics (e.g., betweenness) for the supplier might have shown different scores from those based on the egocentric network, whereby
placing the supplier in a different strategic position with respect to
Honda. Such egocentric network approach, albeit considered a reliable substitute for complete (sociocentric) network data (Marsden,
2002), may not be enough to provide a full understanding or potential of a given supplier, embedded within the larger social network
(Mizruchi and Marquis, 2006).
Third, in quantifying the inter-firm ties, we did not consider the
variances in strength. All the links considered in our analysis were
treated as having the same weight, while the link an OEM has with
the first-tier suppliers should involve more intensive information
exchanges (i.e., kanban system) or a greater amount of materials
(i.e., larger contract size) than those with the second-tier firms, for
instance. Also, we viewed supply networks based on the materials flow and contract connections. However, certainly there are
many other relational connection types that can be considered
in supply networks, such as ownership, technology dependence,
intellectual property, and risk sharing. Network ties could be represented by the number of joint programs or of shared patents, level
of trust, or perceived transactional risks. Future studies therefore
can incorporate the relative strength of supply ties using SNA as the
method can effectively illustrate networks with “weighted” links
(Borgatti and Li, 2009; Battini et al., 2007). Exchange ties involving a multi-level interface will have differential impact compared
to other comparable supply ties based only on a single type of
transaction.
We note that most supply networks are considered a scalefree network, whose degree distribution closely follows a power
law (Albert and Barabási, 2002; Pathak et al., 2007). That is, most
nodes have very few links and only a small number of nodes
(e.g., core firms) have many connections. Future studies may apply
the scale-free network metrics to studying supply networks, such
as clustering coefficient and characteristic path length. Clustering
coefficient measures the degree to which nodes in a network tend
to cluster together around a given node (Barabási et al., 2002), and
it can inform us of how suppliers behave with respect to the final
assembler at both the local and the global level. For instance, it
can tell us how suppliers would come together for better coordination, based on some governance mechanism involving an OEM.
Indicating the system-level “closeness,” characteristic path length
can assist in evaluating whether a given supply network is optimally designed (Braha and Bar-Yam, 2004; Lovejoy and Loch, 2003).
Given a supply network, it can be of considerable interest to know
how the path length compares to the “best” or “worst” possible
configuration for networks with the same number of nodes and
lines. This can provide implications for how effectively the network is designed and how robust it can be to possible supply
disruptions.
Finally, SNA could be applied to advancing existing theories
regarding the structure or topology of supply networks. A range
of SNA metrics can serve as a useful means in this effort. Such
network variables as density and various centralities could be applicable to characterizing typological archetypes of supply network
structures, eventually leading to the development of a portfolio
of contingent approaches to supply management. In conclusion,
we hope that this paper can serve as a call to other operations
and supply management researchers regarding the importance
of framing supply chains as networks and continuing to develop
useful supply network indices. We hope to see more researchers
taking advantage of the usefulness of SNA for untangling and
understanding the complex phenomena embedded in supply
networks.
References
Abrahamson, E., Rosenkopf, L., 1997. Social network effects on the extent of innovation diffusion: a computer simulation. Organization Science 8 (3), 289–309.
Ahuja, G., Polidoro Jr., F., Mitchell, W., 2009. Structural homophily or social
asymmetry? The formation of alliances by poorly embedded firms. Strategic
Management Journal 30 (9), 941–958.
Albert, R., Barabási, A.L., 2002. Statistical mechanics of complex networks. Reviews
of Modern Physics 74, 47–97.
Autry, C.W., Griffis, S.E., 2008. Supply chain capital: the impact of structural and relational linkages on firm execution and innovation. Journal of Business Logistics
29 (1), 157–174.
Barabási, A.L., Jeong, H., Néda, Z., Ravasz, E., Schubert, A., Vicsek, T., 2002. Evolution of
the social network of scientific collaborations. Physica A: Statistical Mechanics
and its Applications 311 (3–4), 590–614.
Battini, D., Persona, A., Allesina, S., 2007. Towards a use of network analysis:
quantifying the complexity of Supply Chain Networks. International Journal of
Electronic Customer Relationship Management 1 (1), 75–90.
Bavelas, A., 1950. Communication patterns in task-oriented groups. Journal of the
Acoustical Society of America 22, 271–282.
Beauchamp, M.A., 1965. An improved index of centrality. Behavioral Science 10,
161–163.
Berry, D., Towill, D.R., Wadsley, N., 1994. Supply chain management in the electronics
product industry. International Journal of Physical Distribution and Logistics
Management 24 (10), 20–32.
Bonacich, P., 1972. Factoring and weighting approaches to status scores and clique
identification. Journal of Mathematical Sociology 2, 113–120.
Borgatti, S.P., Everett, M.G., 2000. Models of core/periphery structures. Social networks 21 (4), 375–395.
Borgatti, S.P., Everett, M.G., 2006. A graph-theoretic perspective on centrality. Social
Networks 28 (4), 466–484.
Borgatti, S., Everett, M., Freeman, L., 2002. UCINET 6 for Windows: Software for Social
Network Analysis. Analytic Technologies, Inc., Natick, MA.
Borgatti, S.P., Li, X., 2009. On social network analysis in a supply chain context.
Journal of Supply Chain Management 45 (2), 5–21.
Bourland, K.E., Powell, S.G., Pyke, D.F., 1996. Exploiting timely demand information to reduce inventories. European Journal of Operational Research 92 (2),
239–253.
Braha, D., Bar-Yam, Y., 2004. Information flow structure in large-scale product
development organizational networks. Journal of Information Technology 19,
244–253.
209
Burkhardt, M.E., Brass, D.J., 1990. Changing patterns or patterns of change: the effects
of a change in technology on social network structure and power. Administrative
Science Quarterly 35, 104–127.
Burt, R.S., 1980. Innovation as a structural interest: rethinking the impact of network
position on innovation adoption. Social Networks 2 (4), 327–355.
Burt, R.S., 1982. Toward a Structural Theory of Action. Academic Press, New York.
Burt, R.S., 1992. Structural Holes: The Social Structure of Competition. Harvard University Press, Cambridge.
Burt, R.S., 1998. The gender of social capital. Rationality and Society 10 (1), 5–46.
Butts, C.T., 2001a. An axiomatic approach to network complexity. Journal of Mathematical Sociology.
Butts, C.T., 2001b. The complexity of social networks: theoretical and empirical
findings. Social Networks 23 (1), 31–72.
Cachon, G.P., 2003. Supply chain coordination with contracts. In: de Kok, A.G., Graves,
S.C. (Eds.), Handbooks in Operations Research and Management Science, vol. 11.
Elsevier, Boston, pp. 229–340.
Cachon, G., Fisher, M., 2000. Supply chain inventory management and the value of
shared information. Management Science 46 (8), 1032–1048.
Cachon, G.P., Lariviere, M.A., 2005. Supply chain coordination with revenue-sharing
contracts: strengths and limitations. Management Science 51 (1), 30–44.
Carter, C.R., Ellram, L.M., Tate, W., 2007. The use of social network analysis in logistics
research. Journal of Business Logistics 28 (1), 137–169.
Cassiman, B., Veugelers, R., 2002. Complementarity in the innovation strategy: Internal R&D, external technology acquisition and cooperation. Discussion paper
series No. 3284. Centre for Economic Policy Research, Cambridge.
Chen, F., Dresner, Z., Ryan, J.K., Simchi-Levi, D., 2000. Quantifying the bullwhip effect
in a simple supply chain: the impact of forecasting, lead times, and information.
Management Science 46 (3), 436–443.
Choi, T.Y., Dooley, K.J., Rungtusanatham, M., 2001. Supply networks and complex
adaptive systems: control versus emergence. Journal of Operations Management
19 (3), 351–366.
Choi, T.Y., Hong, Y., 2002. Unveiling the structure of supply networks: case studies
in Honda, Acura, and DaimlerChrysler. Journal of Operations Management 20
(5), 469–493.
Choi, T.Y., Liker, J.K., 1995. Bringing Japanese continuous improvement approaches
to U.S. manufacturing: the roles of process orientation and communications.
Decision Sciences 26 (5), 589–620.
Choi, T.Y., Krause, D.R., 2006. The supply base and its complexity: implications for
transaction costs, risks, responsiveness, and innovation. Journal of Operations
Management 24 (5), 637–652.
Choi, T.Y., Kim, Y., 2008. Structural embeddedness and supplier management: a
network perspective. Journal of Supply Chain Management 44 (4), 5–13.
Choi, T.Y., Wu, Z., 2009. Triads in supply networks: theorizing
buyer–supplier–supplier relationships. Journal of Supply Chain Management
45 (1), 8–25.
Chopra, S., Sodhi, M.S., 2004. Managing risk to avoid supply-chain breakdown. MIT
Sloan Management Review 46 (1), 53–62.
Coleman, J.S., 1973. The Mathematics of Collective Action. Aldine, Chicago.
Cook, W.J., Cunningham, W.H., Pullyblank, W.R., Schrijver, A., 1998. Combinatorial
Optimization. Wiley, New York.
Cooper, M.C., Lambert, D.M., Pagh, J.D., 1997. Supply chain management: more than
a new name for logistics. The International Journal of Logistics Management 8
(1), 1–14.
Corbett, C.J., Blackburn, J.D., Van Wassenhove, L.N., 1999. Partnerships to improve
supply chains. Sloan Management Review 40 (4), 71–82.
Cox, A., Sanderson, J., Watson, G., 2006. Supply chains and power regimes: toward
an analytic framework for managing extended networks of buyer and supplier
relationships. Journal of Supply Chain Management 37 (2), 28–35.
Croxton, K.L., García-Dastugue, S.J., Lambert, D.M., Rogers, D.S., 2001. The supply
chain management processes. The International Journal of Logistics Management 12 (2), 13–36.
DiMaggio, P., Louch, H., 1998. Socially embedded consumer transactions: for what
kinds of purchases do people most often use networks? American Sociological
Review 63, 619–637.
Dyer, J.H., Nobeoka, K., 2000. Creating and managing a high-performance
knowledge-sharing network: the Toyota case. Strategic Management Journal
21 (3), 345–367.
Ebel, H., Davidsen, J., Bornholdt, S., 2002. Dynamics of social networks. Complexity
(Special Issue: Complex Adaptive Systems: Part I) 8 (2), 24–27.
Ellram, L.M., Tate, W.L., Carter, C.R., 2006. Product-process-supply chain: an integrative approach to three-dimensional concurrent engineering. International
Journal of Physical Distribution & Logistics Management 37 (4), 305–330.
Everett, M.G., 1985. Role similarity and complexity in social networks. Social Networks 7, 353–359.
Everett, M.G., Borgatti, S.P., 1999. The centrality of groups and classes. Journal of
Mathematical Sociology 23 (3), 181–201.
Ferguson, R.J., Paulin, M., Bergeron, J., 2005. Contractual governance, relational governance, and the performance of interfirm service exchanges: the influence of
boundary-spanner closeness. Journal of the Academy of Marketing Science 33
(2), 217–234.
Ford, I.D., 1990. Understanding Business Markets: Interactions, Relationships and
Networks (Ed.). Academic Press, San Diego, CA.
Freeman, L.C., 1977. A set of measures of centrality based on betweenness. Sociometry 40, 35–41.
Freeman, L.C., 1979. Centrality in social networks: conceptual clarification. Social
Networks 1, 215–239.
210
Freeman, L.C., 1980. The gatekeeper, pair-dependency and structural centrality.
Quality and Quantity 14 (4), 585–592.
Freeman, L.C., 1983. Spheres, cubes, and boxes: graph dimensionality and network
structure. Social Networks 5, 139–156.
Frenken, K., 2000. A complexity approach to innovation networks. The case of the
aircraft industry (1909–1997). Research Policy 29 (2), 257–272.
Frohlich, M.T., Westbrook, R., 2002. Demand chain management in manufacturing and services: web-based integration, drivers and performance. Journal of
Operations Management 20, 729–745.
Glanzer, M., Glaser, R., 1959. Techniques for the study of group structure and behavior. I. Analysis of structure. Psychological Bulletin 56, 317–331.
Greve, H.R., 2009. Bigger and safer: the diffusion of competitive advantage. Strategic
Management Journal 30 (1), 1–23.
Grover, V., Malhotra, M.D., 2003. Transaction cost framework in operations and
supply chain management research: theory and measurement. Journal of Operations Management 21 (4), 457–473.
Gulati, R., 1995. Social structure and alliance formation patterns: a longitudinal
analysis. Administrative Science Quarterly 40, 619–652.
Gulati, R., 1999. Network location and learning: the influence of network resources
and firm capabilities on alliance formation. Strategic Management Journal 20
(5), 397–420.
Gulati, R., Nohria, N., Zaheer, A., 2000. Strategic networks. Strategic Management
Journal 21 (special issue), 203–215.
Håkansson, H. (Ed.), 1982. International Marketing and Purchasing of Industrial
Goods: An Interaction Approach. Wiley, Chichester.
Håkansson, H. (Ed.), 1987. Industrial Technological Development: A Network
Approach. Croom Helm, London, UK.
Håkansson, H., Ford, D., 2002. How should companies interact in business networks?
Journal of Business Research 55 (2), 133–139.
Håkansson, H., Snehota, I., 1995. Developing Relationships in Business Networks.
Routledge, London, England.
Hakimi, S.L., 1965. Optimum locations of switching centers and the absolute centers
and medians of a graph. Operations Research 12, 450–459.
Harland, C.M., Lamming, R.C., Cousins, P.D., 1999. Developing the concept of supply
strategy. International Journal of Operations & Production Management 19 (7),
650–673.
Harland, C.M., Lamming, R.C., Zheng, J., Johnsen, T.E., 2001. A taxonomy of supply
networks. The Journal of Supply Chain Management 37 (4), 21–27.
Henderson, R., Clark, K., 1990. Architectural Innovation: the reconfiguration of existing product technologies and the failure of established firms. Administrative
Science Quarterly 35, 9–30.
Hendricks, K.B., Singhal, V.R., 2003. The effect of supply chain glitches on shareholder
value. Journal of Operations Management 21 (5), 501–522.
Hendricks, K.B., Singhal, V.R., 2005. An empirical analysis of the effect of supply
chain disruptions on long-run stock price performance and equity risk of the
firm. Production and Operations Management 14 (1), 35–52.
Human, S., Provan, K., 1997. An emergent theory of structure and outcomes in smallfirm strategic manufacturing networks. Academy of Management Journal 40 (2),
369–403.
Hwarng, H.B., Chong, C.S.P., Xie, N., Burgess, T.F., 2005. Modelling a complex supply
chain: understanding the effect of simplified assumptions. International Journal
of Production Research 43 (13), 2829–2872.
Iansiti, M., 2000. How the incumbent can win: managing technological
transitions in the semiconductor industry. Management Science 46 (2),
169–185.
Ibarra, H., 1993. Network centrality, power, and innovation involvement: determinants of technical and administrative roles. Academy of Management Journal
36, 471–501.
Jarillo, J.C., Stevenson, H.H., 1991. Co-operative strategies: the payoffs and the pitfalls. Long Range Planning 24 (1), 64–70.
Jensen, M., 2003. The role of network resources in market entry: commercial banks’
entry into investment banking, 1991–1997. Administrative Science Quarterly
48 (3), 466–497.
Kauffman, S.A., 1993. The Origins of Order: Self-Organization and Selection in Evolution. Oxford University Press, New York.
Ketchen Jr., D.J., Hult, G.T.M., 2007. Bridging organization theory and supply chain
management: the case of best value supply chains. Journal of Operations Management 25 (2), 573–580.
Kim, W.S., 2009. Effects of a trust mechanism on complex adaptive supply networks:
an agent-based social simulation study. Journal of Artificial Societies and Social
Simulation 12 (3), 4.
Kim, T.Y., Oh, H., Swaminathan, A., 2006. Framing interorganizational network
change: a network inertia perspective. Academy of Management Review 31,
704–720.
Kircherr, W., 1992. Kolmogorov complexity and random graphs. Information Processing Letters 41 (3).
Kleindorfer, P.R., Saad, G.H., 2005. Disruption risk management in supply chains.
Production and Operations Management 14 (1), 53–68.
Klovdahl, A.S., 1985. Social networks and the spread of infectious diseases: the AIDS
example. Social Science & Medicine 21 (11), 1203–1216.
Koehly, M.L., Peterson, K.S., Watts, G.B., Kempf, Kri, K.G.K., Vernon, W.S., Gritz,
R.E., 2003. A social network analysis of communication about hereditary nonpolyposis colorectal cancer genetic testing and family functioning. Cancer
Epidemiology, Biomarkers & Prevention 12, 304–313.
Kotabe, M., Martin, X., Domoto, H., 2002. Gaining from vertical partnerships: knowledge transfer, relationship duration and supplier performance improvement in
the US and Japanese automotive industries. Strategic Management Journal 24
(4), 293–316.
Krackhardt, D., 1990. Assessing the political landscape: structure, cognition, and
power in organizations. Administrative Science Quarterly 92, 142–154.
Kumar, R., Novak, J., Tomkins, A., 2006. Structure and evolution of online social
networks. In: Proceedings of the 12th ACM SIGKDD International Conference
, Philadelphia, PA, USA, pp. 611–617.
Lamming, R.C., Johnsen, T.E., Zheng, J., Harland, C.M., 2000. An initial classification of
supply networks. International Journal of Operations & Production Management
20 (6), 675–691.
Lazzarini, S.G., Chaddad, F.R., Cook, M.L., 2001. Integrating supply chain and network
analyses: the study of netchains. Journal on Chain and Network Science 1 (1),
7–22.
Leavitt, H.J., 1951. Some effects of communication patterns on group performance.
Journal of Abnormal and Social Psychology 46, 38–50.
Lee, H.L., 2002. Aligning supply chain strategies with product uncertainties. California Management Review 44 (3), 104–119.
Lee, H.L., 2004. The triple—a supply chain. Harvard Business Review 83, 102–112.
Lee, H.L., Padmanabhan, V., Whang, S., 1997. Information distortion in a supply chain:
the bullwhip effect. Management Science 43 (4), 546–558.
Lee, H.L., So, R., Tang, C.S., 2000. The value of information in a two-level supply chain.
Management Science 46 (5), 626–643.
Li, M., Vitányi, P.M.B., 1991. Combinatorics and Kolmogorov complexity. In: Proceedings of the Sixth Annual IEEE Conference on Structure in Complexity Theory.
Springer Verlag, Berlin, New York.
Lokam, V.S., 2003. Graph complexity and slice functions. Theory of Computing Systems 36, 71–88.
Lovejoy, W.S., Loch, C.H., 2003. Minimal and maximal characteristic path lengths in
connected sociomatrices. Social Networks 25, 333–347.
Luce, R.D., Perry, A.D., 1949. A method of matrix analysis of group structure. Psychometrika 14, 95–116.
Marczyk, J., 2006. Measuring and tracking complexity. In: Proceedings of the Sixth
International Conference on Complex Systems , June 25–30. New England Complex Systems Institute, Boston, MA, USA.
Marsden, P.V., 2002. Egocentric and sociocentric measures of network centrality.
Social Networks 24 (4), 407–422.
Mills, J., Schmitz, J., Frizelle, G., 2004. A strategic review of “supply networks”. International Journal of Operations & Production Management 24 (9/10), 1012–1036.
Mizruchi, S.M., 1994. Social network analysis: recent achievements and current
controversies. Acta Sociologica 37 (4), 329–343.
Mizruchi, S.M., Marquis, C., 2006. Egocentric, sociocentric, or dyadic? Identifying
the appropriate level of analysis in the study of organizational networks. Social
Networks 28, 187–208.
Nieminen, J., 1973. On the centrality in a directed graph. Social Science Research 2,
371–378.
Nishiguchi, T., 1994. Strategic Industrial Sourcing. Oxford University Press,
Oxford/New York.
Nobeoka, K., 1996. Alternative component sourcing strategies within the
manufacturer-supplier network: benefits of quasi-market strategy in the
Japanese automobile industry. Kobe Economic and Business Review 41,
69–99.
North, M.J., Macal, C.M., 2007. Managing Business Complexity: Discovering Strategic
Solutions with Agent-based Modeling and Simulation. Oxford University Press,
New York.
Park, S., Hartley, J.L., 2002. Exploring the effect of supplier management on performance in the Korean automotive supply chain. Journal of Supply Chain
Management 38 (2), 46–53.
Parker, G.G., Anderson, E.G., 2002. From buyer to integrator: the transformation of
the supply-chain manager in the vertically disintegrating firm. Production and
Operations Management 11 (1), 75–91.
Pathak, S.D., Day, J.M., Nair, A., Sawaya, W.J., Kristal, M.M., 2007. Complexity and
adaptivity in supply networks: building supply network theory using a complex
adaptive systems perspective. Decision Sciences 38 (4), 547–580.
Pollock, T.G., Porac, J.F., Wade, J.B., 2004. Constructing deal networks: brokers as
network “architects” in the US IPO market and other examples. The Academy of
Management Review 29 (1), 50–72.
Powell, W.W., Koput, K.W., Smith-Doerr, L., 1996. Interorganizational collaboration
and the locus of innovation: networks of learning in biotechnology. Administrative Science Quarterly 41 (1), 116–145.
Proctor, C.H., Loomis, C.P., 1951. Analysis of sociometric data. In: Jahoda, M., Deutsch,
M., Cook, S.W. (Eds.), Research Methods in Social Relations. Dryden Press, New
York, pp. 561–586.
Provan, K.G., 1983. The federation as an interorganizational network. Academy of
Management Review 8 (1), 79–89.
Provan, K.G., Milward, H.B., 1995. A preliminary theory of network effectiveness: a
comparative study of four community mental health systems. Administrative
Science Quarterly 40 (1), 1–33.
Pudlák, P., Rödl, V., 1992. A combinatorial approach to complexity. Combinatorica
14, 221–226.
Pudlák, P., Rödl, V., 1994. Some combinatorial-algebraic problems from complexity
theory. Discrete Mathematics 136, 253–279.
Pudlák, P., Rödl, V., Savický, P., 1988. Graph complexity. Acta Informatica 25,
515–535.
Robins, G., Alexander, M., 2004. Small worlds among interlocking directors: network structure and distance in bipartite graphs. Computational & Mathematical
Organization Theory 10, 69–94.
Rowley, T.J., 1997. Moving beyond dyadic ties: a network theory of stakeholder
influences. The Academy of Management Review 22 (4), 887–910.
Rowley, T.J., Baum, J.A.C., Greve, H.R., Rao, H., Shipilov, A.V., 2005. Time to break
up: social and instrumental antecedents of firm exits from exchange cliques.
Academy of Management Journal 48 (3), 499–520.
Sabidussi, G., 1966. The centrality index of a graph. Psychometrika 31, 581–603.
Samaddar, S., Nargundkar, S., Daley, M., 2006. Inter-organizational information sharing: the role of supply network configuration and partner goal congruence.
European Journal of Operational Research 174 (2), 744–765.
Scott, J.P., 1986. Capitalist Property and Financial Power. Brighton, Wheatsheaf.
Scott, J., 2000. Social Network Analysis: A Handbook. Sage Publications, London.
Shaw, M.E., 1954. Group structure and the behavior of individuals in small groups.
Journal of Psychology 38, 139–149.
Spencer, J.W., 2003. Global gatekeeping, representation, and network structure:
a longitudinal analysis of regional and global knowledge-diffusion networks.
Journal of International Business Studies 34 (5), 428–442.
Stam, W., Elfring, T., 2008. Entrepreneurial orientation and new venture performance: the moderating role of intra- and extra-industry social capital. The
Academy of Management Journal 51 (1), 97–111.
Surana, A., Kumara, S., Greaves, M., Raghavan, U.N., 2005. Supply chain network:
a complex adaptive systems perspective. International Journal of Production
Research 43, 4235–4265.
Tichy, N.M., Tushman, M.L., Fombrun, C., 1979. Social network analysis for organizations. Academy of Management Review 4 (4), 507–519.
211
Uzzi, B., 1997. Social structure and competition in inter-firm networks: the paradox
of embeddedness. Administrative Science Quarterly 42, 35–67.
Valente, T.W., 1996. Social network thresholds in the diffusion of innovations. Social
Networks 18 (1), 69–89.
Violino, B., Caldwell, B., 1998. Analyzing the integrators. InformationWeek 709
(November), 45–69.
Wallman, S., 1984. Eight London Households. Tavistock, London.
Wasserman, S., Faust, K., 1994. Social Network Analysis: Methods and Applications.
Cambridge University Press, Cambridge.
Wilding, R., 1998. The supply chain complexity triangle: uncertainty generation in
the supply chain. International Journal of Physical Distribution & Logistics 28
(8), 599–616.
Wu, Z., Choi, T.Y., 2005. Supplier–supplier relationships in the buyer–supplier triad:
building theories from eight case studies. Journal of Operations Management 24
(1), 27–52.
Zack, M.H., McKenney, J.L., 1995. Social context and interaction in ongoing
computer-supported management groups. Organization Science 6 (4), 394–422.
Zaheer, A., Bell, G.G., 2005. Benefiting from network position: firm capabilities, structural holes, and performance. Strategic Management Journal 26 (9),
809–825.
Zhu, Q., Sarkis, J., 2004. Relationships between operational practices and performance among early adopters of green supply chain management practices in
Chinese manufacturing enterprises. Journal of Operations Management 22 (3),
265–289.
MANAGEMENT SCIENCE
informs
Vol. 52, No. 11, November 2006, pp. 1737–1750
issn 0025-1909 eissn 1526-5501 06 5211 1737
®
doi 10.1287/mnsc.1060.0582
© 2006 INFORMS
A Typology of Plants in Global
Manufacturing Networks
Ann Vereecke, Roland Van Dierdonck
Vlerick Leuven Gent Management School, and Faculty of Economics and Business Adminstration,
Ghent University, Reep 1, B-9000 Gent, Belgium {[email protected], [email protected]}
Arnoud De Meyer
Judge Business School, Cambridge University, Trumpington Street, Cambridge CB2 1AG, United Kingdom,
[email protected]
T
he purpose of this paper is to propose a new, empirically derived typology of plants in the international
manufacturing network of multinational companies. This typology is based on the knowledge flows between
the plants. In our research, network analysis has been used as a methodology for understanding the position
of plants in international manufacturing networks. The focus has been primarily on the intangible knowledge
network, and secondarily on the physical, logistic network. Our analysis leads to four types of plants with
different network roles: the isolated plants, the receivers, the hosting network players, and the active network
players. Our analysis shows that the different types of plants play a different strategic role in the company, have
a different focus, and differ in age, autonomy, and level of resources and investments. Also, the analysis suggests
that the evolution of the plant depends to some extent on the network role of the plant. Finally, two scenarios
for the development of a strong network role are identified. The research is useful for the scholar studying the
architecture of knowledge networks, as well as for the practitioner who is in charge of an international network
of manufacturing units.
Key words: manufacturing strategy; knowledge management; international manufacturing; plant networks
History: Accepted by William S. Lovejoy, operations and supply chain management; received May 23, 2002.
This paper was with the authors 2 years and 1 month for 3 revisions.
1.
Introduction
The basic question here is how to design and manage the flows of goods, people, technology, and information in international networks (Chakravarty et al.
1997). Our research contributes to this second category of international operations research.
Competitiveness today is not solely based on the
application of state-of-the-art management techniques
in each of the individual plants, but also on the implementation of an integrative strategy on the network
of plants (Ferdows 1997a). From a logistics perspective, this requires the optimization of the company’s
supply chain. From an organizational perspective, it
requires managing the creation and transfer of knowledge in the network. Plants adopt a different role in
these networks. As plants differ in product allocation
and in focus, they play different roles in the supply
chain (Hayes and Schmenner 1978). As they differ
in the level of creation, sharing, and absorption of
innovations, they play different roles in the intangible
knowledge network in the company (Ferdows 1997b).
The purpose of our research has been to understand
the different roles of plants in this knowledge network. Based on rigorous and in-depth case research,
a new typology of plants has been derived. The plant
types differ in the extent to which they share inno-
In 1964, Skinner warned, “the time has come when
we must begin to sharpen the management of international manufacturing operations” (Skinner 1964,
p. 126). As competition is globalizing and the complexity of the environment in which companies operate is increasing, managing an integrated international network has become an increasingly important task for manufacturing managers (Bartlett and
Ghoshal 1989, Ferdows 1997a). However, despite the
importance attached to it by both academics and practitioners, the field of international operations management is still at a relatively early stage of theory development (Roth et al. 1997) and could be enriched by
insights from empirical research (Chakravarty et al.
1997).
In the field of international operations management, at least two categories of research can be distinguished (Chakravarty et al. 1997). The first category of
research consists mainly of international comparisons.
The basic question here is to what extent models and
concepts in production and operations management
are applicable in different countries or regions. The
second category studies the management of international networks of facilities, suppliers, and markets.
1737
Vereecke et al.: Typology of Plants in Global Manufacturing Networks
1738
vations with the other plants, in the level of visits to
and from the other plants, and in the level of communication with the other plants. The analysis also
shows that different roles in the knowledge network
coincide with different roles in the supply chain.
2.
2.1.
Literature Review
Operations in a Multinational: A Network
Perspective
Over the last two decades, research on the structure
and organization of multinationals has shifted from
a focus on the one-to-one headquarters-subsidiaries
relationships toward a focus on managing a network
of units (Kogut 1989). Ghoshal and Bartlett (1990,
p. 620) claim that the network approach “is particularly suited for the investigation of such differences
in internal roles, relations, and tasks of different affiliated units and of how internal co-ordination
mechanisms might be differentiated to match the variety of sub-unit contexts.”
In the management of these networks, the focus
has often been on the flow of information. Doz and
Prahalad (1991, p. 160), for example, state that differences in the mission of subsidiaries are reflected in
the “pattern and intensity of information flows.” In
their more recent work Doz et al. (2001) argue that the
success of some multinational companies lays in their
ability to “sense” information and knowledge and to
distribute it rapidly throughout the network.
The information flow is only one type of network
relationship between the subsidiaries and headquarters, and among the subsidiaries. The physical flow
of components, semifinished goods or end products,
financial flows, and “flows” of people moving around
in the network are other types of network relationships (Bartlett and Ghoshal 1989).
This trend toward describing the multinational company as a network of units can also be observed in
the manufacturing strategy literature. Work has been
done, for example, in the description of the benefits
and methods of the transfer of best practices across
the manufacturing network. Chew et al. (1990) show
that the improvement of the overall performance of
multisite companies depends on the local innovativeness of the plants, as well as on the interplant transfer
of these local innovations. Flaherty (1986, 1996) adds
to this the importance of coordination. She argues that
the coordination of international operations in a network can improve cost and delivery performance and
enhance the learning from the experiences of units in
the network.
However, the systematic analysis of the relationship
between the plants in the manufacturing network
requires an appropriate methodology. Nohria (1992,
p. 8) claims that, “if we are to take a network perspective seriously, it means adopting a different intellectual lens and discipline, gathering different kinds
of data, learning new analytical and methodological
techniques, and seeking explanations that are quite
different from conventional ones.” Network analysis is a particularly powerful methodology for the
description and analysis of the structure of networks
and the position of the units in the network (Knoke
and Kuklinski 1982). The next section describes the
network relationships between the units in the manufacturing network from a conceptual perspective. The
operationalization of these network relationships and
the application of network analysis techniques are
described in §3.
2.2. Network Position of Plants
The purpose of our research is to understand the position of manufacturing units in international manufacturing networks. Our hypothesis is that distinct
plants play different roles in these networks by having relationships of different type and intensity with
the other plants and with headquarters. Bartlett and
Ghoshal (1989) recognize four types of relationships
between subsidiaries: physical goods, information,
people, and financial resources. The flow of financial
resources in the strict sense of providing capital to
subsidiaries is of lesser importance in our study of
network relationships between plants, and will therefore not be discussed here. The three other types
of relationships—goods, information, and people—
differ in their degree of tangibility. Our interest lies
primarily in the intangible knowledge network of the
multinational, which is explained in the next two sections because we are exploring how the network of
production facilities of the multinational may enhance
the creation of strategic capabilities. The logistics
organization of the multinational, which is reflected
in the focus of the plants and in the tangible transfer of components on semifinished goods through the
network, is discussed in §4.4.
2.2.1. The Information Network. Two types of
information flow can be distinguished: the administrative information flow and the knowledge flow
(Gupta and Govindarajan 1991). In a manufacturing context, the administrative information flows consist of information on inventory levels, purchasing
requirements, forecasts, production plans, etc. These
information flows depend to a large extent on the
degree of centralization of manufacturing tasks, such
as planning, inventory management, and procurement. From a manufacturing strategy perspective, the
knowledge flows are the more interesting ones. It is
commonly accepted that one of the main reasons for
the existence of multinationals is the possibility to
acquire, create, and use technological assets across
1739
national boundaries (Dunning 1993, p. 290). Consequently, the ability to transfer innovations through
the multinational’s network is crucial for attaining
a competitive advantage. Three categories of innovation flows have been studied: the development
and introduction of a new product, the development
and introduction of a new production process, and
the implementation of a new management system
(Ghoshal and Bartlett 1988).
2.2.2. The People Network. The flow of people
in the manufacturing network may take different
shapes. A typical example is the position of a manager having line or staff responsibility in two or more
plants. This can be at the level of the plant manager, as well as the functional levels reporting to
the plant manager. This type of relationship can be
called “interlocking management” by analogy with
the interlocking directorship; i.e., one person being a
member of the board of directors of two or more companies (Gerlach 1992). Of equal importance are the
“dispatched managers,” i.e., the managers who have
been transferred from one operating unit to another,
on a permanent or a temporary basis, by analogy
with the dispatched director. A third shape of the
flow of people refers to the day-to-day operations
of the network. These relations between units are
realized through “coordinators”—managers traveling
frequently between operating units to share information and to accomplish cooperation between the units.
The role of such coordinators has received a lot of
attention in the organization literature. They are specific examples of what Galbraith (1977) and Mintzberg
(1979) have defined as the “liaison devices” of an
organization.
A major advantage of these coordinators is the opportunity they create for personal contact between
people in the organization. Ghoshal et al. (1994) have
shown that the relationship among subsidiary managers and the relationship between managers of subsidiaries and managers of headquarters have a significant influence on the frequency of the intersubsidiary
communication and on the frequency of communication between the subsidiaries and headquarters. Communication plays an important role as a facilitator of
the transfer of innovations in multinationals (Ghoshal
and Bartlett 1988, Gupta and Govindarajan 1991).
We retain from this short discussion three variables
that are particularly relevant for our study: (1) the
flow of innovations between the units in the network;
(2) the extent to which coordination exists in the network through managers traveling between the units;
and (3) the frequency of communication between the
units in the network.
Interlocking management has not been retained as
such in the research because it can be regarded as a
special reason for frequent travels between the two
plants involved. Dispatching has not been retained
either because we assume that this creates a tight relationship between the dispatching and the receiving
unit only if the dispatched manager keeps in touch
with his original unit. Measuring the communication
between the two units then captures this.
3.
Research Methodology
3.1. Case Research
The research reported here is part of a larger research
study on the international plant configuration. The
research was exploratory, i.e., we wanted to understand the “how” and “why” of the international plant
network. Thus, case study research has been preferred
over other research methodologies (Yin 1984).
To achieve precision and rigor, we followed the
methodological guidelines proposed by Eisenhardt
(1989), Miles (1994), and Yin (1984). Without being exhaustive, we mention that a strict research protocol
has been designed, a questionnaire with both closedand open-ended questions has been developed as
guidance for the interviews, accommodations have
been made to avoid interview fatigue, and both qualitative and quantitative data have been collected in a
rigorous and structured way and have been analyzed
in a systematic way. Several variables have been measured through multiple item measures. The reliability
of these variables has been assessed by calculating the
Cronbach alpha, and factor analysis has been used to
reject or confirm the assumption that some theoretical constructs underlie the items (Carmines and Zeller
1979, DeVellis 1991).
To enhance construct validity, multiple raters have
been used. This tactic avoids the risk that data comes
from a single respondent with a biased view or with
limited access to information (Speier and Swink 1995,
Boyer and Verma 1996). The intraclass correlation
(ICC) method has been used to assess the interrater
reliability of the variables. The ICC index measures
the variance of the scores of the raters within a plant,
relative to the between-plant variance. Data on the
ICC for all variables used in the analyses can be found
in Appendix 1.
3.2. Data Collection
The case research has been carried out in eight manufacturing companies headquartered in Europe, in
different industries: food products (two companies),
textile goods, plastic products, leather products, primary metal, fabricated metal, and electrical goods.
Thus, no single industry dominates the sample. The
companies had between four and 10 manufacturing
plants. The primary selection criterion for the cases
has been diversity, at the level of the company as
well as the plant. At the company level, it is important to have diversity in terms of the international
1740
environment in which the company operates because
one of the research objectives was to explore the link
between the characteristics of the company’s international environment and the plant configuration in the
company. Consequently, the cases are distributed over
the integration/responsiveness grid, as defined by
Bartlett and Ghoshal (1989). Two of the cases are classified as “global,” two as “transnational,” and four as
“multinational” (Vereecke and Van Dierdonck 1999c).
Diversity at the plant level has been obtained by
selecting companies with a minimum of four plants,
spread over a broad geographical region—the rationale being that with three plants or less, companies
have few opportunities for differentiating the role and
focus of their plants. A geographical spread of the
plants (pan-European or even global) was expected
to result in a broad range of drivers for establishing the plant, and therefore also in a broad range
of plant roles (Ferdows 1997b). The sample was limited to companies with their headquarters in Western
Europe.
Data have been gathered at two levels of analysis:
the plant and the company.
• Interviews have been conducted with the general
manager and with manufacturing managers at headquarters. In total, data has been collected on 59 manufacturing plants, through 37 interviews (with a total
duration of approximately 120 hours). The number
of interviews varied between two and six per case.
A structured questionnaire with closed- and openended questions has been used as a guide through the
interviews.
• A second questionnaire has been sent to the plant
managers and/or the manufacturing managers in the
distinct production plants. One hundred fourty four
questionnaires have been sent to 54 out of the 59
plants. For five of the plants, headquarters asked us
not to send a questionnaire to the plant managers.
Eighty three percent of the questionnaires have been
returned from 50 plants. This implies that in total we
have received data from the plant managers on 50 out
of the 59 plants (85%). The number of questionnaires
returned from the plants varied between one and five
per plant.
• Information has also been obtained from desk research on company brochures, publications, and company archives.
Fourty-two plants were located in Europe, spread
over 14 different countries. The other 17 plants were
spread over 10 different countries in East Asia and
the Middle East, the United States and Canada, and
South Africa and Australia. We thus have a truly
international sample. The number of years the plant
had been part of the company ranges between 0 (this
plant was starting up at the moment of the research)
and 50 years, with an average of 17 years. The number of employees in the plants ranges between 77 and
1,100 with an average of 340.
3.3.
Operationalization of the Network Position of
the Plants
In describing the manufacturing network of a multinational company as an information and people network, the network units considered are all the plants
and the group of managers in headquarters responsible for manufacturing (in this paper, referred to
as “headquarters”). As discussed earlier, the network
relationships considered in this research are the flows
of innovation, the use of coordinators, and the communication between the units in the network.
The innovation transfers have been measured by
asking managers in the plants (through the mail questionnaires) and in headquarters (through the interviews) to enumerate and describe the transfers of
product, process, and managerial innovations they
know of over the past three years. A similar operationalization has been used by Ghoshal and Bartlett
(1988). The information that has been gathered from
these different sources has been checked, complemented, and corrected by at least one manager in
headquarters, in the course of the in-depth interviews.
The presence of coordinators has been operationalized as the extent to which people are traveling from
one unit to another. This information on people flows
has been collected through the mail questionnaire to
the plants. The measurement is based on the tool used
in the research by Ghoshal (1986). The respondents
had to report the number of days they had spent, over
the previous year, in headquarters and in each of the
plants in the company’s network.
One of the questionnaire items measures the communication between the managers in the plants and
in headquarters. However, such self-reported answers
may suffer from recollection problems. This problem
is severe if the data collection method consists of
an interview or questionnaire asking the respondent
to name the persons he/she communicates with frequently. This approach has been used in early studies of communication networks in R&D laboratories
(Allen 1977). An alternative approach is to provide a
list of people, and to ask the respondent with whom
on this list he/she has communicated, rather than letting the respondent name the people he communicated with (Knoke and Kuklinski 1982). This approach
has been followed in our research. A score of 3, 2,
and 1 has been given to daily, weekly, and monthly
communication, respectively. Bartlett and Ghoshal
(1989) have also preferred this scoring system.
The primary network measure used in our research
is the centrality of the plant in the network. If network relations are mutual (as is the case for the communication network), we measure centrality of the
1741
unit through its degree. The degree of a unit is defined
as the proportion of other units with which a unit
has a direct relationship (Knoke and Kuklinski 1982).
If network relations are not mutual (as is the case
for the flows of people and innovations), two degree
measures are used: the unit’s indegree and outdegree
(Knoke and Kuklinski 1982). The indegree of a unit is
defined as the proportion of relations received by the
unit from all other units. The outdegree of a unit is
defined as the proportion of relations from that unit
to all other units.
Based on these definitions of centrality, the following network variables have been defined:
• The communication centrality of plant i captures
the frequency of communication of the manufacturing
staff of plant i with the manufacturing staff of the
other units in the network.
• The innovation indegree of plant i captures the
intensity of the innovation flow transferred (and implemented) from the other units to plant i.
• The innovation outdegree captures the intensity of
the innovation flow transferred (and implemented)
from plant i to the other units.
• The people indegree of the plant captures the number of days plant i has received visitors from the manufacturing staff team of the other plants.
• The people outdegree of plant i captures the number of days manufacturing staff people of plant i have
been visiting other plants in the plant configuration
In network analysis, the consequences of missing
data are severe because the lack of data from a single
unit implies the lack of data on the N − 1 possible
relationships of this unit with the other units in the
network. Estimates such as centrality can therefore be
distorted if data are missing. Consequently, great care
has been taken so as to maximize the response rate
(Vereecke and Van Dierdonck 1999b).
As suggested by Ketchen and Shook (1996), the
number of clusters has been determined through the
use of multiple techniques.
• Upon visual inspection of the dendogram, we
recognize a structure with four clusters.
• A four-cluster classification accounts for 56% of
the variance in the data. Disaggregation into five, six,
and seven clusters adds approximately 6% to the variance explained at each step. After seven clusters, the
increases in R2 are low (below 3%). This observation
points at a classification into four or seven clusters.
• The cubic clustering criterion (CCC) points at
nine clusters. However, tests have indicated that the
CCC may suggest too many clusters (Milligan and
Cooper 1985).
• We have used the analytics software SAS to perform a number of the tests that have been put forward
by Milligan and Cooper as most effective (Milligan
1996). The pseudo F statistic, developed by Calinski
and Harabasz (1974), has local peaks at two and seven
clusters. The pseudo t 2 statistic, based on Duda and
Hart (1973), indicates a clustering of the data in two,
four, or seven clusters.
We conclude that the different test routines point at
a clustering into two, four, or seven clusters. Because
there is partial agreement among the test results,
Milligan (1996) suggests opting for the larger number,
that is, seven. However, when going from the four to
the seven-cluster solution, we see that the pattern of
three clusters is roughly maintained, while the fourth
cluster falls into four smaller clusters (including a
cluster of one unit), which are difficult to distinguish.
Consequently, the seven-cluster solution merely adds
complexity without providing revealing insights. We
have therefore opted for a classification into four
clusters.
3.4. Clustering of the Data
To ensure the validity of the network typology, a twostage procedure has been followed to cluster the data
(Ketchen and Shook 1996). We had sufficient data on
49 of the plants to involve them in the cluster analysis. Ward’s hierarchical clustering method has been
used to define the number of clusters. This number of clusters has then been used as the parameter in the nonhierarchical K-means clustering method
with Euclidian distance measure. K-means clustering
is preferred over the hierarchical cluster methods for
the development of the typology because it is an iterative partitioning method and thus is compensating
for a poor initial partitioning of the cases. Because the
units of measurement for the network relationships
differ substantially and Euclidian distance is used as
the distance measure in the cluster analysis, the variables have been standardized prior to the clustering
(Aldenderfer and Blashfield 1984, p. 21).
4.
Empirical Results
4.1. A Network Typology of Plants
The four clusters represent different positions of plants
in the plant network of information and people. The
average of the network variables in each of the clusters is represented graphically in Figure 1.
The typology of plants resulting from this cluster analysis is summarized in Table 1. We distinguish three levels for each of the variables: “low”
for average value below 0; “medium” for average
level between 0 and 1; and “high” for average value
above 1. These cut-off values are defined on the standardized variables.
Plants in Cluster A occupy an “isolated” position in
the plant network. Few innovations reach the plant,
few innovations are transferred to other units, few
manufacturing staff people come to visit such a plant,
few manufacturing staff people from this plant go
1742
Figure 1
Whereas in type C plants the inflow of visitors is significantly higher than the outflow p < 1%, in type D
plants the outflow is higher than the inflow p < 5%.
The D plant is thus highly involved in the network,
and takes a more active role than the C plant. We label
them as the “active network players.”
Network Clusters: Graphical Representation
Plot of means for each cluster
3
2
1
0
–1
Cluster D
Cluster C
Cluster B
Cluster A
–2
Communication centra
Innovation outdegree
Innovation indegree
People indegree
People outdegree
Variables
visit other plants, and there is little communication
between the manufacturing staff people of this plant
and the other manufacturing managers in the network.
A plant in Cluster B is comparable to the isolated
plant on all but one variable: it receives more innovations from the other units in the network. We will
therefore label these plants as “receivers.” Clusters A
and B thus consist of plants that are only weakly
embedded in the manufacturing network. They represent 37 out of the 49 plants in the sample.
Clusters C and D consist of plants that are true network players. A type C plant frequently exchanges
innovations, both ways, with the other units and its
manufacturing staff communicates extensively with
the other manufacturing managers in the network.
A C plant is also frequently hosting visitors from
other units in the network. In the network, the C plant
thus takes the role of the “hosting network player.”
The type D plants differ from the type C plants
in two aspects: First, the level of communication
centrality and the outflow of innovations are even
higher in the type D than in the type C plants (significantly different at p = 10% for communication centrality and at p = 5% for innovation outflow). Second,
the major flow of visitors is in the opposite direction.
Table 1
Network Typology of Plants
Network variable
Number of plants in cluster
Communication centrality
Innovation indegree
People indegree
People outdegree
Cluster A
Isolated
Cluster B
Receiver
Cluster C
Hosting
network
player
11
Low
Low
Low
Low
Low
26
Low
Medium
Low
Low
Low
8
Medium
Medium
Medium
High
Medium
Cluster D
Active
network
player
4
High
High
High
Medium
High
4.2. Cluster Validation
Analysis of variance on the variables used to generate
the cluster solution is frequently used to test the validity of the cluster analysis solution. The test results are
summarized in Table 2.
However, we do not want to overemphasize the
value of this analysis of variance. Because the clustering method attempts to minimize variance within
the clusters, it is logical that the F -test is significant
(Aldenderfer and Blashfield 1984, p. 65). External criteria analysis is more appropriate. Such analysis is
based on statistical tests on variables that have not
been used to generate the cluster solution, and yet are
relevant (Aldenderfer and Blashfield 1984, Milligan
and Cooper 1985).
A variable that is strongly related to the typology
discussed here is the concept of the “strategic role”
of the plant. Building on the work done by Ferdows
(1989), we define the importance of the strategic role
of the plant as the extent to which the plant contributes to the other units in the manufacturing network (Vereecke and Van Dierdonck 1999a). We have
measured the importance of the strategic role of the
plant on a nine-point Likert scale, describing plants
which have as their main goal “to get the products
produced” at the lowest extreme, to plants that are a
“center of excellence, and serve as a partner of headquarters in building strategic capabilities in the manufacturing function” at the highest extreme. Given our
definition, the importance of the strategic role of the
plants in Cluster D should be high. The importance of
the strategic role of the plants in Clusters A and B, on
the other hand, should be low because these plants
make little contributions to the plant network. The
plants in Cluster C are expected to play a strategic
role of medium importance.
The average and median of the importance of
strategic roles are shown in Table 3. We should note
here that for the importance of the strategic role, as
Table 2
Analysis of Variance on Four-Means Cluster
Solution
Network variable
Communication centrality
Innovation indegree
People indegree
People outdegree
F
p-level
12, 18
17, 38
21, 69
47, 81
14, 76
0.000006
0.000000
0.000000
0.000000
0.000001
Table 3
Cluster A
Cluster B
Cluster C
Cluster D
Overall
Importance of Strategic Role of the Plants
Valid N
Mean
Median
11
26
8
4
49
4.80
4.52
5.76
7.97
5.07
4.67
4.69
6.44
8.10
4.80
Median test: obs-exp
below median∗
039
173
−008
−204
∗
Number of cases observed minus number of cases expected below the
overall median level of strategic role, that is, below 4.80. A positive number
shows the number of cases observed below the overall median, and consequently indicates a relatively low level of strategic role in the cluster.
well as for most of the plant characteristics that will
be discussed later, the assumption of normality is violated. For those variables, the nonparametric alternatives to the ANOVA, the Kruskall-Wallis and Median
Tests, have been used.
The Kruskal-Wallis test indicates a significant difference in the level of the strategic role between the
clusters p < 10%. The Median Test confirms that the
difference in strategic role follows the hypothesized
pattern, as can be seen in Table 3. Cluster B contains
slightly more cases below the median level of strategic
role than could be expected if the strategic role were
evenly distributed over the four clusters, indicating a
relatively low level of strategic role. Cluster D contains more cases above the median level of strategic
role than could be expected if the strategic role were
evenly distributed over the four clusters, indicating
a relatively high level of strategic role. The MannWhitney U-Test confirms that the level of strategic
role in Clusters A and B is significantly lower p < 5%
than in Cluster D.
4.3. Future Strategic Role of the Plant
We have discussed the relationship between the network position of the plant and the importance of
the strategic role played by the plant. Our research
also provides information on the expected changes in
the strategic role of the plant. The interviewees were
asked to estimate the importance of the strategic role
of the plant as they expect it to be in five years on the
nine-point Likert scale described above. The data suggests that in Clusters C and D, only a few marginal
increases and decreases in strategic role are expected.
This suggests that the plants which occupy an integrated position in the network (Clusters C and D) are
fairly stable in terms of the importance of the strategic
role they play in the company. Several of the A and B
plants, on the other hand, are expected to experience
an increase in strategic role. For some, the expected
increase is quite substantial. Given the relationship
that we observed between the role of the plant and its
network position, it is fair to expect that these plants
will probably be moving from Clusters A or B toward
1743
Clusters C or D. Several of the other plants in Clusters A and B are expected to experience a decrease in
strategic role. Again, for some, the expected decrease
is quite substantial. It is clear that these two clusters
of nonintegrated plants are less stable than the two
clusters of integrated plants.
An example illustrates our point. Two of the “receiver” plants in the sample have been closed since
we started the case research. We do not want to infer
here that the plants in the “isolated” or “receiver”
clusters are on the waiting list for closure. The examples of plants with a positive expectation in strategic role would certainly contradict this point. Our
hypothesis is that the plants in these two clusters are
in a variable position, and that this variability may
lead toward an increase as well as a decrease in terms
of the importance of the role the plant plays in tomorrow’s network. These plants seem to provide strategic
flexibility in the network.
It is interesting to mention that the decrease in
strategic role that is predicted by headquarters for
some of the isolated plants and the receivers is not
expected by the managers in the plants. The lack of
network relationships for the isolated plants and the
receivers seems to create a gap between the expectations of plant management and the considerations in
headquarters. It may also suggest that the managers
in A and B plants are less involved in strategic decision making and, thus, are less well informed.
4.4. Characteristics of the Plant Types
To better understand the network typology of plants,
the four types of plants have been compared on a set
of plant characteristics. We have analyzed:
• The age of the plant (number of years the plant
has been part of the company).
• The size of the plant (expressed in number of
employees).
• The focus of the plant (Hayes and Schmenner
1978, Collins et al. 1989):
Product focus: the extent to which the plant
focuses on a narrow portion of the company’s product
range, and
Market focus: the extent to which the plant
focuses on a narrow portion of the geographical market served by the company.
• The supplier/user relationship with other plants in
the network: the extent to which a plant supplies
components or semifinished goods to or uses components or semifinished goods from another plant in
the network. It has been measured as the centrality
(outdegree and indegree) of the plant in the physical
network of goods. The outdegree of plant i captures
the portion of plants in the plant configuration, to
which plant i supplies components or semifinished
goods. The indegree of plant i, (analogously) captures
1744
the portion of plants in the plant configuration, from
which plant i receives components or semifinished
goods.
• The level of investment: A list of 14 potential investments has been included in the questionnaires.
From this list of 14 items, four types of investment
have been identified through factor analysis:
(1) Investments in the production process, that is,
in setup time reduction, plant automation, process
analysis, productivity improvement, and throughput time reduction (Cronbach alpha of the resulting
factor = 077).
(2) Investments in planning, that is, in material
and/or capacity planning and just-in-time systems
(Cronbach alpha of the resulting factor = 079).
(3) Investments in managerial improvement programs, that is, in statistical process control, supplier
partnerships, total quality management, and employee participation programs (Cronbach alpha of the
resulting factor = 073).
(4) Investments in new product development.
• The autonomy of the plant. Both strategic autonomy and operational autonomy have been measured
through questionnaires administered in the plants.
A similar approach has been followed by Ghoshal
(1986), Bartlett and Ghoshal (1989), and De Bodinat
(1980). Two dimensions of strategic autonomy have
been identified, through factor analysis:
(1) Strategic autonomy in decisions concerning
the operations of the plant, that is, the decision to develop a new product or to introduce a new planning
system and the selection of a new supplier (Cronbach
alpha of the resulting factor = 081).
(2) Strategic autonomy in decisions concerning
the design of the plant, that is, the decision to develop a new production process and the choice of
a new technology (Cronbach alpha of the resulting
factor = 085).
Two dimensions of operational autonomy have
been identified, through factor analysis:
(1) Operational logistics autonomy, that is, in
developing a production plan, placing purchasing
orders, managing inventories (Cronbach alpha of the
resulting factor = 084).
(2) Operational autonomy in design and engineering, that is, in developing new products and processes (Cronbach alpha of the resulting factor = 088).
• The level of capabilities in the plant. Two types
of capabilities are distinguished: the capabilities
to develop new products and managerial capabilities. They have been measured in the headquarters
interviews through a 1–9 Likert scale. The Cronbach
alpha for this construct was 0.85.
• The performance of the plant. Performance has
been measured relative to the target set for the plant.
Performance data has been obtained from a list of
nine performance items, included in the questionnaire
sent to the plant management teams. Because this performance data is self-reported, it is important to have
data from multiple respondents per plant, and to evaluate the interrater reliability. Two dimensions of performance have been identified through factor analysis
(see Appendix 1):
(1) Performance on time measures, that is, performance relative to the target set for manufacturing throughput time, delivery lead time, and on-time
delivery to customers (Cronbach alpha of the resulting factor = 085).
(2) Performance on cost and quality measures,
that is, performance relative to the target set for unit
production cost, productivity of direct workers, defect
rates, and overall product quality (Cronbach alpha of
the resulting factor = 083).
The results of the (mostly nonparametric) comparisons of the four clusters on these variables are listed
in Table 4. For those variables that showed a significant difference across the four clusters (with significance level p < 10%), pairwise comparison of the
mean or median is reported in Table 4.
We conclude from these comparisons that
(1) Plants in Cluster C are significantly older than
plants in Clusters A and B.
(2) Plants in Cluster A are significantly more market focused than plants in Clusters C and D; and
plants in Cluster B are significantly more market
focused than plants in Cluster C.
(3) The outflow of components and semifinished
goods is significantly lower for plants in Clusters A
and B than for plants in Cluster D.
(4) The inflow of components and semifinished
goods is significantly lower for plants in Cluster A
than for plants in Cluster B; and is significantly lower
for plants in Cluster B than for plants in Clusters
C and D.
(5) The level of strategic autonomy in plant design
for plants in Cluster A is significantly lower than for
plants in Clusters B, C, and D. Plants in Cluster B
have a significantly lower level of strategic autonomy
in plant design than plants in Cluster D.
(6) The level of process investment in plants in
Cluster D is significantly higher than in plants in
Clusters A, B, and C.
(7) Plants in Cluster A invest significantly more
in managerial improvement programs than plants in
Clusters B and C .
(8) The level of capabilities in plants in Cluster B
is significantly lower than in plants in Clusters A, C,
and D.
Table 5 summarizes the characteristics of the clusters that result from these comparisons. The comments made throughout the interviews provide some
additional insights in the profile of the clusters. These
comments are listed in Appendix 2.
1745
Table 4
Statistics on Plant Characteristics by Cluster
Mean/median
Plant
characteristic
Variable
A
B
C
D
Difference between clusters
Age
Number of years plant is
part of company
111
168
306
197
Anova p < 1%
A < Bns /A < C∗∗ /A < Dns /B < C∗∗ /B < Dns /C > Dns
Size
Number of employees
Number of workers
Number of salaried workers
Number of manufacturing
staff people
154
111
43
13
240
165
43
21
362
251
126
41
533
308
226
40
Not significant
Not significant
Not significant
Not significant
Market focus
Proportion of market range
supplied by the plant
0 18
0 63
0 90
0 89
Kruskal-Wallis Anova with p < 5%
Mann Whitney U-test
A < Bns /A < C∗∗ /A < D† /B < C∗ /B < Dns /C ≈ Dns
Product focus
Proportion of product range
0 15
0 22
0 30
0 38
Not significant
Supplier/user
relationship
Outdegree
0
0
0
0 47
Mann Whitney U-test
A ≈ B/A ≈ C/A < D∗∗ /B ≈ C/B < D† /C < Dns
Indegree
0
0 11
0 22
0 42
Mann Whitney U-test
A < B† /A < C∗∗ /A < D∗ /B < C∗ /B < D† /C < Dns
Operational
autonomy
Logistics
Development and engineering
62
44
69
48
64
58
58
62
Not significant
Not significant
Strategic
autonomy
Operations of the plant
Design of the plant
41
37
52
48
51
57
54
63
Not significant
Anova p < 5%
A < B∗ /A < C∗∗ /A < D∗∗ /B < Cns /B < D† /C < Dns
Investment
Process investment
55
53
51
68
Investment in planning
Managerial investment
44
65
49
49
46
49
63
57
Anova p < 10%
A > Bns /A > Cns /A < D† /B > Cns /B < D∗ /C < D∗
Not significant
Anova p < 5%
A > B∗∗ /A > C∗ /A > Dns /B ≈ C/B < Dns /C < Dns
Not significant
New product investment
49
52
57
70
Plant capabilities
Level of resources
64
53
64
75
Anova p < 5%
A > B† /A ≈ C/A < Dns /B < C† /B < D∗∗ /C < Dns
Performance
relative to target
Time performance
Cost and quality performance
10
10
072
0 63
084
0 02
082
0 69
Not significant
Not significant
Notes. Variables for which the assumption of normality is rejected are in italic. For those variables, the median value is mentioned (in italic). For the other
variables, the mean value is mentioned.
∗∗
Significant at p < 1%; ∗ significant at p < 5%; † significant at p < 10%; n.s.—not significant at p < 10%.
5.
Discussion
Some general lessons can be drawn from the plant
typology and the characteristics of the four types of
plants.
First, the plants providing innovations to the manufacturing network, the “hosting network players” and
the “active network players,” are at the same time receivers of innovations from other units in the network.
Apparently, transferring knowledge is beneficial, not
only for the receiver, but also for the provider. An
explanation may be that the quality of the relationship
between two units is a major factor in the exchange
of innovations, or as Szulanski (1996, p. 36) has put
it, “the relationship serves as a conduit for knowl-
edge.” Once such a relationship has been established,
it works in both directions.
Second, the analyses show that there is a strong link
between the position of the plant in the intangible network of ideas and in the tangible network of goods.
This is in line with Nonaka and Takeuchi (1995),
who argue that codified and noncodified knowledge
complement and reinforce each other. The “isolated”
plant, which is not actively taking part in the network
of ideas, is also isolated in the physical sense: we
observed very little flows of components or semifinished goods from these plants to the other plants
in the network, and vice versa. The network players
(type C and D), on the other hand, are typically sup-
1746
Table 5
Summary of Plant Characteristics by Cluster
Plant characteristics
A
Relatively young; market focused; little inflow and outflow of
components and semifinished goods; relatively low level of strategic
autonomy in plant design; relatively high level of managerial
investment
B Relatively young; little outflow of components and semifinished goods;
relatively low level of managerial investment; relatively low level of
capabilities
C Relatively old; broad market; high inflow of components and
semifinished goods; relatively low level of managerial investment
D High inflow and outflow of components and semifinished goods;
relatively high level of strategic autonomy in plant design; relatively
high level of process investment
pliers to the other plants (in the case of Cluster D) or
customers of the other plants (in the case of Cluster C)
for components or semifinished goods. Kobrin (1991,
p. 19) argued that “the two most important intrafirm
flows are products and technology, and the latter is
often embodied in the former,” and also observed
this link between knowledge and physical flows. Our
research suggests that the product is not only a carrier of technological product and process innovation,
but also of managerial innovations.
Third, we see that building network relations takes
time. The average age of the networked plants (type C
and D) is 28 years, whereas the average age of the two
more isolated types of plants (type A and B) is only 15
years. The difference in age between these two groups
is significant p < 1%. Networks apparently develop
over a long period of time.
A fourth conclusion is that the four different network roles reflect very different plant characteristics.
The “isolated” plant in Cluster A is very independent. In its isolated position, it does not contribute to
the network, but on the other hand, it also does not
depend on the other network units for its components
or for maintaining or improving its manufacturing
capabilities. Plant management has the capabilities to
run the plant independently. The receivers in Cluster B typically are local players that need support—
technical and/or managerial—of headquarters or the
other plants in the network for their survival. They
need this support either because of the negative attitude and lack of skills in the plant, or because of the
strategic decision of headquarters to keep investments
in the plant relatively low. The hosting network players (Cluster C) are typically fairly old, they supply
a broad market, and they are characterized by a low
level of managerial investment. The hosting network
player has been observed in seven of the eight companies studied. With one exception, this role is played
by only one of the plants per company. It is interest-
ing that half of the eight C plants are the “mother
plant,” the earliest plant in the network, located close
to headquarters. We hypothesize that because of its
age, the broad market it supplies, and its easy access
to headquarters, the plant has gained a lot of experience, which explains why the plant is seen as a
competence center by other plants. The other four
C plants are located close to another plant with which
they have established tight relationships. The inflow
of people in these C plants dominantly comes from
this neighboring plant, which also has a higher level
of strategic role. In two cases, the neighboring plant
happens to be a D plant. The profile we see here is
one of a satellite plant that is heavily influenced by
the presence of another network unit. We conclude
that the scenario which leads to a C-type plant seems
to build on heritage: the network relationships exist
because the plant has been in the network for a very
long time and is located close to headquarters or to an
active network player. The C plant seems to undergo
this scenario in a passive way, rather than to play an
active role in it.
The scenario that emerges from the characteristics
of the type-D plants is more dynamic and active.
These plants build capabilities through investments
under a relatively high level of autonomy. Such plants
are actively building network relationships by sending manufacturing staff to other plants and through
extensive communication. It is their enthusiasm and
their technical specialization that makes them an
important network player.
From interaction with managers about the typology,
we have noticed that the D cluster is an intriguing
category for plant managers. The D plants are typically plants that act as a center of excellence or as
a pilot plant for new products, they are regarded as
the “think tank” or “engine” in the network, and are
known as the technical “specialist” plant in the network (see Appendix 2). This intriguing profile raises
questions as to the further evolution of these plants,
which makes it an issue for future research.
Fifth, there is no significant difference in performance between the clusters. Reaching the targets on
cost, quality, or time measures does not appear more
or less difficult in the distinct clusters. This suggests
that there is not a unique optimal network position
for a plant. Rather, the network position of the plant
should be regarded from a contingency perspective.
Finally, the analyses suggest that the future perspectives of the plant depend on the plant’s network position. Plants that are strongly embedded in
the production network are expected to maintain the
high level of strategic role they are already playing in the network. The future of plants in rather
isolated positions has been predicted to be in two
opposite directions: some plants are expected to grow
1747
in strategic importance and are assumed to develop
network relationships; others are expected to become
less important and may even disappear from the manufacturing network. A possible explanation may be
that in case of overcapacity and cost cutting, an “isolated” or “receiver” plant is a welcome candidate for
disinvestment or closure. Closing such a plant implies
a reduction of overall capacity, which is exactly what
is aimed at. It does not imply, however, an important
reduction in knowledge transfers because these plants
do not contribute considerably to the other plants in
the network. This is apparently a headquarters’ decision plant managers are not aware of.
6.
Contributions to Researchers and
Practitioners
Previous classifications of plants have focused on
the tangible characteristics of plants: the products
the plant produces, the processes it has in place, the
markets it serves, and the parts it supplies to other
plants in the network (Hayes and Schmenner 1978).
The typology developed in our research differs, as
it classifies plants on the basis of their position in
the intangible knowledge network. We focus primarily on flows of knowledge, rather than flows of goods.
The conclusions of our work are therefore useful to
any scholar who wants to study the architecture of
knowledge networks in manufacturing, and eventually in other environments such as R&D or service
operations.
The research has allowed us to identify, among the
plants we studied, 12 network players, i.e., plants that
showed a strong interaction with other units in the
network. This interaction between plants is a fairly
new trend, or at least, a trend not previously well
documented. Moreover, our research offers a methodology for identifying network relationships in manufacturing networks.
To the manager in charge of a multinational network of manufacturing plants, the typology serves as
a “toolbox” for drawing a map of the plant network.
In our multiple discussions with managers about the
typology, we learned that the typology has high face
validity to them, and allows them to classify their
plants, even without actually measuring the in- and
out-flows of the plants. An evaluation of this map
may help them in identifying possible gaps or unbalances. Because the position of the plant in the network does not impact the plant’s performance, any
of the types of plants can be effectively present in
the network. If managers believe that their network
would benefit from plants spreading best practices,
they should identify which plants have active and
hosting capabilities, and foster these plants in their
network. However, the hosting network players seem
to be a result of the past, while the existence of
the active network players can be stimulated. Our
research indicates what it takes to develop an active
network player. On the other hand, the manager may
find it wise to have some isolated plants or receivers
(types A and B). These are quite mobile building
blocks of the network. Reducing the number of isolated plants or receivers does not impact the potential
for transferring knowledge. As such, the presence of
isolated plants and receivers gives the manager some
structural flexibility in managing his network.
To the plant manager, the research shows the danger of a protective attitude toward the exchange
of knowledge. The isolated position taken by these
plants may well result in a difference in view between
plant managers and company managers about the
strategic future of the plant.
7.
Limitations and Future Research
An important limitation of the research is the focus
on the intracompany network relationships. While
we acknowledge that intercompany network relationships are important in creating sustainable competitive advantage, we have limited our research to the
network relationships between units of the same company. Whether the hosting and active network players
are also tightly embedded in the external, intercompany network with suppliers, customers, and other
network partners remains to be studied.
Second, our research describes the strategic role
played by plants in international plant networks. It
identifies those plants that develop knowledge and
capabilities and that transfer this knowledge to the
other plants in the network. The research does not
explain how this knowledge is developed, nor does it
describe the mechanisms used for the diffusion of this
knowledge and their effectiveness. Also, as stated earlier, the research is static and raises questions as to the
further evolution of the plants and of their position in
the network. This is an area of future research.
The absence of significance in the difference in performance in the network typology may point at a lack
of difference in performance. However, it may as well
be a consequence of the performance measure that
has been used, which is static and rather restrictive
(that is, performance relative to the target set for the
plant) and operationalized as a perceptual measure.
There is definitely still a need to study the relationship between plant performance and the position the
plant plays in the manufacturing network.
Also, we did not make any assertions about the
relationship between the portfolio of plants in terms
of their network type and the performance of the
company. We hypothesize that the optimal portfolio
1748
of plants is contingent on the company’s competitive environment. However, this needs to be studied.
As mentioned in the Methodology section (§3), this
paper is based on case research. While one of the
major advantages of case research is the depth of the
information that can be collected, its major disadvantage is the limitation in sample size, and therefore the
potential limitation in external validity. However, we
are convinced that the careful selection of the cases
from a diversity of industries improves the external
validity of the work.
The cases have been limited to companies headquartered in Europe to avoid cultural differences between the cases. Whether the conclusions still hold
in multinationals headquartered in other continents is
unexplored and can be subject to future research.
Finally, the research focuses on manufacturing companies only. Whether a similar typology can be developed for service companies is an open question.
8.
Conclusion
In the research, network analysis has been used as
a methodology for understanding the position of
plants in international manufacturing networks. The
focus has been primarily on the intangible knowledge
network, and secondarily on the physical, logistic network. A typology of plants in a manufacturing network has resulted from the research. Four types of
plants, with a different strategic role, different characteristics, and different perspectives for the future have
emerged. The typology indicates that flows of knowledge between plants seem to be reciprocal, and that
there is a clear correlation between tangible and intangible flows in the network. The driver behind intensive network relations may be either heritage or a
deliberate investment in capabilities inside the plant.
Anyhow, building network relationships takes time.
We have also observed that the future of plants that
are tightly embedded in the network is more stable
and secure.
Overall, this leads us to believe that in managing international networks of plants, managers can
balance long-term knowledge development and medium-term flexibility. In approving investments in the
network relationships, they allow some of the plants
to play an active role in the creation and diffusion
of knowledge in the network, thus creating long-term
competitive advantage. The other plants provide the
manager with strategic flexibility. Their role in the network can be adapted in the medium term, according
to the changing needs of the business.
Appendix 1. Interrater Reliability Scores on Perceptual Measures
Construct
Factor
Item
ICC
Strategic role today
0.85
Strategic role 5 y Ahead
0.83
Operational autonomy
Strategic autonomy
Logistics
Developing a master production schedule
Developing material and capacity plans
Developing the shop floor schedule
Developing sales forecasts
Placing purchasing orders
Managing inventories
0.81
0.78
0.70
0.89
0.80
0.70
Development and engineering
Developing new products
Making changes to existing products
Developing new production processes
Making changes to existing production processes
0.74
0.77
0.78
0.79
Operations of the plant
Decision to develop a new product
Decision to make changes to an existing
product design
Selection of a new supplier
Decision to introduce a new planning and
control system
Choice of standards, goals, and performance
measures for quality management
0.69
0.76
Design of the plant
Decision to develop a new production process
Decision to make changes to an existing
production process
Choice of technology
0.77
0.80
0.70
0.76
0.78
0.73
1749
Appendix 1. Continued.
Construct
Investment
Factor
Process investment
Investment in planning
Managerial investment
New product investment
Plant capabilities
Level of resources
Not included in the analyses
Performance relative to target
Time performance
Cost and quality performance
Not included in the analyses
Item
ICC
Setup time reduction
Plant automation
Process analysis
Productivity improvement
Throughput time reduction
Material and/or capacity planning
Just-in-time systems
Statistical process control
Supplier partnerships
Total quality management
Employee participation programs
New product development
0.67
0.73
0.73
0.75
0.85
0.57
0.72
0.87
0.81
0.89
0.76
0.77
Capabilities in developing new products
Managerial capabilities
Level of technical resources
0.66
0.62
0.34
Manufacturing throughput time
(from start until finish of production)
Service level (on-time delivery to customers)
Delivery lead time (from customer’s order
until delivery)
Average defect rates at the end of manufacturing
Average unit production costs for a typical product
Productivity of direct production workers
Overall product quality as perceived by the customers
Rate of new product introduction
Equipment setup time
0.61
0.75
0.60
0.75
0.80
0.78
0.69
0.47
0.54
Note. For most of the items the ICC exceeds 0.60, which is the cutoff value suggested by Boyer and Verma (2000). For the item “investments in materials
and/or capacity planning,” the ICC reaches 0.57. However, because this is very close to the cutoff level, the item has been retained in the analyses. The ICC
cutoff level of 0.60 was not reached for the items “level of technical resources,” “rate of new product introduction,” and “equipment setup time.” Consequently,
these items have been omitted from the analyses.
Appendix 2. Overview of Interview Comments
Cluster A
Independent 2×
Local 2×
Improved/learning 4×; problem solvers 2×
Manufacturing capabilities 7×
Motivated 2×; creative
Development for their own
Product focused
Cluster B
Only executes 2×
Some development 9×; a lot of development 1×
Expert (in quality, service, material handling, CIM,
energy savings)
Limited; simple
Complex; difficult; below expectation
Local market 7×; Local improvements; local culture
Needs help 2×; receives (technical) support 3×
Lives its own life; goes its own way; distance
Lack of motivation; inflexible management;
management problems 2×; lack of skills 4×;
Insufficient experience; negative mentality;
counterproductive mentality; mentality is to accept
Problems as they come; social climate has improved
Crew of Belgian managers; no own management;
group of expatriates; satellite plant 2×;
Management input from HQ; strong liaison manager
in other plant
Cluster C
Pilot plant 1×; test site 1×; development site 4×
Center of excellence 3×; center of competence
Product know-how 3×; process know-how
Training center
Supports other plants 2×; motor for all products 1×
Close to HQ; home player; mother plant 1×
Quite motivated to experiment
Lack of focus
Lack of investment
Satellite 1×
Product specialist
Cluster D
Center of excellence 2×; development center; pilot
plant 3×
Think tank; generator of ideas
Atmosphere of activity; do-spirit; enthusiasm; happy to
experiment
Motor of the other plants; engine
Gives technical assistance; high tech; specialist; process
know-how
Close to HQ 1×
1750
References
Aldenderfer, M. S., R. K. Blashfield. 1984. Cluster Analysis. SAGE
Publications, Beverly Hills, CA.
Allen, T. J. 1977. Managing the Flow of Technology. MIT Press, Cambridge, MA.
Bartlett, C. A., S. Ghoshal. 1989. Managing Across Borders: The Transnational Solution. Hutchinson Business Books, Boston, MA.
Boyer, K. K., R. Verma. 1996. A note on the use of multiple raters
in survey-based operations strategy research. Proc. Decision Sci.
Annual Meeting, Orlando, FL, 1344–1346.
Boyer, K. K., R. Verma. 2000. Multiple raters in survey-based operations management research: A review and tutorial. Production
Oper. Management 9(2) 128–140.
Calinski, T., J. Harabasz. 1974. A dendrite method for cluster analysis. Comm. Statist. 3 1–27.
Carmines, E. G., R. A. Zeller. 1979. Reliability and Validity Assessment.
Sage Publications, Beverly Hills, CA.
Chakravarty, A., K. Ferdows, K. Singhal. 1997. Managing international operations versus internationalizing. Production Oper.
Management 6 100–101.
Chew, B. W., T. F. Bresnahan, K. B. Clark. 1990. Measurement, coordination and learning in a multiplant network. R. S. Kaplan,
ed. Measures for Manufacturing Excellence. Harvard Business
School Press, Boston, MA, 129–162.
Collins, R. S., R. W. Schmenner, D. C. Whybark. 1989. Pan-European
manufacturing: The road to 1992. Eur. Bus. J. 1 43–51.
De Bodinat, H. 1980. Influence in the Multinational Corporation: The
Case of Manufacturing. Arno Press, New York.
DeVellis, R. F. 1991. Scale Development. Sage Publications, London,
UK.
Doz, Y. L., C. K. Prahalad. 1991. Managing DMNCs: A search for a
new paradigm. Strategic Management J. 12 145–164.
Doz, Y., J. Santos, P. Willimanson. 2001. From Global to Metanational.
Harvard Business School Press, Boston, MA.
Duda, R. O., P. E. Hart. 1973. Pattern Classification and Scene Analysis.
John Wiley & Sons, New York.
Dunning, J. H. 1993. Multinational Enterprises and the Global Economy.
Addison-Wesley, Reading, MA.
Eisenhardt, K. M. 1989. Building theories from case study research.
Acad. Management Rev. 14 532–550.
Ferdows, K. 1989. Mapping international factory networks. K. Ferdons, ed. Managing International Factory Network. Elsevier Science Publishers, Amsterdam, The Netherlands, 3–21.
Ferdows, K. 1997a. Made in the world: The global spread of production. Production Oper. Management 6 102–109.
Ferdows, K. 1997b. Making the most of foreign factories. Harvard
Bus. Rev. 75(2) 73–88.
Flaherty, M. T. 1986. Coordinating international manufacturing and
technology. M. C. Porter, ed. Competition in Global Industries.
Harvard Business School Press, Boston, MA, 83–109.
Flaherty, M. T. 1996. Global Operations Management. McGraw Hill,
New York.
Galbraith, J. R. 1977. Organization Design. Addison-Wesley, Reading,
MA.
Gerlach, M. L. 1992. The Japanese corporate network: A blockmodel
analysis. Admin. Sci. Quart. 37 105–139.
Ghoshal, S. 1986. The innovative multinational: A differentiated network of organizational roles and management
processes. Doctoral dissertation, Harvard Business School,
Cambridge, MA.
Ghoshal, S., C. A. Bartlett. 1988. Creation, adoption, and diffusion
of innovations by subsidiaries of multinational corporations.
J. Internat. Bus. Stud. 19(3) 365–388.
Ghoshal, S., C. A. Bartlett. 1990. The multinational corporation
as an interorganizational network. Acad. Management Rev. 15
603–625.
Ghoshal, S., H. Korine, G. Szulanski. 1994. Interunit communication
in multinational corporations. Management Sci. 40 96–110.
Gupta, A. K., V. Govindarajan. 1991. Knowledge flows and the
structure of control within multinational corporations. Acad.
Management Rev. 16 768–792.
Hayes, R. H., R. W. Schmenner. 1978. How should you organize
manufacturing? Harvard Bus. Rev. 56(1) 105–119.
Ketchen, D. J., C. L. Shook. 1996. The application of cluster analysis
in strategic management research: An analysis and critique.
Strategic Management J. 17 441–458.
Knoke, D., J. H. Kuklinski. 1982. Network Analysis. Sage Publications, Beverly Hills, CA.
Kobrin, S. J. 1991. An empirical analysis of the determinants of
global integration. Strategic Management J. 12 17–31.
Kogut, B. 1989. Research notes and communications: A note on
global strategies. Strategic Management J. 10 383–389.
Miles, M. B., A. M. Huberman. 1994. Qualitative Data Analysis. Sage
Publications, Thousand Oaks, CA.
Milligan, G. W. 1996. Clustering validation: Results and implications for applied analyses. P. Arabie, L. J. Hubert, G. De Soete,
eds. Clustering and Classification. World Scientific Publishing,
Singapore, 341–375.
Milligan, G. W., M. C. Cooper. 1985. An examination of procedures
for determining the number of clusters in a data set. Psychometrika 50 159–179.
Mintzberg, H. 1979. The Structuring of Organizations: A Synthesis of
the Research. Prentice Hall, Engelwood Cliffs, NJ.
Nohria, N. 1992. Is a network perspective a useful way of studying organizations? N. Nohria, R. G. Eccles, eds. Networks and
Organizations. Harvard Business School Press, Boston, MA,
1–22.
Nonaka, I., H. Takeuchi. 1995. The Knowledge-Creating Company:
How Japanese Companies Create the Dynamics of Innovation.
Oxford University Press, New York.
Roth, A. V., A. E. Gray, J. Singhal. 1997. International technology
and operations management: Resource toolkit for research and
teaching. Production Oper. Management 6 167–187.
Skinner, W. C. 1964. Management of international production.
Harvard Bus. Rev. 42(5) 125–136.
Speier, C., M. Swink. 1995. Manufacturing strategy research: An
examination of research methods and analytical techniques.
Working paper, College of Business Administration, University
of Oklahoma, Norman, OK.
Szulanski, G. 1996. Exploring internal stickiness: Impediments to
the transfer of best practice within the firm. Strategic Management J. 17 27–43.
Vereecke, A., R. Van Dierdonck. 1999a. Design and Management of
International Plant Networks: Research Report. Academia Press,
Gent, Belgium.
Vereecke, A., R. Van Dierdonck. 1999b. Network analysis: A powerful tool in global manufacturing research. Proc. Decision Sciences
Annual Meeting, New Orleans, LA, 1250–1252.
Vereecke, A., R. Van Dierdonck. 1999c. A typology of international plant networks: Empirical study. E. Bartezzaghi, R. Filippini, G. Spina, A. Vinelli, eds. Managing Operations Network.
SGEditoriali, Padua, Italy, 653–660.
Yin, R. K. 1984. Case Study Research: Design and Methods. Sage Publications, Beverly Hills, CA.

Strategy in Industrial Networks:

Transcription

Similar documents

PDF.

Wavelength 201403 London Snapshots

IKEA-The Global Retailer - Berkeley-Haas

sIMpLE LIVIng - vr

IKEA Kitchens

IKEA helps customers find their way with indoor Google Maps

check it out

november