strategy in industrial networks - welcome to...

Strategy in Industrial Networks:EXPERIENCES FROM IKEA

Enrico Baraldi

I KEA was founded over 60 years ago in Southern Sweden. It has sincegrown to become the world’s largest furniture retailer; in 2006, it hadsales of over Euro17 billion, as well as 12,000 product items and104,000 employees. The company’s focus has consistently been on mar-

keting products at extremely low prices. Its first purchases in the 1950s weremade from producers’ unsold stocks, in order to keep costs low. However, largesales success soon allowed IKEA to start ordering models of its own design fromlocal 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 itspurchasing strategy, relying on long-term relationships with selected suppliers asexternal sources for its offerings. Today, its supply network spans the entireworld and has become increasingly complex. However, the use of this network isstill in accordance with the same basic strategy as in the 1960s: to design andpurchase products that entail low production and transportation costs. IKEAachieves this by carefully taking into account, in its design and purchase strat-egy, all the activities performed in the network, from raw materials to customerhomes. Remaining faithful to its original external orientation, IKEA performsonly a few of these activities internally, while it intensively uses its relationshipswith suppliers to combine its internal and their external resources for the sake ofboth efficiency and development. For instance, products are developed in closeinteraction with suppliers while taking into consideration the impact of the rawmaterials, components, and facilities involved, since all these resources entail

99CALIFORNIA MANAGEMENT REVIEW VOL. 50, NO. 4 SUMMER 2008 CMR.BERKELEY.EDU

The author would like to thank Jan Gardberg of IKEA and all other managers that kindly provided theempirical material for this article, and two anonymous reviewers for their insightful comments.

costs and have an impact on quality, design, and function. In fact, next to lowcosts, reasonable quality, appealing designs, and adequate product functionalityare major goals for IKEA. These goals induce the company to promote a constantproduct and technical development, which contributes to its image as an innov-ative and fashion-oriented firm, but which depends heavily on the contributionof its entire network of suppliers.

To cope with such tasks, IKEA needs advanced skills in marketing, retail-ing, logistics, purchasing, product development, and technologies. This need ofcompetence is reflected by IKEA’s complex organization, which consists of over550 business units specializing in these fields and spread over more than 50countries. However, the complexity of IKEA’s organization is overshadowed bythat 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 IKEAstores are located in 30 countries includingEurope, Australia, the U.S., and China.Between IKEA’s stores and suppliers stands avital, but less visible part of IKEA’s network:

its wholesale and logistic operations, comprising 26 Distribution Centers spreadover 12 countries. Since IKEA does not own any transport facilities, this net-work is physically connected via another group of external actors, a few hun-dred logistic partners.

A pivotal role in this network is played by “IKEA of Sweden,” a leadingbusiness unit that not only manages IKEA’s product range, but also supervises theentire IKEA universe and develops long-term marketing, logistics, and purchas-

Strategy in Industrial Networks: Experiences from IKEA

UNIVERSITY OF CALIFORNIA, BERKELEY VOL. 50, NO. 4 SUMMER 2008 CMR.BERKELEY.EDU100

Enrico Baraldi is an Associate Professor atUppsala STS Center and at the Department ofBusiness Studies, Uppsala University, Sweden.<[email protected]>

FIGURE 1. IKEA and Its Industrial Network

Sub-suppliers (10,000)

Suppliers(1,300) Logistic Partners

(500)

IKEA ofSweden

IKEA’s Boundary

IKEA TradingOffices (40)

IKEADistributionCenters (26)

IKEA Stores(220)

ing strategies. In fact, whereas most IKEA units are rather specialized (e.g., localpurchasing for IKEA’s 40 Trading Offices), IKEA of Sweden has both an overallresponsibility and a coordinating role in the development, purchase, distribu-tion, and marketing of each single product.

This article analyzes the experience of IKEA in dealing with its industrialnetwork and discusses the structural components and dynamic interactions of a“network strategy,” that is, a strategy that considers and uses the external net-work 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 Swe-den, Poland, and Italy.1

The Pervasiveness of Industrial Networks

Networks are widely publicized and researched phenomena, especially inhigh-tech sectors,2 where the likes of Dell, Microsoft, or Genentech pursue theirnetwork strategies through R&D joint ventures, cross-licensing, or strategic alliances.However, these fashionable terms do not fit the type of networking going onwithin the furniture industry or other low-tech industries. However, traditionalsectors also present network-like structures. Examples include the tile industry,3

the apparel industry,4 and the Italian districts5—which specialize, for instance, inknitwear (Carpi), packaging machines (Bologna), or textiles (Prato). These sec-tors are composed of many small and medium-sized firms that develop closelinks to their suppliers and partners, and often also cooperate intensively withthem on technical issues.

Networks are not only important for small firms that need to interactwith their peers to supplement their limited resources, but networks are funda-mental for large companies as well. For instance, multinationals in the steel,6

paper,7 and automotive8 industries interact tightly with their suppliers, sub-sup-pliers, distributors, and customers to develop new technologies or increase effi-ciency. There are many examples of large firms from several sectors that reliedstrongly on networks for their rapid growth: Apple, Benetton, Toyota, Corning,and McDonald’s, to name a few.9 Finally, networks of stable relationships are thenorm in several industries, including construction, publishing, textiles, and cul-tural 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 bymainstream management research until around the 1980s. This neglect of inter-firm relationships stems from both the actual behavior of firms—in periods char-acterized by arm’s-length relations,12 close interactions and networks are moredifficult to discern—and the dominant research paradigms, which either haddifferent units of analysis or viewed relationships simply as odd exceptions.13

Although Richardson14 recognized the importance of business relation-ships as early as 1972, it took time for these ideas to enter mainstream strategy


CALIFORNIA MANAGEMENT REVIEW VOL. 50, NO. 4 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 net-works of social relations where trust matters,15 and Powell attributed networks astatus equal to markets and hierarchies, viewed as three alternative forms fororganizing economic activity.16 Eventually, Swedberg suggested that all marketscan be viewed as social structures filled with interactions, rather than as pureprice-driven mechanisms.17

In sum, there exists compelling evidence for the diffusion and persistenceof business relationships and networks, in all sectors and for all sizes of firms.Mainstream strategy research was late in recognizing the importance ofnetworks, and did not take them into account until the influence of sociologyand widely publicized strategic alliances or joint ventures made it impossible toneglect this interactive side of business life. However, there is still a risk thatfocusing on these special and conspicuous networking episodes may hide thebulk of networking activities and interactions that go on silently under the sur-face of daily business activity.18 To avoid such a risk, one can rely on a theoreti-cal perspective that views industrial networks and relationships as essentialphenomena, that is, as the norm rather than the exception: this is the viewpointof 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 merelyexceptions in economic organizing, in a world where firms can only choosebetween pure market exchanges or hierarchical control? What happens if weoverstate the argument of Swedberg and start considering instead networks asthe norm, that is, as the most natural and normal form of economicorganizing?19 Following this reasoning, networks came first, while markets andhierarchies are both human constructions, that is, structures imposed on net-works for the sake of transparency (markets) or control (hierarchies).20 Accept-ing networks as the norm can also help focus on the effects that they produce.Examples of such “network effects” are unexpected product failures,21 irrationalpatterns of electricity consumption,22 and imbalances spread by IT systems.23

However, to enable researchers and managers make sense of networks and theireffects, 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” (Indus-trial Marketing and Purchasing).24 This approach, sometimes referred to as“Markets-as-Networks,” grew out of extensive empirical studies of industrialbuyer and seller relationships conducted in Europe in the 1960s and 1970s, thatis, well before the modern frenzy over networks.25 The early empirical findingswere then related to sociological theories of exchange26 in order to develop aseries of models of the dyadic interaction between firms. These models stressedthe importance of power/dependence, cooperation, closeness, and expectationsin the daily interactions between buyer and seller.27 It was only a short step from



single relationships to networks of relationships: when IMP researchers recognizedthat single relationships are related to each other via technical, economic, andsocial interdependencies, the way was paved for more complex models stretch-ing to the entire network surrounding a firm.28

The Markets-as-Networks approach has gained currency within the fieldof industrial marketing29 and international business,30 but it can also be used tocast new light on strategic issues of efficiency and development.31 To summarize,the Markets-as-Networks approach focuses on all types of inter-firm interac-tions. The starting point is that firms constantly interact with counterparts suchas key suppliers and customers through business relationships that are related ina network structure. Consequently, firms are embedded within networks thatexist beyond their will. From a strategic point of view, firms need to be aware ofthese networks and of how to use them actively. Companies vary widely in theirwill 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 areembedded in a network, which can be both good and bad.33 It is therefore advis-able for firms to understand how networks work and how they can beapproached 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 andexternal resources needs to emerge before a company can use the network in itsdaily interactions with counterparts. Therefore, the analytical frame that we willapply to discuss IKEA’s network strategy comprises three structural componentsand two types of dynamic interactions (see Table 1). The structural componentsconcern 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



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 arelationship’s goals, role and function).

2. Forming the Network Structure:This structure is based on the number of businessrelationships, their roles in the network, hierarchy, andgeographical location.

3. Evaluating Goal Matching with theNetwork:

Comparing a firm’s goals/resources with those of thenetwork.Which external resources are needed andavailable? What are the goals of other actors?

Combining Resources:Concrete and complex resource combinations createdthrough interaction processes with external actors:inter-organizational routines/joint projects

1. Interacting via Inter-OrganizationalRoutines:

Repetitive and formal processes that mobilizerelationships and enable resource combinations tomaintain efficiency.

2. Interacting via Joint Projects:Ad hoc and informal processes that mobilizerelationships and enable resources combinations tofoster development.

involved actors), and the configuration of external resources (e.g., their distribu-tion and technical connections). These structural elements are relatively stableand do not change overnight in a network, as stressed by the Markets-as-Net-works approach.34 The dynamic interactions concern instead the processes thatgo on daily in any business relationship in terms of the activities performed, thecommunications among actors, and the concrete combinations and adaptationsof resources across organizational boundaries. These interactions are termed“dynamic” because the underlying processes change more frequently and cancontribute to changes in the structural components (e.g., to the emergence ofnew long-term goals or to the inclusion of new actors in the network structure).

The three structural components are: definition of the content of eachbusiness relationship; formation of the network structure; and evaluation of thematching of IKEA’s goals and resources with those of the network. The logic ofComponent 1 is that IKEA strives to influence such relationship contents asexchanged volumes, commitments, trust and learning depending on how eachrelationship can contribute to achieving IKEA’s goals. However, Component 2stresses that focusing on a single relationship would not be enough becauseIKEA’s goals can be reached only by connecting several relationships into abroader network structure, including the establishment of new relationships andthe assignment of specific roles to certain counterparts within a hierarchy ofrelationships. Finally, Component 3 suggests that a firm needs to evaluate howits 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 asestablishing new relationships or changing their contents.

The three structural components provide the basic network structurewithin which dynamic interactions unfold continuously. At a general level, thesedaily interaction processes entail combinations of resources by IKEA and otheractors in the attempt to match IKEA’s goals and resources with those in the net-work. At a specific level, there are two types of dynamic interactions: interactingvia inter-organizational routines for efficiency purposes; and interacting via jointprojects for development purposes. Whereas inter-organizational routines are,for efficiency’s sake, rigid scripts executed repetitively, joint projects are develop-ment processes that can stretch over several years, thus becoming less control-lable by IKEA and more uncertain.

IKEA,The Interacting Company:Handling a Network of Relationships

IKEA’s relationships and network are pivotal in fostering development ofIKEA’s products and technologies, and in sustaining efficiency in its daily opera-tions. IKEA is certainly not the only firm that relies on extensive network inter-actions for its strategy. In particular, other retailing firms such as Benetton andHennes & Mauritz apply a similar approach to complex webs of partners. How-ever, IKEA does it in a special way. A major difference in comparison, for



instance, with Wal-Mart,35 is that IKEA stretches its interactions as far upstreamas possible in the network, all the way to raw material suppliers. IKEA inten-sively cooperates with these suppliers in order to ensure the quality and theenvironmental friendliness of inputs to its direct suppliers downstream. Anotherimportant difference is the extent of IKEA’s cooperation with its partners: thisextends to complex and enduring development projects whereby IKEA’s prod-ucts and technologies are co-developed with suppliers.36

Finally, IKEA’s interaction mechanisms strongly differ from both hierar-chically steered networks, such as the Japanese Keiretsu,37 and looser arm’s-length 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 largebuyer, IKEA takes a long-term approach and strives to build lasting relationshipsbased on mutuality. This means that it also explicitly considers the interests ofsuppliers, who would otherwise lose the motivation to interact with IKEA.Moreover, IKEA does not strive to unilaterally control these relationships, butrelies on extensive delegation of tasks to its suppliers, and even accepts beingdependent on some of them. Thus, for IKEA, mutual trust and commitment aremore important interaction mechanisms than power.

IKEA’s Background and the Structure of Its Network

IKEA’s concern with providing low-price products characterizes both itscurrent strategy and its history. The introduction of flat packs in the 1950sallowed important savings in transportation and production costs. In fact, IKEA’scustomers took over assembly activities, and suppliers only needed to deliverun-assembled furniture components. Later on, selling costs could be containedthanks to “showroom-warehouses”: retail stores were redesigned so as to com-bine a large exhibition area with an adjacent self-service warehouse. IKEA couldafford such low retail prices that its sales rocketed in the 1960s, signaling thestart of its expansion, with several stores in Sweden and abroad. However, astrong reaction soon followed from Swedish furniture retailers, who tried tostrangle IKEA’s purchase sources by requiring that all Swedish producers stoppedsupplying IKEA. IKEA’s countermove was to go looking for suppliers abroad.Thus, the first agreements with Polish producers were signed in the 1960s, lay-ing the groundwork for many business relationships that still exist today.

Long-lasting relationships with selected key suppliers are still the hall-mark of IKEA’s purchasing and product development strategy. As IKEA isdirectly involved only in conceiving, distributing, and selling its products, itneeds partners that can concretely develop and produce a total of 12,000 itemsthat meet its cost, quality, and design goals. Still, extensive knowledge of thenetwork, from raw materials to customer homes, is pivotal for IKEA to conceiveproducts that are not only “cool” and functional enough to sell, but that can alsobe produced according to set cost and quality goals.

Figure 1 gave a simplified idea of the extension and complexity of IKEA’snetwork, encompassing about 10,000 organizations, from sub-suppliers to IKEAretail stores. Connecting this network geographically requires a very advanced



logistic system, including 20,000 transport corridors. However, what countsmore than these physical connections are the organizational connections thatIKEA needs in order to interface with this complex network. In fact, IKEA cre-ated specific organizational interfaces to handle relationships with the variousactors: not only the 1,300 direct suppliers and 500 carriers, but also hundreds ofthe 10,000 sub-suppliers, especially large firms supplying core materials such aspackaging (e.g., SCA) or coatings (e.g., Akzo-Nobel).

The strategic role of the unit known as IKEA of Sweden has already beenmentioned. From IKEA’s hometown of Älmhult, its 700 employees performanother fundamental role: interacting from a central position with key logisticpartners and suppliers on such strategic issues as long-term capacity planning andmajor technical development projects. Conversely, at a local level, the main interfaceswith suppliers are IKEA’s 40 Trading Offices, employing 3,000 people worldwide.Even if these local purchasing offices replicate the structure and competence ofthe central unit, IKEA of Sweden (with purchase strategists, product categoryspecialists, technicians, order managers, and logisticians), their interactions withsuppliers mostly concern daily issues (e.g., orders and deliveries) and occasion-ally development or tendering for new product assignments. IKEA pays its Trad-ing Offices a percentage on the purchases made from the suppliers they“represent.” Finally, the daily logistic coordination with all suppliers and carriersis handled by IKEA’s 26 Distribution Centers.

IKEA’s Approach to Business Relationships

IKEA’s approach to supplier relationships depends on the productinvolved. Complex products, both in terms of construction (e.g., sofas) and ofproduction technology (e.g., the “Lack” table, featured below), are assigned tosuppliers with which extensive mutual trust, commitment, and knowledge havebeen established through long-term relationships. These strong relationshipsentail extensive joint investments in facilities. On the other hand, productswhose technical simplicity means they are easily interchangeable (e.g., rugs) areusually purchased through shorter-term relations. A similar variation exists inthe relationships with logistic partners: out of over 500 such partners, IKEA hasdeveloped 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 establishedrelationships. A common trait is IKEA’s attempt to avoid abusing its power posi-tion. The focus is instead on the mutual benefits accruing to both IKEA and itssuppliers. Cooperation with IKEA should ideally bring suppliers advantages suchas stable and long-term orders or technical development projects where IKEAcan “pay” for thousands of tests. In fact, IKEA is well aware that those actorsthat no longer have advantages may end their cooperation (as did the coatingsupplier Becker-Acroma in the episode reviewed in Appendix A).

Moreover, IKEA does not unilaterally control these relationships butaccepts that it may sometimes be strongly dependent on its suppliers, as in thecase of key logistic partners and those suppliers that daily refill IKEA’s stores



through “Vendor-Managed Inventory.” IKEA does not even unilaterally controlsuch key sources of its innovativeness as product and technology developmentprojects. Here, IKEA delegates much responsibility to the most competent part-ners, either those who have long been in charge of manufacturing a certainproduct, 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 ofthe actors involved. There are many small suppliers (and especially sub-suppli-ers) that are highly dependent on IKEA, and which IKEA can more directlyinfluence with its powerful position. However, IKEA exerts its influence notonly by holding down prices, but also by inducing suppliers to upgrade theirtechnologies in ways that eventually benefit the suppliers themselves. This mutu-ality is in fact the hallmark of all of IKEA’s business relationships, even of thosewhere IKEA could exploit a power position due to the overdependence of asupplier.

On the other hand, there are also larger counterparts with which IKEAhas a much more balanced power relation. These large actors are not easilyinfluenced by IKEA, and they comply with IKEA’s requests only if they gainsomething from a specific cooperation. For instance, Akzo-Nobel, a sub-supplierof coatings, is a 15,000 employee and €6 billion chemical group. IKEA is cer-tainly an important customer for Akzo-Nobel, but not to the extent that Akzowill blindly comply with any requests. Instead, Akzo-Nobel chooses to engage inIKEA’s development efforts when this provides it with specific advantages, suchas learning a new technology. Similarly, the key logistic partner Maersk hasthousands of trucks and employees, alongside hundreds of vessels and localoffices: in this case, it is more IKEA’s transportation routines that need to fit intothe logistic network of this partner rather than vice versa. Relationships withsuch large actors directly involve IKEA of Sweden for central negotiations; andeven 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 pointof view,39 because they are spread over the regions that provide specificresources or location advantages, such as nearness to IKEA’s major markets(Germany and Central Europe). Geographical location is one of the key factorswhen selecting new suppliers, because it strongly affects costs, competences, anddelivery times. The resulting geographic pattern is as follows: Chinese suppliersrank 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 compe-tence, and nearness to Central Europe; and Swedish suppliers, despite highcosts, still rank third thanks to their advanced technical competence.

Geography is an important factor, but not the only factor, in supplierselection and hence in constructing the structure of IKEA’s network. During theselection process, IKEA of Sweden and the local Trading Offices that propose asupplier also evaluate these other factors, in order of importance: total costs,



provided that IKEA’s quality and environment requirements are respected; cur-rent and planned production capacity; technical competence; and readiness tomake investments for and with IKEA. An additional rule is that suppliers shouldnot depend on IKEA for more than 50% of their turnover. The rationale here isnot only the avoidance of over-dependent suppliers that would be hurt toomuch if IKEA’s order volumes should decrease, but also reliance on the learningand development achieved by a supplier thanks to interactions with customersother than IKEA.

Then, with some suppliers, IKEA develops long-lasting and complex-content relationships, which entail large volumes and commitments. IKEA ofteneven purchases machinery for these suppliers and trains their personnel. Aneven more restricted group of highly trusted suppliers is then invited to take partin complex technical development projects with IKEA: these suppliers are thosewith greater competences (e.g., in logistics or coating technologies) and thosewilling to become more committed to IKEA due to a positive history of interac-tions or strong expected benefits. IKEA already understood in the 1960s thatusing a long-term approach to purchasing actually favors IKEA itself, by allow-ing lower production costs and purchase prices and faster and improved devel-opment for its products. This approach produces even better results if IKEAfollows a philosophy of mutuality and also takes into account the interests of itssuppliers: this is reflected in IKEA’s attempt to balance production volumesamong several suppliers and in IKEA’s investment programs to upgrade a sup-plier’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 fulfill-ment of IKEA’s orders to “Vendor-Managed Inventory”) must correspond both toincreased IT integration with IKEA and to improved logistics capabilities. How-ever, this model does not build only on improved routines and IT, but also onthe development of a stronger business relationship, entailing more trust andcommitment between IKEA and a supplier. This also implies better informationflow and the commitment of concrete resources, ranging from dedicated person-nel to building new warehouses at a supplier’s site.

The deepening of a supplier relationship around IT and delivery issuesstarts from consistent delivery performances, and proceeds along a supplier’simprovements on these dimensions:

▪ increased experience of IKEA’s ordering routines or retail sales patterns;

▪ logisticians and order management teams dedicated to handling IKEA’sorders;

▪ improved communication with IKEA’s units such as retail stores andIKEA of Sweden;

▪ improved IT competences acquired from daily use of advanced IT toolssuch as ERPs;

▪ the willingness to invest in expensive new IT solutions;

▪ a direct knowledge of the IT ordering systems located at IKEA; and



▪ improved manufacturing flexibility and willingness to make physicalinvestments (in production or warehousing capacity) in order to copewith fluctuating order volumes.

Suppliers who achieve the aforementioned capabilities (often thanks tosupport and training provided by IKEA) and make the required commitmentscan move up along the ladder depicted in Figure 2. However, IKEA is also cost-driven 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



FIGURE 2. IKEA’s Ladder Model for Supplier Interactions and IT/Logistics Capabilities

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.

Call-Off

OPDC(Order-Point

Distribution Center)

VMI(Vendor-Managed

Inventory)

Capability Requirements:(entry step: all suppliers)

• delivery precision enough tofulfil only 13 orders per yearwithin 4 weeks;

• simple IT connection via non-online system ECIS(Electronic Commerce forIKEA Suppliers);

• no need to be sole itemsupplier in the DC area.

Capability Requirements:(almost 30% of supplies)

• increased delivery precisionto fulfil daily orders within 12days;

• high production flexibility -expanded warehousingfacilities;

• IKEA-dedicated ordermanagers team,

• online EDI-ERP connectionwith IKEA’s order systemINOS;

• need to be sole item supplierin the DC area.

Capability Requirements:(1% of suppliers qualify:trustworthy and proactive)

• Same as OPDC, plus…

• highest delivery precision;

• ability to manage IKEA stores’inventories;

• forecast retail sales;

• set security stock levels;

• plan/organize transport;

• interact with all IKEAdistribution units and withIKEA of Sweden for plannedsales campaigns

• deep knowledge of IKEA ITsystems for inventory,ordering, and transport

• direct online access to all theabove IT systems.

network is heterogeneous, a feature that IKEA views as a key source ofdevelopment.

What Goes on in IKEA’s Network? Two Interaction Processes

The rich structure of relationships and the widespread competencesaccessed by IKEA through its network would be useless if IKEA did not have theability to combine these resources to achieve efficiency and foster development.At a general level, this ability relies on IKEA’s organizational structure, on itsstrong competence in such areas as product and technology development orlogistics, and on its long-term approach and network-oriented culture. At amore concrete level, the combination of external resources and competencesrelies on two managerial tools: detailed routines performed repetitively by IKEAand its suppliers in such efficiency-driving processes as order management; andad hoc projects that tackle specific product and technical development issues andoften involve up to 20 firms. Inter-organizational routines and projects areessential mechanisms to promote and handle the numerous interactionprocesses 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 commu-nicates them to the supplier in charge of each specific product. Setting exactorder quantities is critical in order to avoid stock-outs in retail stores or extrainventory costs. Timing is essential too, because suppliers are bound to givenlead-times and cannot react to a delayed order with immediate deliveries. IKEAachieves efficiency in the order-management process through an advanced pro-cedure that estimates its own replenishment needs and very structured routinesthat transfer these needs to each supplier, stating such details as the exactresponse time and modes required from them. Nothing is left to chance, andIKEA considers many variables (such as stock levels, lead-times, and goods-in-transit) 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 prod-ucts, with very different production approaches and lead-times. For instance, asofa will be finished after a customer order is placed, because the producer needsto wait until the customer has chosen the fabric; whereas standard productssuch as coffee tables are produced before orders, against sales forecasts. There-fore, IKEA’s orders of sofas and other customized products are steered by end-customer orders, and so need longer lead-times; whereas orders of standardproducts are triggered by automatically preset reorder points at IKEA stores andDistribution Centers, and have shorter lead-times. To be able to fulfill an order, asupplier needs to master the details of the ordering routine specific to each ofthe products it delivers to IKEA. Manufacturers supplying IKEA with volumeshigher than Euro50 million yearly have large teams of IKEA-dedicated ordermanagers, fully engaged in handling IKEA’s orders.

To further complicate things, not all IKEA suppliers are equally able torespect short lead-times and to fulfill orders coming in daily. Therefore, different



suppliers interact with IKEA by means of different ordering routines, dependingon their ability and type of product. For instance, IKEA cannot implement withall suppliers the ordering routine “OPDC” (Order-Point Distribution Center),which involves issuing orders every day and requiring fulfillment within 12days. Instead, many suppliers follow the “Call-Off” routine, which involves issu-ing orders every fourth week and requiring deliveries within the following 4weeks. The OPDC modality is clearly more precise for IKEA’s receiving units andreduces stock levels and goods-in-transit within IKEA. However, OPDC createsstronger pressures on suppliers not only to learn a new ordering routine, butalso to become more flexible and speculative, or even to increase finished goodsstocks 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 toIKEA’s stock data and assigns them the responsibility to decide when and howmuch to deliver. In this way, a supplier is empowered to exploit its knowledge ofIKEA’s ordering patterns. However, this knowledge, essential to forecastingIKEA’s needs, only comes after having interacted daily with IKEA for someyears. Additionally, with VMI, close interaction and joint planning between asupplier and IKEA stores become necessary to secure product coverage for spe-cial events such as store openings and sales promotions. All order managementmodalities are highly interactive routines performed daily by IKEA and its sup-pliers according to rigid scripts, mostly decided unilaterally by IKEA. At the sametime, IKEA strives to introduce new ordering routines such as OPDC with keysuppliers, according to the aforementioned ladder model that requires directsupplier involvement and large mutual investments.

The “Printed Veneer” Project for the “Lack” Table

An important type of development projects in IKEA’s network concernsnew technologies. In fact, IKEA searches constantly for technologies that canreduce costs, improve quality, or allow new designs. Coating technology, forinstance, is pivotal in furniture manufacturing, because it strongly affects bothdesign and quality, but it is also costly and, if badly handled, can become haz-ardous to health. IKEA is therefore very concerned with coating technologiesand has promoted hundreds of projects to improve them at its suppliers. One ofthese projects is the “printed veneer” project, which addressed the high cost ofthe veneers used for IKEA’s best-selling “Lack” table, with more than 2.5 millionunits sold yearly.

In 2000, IKEA of Sweden raised some concerns about the high cost ofveneers, which accounted for about 20% of the material costs for the veneeredversions of Lack. These concerns triggered a series of discussions and an evalua-tion of potential solutions at Swedwood Poland, the manufacturer of Lack tables.Akzo-Nobel, one of their major coating suppliers, proposed to Swedwood a newcoating technology that would allow substituting real veneers with a printedpattern. Thus, Swedwood and Akzo-Nobel initiated a large technology develop-ment project with the aim of printing veneer on wood. This project was particu-



larly important not only for cost reduction, but also for product design and aes-thetics. Veneers are very visible on Lack tables and so the need to respect theperceived-quality requirements made this project particularly complex. The aes-thetic target was that consumers should not notice any major difference in theprinted pattern compared to real veneers. Akzo-Nobel took a leading role inperfecting the technology and introducing it into Swedwood’s plants. Leadingthe project meant identifying suppliers of the necessary equipment and supervis-ing the many tests required to fine-tune the technical solution. Many technicaland organizational resources had to be combined during this project (see Figure3, which shows the products, equipment, business units, and relationships thatwere involved).

The involved business units were all related by long-term relationships. Inparticular, Akzo had been supplying Swedwood for 20 years and closely inter-acted with IKEA of Sweden to negotiate the prices and conditions applicable toIKEA’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.



FIGURE 3. The Network of Resources Combined in the “Printed Veneer” Project

Akzo-Nobel

IKEA of Sweden

Coating Line

Coatings/Inks

Swedwood

"Lack"

Product

Equipment

Business Unit

Business Relationship

SorbiniBürkle

IKEA

HDF

To start with, Akzo had to develop new coatings and inks. Moreover, many tech-nical resources fell outside Akzo’s competence: new coating lines, resemblingprinting presses instead of traditional coaters, were necessary. Therefore, Bürkleand Sorbini, two coating lines suppliers who had previously supplied Swedwoodand cooperated with Akzo, modified their equipment to suit the new process.Following Akzo and Swedwood’s specifications, Bürkle and Sorbini adapted twocoating lines that were installed at Swedwood’s plants. However, the mostdemanding part of the project was getting all the technical resources to worktogether to satisfy IKEA’s requirements. This took over a year of tests, led byAkzo at Swedwood’s plants. On several occasions, Akzo had to modify its coat-ings and inks to allow them to work together with HDF and the new coatinglines.

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 ofinexpensive and high-quality veneers purchased by IKEA’s Trading Offices. Still,the new technology was now available, and IKEA decided to apply it to otherproducts manufactured by Swedwood, namely, a series of shelves that wereofficially launched in 2002. In initiating this project, IKEA had justified its highcosts 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 relyon the fact that Swedwood manufactured several other IKEA products withsimilar 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’snetwork. Moreover, in 2006 the aforementioned large supply of inexpensiveveneers 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 printdirectly on IKEA’s furniture any pattern developed by IKEA’s designers. Thus, atechnical development initiated for cost reduction has opened up another way tosustain IKEA’s image as an innovative and cool furniture company.

Interaction processes such as ordering routines and the printed veneerproject would not be possible without two important premises: IKEA’s internalstructure and competences and its well developed interfaces for interacting with thenetwork. For instance, IKEA of Sweden employs 700 people specializing in fur-niture technologies, logistics, and so on. That unit includes 50 “purchase strate-gists” and 100 technical experts. Moving from IKEA’s center to its periphery, weencounter 40 local Trading Offices employing 3,000 people who deal daily withsuppliers. Therefore, IKEA’s multifaceted organization, with its many internalcompetences and external interfaces, is necessary for IKEA to pursue its networkstrategy and interact with the actors in its network.



Discussion: Structures and Dynamic Interactions in a Network Strategy

What sustains IKEA’s network strategy? In other words, what are the keyfactors that enable IKEA to interact with its network in order to fulfill its effi-ciency and development goals? The framework introduced earlier contains bothstructural components and dynamic interactions. The idea is that before interactingwith the network and using it dynamically, a network structure needs toemerge. This is clearly a simplification, because network structure and dynamicsare tightly related. Indeed, a network structure can very well emerge more fromongoing interactions (inter-organizational routines or ad hoc projects)40 thanfrom IKEA’s active choices to shape it. Still, as the purpose of this framework isto 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 struc-ture and composition of this network are strategically important because theycreate both opportunities and restrictions for the actions that IKEA can undertake.Therefore, foresight is necessary over the goals to be reached, the resourcesneeded to accomplish them, and the actors to be mobilized to access externalresources. The actors and resources in IKEA’s network enable IKEA to achieve itsmain strategic goal: “providing inexpensive good-quality products to as manycustomers as possible.” Moreover, this network structure reflects more specificefficiency goals, such as low sourcing costs or smooth product flows, and develop-ment goals, such as new products or new technologies.

Structural Component 1: Defining Relationship Contents

Because of the importance of specific counterparts for achieving its goalsIKEA needs to define the content of each relationship, in terms of exchange vol-umes, learning, trust, commitment, and duration. However, the content of onerelationship cannot be defined in isolation from other relationships because it isrelated to the role played by that relationship in the overall network: forinstance, compared to an interchangeable sub-supplier, a first-tier unique sup-plier often receives larger volumes, sustained by the high trust and commitmentemerging from a long-term interaction. Not surprisingly, IKEA aims tostrengthen the latter type of relationships. But among first-tier suppliers thereare even more subtle differences based on the degree of IT integration withIKEA and on the operational responsibilities assigned to each supplier. Toincreasing supplier capabilities and commitment correspond increasing logisticresponsibilities, according to a ladder model (see Figure 2). This increase in sup-plier responsibility entails an increase in IKEA’s dependence on selected suppli-ers. Another dimension to categorize relationship content is the initiative andleadership taken by a partner in development processes. In fact, partners such asAkzo-Nobel, who are formally second-tier suppliers, often display strong initia-tive and take a leading role because of their advanced technical competences



and their own network of contacts. This type of relationships entails importantexchanges of technical knowledge with IKEA and is oriented towards the longterm since many development processes are time-consuming.

All the above forms of relationship content are issues that IKEA canrarely shape unilaterally. Moreover, IKEA’s ability to control and define a rela-tionship’s content is even more limited for its indirect relationships, such as thosebetween its direct suppliers and their sub-suppliers. Therefore, IKEA must agreeto delegate, for instance, project leadership to a key partner that can select otheractors and define the contents of their own relationships (as Swedwood andAkzo-Nobel do with their own partners). This delegation is extensive in develop-ment projects, where IKEA’s trusted key partners can partly affect the projectgoals and have ample freedom to mobilize other actors that have the requiredtechnical capabilities. IKEA’s delegation and a partner’s freedom in defining arelationship’s content are instead more restricted when it comes to inter-organi-zational routines (e.g., order management), which for efficiency’s sake need tobe more controllable and uniform.

However, even within IKEA’s direct relationships (e.g., with Swedwood),IKEA cannot unilaterally decide the relationship content. Content such asexchanged volumes, technical learning, and trust emerge instead from a two-party game that requires the capabilities, engagement, and commitment ofIKEA’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 technicallyor volume dependent on certain counterparts that it must partly accept theirefforts to shape the relationship content: for instance, exchanged volumes andtechnical investments are typically negotiated with key partners such as Swed-wood.

Structural Component 2: Forming the Network Structure

Each relationship plays a well-defined role within a partly hierarchicalstructure. IKEA’s network includes, for instance, first-tier as well as second-tiersuppliers, some of which play the role of “major” as opposed to “reserve” suppli-ers. Moreover, IKEA has direct relationships to certain actors (e.g., Swedwood)and indirect relationships to others (e.g., Sorbini). IKEA strives to shape thisstructure 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 sup-pliers from 2,400 to 1,300.41 Therefore, the remaining direct suppliers weregiven more responsibility. This implies not only larger volumes, but also thateach direct supplier now handles more complex relationships with an increasednumber of sub-suppliers. Each relationship also has a specific function for theentire network because it connects other relationships: for instance, the relation-ship IKEA-Akzo assigns a leading role to Akzo in setting the technical agenda ofits relationships with Sorbini and Bürkle (see Figure 3).

Another network structure feature essential for a global player is the geo-graphical spread of relationships. Suppliers are selected depending on the place-related features of their resources (e.g., labor costs, raw materials, and proximity



to end markets) and on their role in IKEA’s network (e.g., project-leading part-ners need to be located near other involved units). The geographical structure ofIKEA’s network is evolving towards a stronger emphasis on Chinese suppliers,who increased their share of IKEA’s purchases from 14% to 18% between 2003and 2006. The macro-reasons behind this trend are not only easier communica-tion and lower transportation and production costs, but also the technicaladvances made by Chinese suppliers, who now guarantee quality levels compa-rable with European ones. IKEA’s retail expansion in China further motivatesselecting local suppliers, for both political and transport reasons. However,despite this geographic trend, suppliers from Sweden, Italy, and Poland still playa major role when it comes to sophisticated technical developments that requirevery specific competences, tight coordination, and co-location (see the Lackexample).

Although IKEA strives to shape the configuration of its network, its con-trol over this structure is never complete: partners’ goals matter too. Further-more, despite IKEA’s will, some actors take steps either to leave the network orto change their position within it (see the case of Becker-Acroma in AppendixA). IKEA’s control and overview of some parts of the network are also limitedbecause some partners include in a project wholly new actors that IKEA had notconsidered. This happened, for instance, during the “printed veneer” project,whereby Akzo-Nobel involved a new equipment supplier, an Italian cylinderengraver, in IKEA’s network. In such cases, IKEA agrees to delegate to reliablepartners not only the performance of key tasks, but also the selection into IKEA’snetwork of other relevant actors: this clearly affects the structuring of the net-work 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 shapea 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 thatenables 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 notalways match each other. IKEA certainly seeks firms with which it can sharegoals at a general level: for instance, IKEA and Akzo-Nobel share the goal ofdeveloping coating technologies that improve furniture quality or costs. How-ever, goals can sometimes be matched only after intense negotiations and dis-cussions. Moreover, unexpected turns of events can change the initial goalcongruence: this happened for example when IKEA decided not to launch the “printed on” Lack tables, which had been the main development goal ofSwedwood and Akzo-Nobel. However, at a general level, Akzo-Nobel’s goal ofdeveloping new coating technologies could still be accomplished within its rela-tionship to IKEA. In fact, IKEA is a perfect test customer that can subsidize



Akzo’s own developments: how many furniture retailers could afford to sacrificethousands of real products just to perform technical tests?

IKEA’s goals are more easily matched with those of the network in rou-tine interactions, such as order management routines, which are applied dailywithout being questioned. However, goal convergence becomes more problem-atic in situations requiring development efforts and sacrifices from IKEA’s coun-terparts: for instance, rising conflicts over technical developments for IKEA’sBilly bookshelf induced the coating supplier Becker-Acroma to withdraw froman unwanted joint venture with IKEA (see Appendix A). Perfect goal congru-ence in the network cannot be achieved, simply because the network is notcomposed of firms that IKEA can unilaterally control. Despite IKEA’s size andpower, these firms remain independent and hence are relatively free to set theirown goals. More precisely, the goals of IKEA and all firms in the network areinterdependent. This means that achieving one’s goal depends on the goals ofother firms. Håkansson and Ford stress that managing the network is impossibleor even undesirable.42 Instead, all firms need to manage in networks, whichrequires carefully taking into account the goals and resources of all the firmsinvolved. Therefore, to enable “managing in the network,” a key component ofa network strategy is assessing the matching between one’s own goals and thoseof the other firms involved.

The Dynamic Interactions of a Network Strategy

Next to evaluating and shaping the structure of the industrial network, itis necessary to understand what happens within this structure. Thus, the analyt-ical focus shifts to the interaction processes that unfold in the network and thatstand for its dynamics. These processes are particularly important becauseachieving strategic goals depends on how the interaction processes unfold, whichnow becomes more important than who performs them (i.e., the structural com-ponents). In strategic jargon, dynamic interactions are more concerned withimplementation than are structural components, but they are essential toachieving the specific efficiency and development goals associated with a givennetwork structure.

While forming the network structure can rely on simply assessing thematching between internal and external resources, implementing a networkstrategy with specific goals requires combining dynamically and more profoundlythe resources in the network. These concrete resource combinations are createdthrough continuous interaction processes. For instance, through the order man-agement routines, IKEA achieves efficiency when all firms involved combinedaily such resources as 12,000 products, countless components, transport means,and their specific competences. Resource combinations are particularly dynamicwhen the goal is to develop new technologies: during the “printed veneer” pro-ject, coating-related resources were combined and recombined in complex ways,as witnessed by the many tests required to find out how new resources couldwork together.



Inter-organizational routines and projects drive a network strategy becausethey are the concrete mechanisms through which relationships are practicallymobilized and they provide the context in which dynamic resource combina-tions occur. However, these two types of interaction processes differ in theirfunctions and applicability to efficiency goals, for which routines are more appro-priate, as opposed to development goals, for which projects are a better mecha-nism.

Dynamic Interactions 1: Interacting via Inter-Organizational Routines

Compared to development projects, inter-organizational routines aremore structured and rigid mechanisms because they clearly specify in advance thenature and timing of the activities performed by IKEA and its partners. For thesereasons, inter-organizational routines are adequate for daily interactions thataim to maintain efficiency. Examples of key efficiency goals for IKEA includereduced lead-times and inventory costs as well as more precise deliveries. Toachieve such goals, order management routines, for instance, clearly specify thetime frames, each actor’s roles, the interaction patterns, and the resourcesinvolved.

The continuous performance of inter-organizational routines also bringsabout adaptations in the firms’ activities that can be better linked for mutualefficiency gains, as stressed by Håkansson and Snehota.43 Moreover, according toZollo, Reuer, and Singh, inter-organizational routines improve the performanceof cooperative agreements.44 Still, compared to the open-ended nature andample delegation of joint projects, IKEA strives to control more unilaterally theperformance of critical routines such as order management, for the sake of cer-tainty and homogeneity across its network. Therefore, these inter-organizationalroutines 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, andare therefore employed for interaction processes that aim to foster product andtechnology development.45 For instance, the “printed veneer” project had ratherbroad goals (i.e., tackling the issue of costly veneers), timeframes, roles, andinteraction patterns, and many unplanned resources could be involved as theproject unfolded. In comparison to routines, projects entail less control and moreuncertainty for IKEA. For instance, the “printed veneer” project was highlyuncertain because it was rather open-ended and technically challenging and hadan open timeframe. Moreover, IKEA was not able to control the selection of newparticipants in this project, as the technical leadership was assigned to Akzo.

Nevertheless, no matter how strongly IKEA can control a project, inter-organizational projects are important drivers of development for at least threereasons: they organize IKEA’s interactions with its network by allocating tasksand responsibilities, they focus attention and efforts on a specific developmentproblem, and they select and mobilize key actors who can contribute relevantknowledge to tackle the problem at hand. Similar advantages of projects as tools



to foster innovation have been recognized (although in an intra-organizationalcontext) by Bowen, Clark, Holloway, and Wheelwright.46 Moreover, as stressedby Brady and Davies,47 projects facilitate “explorative” learning, that is, theyenable learning by exploring new areas of knowledge. Such exploration is par-ticularly important when IKEA does not know what technology needs to bedeveloped or who can develop it, as in the case of “printed veneer.” Then, theflexibility of projects as coordination mechanisms comes in handy, together withIKEA’s delegation to selected actors, who are in turn free to engage their owntrusted partners. Furthermore, if the new actors remain permanently engaged,projects assume an even broader renewal potential by affecting the structure andcomposition of IKEA’s network.

Granted the importance of the external network for IKEA, a natural ques-tion is: What does IKEA need to have itself in order to interact successfully withthis network? IKEA’s competence, organizational structure, and culture are importantprerequisites for IKEA to be able to initiate and take part in the interactionprocesses that drive its network strategy. Those internal features strongly affecthow the focal firm interacts with specific partners, via inter-organizational rou-tines and joint projects,48 and are fundamental for any strategic center to suc-cessfully interact with a web of partners.49 Firstly, IKEA possesses strongcompetences in all areas necessary to interact with and support suppliers: IKEAof Sweden employs about 100 experts in important areas for its product andtechnology development, such as wooden materials, coatings, and surface treat-ments; while to support interactions with manufacturers and carriers concerningdistribution routines, IKEA of Sweden employs about 100 logistics and ordermanagement experts. These technical and logistic competences are also repro-duced on a local level within IKEA’s Trading Offices.

Secondly, IKEA has created both central and local organizational interfacesto interact with suppliers. The central business unit, IKEA of Sweden, bothnegotiates with major suppliers the worldwide prices and discusses long-terminvestments and capacity plans. IKEA of Sweden also intervenes during supplierselections and extensive development projects. Alongside these central inter-organizational interfaces, IKEA also interacts locally with single suppliersthrough its 40 Trading Offices. These business units are in charge, first and fore-most, of daily issues such as ordering and logistic coordination; but they alsointervene in yearly capacity planning and supplier selection meetings, wherethey represent their own local suppliers against other Trading Offices. Thirdly,IKEA’s culture may be inspired by cost-containment, but there is a stronglyrooted understanding that this goal requires mutual long-term commitments withkey external partners. These values inspire IKEA’s criteria for supplier selection(which stress a supplier’s readiness to make long-term investments for and withIKEA) and management systems (which reward Trading Offices based on thevolumes that they, together with their local suppliers, supply to IKEA worldwide).Such mechanisms strengthen the specific Trading Office-supplier relationshipand facilitate the emergence of mutual trust. Thus, IKEA can accept the neces-



sity of being dependent on certain highly trusted partners without the need tounilaterally control the relationship.

However, despite IKEA’s positive experiences, its competences and organi-zational structure often need to change to fit specific partners and interactionprocesses that unfold in the network. Changes in IKEA’s competences may benecessary to exploit new opportunities and are often driven by joint projects: forinstance, IKEA learned new coating technologies and “VMI” by interacting withits suppliers and logistic partners. A specific joint project may even lead IKEA tocreate special business units (see the joint venture GIAB presented in AppendixA). On a broader scale, IKEA currently deals with three major organizationalchanges stimulated by developments in the network: opening new TradingOffices in countries offering interesting sourcing opportunities; increasing theresponsibilities and competences of Trading Offices in Asia, as a reaction to theincreased volumes and know-how of Asian suppliers; and creating direct inter-faces between retail stores and selected suppliers, in order to improve productdevelopment and deliveries during sales promotions.

Conclusions and Implications from the IKEA Case

Industrial networks and business relationships play key roles for the strat-egy of IKEA and of most firms. Therefore, firms need a “network strategy,” thatis, they need to consider and use the external network in order to accomplishtheir own goals. The framework presented in this article helps to systematize thefactors that can improve interaction with the external network. The structuralcomponents and the dynamic interactions were elaborated on the basis of IKEA’snetwork strategy; the key implications of IKEA’s experiences of network interac-tions can be summarized as follows:

▪ In order for a firm to implement a network strategy and achieve its owngoals, the focal firm’s resources must be combined with those of externalactors. This combination is made concrete through two types of interac-tion processes: inter-organizational routines—well suited for achievingefficiency goals—and joint projects, aptly addressing development goals.

▪ These interaction processes are facilitated if the goals and resources of thevarious parties match each other. However, perfect goal congruence neverexists in a network of independent firms. Therefore, evaluating the goaland resource matching with specific counterparts can help in two majorways: in choosing from the beginning partners with more attunedresources and goals (e.g., those willing to make long-term commitments);and in supporting the negotiations necessary to increase the goal congru-ence with specific partners.

▪ Even if a firm can improve its “network matching” by forming the struc-ture of the network (e.g., by selecting suppliers), no absolute control canbe established over the network. Some counterparts may not accept therelationship content expected by the focal firm, or may even choose tointerrupt their interactions and leave the network. Therefore, business



relationships with key partners need to be carefully handled to establishstrong outposts in the network. In fact, the limits to controlling the net-work structure and processes suggest the need to delegate responsibilitiesto trusted and competent external actors who are capable, in turn, ofengaging other relevant actors. This is what IKEA does, and it is especiallyimportant in technical development projects that involve very widespreadcompetences.

▪ The external network cannot compensate for the gaping weaknesses ofunprepared firms: the strategic center cannot be “hollow.” In fact, form-ing the network by attracting counterparts and continuously interactingwith them requires that a firm is capable and prepared to “meet the net-work” in three main ways: by possessing extensive and specialized compe-tences (see IKEA’s considerable logistic and technical competences); bycreating appropriate inter-organizational interfaces (see IKEA’s many special-ized purchasing units and its product developers, who travel 200 days ayear to meet suppliers); and by promoting a network-oriented culture thatfavors a long-term approach and the creation of mutual trust instead ofthe abuse of power over partners.

▪ While being prepared to “meet the network” also means being flexibleenough to change internal competences and inter-organizational inter-faces to better interact with a changing network, a network-oriented cul-ture is instead more of a stable pillar. In fact, IKEA’s entire networkstrategy strongly relies on mutual trust and commitment to selected part-ners as substitutes for absolute control: its is only if you trust a committedpartner that you can accept dependence on it and delegate essential taskssuch as technical developments. However, trust and commitment do notappear overnight in a business relationship, but require a long-termapproach. This is why IKEA takes so seriously its supplier developmentprograms to improve selected partners’ IT and logistics capabilities (seeFigure 2) or its joint technical development projects. These initiatives taketime, but they eventually pay off when these suppliers can be moretrusted for IKEA’s daily replenishments or for technology issues.

APPENDIX AFrom Saving the “Billy” Bookshelf to Conflicts with Suppliers

The “Billy” bookshelf is one of IKEA’s best sellers; it sells over one millionunits 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-friendlycompany. However, this situation came under threat in 1991, when traces offormaldehyde were found in Billy packs in Germany. In IKEA’s largest market,the media started accusing IKEA of “poisoning” customers: a commercial cata-strophe and a complete loss of reputation were near, and IKEA had to solve theproblem rapidly in order to restore its goodwill. IKEA of Sweden began a close



scrutiny of the whole production process at Sydpoolen, a group of Swedish sup-pliers that manufactured Billy. The coating process soon became the key suspect,since hazardous chemicals were used in this production phase, and there was arisk that these chemicals might remain on the surface of the furniture.

In a state of emergency, IKEA of Sweden mobilized Becker-Acroma, oneof its major coating suppliers. In consultation with Sydpoolen’s technicians, theywere able to relate the formaldehyde emissions to two causes: the use in Syd-poolen’s factories of acid-curing coatings known to emit this gas; and too short atime between curing and packaging, which did not allow all the formaldehyde todissipate before packing took place. Therefore, the first action taken was pro-hibiting the use of acid-curing coatings at all IKEA’s suppliers. Meanwhile,Becker and other coating producers started supplying an older technologyknown as water-borne lacquers. IKEA, however, saw this solution as only tem-porary, and started searching for a robust solution. Becker suggested the use ofUV-curing, in which ultra-violet lamps are used to attach coatings to wood pan-els.

After a few months, IKEA also founded GIAB, a new firm with the spe-cific purpose of making this UV technology more economically viable and intro-ducing it to all IKEA’s suppliers. At the same time, IKEA made it clear to all thefirms involved with this technology that they were required to take shares inGIAB, which was then transformed into a joint venture. Among these actorswere Becker-Acroma, Sydpoolen, and the sub-supplier of wood panels. GIABwas equipped with a UV-based coating line that functioned as a full-scale testingfacility around which the above actors gathered to perform tests. The goal was tofind out how coatings needed to be modified and which parameters should bereproduced in the coating lines of other IKEA suppliers. These resources areshown in Figure A1.

IKEA’s goal with the joint venture GIAB was then extended to making it apowerhouse for the development of furniture coating technologies in general.However, even though cooperation was achieved on the factory floor, the pro-ject was ridden with conflicts. Becker-Acroma felt that the coating suppliers hadbeen used by IKEA as scapegoats for the Billy “scandal,” and that they were nowobliged to do most of the work of finding a solution. Becker-Acroma also sawthe establishment of GIAB as an accusation that these suppliers were incapableof developing new solutions by themselves. Despite these hard feelings, manytests were performed, which eventually led to finding the right combination ofUV coatings, particle boards, and coating lines. During the 1990s, this technol-ogy was rolled out across all IKEA suppliers, which today use UV curing almostexclusively. However, the forced cooperation eventually became unacceptable,especially to larger actors such as Becker-Acroma, who more or less stoppedcontributing to GIAB’s development. Therefore, at the end of the 1990s, despiteIKEA’s plans to turn GIAB into a consultant with its own customers outside theIKEA universe, GIAB lost momentum, firstly as it had fulfilled its task, and sec-ondly as it had never been accepted by coating producers.



Notes

1. Next to 70 in-depth interviews, the empirical material was collected via a dozen visits andsite 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 clipsand other publications related to IKEA.

2. For examples in the software sector, see B. Iyer, C.H. Lee, and N. Venkatraman, “Managingin a ‘Small World Ecosystem’: Lessons from the Software Sector,” California ManagementReview, 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 Inno-vation: Networks of Learning in Biotechnology,” Administrative Science Quarterly, 41/1 (March1996): 116-145; W.W. Powell, “Learning From Collaboration: Knowledge and Networks inthe 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 IndustrialDistrict: The Role of Interfirm Relations in the Growth and Transformation of Ceramic TileProduction in Italy,” Research Policy, 14/6 (December 1985): 329-343; M. Russo, “Comple-mentary Innovations and Generative Relationships: An Ethnographic Study,” Economics ofInnovation & 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 andCompetition in Interfirm Networks: The Paradox of Embeddedness,” Administrative ScienceQuarterly, 42/1 (March 1997): 35-67.

5. See M.H. Lazerson and G. Lorenzoni, “The Firms that Feed Industrial Districts: A Return tothe 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 MultidimensionalView,” Strategic Management Journal, 23/9 (September 2002): 795-806; E.H.M. Moors and P.J.



FIGURE A1. The Network of Resources Combined during Billy’s Coating Project

Test Facility

GIAB

IKEA

Sydpoolen

Coating Lines

Coatings

Becker-Acroma

Wood Panel

Wood Panel Producer

“Billy”

Vergragt, “Technology Choices for Sustainable Industrial Production: Transitions in MetalMaking,” International Journal of Innovation Management, 6/3 (September 2002): 277-299.

7. T. Wedin, Networks and Demand: The Use of Electricity in an Industrial Process (Uppsala, Sweden:Department of Business Studies, Uppsala University, 2001).

8. 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 (December1995): 347-365.

9. The list is presented on page 148 of G. Lorenzoni and C. Baden-Fuller, “Creating a StrategicCenter to Manage a Web of Partners,” California Management Review, 37/3 (Spring 1995):146-163.

10. See W.W. Powell, “Hybrid Organizational Arrangements: New Form or Transitional Develop-ment?” California Management Review, 30/1 (Fall 1987): 67-87.

11. 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 theFordistic model emerged in the early 1900s from a highly networked structure in the carindustry. See pages 365-368 in R.N. Langlois and P.L. Robertson, “Explaining Vertical Inte-gration: Lessons from the American Automobile Industry,” Journal of Economic History, 49/2(June 1989): 361-75.

12. These types of adversarial postures were for instance typical in the car industry until theearly 1980s, as discussed by L.-E. Gadde and H. Håkansson, Professional Purchasing (London:Routledge, 1993).

13. Dominating paradigms within strategic management have been for decades the microeco-nomics-inspired Industrial Organization [see J.S. Bain, “The Theory of Monopolistic Compe-tition 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 Industriesand Competitors (New York, NY: Free Press, 1980)] and Transaction Cost Economics [see O.E.Williamson, “The Economics of Organizing: The Transaction Costs Approach,” AmericanJournal of Sociology, 87/3 (November 1981): 548-577; “Strategizing, Economizing, and Eco-nomic Organization,” Strategic Management Journal, 12/8 (Winter 1991): 75-94]. WhereasIndustrial Organization could not consider networks because it used as key unit of analysiswhole sectors, Transaction Costs Economics views inter-firm relationships and networks asexceptions in economic organizing.

14. G.B. Richardson, “The Organization of Industry,” Economic Journal, 82/327 (September1972): 883-896.

15. M.S. Granovetter, “Economic Action and Social Structure: The Problem of Embeddedness,”American Journal of Sociology, 91/3 (November 1985): 481-510.

16. 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.

17. R. Swedberg, “Markets as Social Structures,” in N.J. Smelser and R. Swedberg, eds., TheHandbook of Economic Sociology (Princeton, NJ: Princeton University Press, 1994), pp. 255-282.

18. Even scholars studying strategic alliances are dissatisfied with the research bias towardsapplying atomistic views to interactive phenomena such as alliances. For instance, RanjayGulati called for a paradigm shift, from an atomistic to a social network-embedded view ofalliances. See page 295 in R. Gulati, “Alliances and Networks,” Strategic Management Journal,19/4 (April 1998): 293-317.

19. This view is presented for instance by H. Håkansson, “Organization Networks,” in A. Sorgeand M. Warner, eds., The IEBM Handbook of Organizational Behaviour (London: ThomsonBusiness Press, 1997), pp. 232-239; M.J. Piore, “Fragments of a Cognitive Theory of Techno-logical Change and Organizational Structure,” in N. Nohria and R.G. Eccles, eds., Networksand Organizations: Structure, Form and Action (Boston, MA: Harvard Business School Press,1992), pp. 430-444.

20. 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 con-strained either inside a strict hierarchy (based on command and bureaucratization) or by aperfect market (composed of autonomous firms that coordinate solely via price signals).

21. See H. Håkansson, “Product Development in Networks,” in H. Håkansson, ed., IndustrialTechnological Development—A Network Approach (London: Croom Helm, 1987), pp. 84-115.



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, UppsalaUniversity, 2003).

24. A comprehensive literature overview over the IMP tradition is presented on pages 25-30 ofH. Håkansson and A. Waluzsewski, Managing Technological Development: IKEA, the Environmentand Technology (London: Routledge, 2002).

25. These studies included both a large-scale questionnaire studying hundreds of business rela-tionships in Europe in the late 1970s and numerous extensive, deep case studies. The latterresearch approach still characterizes current IMP research: since the 1970s, hundreds of casestudies have traced the industrial network around a focal firm or a technical solution bymeans 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 ofAuthority 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 andPurchasing of Industrial Goods: An Interactive Approach (Chichester: Wiley, 1982).

28. An example of these network-level models is the so-called “ARA” model, which decomposesthe substance of industrial networks into three layers: inter-firm Activity patterns, Resourceconstellations, and webs of Actors. This Activity-Resource-Actor model is elaborated inH. 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 Multina-tional Corporation,” International Business Review, 5/5 (October 1996): 487-508; and U.Andersson, M. Forsgren and U. Holm, “The Strategic Impact of External Networks: Sub-sidiary 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, “NoBusiness is an Island: The Network Concept of Business Strategy,” Scandinavian Journal ofManagement, 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 Busi-ness Markets (London: Thompson Learning, 2002), pp. 88-94. Networks do not only producepositive effects for development and efficiency, but they are also ridden with conservativeforces that embed (in the sense of “constraining”) the firm. Too close business relationshipscan 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: networkscan 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 conti-nuity and can stretch over decades (as is the case for several of IKEA’s relationships). Thiscontinuity includes institutionalization and dependencies that make relationships hard tochange or terminate. These combined effects create stability in a network, where theinvolved actors, their goals, and their resources change only slowly. For the structural char-acteristics 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: TheFace 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: “Wal-Mart has the power to squeeze profit-killing concessions from suppliers...Wal-Mart pricepressure can leave so little profit that there is little left for innovation.” The quote is found inC. 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 HiddenJapanese Conglomerates (New York, NY: McGraw-Hill, 1994); H. Kim, R.E. Hoskisson and



W.P. Wan, “Power Dependence, Diversification Strategy, and Performance in Keiretsu Mem-ber Firms,” Strategic Management Journal, 25/7 (July 2004): 613-636.

38. See Fishman, op. cit.; C. Fishman, “The Wal-Mart Effect and a Decent Society: Who KnewShopping Was So Important?” Academy of Management Perspectives, 20/3 (August 2006): 6-25,at pp. 13-14.

39. 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,” inE. 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.

40. Development projects, regarded here as dynamic interactions, entail also strong structuringelements: 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 itspartners, as it partly did in the Billy episode among IKEA and Becker-Acroma (see AppendixA). Moreover, IKEA’s technical projects are often an occasion for new actors to be includedin the structure of IKEA’s network: the “printed veneer” project, for instance, involved forthe first time an Italian firm that is now a stable supplier of engraved cylinders.

41. Meeting with Anders Brorström, Deputy Procurement Director, IKEA Russia, Moscow,February 28, 2007.

42. 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 stressthat achieving total control over a network, however unlikely it is, is undesirable because itwould transfer all burden and source of innovation and wisdom to the controlling companyitself.

43. On the concept of “activity links,” one of the key dimensions in the IMP network “ARA-model,” see H. Håkansson and I. Snehota, eds., Developing Relationships in Business Networks(London: Routledge, 1995), pp. 28-30.

44. See M. Zollo, J.J. Reuer, and H. Singh, “Interorganizational Routines and Performance inStrategic Alliances,” Organization Science, 13/6 (November/December 2002): 701-713.

45. These advantages of both intra- and inter-organizational projects as coordination mecha-nisms are recognized, for instance, by J. Sydow, L. Lindkvist, and R. DeFilippi, “Project-Based Organizations, Embeddedness, and Repositories of Knowledge: Editorial,” OrganizationStudies, 25/9 (November 2004): 1475-1489.

46. H.K. Bowen, K.B. Clark, C.A. Holloway, and S.C. Wheelwright, “Development Projects: TheEngine of Renewal,” Harvard Business Review, 72/5 (September/October 1994): 110-120.

47. See T. Brady and A. Davies, “Building Project Capabilities: From Explorative to ExploitativeLearning,” Organization Studies, 25/9 (November 2004): 1601-1621, at 1605-1607.

48. A routine such as OPDC strongly depends on IKEA’s competences and organization and isimpossible to perform without adequate external interfaces. The effect of IKEA’s internalstructure is present, although less evident, on development projects, whose success certainlydepends on IKEA’s competence and organizational structure, but where much responsibilityand control is typically delegated to actors farther away from IKEA.

49. See Lorenzoni and Baden-Fuller, op. cit.



MANAGEMENT SCIENCEVol. 56, No. 3, March 2010, pp. 468–484issn 0025-1909 �eissn 1526-5501 �10 �5603 �0468

informs ®

doi 10.1287/mnsc.1090.1117©2010 INFORMS

The Impact of Misalignment of OrganizationalStructure and Product Architecture on Quality in

Complex Product Development

Bilal GokpinarDepartment of Management Science and Innovation, University College London,

London WC1E 6BT, United Kingdom, [email protected]

Wallace J. HoppStephen M. Ross School of Business, University of Michigan, Ann Arbor, Michigan 48109, [email protected]

Seyed M. R. IravaniDepartment of Industrial Engineering and Management Sciences, Northwestern University,

Evanston, Illinois 60208, [email protected]

Product architecture and organizational communication play significant roles in complex product develop-ment 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. Specifi-cally, we study the vehicle development process of a major auto company and use vehicle quality (warrantyrepairs) as our NPD performance metric. Our empirical results indicate that centrality in a product architecturenetwork is related to quality according to an inverted-U relationship, which suggests that vehicle subsystemsof intermediate complexity exhibit abnormally high levels of quality problems. To identify specific subsystemsin danger of excessive quality problems, we characterize mismatches between product architecture and organi-zational structure by defining a new metric, called coordination deficit, and show that it is positively associatedwith quality problems. These results deepen our understanding of the impact of organizational structure andproduct architecture on the NPD process and provide tools with which managers can diagnose and improvetheir NPD systems.

Key words : new product development; product architecture; organizational structure; complex networksHistory : Received June 23, 2008; accepted August 30, 2009, by Christoph Loch, R&D and productdevelopment. Published online in Articles in Advance January 12, 2010.

1. Introduction and Literature ReviewProduct innovation is central to business creation andgrowth. Firms that are able to bring a steady streamof timely and well-executed products to market arelikely to enjoy long-term financial success. Designingproducts and bringing them to market, however, isnot a straightforward process. Product developmentefforts usually involve many design iterations. Theability of the organization to manage these inevitabledesign changes has a major impact on product qualityand firm competitiveness.The development process is particularly challeng-

ing for complex products such as automobiles orairplanes, which involve thousands of engineersspending years designing, testing, and integratinghundreds of thousands of parts. Consequently, a keychallenge in these product development processesis matching the organization to the product beingdeveloped. This involves two fundamental problems:(1) how to assign people to the parts and subsystems

that make up the product, and (2) how to ensure thatpeople communicate/collaborate effectively in theperformance of design tasks. As evidence that theseproblems are universal and difficult, a recent jointstudy by BusinessWeek and the Boston ConsultingGroup, reported that 1,000 senior managers aroundthe globe cited a lack of coordination as the second-biggest barrier to innovation (McGregor 2006).From an operations management standpoint, we

can view the new product development (NPD) pro-cess as a network of engineers designing a networkof parts. Consequently, in this paper, we study theproblem of coordinating parts and people in an NPDprocess by means of network analysis. As such, thispaper is part of a growing literature that makes useof networks to represent both product architecture(Krishnan and Ulrich 2001, Henderson and Clark1990, Ulrich 1995) and organizational structure (Clarkand Fujimoto 1991, Brown and Eisenhardt 1995). Inthe work closest to our own, Sosa et al. (2004) adopted

468

Gokpinar, Hopp, and Iravani: Impact of Misalignment of Organizational Structure and Product ArchitectureManagement Science 56(3), pp. 468–484, © 2010 INFORMS 469

a combined perspective in a study of the alignmentof design interfaces and communication patterns. Ina subsequent paper, they identified factors that makesome teams better than others at aligning their cross-team interactions with design interfaces (Sosa et al.2007). Although the insights from these studies areinteresting and potentially useful in improving NPDprocesses, they are premised on a basic assumption,namely that misalignment of the design organizationand the product architecture is detrimental to perfor-mance. But, because none of these studies actuallymeasured performance, they could not corroboratethis 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 archi-tecture and organizational structure, and (b) empiri-cally investigating the effect of coordination deficit onproduct quality.In a broader sense, our work builds upon and

integrates two streams of research: (i) operations ofcomplex product development and (ii) social networkanalysis of organizational performance. Althoughthese areas are very broad and have been studiedfrom a variety of perspectives, they have been studiedtogether under two major research headings: knowl-edge networks (see Nonaka and Takeuchi 1995, Con-tractor and Monge 2002) and modularity (see Ulrich1995, Baldwin and Clark 2000).Researchers have investigated various aspects of

knowledge networks in the context of product devel-opment and have provided critical insights into whysome business units are able to make effective use ofknowledge from other parts of the company, whileother units find knowledge to be a barrier to inno-vation (Hansen 2002, Carlile 2002). In this paper,we construct a very specific knowledge networkthat characterizes collaboration and communicationbetween design engineers and identify structural fea-tures of this network that are correlated with qualityproblems in the final product.Modularity refers to methods for reducing the num-

ber of interactions and interfaces among parts andcomponents in product design (Ulrich 1995, Baldwinand Clark 2000). Organizational implications of mod-ularity, as well as the organizational factors that sup-port the use of modularity, have been studied byseveral researchers (see Sanchez and Mahoney 1996,Schilling 2002, Ethiraj and Levinthal 2004, Flemingand Sorenson 2004). Unlike other network studiesof modularity, which represent interfaces as eitherpresent or not present, we make use of engineeringdata to characterize the strength of interfaces betweencomponents. This gives us a more detailed represen-tation of the product architecture, which we compare

to the organizational structure to quantitatively mea-sure the degree of misalignment.This paper also contributes to the literature on the

use of social network tools in empirical studies oforganizations (see, e.g., Wasserman and Faust 1994).A distinctive feature of our study is that we make useof archival data, rather than surveys, to construct asocial network. Because such data is readily availablein NPD environments, this approach may ultimatelymake network analysis more practical as a manage-ment tool.Finally, from a practical perspective, our work can

help managers to systematically identify and quan-tify potential problem areas that can be addressed toimprove the quality of the resulting products. Ourmetric of organizational misalignment (i.e., coordina-tion deficit) can be computed using standard datafrom an engineering change order system. As such, itprovides a way to highlight opportunities for improv-ing coordination among design engineers without col-lecting additional data. This should be particularlyvaluable in environments where evolution of productarchitectures changes coordination needs over timeand makes static organizational policies ineffective.The remainder of this paper is organized as fol-

lows. In §2, we provide the theoretical backgroundand frame our hypotheses. In §3, we present a detaileddescription of the data and the system in which we testthe hypotheses. In §4, we describe the model develop-ment, and in §5, we present the analysis and results.We discuss our results in §6 and conclude in §7.

2. Theory and HypothesesUlrich (1995, p. 420) defined product architectureas “(i) the arrangement of functional elements, (ii) themapping from functional elements to physical com-ponents, and (iii) the specification of the interfacesamong interacting physical components.” All three ofthese dimensions may influence ultimate product per-formance at either the local (component) level (Ulrich1995, Baldwin and Clark 2000, Mihm et al. 2003) orthe global (product) level (Clark and Fujimoto 1990).An intermediate level between the component andproduct levels is the subsystem level, which is widelyused by firms to describe product architectures formanagement purposes. From a network perspective,product architecture can be characterized by repre-senting subsystems as nodes and interfaces betweensubsystems as links. Network metrics can then beused to describe the nature and position of prod-uct subsystems. For example, a subsystem with many(physical and functional) interfaces will have highcentrality in the product architecture network.Given this interpretation, the centrality of a sub-

system can serve as a proxy for complexity, because

Gokpinar, Hopp, and Iravani: Impact of Misalignment of Organizational Structure and Product Architecture470 Management Science 56(3), pp. 468–484, © 2010 INFORMS

more interfaces imply more design issues. So, ifmanagement were to devote equal resources andattention to all subsystems, we would expect highlycentral subsystems to exhibit worse performance(e.g., more quality problems). But management doesnot usually do this. Previous research on productdevelopment has shown that architectural/technicalinterdependence is associated with organizationalcommunication (Henderson and Clark 1990, Brownand Eisenhardt 1995, Adler et al. 1995). Highly centralsubsystems, which are heavily connected to other sub-systems, logically receive intense attention from theorganization, which offsets the potential quality prob-lems resulting from their complexity.If management were to perfectly correlate resources

and attention to the complexity of each subsystem, wewould not expect to see any correlation at all betweencentrality and performance (e.g., quality). But we donot think this is realistic either. Complexity is notdirectly observable. Therefore, mismatches betweenorganizational attention and subsystem interconnec-tivity, of the type observed by Sosa et al. (2004), arelikely to occur. We conjecture that they are most likelyto occur for subsystems of intermediate centrality. Thereason is that highly central subsystems are obviouslyhighly complex, and hence receive substantial orga-nizational attention. Indeed, they may receive evenmore attention than they need because they presentsuch manifest design challenges. At the other endof the scale, low centrality subsystems, which havefew interfaces, require relatively little coordinationeffort and so are unlikely to be underattended. Eventhe minimal amount of organizational coordinationbuilt into standard design practices is likely to beenough for these subsystems. But intermediate cen-trality subsystems are neither complex enough to beobvious nor simple enough to be easy. These are thesubsystems where a delicate matching of resourcesand attention to the design complexity is most dif-ficult. Therefore, this is where we expect to find themost mismatches, as we conjecture in the followinghypothesis:

Hypothesis 1. The centrality of a product subsystemin the product architecture has an inverted-U associationwith the quality problems observed in that subsystem.

The earlier discussion provides a very generalsense of where a lack of organizational coordinationmight lead to excessive quality problems. But becausecentrality only characterizes subsystems in a coarsemanner, it would not provide any explicit manage-rial guidance on which subsystems are most proneto quality problems. Henderson and Clark (1990)established a relationship between product architec-ture and design organization, and pointed out theimportance of matching team interfaces to technical

interfaces. Sosa et al. (2004) observed that productdevelopment teams tend to ignore certain types oftechnical interfaces. In a more recent study, Sosa et al.(2007) presented anecdotal evidence from industrywhere the shortage of organizational attention to tech-nical interfaces resulted in poor performance. Basedon these findings and the intuition of NPD man-agers in our client firm, we conjecture the followinghypothesis:

Hypothesis 2. Mismatches between product architec-ture and organizational coordination in subsystems arepositively associated with the quality problems observed inthese subsystems.

We note, however, that neither Henderson andClark (1990) nor Sosa et al. (2004) actually measuredNPD process performance to test the above hypoth-esis. To do this, we will first establish a quantita-tive measure of the degree of mismatch between theorganization and the product and then correlate thismetric with empirically observed performance (i.e.,warranty claims).

3. Overview of the VehicleDevelopment Process

Our empirical analyses are based on a detailed studyof the new vehicle development process of a largeU.S. auto manufacturer. Because it involves manyinterdependent tasks over an extended period, auto-motive design is a prototypical example of complexproduct development. To create a useful model, oneof the authors spent two summers (about six months)on site for data collection and analysis. This allowedus to gain a good understanding of the product devel-opment process through observation of common prac-tices and obstacles. We also collected an extensivedata set from the engineering change order (ECO) sys-tem, which is used by the firm to manage and docu-ment the design process.

3.1. Vehicle Development ProcessOur main unit of reference regarding the vehicledevelopment process is a vehicle program. For a largecompany, such as the one we studied, there are typ-ically multiple models with different brand nameswithin the same vehicle program. A model refersto the end product that is sold to the customers indealerships (e.g., Chevrolet Malibu, Toyota Camry,etc.). Although models under the same vehicle pro-gram may be sold under different brand names, theirunderlying architectural structure and product devel-opment effort is similar. Because vehicles are built offof platforms, there is a good deal of component com-monality across models. Some of these componentsare entirely new to the program, whereas some arecarried over from previous programs.


It takes two to three years to complete the entireproduct development process for a program. Aprogram provides a platform for several vehicles,which are typically launched in a staggered fashionto smooth demands on engineering and marketingresources. Once in the market, vehicles are usuallygiven major refreshes (redesigns) every five to sixyears, with a minor update at about the midpoint ofa model life cycle. Most models are eventually retiredafter being in the market for several design cycles.For practical reasons of data availability, we fol-

lowed our client in dividing an automobile into 243architectural subsystems, which contain roughly 150,000parts that interact with each other. Consequently, wewill construct the product architecture network bydescribing interfaces between subsystems.

3.2. OrganizationThe primary actors in a vehicle development orga-nization are design engineers. Although other typesof engineers (e.g., materials engineers, quality-controlengineers, and testing engineers) are employed by theorganization, their direct involvement in the designprocess is limited. Therefore, we focus our attentionexclusively on the design engineers. In the systemwe studied, there were about 10,000 engineers whoparticipated in the engineering design work. Theseengineers are responsible for creating the parts, mak-ing sure that they meet design specifications andcoordinating interfaces with other parts. Design engi-neers typically work in groups of 5–10 people, ledby a manager who is responsible for supervising thedesign of parts and components, as well as coordi-nating efforts within and beyond the group. Withinthe product development system, design engineerscoordinate with each other through distribution lists.Whenever there is activity related to a part, desig-nated engineers who are directly involved (e.g., anengineer whose parts share a direct physical inter-face with a modified part) or indirectly involved (e.g.,an engineer whose part shares an indirect functionalinterface) are notified via the distribution list. Indi-vidual engineers are placed on these distribution listsas a result of both management policy and requestsby engineers. As such, distribution lists capture bothformal connections inherent in the organization chartand informal connections that emerge from the expe-rience of engineers.

3.3. Information System andEngineering Change Orders

Because of significant advances in computer data stor-age and processing capacities, firms are able to accu-mulate and track large amounts of data about theproduct development process. In our study, we madeuse of an ECO system like that used in most prod-uct development processes as a tool to control anddocument the product development process. (For a

detailed overview of ECO systems, see Loch andTerwiesch 1999.) An ECO is filed by a design engi-neer every time a new part is released or an existingpart is changed in any way. Although the details varyfrom one company to another, the basic features ofmost ECO systems are similar.In our client’s vehicle design process, the ECO

database contained approximately 100,000 separateECOs for one model year. A typical ECO contains theidentity of the engineer who initiated it, a unique rea-son code that explains why the ECO was issued, theidentities of other engineers to be notified as part ofa distribution list about activity related to the ECO,part numbers associated with it, and the targeted andactual dates of completion. Figure 1 shows a simpli-fied version of our client’s ECO process. Note thatthere are several different situations for which ECOsare 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., becauseof a government regulation, styling change, or sup-plier 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 theirreason codes: (1) new release ECOs, which are filed forall parts of a new model (note that some of these partsare new, whereas others are existing parts from a pre-vious model that have been renumbered for the newmodel); (2) problematic ECOs whose reason codes wereidentified by several design engineers with whom weconsulted as indicating problems in the design pro-cess; and (3) other ECOs, which include all ECOs notcontained in the above categories (e.g., ECOs due toa cost reduction initiative or a change in governmentregulations). The role of each of these ECO types inthe design process are illustrated schematically in Fig-ure 2. We use this classification to create variables inthe empirical model in the next section.There have been several studies (see Clark and

Fujimoto 1991, Huang and Mak 1999, Terwiesch andLoch 1999, Loch and Terwiesch 1999) of ECOs inthe design process. These examined the broad signif-icance of ECO generation without specifically captur-ing product architecture information or organizationalstructure. Because ECOs are filed when an individ-ual part fails to meet specifications, two or more partshave interface problems, or product changes are madethat affect part designs, the ECO database containsa great deal of information. To our knowledge, thisstudy is the first attempt to use the ECO system tocapture product and organization interactions.Previous studies (e.g., Sosa et al. 2004, 2007) have

relied on surveys to construct networks for both prod-uct architectures and organizational structures. This is(a) time consuming, which may limit use in practice,


Figure 1 A Schematic of ECO Flows in the Vehicle Design Process

Discuss, draft and

• Identify engineers who need to be notified in the distribution list

• List related part numbers

• Specify dates of completion

Design engineer /group manager

Responsible engineer

Incorporate necessary changes toowned parts and sign off

Other engineers on the distribution list

Approve changes andsign off

Group manager

Perform necessary changesand modifications

Responsible engineer

Change approval: cross system,cross functional board

Resolved

Not resolved

Sample reasons for issuing an ECO• New release• Performance test failure• Cost reduction• Government regulation, etc.

pre-approve ECO

and (b) subject to people’s memories (e.g., a vehicleprogram lasts several years, so people must rememberwith whom they collaborated years ago to constructa relevant organizational structure network through asurvey). Because the ECO system contains informa-tion about both parts and the engineers who work onthem, we can use it instead of surveys to construct the

Figure 2 Modification of a Part Through the ECO System

Partrelease

Design LaunchNew release

ECO

OtherECO

ProblematicECO

Test

Design/Integrationproblems

Other problems

Pass

product architecture and organizational coordinationnetworks.

4. Model DevelopmentIn this section, we describe how we created the prod-uct architecture and organizational coordination networksfrom the ECO data described above. These networks


Figure 3 Product Architecture Network: A Network of Vehicle Subsystems

are the basis for the key independent variables in ourempirical study of vehicle quality. So, once we havedescribed the networks, we discuss the constructionof the dependent variable, independent variables, andcontrol variables in our regression model.

4.1. Creating the Product Architecture NetworkWe constructed the product architecture network bydefining vehicle subsystems as nodes. We definedlinks between these nodes by looking only at newrelease ECOs. Note that these new release ECOs arenot a result of a problem or later changes, but purelya result of initiating all parts of a new vehicle pro-gram. As such, they provide an unbiased summary ofthe linkages between parts. For example, when a partin the steering wheel subsystem is newly released, allparts related to it, which may be in the steering wheel,electrical traction, or other subsystems, will be auto-matically listed on the new release ECO for that part.The logic behind this construction is straightfor-

ward: When a part is initiated by issuing a newrelease ECO, all parts that share some sort of physicalor functional interface with that part are also listedin the ECO. Therefore, if we look at all such ECOsand count how many times two subsystems appearin the same ECO, we can get a proxy for the strengthof the architectural interaction (number of interfaces)between the two subsystems. Specifically, we use thenumber of new release ECOs that include parts bothfrom subsystems i and j as the weight for the linkbetween nodes i and j in the product architecturenetwork.This network reveals that the various subsystems

differ substantially in terms of their connections to

other subsystems. For example, in a car, the wiringharness subsystem has physical connections to almostevery other subsystem, while the air cleaner subsys-tem has only a limited number of physical connec-tions with the rest of the vehicle. Figure 3 depicts avisual representation of the product architecture net-work, which shows that the network is too large andcomplex to analyze visually. Clearly, we need quanti-tative metrics to characterize the product architecturein a useful manner.

4.2. Creating the OrganizationalCoordination Network

Many organizational studies (see Ibarra 1993, Krack-hardt and Hanson 1993, Burt 2004) have studied com-munication and advice networks of individuals byusing empirical data sets that are usually collectedthrough surveys or questionnaires. Our study differsfrom these by making use of formal institutional con-nections, rather than informal social ones. One benefitof this approach is that it permits organizational anal-ysis with data already being recorded, and so does notsubject the organization to the burden of a detailedsurvey. A second benefit is that it focuses on links overwhich management has a great deal of influence (i.e.,who is listed on which distribution list). Hence, anylevers indicated by this analysis can be translated intoconcrete management policies.To construct the organizational coordination net-

work, we again used vehicle subsystems as nodes andproceeded in two steps. In the first step, we used onlythe new release ECOs to determine which engineersare associated with which subsystems. We did thisbecause our client indicated that only key engineers


involved in the design of the parts (and hence subsys-tems) are listed in the new part release ECOs. (Notethat an engineer may be associated with more thanone subsystem, while a subsystem always involvesmore than one engineer.) In the second step, we usedall ECOs to characterize communication between sub-systems, to capture the full range of communicationover the duration of the project. Specifically, we usedthe number of distribution lists that include engineersfrom both subsystem i and subsystem j as the weightof the link between nodes i and j in the organizationalcoordination network. Note that each distribution listcorresponds to an issue in the product developmentwork, so we count the number of distribution listsrather than the number of people in establishing thelinks between subsystems.

4.3. Scope of the ModelWe tested our hypotheses by developing a regressionmodel. We examined 13 vehicle programs, with 243subsystems in each, giving us a total of n = 243 ×13= 3�159 observations in the model. Each of the 13programs corresponded to a 2005 model-year vehicledesigned in the United States and sold solely to U.S.customers. Note that these programs correspond toplatforms from which many models are introduced.For example, our client launched 32 distinct modelsin the 2005 model year.As the dependent variable in the model, we used

warranty claims data aggregated from roughly 17,000unique problem codes up to the subsystem level. Wefollowed our client in using IPTV (incidents per thou-sand vehicle) as a measure of quality. We used thenumber of warranty incidents (IPTV) reported duringthe first 12 months after the vehicle launch. Note thatwe observed warranty data during the first year of thevehicle use (i.e., in calendar years 2005 and 2006), butthe ECOs that describe the product and organizationnetworks for these programs were initiated duringcalendar years 2002–2005. Therefore, collecting designand quality data for one model year requires examin-ing over four years of data within the company.We conducted a similar study by examining the

vehicles that were launched in the 2006 model yearto check the robustness of the model. As before, wefocused on 13 vehicle programs, which correspondto 26 distinct models. Because the procedure and theresults are very similar to those for the 2005 modelyear, we present them in the online appendix (pro-vided in the e-companion).1

1 An electronic companion to this paper is available as part of the on-line version that can be found at http://mansci.journal.informs.org/.

Figure 4 Calculating Degree Centrality in the Product ArchitectureNetwork

Electricaltraction Steering

wheel

Doortrim Wiring

harness Battery

19

25434

139

Centrality score = 66

Centralityscore = 17

Centrality score = 44

Centralityscore = 9

Centrality score:13 + 43 + 25 + 9 = 90

4.4. Independent Variables

4.4.1. Centrality in the Product Architecture Net-work. After creating the product architecture networkas outlined in the previous section, we calculatedthe centrality scores of the nodes (subsystems) usingUCINET 6,2 Borgatti et al. (2002). We use degree cen-trality, which is computed as the sum of the weightsof the links emanating from a node to characterize thelevel of connectivity of a subsystem. Subsystems withhigher degree centrality have more interfaces and aretherefore, presumably, more complex. Figure 4 illus-trates this by showing partial centrality scores for aportion of the product architecture network.To look for the U-shaped relationship conjectured

in Hypothesis 1, we also included the square of thedegree centrality as an independent variable. A posi-tive coefficient for the linear variable and a negativecoefficient for the squared variable would suggest aninverted U-shaped relationship between degree cen-trality and warranty claims.

4.4.2. Coordination Deficit. Hypothesis 2 conjec-tures that misalignment between the product archi-tecture and organizational structure is associated withquality problems. The product architecture and orga-nizational coordination networks defined above pro-vide a means for quantifying misalignment. But thereare many ways to specify and measure mismatchesbetween the two networks. As long as (a) the met-ric is computed from the data contained in theproduct architecture and organizational coordinationnetworks, and (b) the metric monotonically increasesin the extent to which the two networks are mis-aligned, then it can be considered as a possible met-ric. Below, we discuss one such metric that we feelfits the NPD process, along with three other plausiblemetrics.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 graph-ically displays networks and computes most standard networkmetrics.


product architecture network imply a certain numberof design issues that must be resolved. This numbermay be uncertain, but we assume that it is pro-portional in expectation to the number of interfacesindicated by the product architecture network. Wefurther assume that each issue requires some num-ber of communications to resolve, which again maybe uncertain. The product of these two numbers isthe number of communications required to success-fully coordinate the two subsystems. If the num-ber of communications falls short of this limit, thenunresolved issues may result in design flaws thatlead to warranty claims. Because each unresolvedissue represents an additional flaw, we conjecturethat the expected number of warranty claims is lin-early related to the difference between the actualand required number of communications. However,if the number of actual communications exceedsthe required number, then no additional benefit isgained, because communications about the interfacesbetween subsystems i and j will not impact designissues involving interfaces between other pairs ofsubsystems.We quantify the above reasoning into a metric

that we call the coordination deficit metric. To do this,we let WA and WC represent the product architec-ture and coordination networks, respectively, whereWA = �WA

ij , and WAij represents the weight of the link

between nodes i and j in the product architecture net-work; and WC = �WC

ij , and WCij represents the weight

of the link between nodes i and j in the organiza-tional coordination network. Because these weightsmay have different magnitudes, we normalize themby dividing by the total weight of the links in eachnetwork. This yields Aij =WA

ij /�∑

i� j WAij /2� and Cij =

WCij /�

∑i� j W

Cij /2�, where Aij and Cij represent the pro-

portion of total links that are from subsystem i tosubsystem j in the product architecture and organiza-tional coordination networks, respectively. With these,

Figure 5 An Example of Calculating Coordination Deficit for a Subset of Nodes

Product architecture network

Organizational coordination network

0.169

0.2210.3800.036

0.115 0.079


wheel

Doortrim Wiring

harnessBattery

0.134

0.397

0.3250.024

0.084

0.036


wheel

Doortrim Wiring

harness Battery

3

11

27

2

19

4

9

2543

13

7

33

we define i as the coordination deficit for node (sub-system) i as

i =∑j

max�Aij −Cij�0�� (1)

Note that this metric includes only links whereAij −Cij is positive (i.e., the connection between nodesi and j is stronger in the product architecture net-work than in the organizational coordination net-work) to capture undercoverage of subsystem link-ages. Because problems from lack of coordinationcannot be reduced below zero, we would not expectexcess coverage along one link to offset inadequatecoverage along another. Hence, we omit links whereAij −Cij is negative.Figure 5 illustrates calculation of the coordination

deficit metric for a subset of the nodes in the vehi-cle development system. In this example, the wiringharness subsystem has four links in the product archi-tecture network to the door trim, electrical traction,steering wheel, and battery subsystems, with weightsof 13, 43, 25, and 9, respectively (see Figure 5).Because the total weight of all the links in the networkis 4+ 19+ 13+ 43+ 25+ 9= 113, these links representthe following fractions of the total: 0.115, 0.380, 0.221,0.079. In the organizational coordination network, thewiring harness subsystem has four links to the samenodes as in the product architecture network, withweights of 2, 27, 33, and 7. These represent the follow-ing fractions of the total: 0.024, 0.325, 0.397, and 0.084.For each link, we compute the difference between thefraction of weight in the product architecture networkand the fraction of weight in the organization network(inserting a zero if this difference is negative). Thisyields a coordination deficit for the wiring harnesssubsystem of �0�115−0�024�+ �0�380−0�325�+0+0=0�146. Once we have computed coordination deficit inthis manner for all subsystems (nodes), we can use itas an independent variable in our model.


Although our coordination deficit metric is reason-able, it is not the only way to measure mismatchesbetween the product architecture and organizationalcoordination networks. To see if another measuremight work better, we considered three alternativesthat also satisfy the two criteria we defined above fora metric to measure misalignment:1. The ratio metric is computed as the ratio of the

percentage of links in the two networks. That is, wefirst calculate the percentage of the entire networkflow at each link for both architectural and coordina-tion networks as we did for the coordination deficitmetric (i.e., calculating the Aij and Cij ). However,unlike the coordination deficit metric, which calcu-lates the difference between the flow at links in twonetworks, this metric calculates the ratio between theflow at links in two networks. After calculating theratios, it proceeds similar to the coordination deficitmetric, 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 degreeof mismatch between the product architecture andorganizational coordination networks, it implies thatreducing the number of mismatches will affect quality(warranty claims) in a nonlinear fashion.2. The node difference metric is computed as the dif-

ference between the centrality score of the subsystem(node) in the product architecture network and that inorganizational coordination network. That is, if we letAi =

∑j Aij be the centrality of node i in the product

architecture network, and Ci =∑

j Cij be the centralityof node i in the organizational coordination network,the node difference metric for node (subsystem) i isgiven by

Di =max�Ai −Ci�0�� (3)

As such, this metric considers node differencesbetween the two networks, rather than link differ-ences.3. The local deficit metric is obtained by calculat-

ing the percentage of flow along each link emanatingfrom a node. After calculating these flow percent-ages at each node for both networks, it proceeds in afashion similar to the coordination deficit metric andcalculates the aggregated deficit scores. Formally, Aij

and Cij are now calculated as Aij =WAij /�

∑j W

Aij � and

Cij =WCij /�

∑j W

Cij �. We then aggregate these for node

(subsystem) i as

Li =∑j

max�Aij −Cij�0�� (4)

Because it normalizes flows at each link by the totalflow from that node, rather than total network flow,the local deficit metric is not sensitive to the totalamount of coordination effort associated with a sub-system. For example, 1 unit of flow between nodes iand j out of a total flow of 10 units from node i isregarded as equivalent to 10 units of flow betweennodes i and j out of a total flow of 100 units fromnode i.We examined both the original coordination deficit

metric and these three alternate metrics in our regres-sion analysis, as we discuss in §5.

4.5. Control Variables

4.5.1. Previous Year’s Warranty Claims. Al-though we control for all relevant factors for whichwe could obtain data, there may still be unobservedfactors, such as subsystem characteristics or engineercapabilities, that could bias the results. According to(Wooldridge 2002), “Omitted variables bias can beeliminated, or at least mitigated, if a proxy variableis available for the unobserved variable” (p. 63), and“often the outcome of the dependent variable from anearlier time period can be a useful proxy variable” (p.66). Nerkar and Paruchuri (2005) and Heckman andBorjas (1980) used this approach by introducing previ-ous performance as an independent variable to predictcurrent performance. To control for unobserved fac-tors, we used warranty claims in the previous year asan independent variable.

4.5.2. Fraction of Problematic ECOs. We includedthe fraction of problematic ECOs as a measure of inter-nal quality problems. The rationale is that the rate ofinternal quality problems could be a signal of exter-nal warranty issues. Because problematic ECOs are aresult of design related mistakes, a high percentage ofproblematic ECOs is a reasonable proxy for the rate ofinternal quality problems.

4.5.3. Fraction of New Parts. Following Clarket al. (1987), who adjusted for the fraction of new partsin a vehicle to compare the productivity of differentauto makers, we include the fraction of new parts (rel-ative to the previous model year) in a subsystem as acontrol variable. We would expect to experience morequality problems with new parts than old ones.

4.5.4. Other Controls. We also controlled for thefollowing additional factors:• Number of parts: This is the total number of parts

in a subsystem, which may be a proxy for the subsys-tem size or complexity.• Number of engineers: This represents the total

number of engineers that appear in the distributionlists associated with a subsystem, which is anotherpotential proxy for the complexity of that subsystem.


• Number of ECOs: This is the total number of ECOsthat are generated in a subsystem, which may be yetanother indicator of the complexity of a subsystem.• Average ECO tardiness: This variable is calculated

using the targeted completion dates and actual com-pletion dates of ECOs. Specifically, it calculates thetardiness for each ECO and then averages it across allECOs in a subsystem. Tardiness could indicate troublein the design process (bad for quality) or additionaltime spent resolving problems (good for quality) andso does not have an obvious expected relationshipwith warranty claims.

5. Analysis and ResultsTable 1 shows descriptive statistics and bivariatecorrelations between the variables in our models.Warranty incidents for the 2004 and 2005 model yearsare highly correlated as expected. Furthermore, wenote that both centrality of a subsystem and coordina-tion deficit have positive correlations with 2005 war-ranty claims.Our study examines a total of 243 subsystems

across 13 vehicle programs. Because we have all sub-systems present in all programs, we have the repeatedobservations for each of the 243 subsystems. Thispanel structure of our data set (i.e., a cross-sectionof 243 subsystems observed 13 times) allows us toexplore both within and between subsystem varia-tion. By using panel data methods, we can controlfor the unobserved subsystem characteristics, whichcould pose a major problem for the ordinary leastsquares estimates (Petersen and Koput 1991).A fixed-effects model could address the problem

of unobserved heterogeneity by including an errorterm that is assumed to be constant over vehicle pro-grams for each subsystem, whereas a random-effectsmodel could address this by inserting an error term

Table 1 Descriptive Statistics

Variable description Mean SD Min Max (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

(1) Warranty incidents 4�115 2�398 0�086 11�830 1for 2005

(2) Warranty incidents 4�372 2�468 0�161 13�652 0�728 1for 2004

(3) Number of parts 230�2 13�489 164 275 0�033 0�014 1(4) Number of design 104�5 28�905 17 221 −0�074 −0�085 0�202 1

engineers(5) Number of ECOs 232�2 16�861 132 294 0�059 0�021 0�317 0�389 1(6) Fraction of new parts 0�332 0�158 0�076 0�714 0�165 0�143 0�034 0�108 0�065 1(7) Fraction of 0�284 0�053 0�109 0�570 0�133 0�117 0�067 0�074 0�087 0�314 1

problematic ECOs(8) Average ECO tardiness 23�68 5�058 0�000 82�000 −0�112 −0�094 0�092 −0�146 0�105 0�103 −0�116 1(9) Centrality of a subsystema 0�219 0�173 0�033 0�572 0�365 0�211 0�285 0�085 0�173 0�007 0�079 −0�084 1

(10) Centrality squared 0�078 0�148 0�001 0�760 −0�288 −0�192 −0�109 −0�022 −0�111 −0�008 −0�071 0�069 −0�173 1of a subsystema

(11) Coordination deficit 0�081 0�047 0�001 0�244 0�269 0�187 0�045 −0�089 0�025 0�093 0�124 −0�118 0�461 −0�297 1

aNormalized scores.

that varies randomly over programs for each subsys-tem. While random-effects models make use of the(seldom met) assumption that individual effects areuncorrelated with the regressors, and model individ-ual constant terms as randomly distributed acrosscross-sectional units, fixed-effects models impose themost powerful control on unobserved heterogene-ity 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) Fixed-effects models can produce biased estimates for pan-els over short time periods (Greene 2008, Hsiao 1986).Because we only have 13 programs (similar to having13 time units) and a large number of cross-sections(N = 243), a fixed-effects model may not be appropri-ate. (ii) Fixed-effects models provide poor estimatesof the effects of the variables that vary only slightlyover time (i.e., over the 13 programs) (Kraatz andZajac 2001). In our panel data, some of the key vari-ables such as number of design engineers and fractionof new parts change only slightly across programs.Random-effects models do not share these limitations.They allow us to examine both within and betweensubsystem variance in independent and dependentvariables.Nevertheless, we fitted both the random-effects

model and fixed-effects model, and conducted aHausman test to determine which specification ismore appropriate (Hausman 1978). In this test, underthe null hypotheses, the two estimates do not dif-fer significantly, and, therefore, the more efficient andconsistent random-effects model is preferable. TheHausman test resulted in a test statistic of �2 = 11�74,which is well below the critical value of 15.51 fromthe chi-squared table. Therefore, the null hypothesisof the “no statistical differences” is not rejected, whichimplies that the random-effects model is the appropri-ate specification for our data.


Table 2 Models of Warranty Incidents for Product Subsystems

Model 1 Model 2a Model 2b Model 3a Model 3bEstimation method: Random-effects Random-effects Fixed-effects Random-effects Fixed-effectsVariable (controls) (architecture) (architecture) (deficit) (deficit)

Warranty incidents for 2004 0�6943∗∗∗ 0�6881∗∗∗ 0�8319∗∗∗ 0�7024∗∗∗ 0�8608∗∗∗

�0�077� �0�077� �0�126� �0�079� �0�131�Number of parts 0�0038 0�0053 0�0084 0�0049 0�0077

�0�013� �0�010� �0�019� �0�014� �0�017�Number of design engineers −0�0098 −0�0087∗ −0�0125 −0�0091∗ −0�0098

�0�006� �0�005� �0�008� �0�005� �0�007�Number of ECOs 0�0061 0�0066 0�0039 0�0064 0�0043

�0�004� �0�004� �0�006� �0�004� �0�006�Fraction of new parts 3�752∗∗ 3�744∗∗ 3�108 3�719∗∗ 3�325

�1�440� �1�472� �2�140� �1�465� �2�144�Fraction of problematic ECOs 10�16∗∗ 9�653∗∗ 6�741∗∗ 9�462∗∗ 6�722∗∗

�5�041� �4�890� �3�127� �4�851� �3�125�Average ECO tardiness −0�153∗∗ −0�147∗∗ −0�219∗ −0�116∗∗ −0�236∗

�0�059� �0�064� �0�115� �0�051� �0�121�

Centrality of a subsystem 3�78∗∗∗ 2�92∗∗∗

�0�942� �0�874�Centrality squared of a subsystem −6�07∗∗∗ −5�13∗∗∗

�1�186� �1�143�Coordination deficit 2�6975∗∗∗ 2�3494∗∗

�0�997� �0�922�

R-squared (%) 71.00 72.55 29.53 73.95 30.76Adjusted R-squared (%) 70.93 72.46 29.35 73.88 30.61N 3,159 3,159 3,159 3,159 3,159

∗p < 0�1; ∗∗p < 0�05; ∗∗∗p < 0�01.

Table 2 presents the results of our panel model.Model 1 consists of only the control variables. As wewould expect, this shows that warranty incidents in2004 is significant (p < 0�01) as a predictor of warrantyincidents in 2005. This confirms that it is an effectiveproxy variable. Both the fraction of new parts andthe fraction of problematic ECOs also have signifi-cant positive coefficients (p < 0�05), which indicates apositive association between these variables and war-ranty incidents. Average ECO tardiness is significant(p < 0�05) with a negative coefficient, which suggeststhat ECOs that take longer to resolve tend to resultin fewer quality problems in the field. Finally, thenumber of parts and the number of ECOs in a sub-system are not significant. This agrees with our on-site observations that (i) subsystems with more partsare not necessarily more complex, because some ofthe simplest subsystems involve many tiny parts, and(ii) total number of ECOs itself is not a good qual-ity indicator because many ECOs are not problemrelated. Note that Model 1 explains almost 71% of thevariation in 2005 warranty incidents.

5.1. Inverted-U relationshipModel 2a adds the linear and quadratic terms forsubsystem centrality to investigate Hypothesis 1. Wenote that both terms are significant, but that the coef-ficient is positive for the linear term and negative

for the quadratic term. Although this is consistentwith the conjectured inverted U-shaped relationship,it is not sufficient to demonstrate it. We must alsoshow an appropriate distribution of the independentvariable around the maximum. Without this, the coef-ficients might indicate a monotonic concave relation-ship instead. The Box-Whisker plot in Figure 6, whichprovides a simple visualization of the data by divid-ing the sample into deciles and box-plotting each sub-sample, also supports the inverted-U relationship.To further check the inverted-U relationship, we

performed two calculations (see the online appendixfor details): (i) First, we calculated the location ofthe inflection point (i.e., the maximum), which cor-responds to (3�78 − 2 × 6�07 × x = 0) or x = 0�311.This value is about half a standard deviation abovethe mean, which supports the inverted-U relationship.(ii) We divided the data into deciles, conducted sepa-rate regressions within each subsample, and observedthe pattern of the coefficient of subsystem centrality.The estimated coefficients in these separate regres-sions confirm the inverted-U relationship with themaximum in the seventh decile. We also checked foroutliers, because a significant curvilinear relationshipbetween subsystem centrality and warranty incidentsmight be attributed to a few outliers in subsystemcentrality. We did not detect any influential outliersusing Cook’s distance (Cook and Weisberg 1982).


Figure 6 Box-Whisker Plot for Subsystem Centrality and Warranty Incidents

0

2

4

6

8

10

12

14

War

rant

y in

cide

nts

Centrality deciles

q1MinMedianMaxq3

Decile 2Decile 1 Decile 3 Decile 4 Decile 5 Decile 6 Decile 7 Decile 8 Decile 9 Decile 10

Note that the behavior of the control variables isquite similar in Models 1 and 2a. Also note thatwith the addition of subsystem centrality variables(both the linear and quadratic term), the adjustedR-squared improves from 70.93% to 72.46% despitethe loss of two degrees of freedom. Hence, we con-clude that Model 2a supports Hypothesis 1 and indi-cates an inverted-U relationship between subsystemcentrality and warranty incidents.

5.2. Coordination DeficitModel 3a replaces the two subsystem centrality vari-ables with our coordination deficit metric as a predic-tor and shows coordination deficit metric to be highlysignificant (p < 0�01) with a positive coefficient. Notethat instead of including both subsystem centralityand coordination deficit in the model simultaneously,the coordination deficit variable replaces the subsys-tem centrality variable.The reason for this is that it is our theory that

both variables are proxies for the same effect, namelythat mismatches between the organizational coordi-nation network and the product architecture networkincrease the likelihood of warranty claims. How-ever, subsystem centrality captures this effect onlyvery roughly, by suggesting through the observedinverted-U relationship that intermediate centralitysubsystems tend to exhibit higher levels of warrantyincidents. In contrast, coordination deficit measuresthe mismatches much more directly by incorporat-ing information about the organizational coordinationnetwork, as well as the product architecture network.

That the two variables overlap in their predictive roleis supported by the fact that they are correlated (cor-relation coefficient= 0.461). That they are not identicalis supported by the fact that a regression includingboth subsystem centrality and coordination deficit hasboth variables significant (at the 5% level).We observe that although the majority of the

explanatory power of the model comes from the pre-vious year’s warranty claims, adding the coordina-tion deficit metric to the original control variablescauses adjusted R-squared to improve from 70.93% inModel 1 to 73.88% in Model 3a. Hence, Model 3a sup-ports Hypothesis 2 by suggesting that coordinationdeficit and warranty incidents are positively associ-ated. The fact that R-squared is higher in Model 3athan in Model 2a suggests that coordination deficitis a better predictor of warranty claims than is sub-system centrality. This makes sense because coordi-nation deficit contains much more information aboutthe system than does subsystem centrality. Finally,note that coefficients of the variables are quite stableacross models. This supports our earlier observationthat multicollinearity is not a problem. But, it alsosuggests that the magnitudes of the coefficients are agood gauge of the effects.In addition to the three models presented in Table 2,

we ran versions of Model 3a using the ratio, node dif-ference 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 notsurprising. The ratio metric assumed a nonlinearrelationship between reduction in mismatches and


warranty claims, which is hard to defend consider-ing the observation in §4.4.2 that architectural inter-faces generate issues requiring communications toresolve, implying a linear relationship between war-ranty claims and mismatches. The node differencemetric is a coarser measure of mismatches than thecoordination deficit metric because there are manyfewer nodes than links, which explains why it did notwork as well in predicting quality problems. Finally,the local deficit metric used the normalization offlows at each node, and did not consider the flowsin the other parts of the network. Hence, this met-ric cannot detect some kinds of misalignment that aredetected by the coordination deficit metric, and, con-sequently, it was less effective.

5.3. Checking for Endogeneity and theRobustness of Results

Because some of the covariates in our modelsinclude measures that are potentially endogenous(e.g., assignment of distribution lists, etc.), it is impor-tant to examine potential endogeneity issues and theirimpact on the results. Although the Hausman testsuggested that a random-effects model is appropri-ate for our data, we also examined the results ofthe fixed-effects model, which specifically controls forthe unobserved heterogeneity. We present the resultsof the fixed-effects regressions in Models 2b and 3b.Note that the coefficients and significance levels ofthe variables in these fixed-effects models are differ-ent than those in the earlier random-effects modelsdue to different model assumptions. Also, because thefixed-effects models only exploit within subsystemvariation, the R2 values are significantly lower thanthe random-effects models. The main conclusion fromModels 2b and 3b is that our main variables of inter-est, 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. Toexamine the endogeneity of coordination deficit, wecreated a two-stage least squares (2SLS) random-effects model (Baltagi 2001). In this procedure, inthe first stage we regress all exogenous variables onthe suspected endogenous variable (i.e., coordinationdeficit) and get the fitted values. Here we use num-ber of new release ECOs, number of new release parts,and number of engineers working on new release partsas the instruments. We expect these variables to havea direct effect on coordination deficit, but no effecton warranty claims other than their indirect effectthrough coordination deficit. We then calculate thesecond stage model using the fitted values created inthe first stage. The results of the 2SLS random-effectsmodel showed that all pairs of coefficients are within

Figure 7 Organizational Coordination vs. Product ArchitectureCentrality

y = 1.114x 2

R2 = 0.471

0

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0.40

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

Product architecture centrality

Org

aniz

atio

nal c

oord

inat

ion

cent

ralit

y

– 0.249x + 0.072

a 90% confidence interval of the original random-effects model, suggesting that endogeneity does notpose a threat for our analysis.Results of Model 2a in Table 2 support the proposed

inverted-U relationship between product architecturecentrality and warranty claims. As mentioned in §2,one potential explanation of this relationship could beas follows: It may be hard for organizations to gaugeand provide the right amount of attention to interme-diate central subsystems, but it is probably easier toidentify highly central subsystems and provide suf-ficient resources accordingly. If this is the case, thenwe should observe fewer warranty claims on highlycentral subsystems than on subsystems with interme-diate centrality. To explore this further, we investi-gated the relationship between product architecturecentrality and organizational coordination centrality.As we see in Figure 7, there is very little differencebetween the organizational coordination centrality forsubsystems with product architecture centrality below0.25, suggesting that the firm does not distinguishbetween the complexity of these subsystems whenmaking decisions that affect coordination activities.However, for subsystems with product architecturecentralities above 0.25, organizational coordinationcentrality increases very rapidly (i.e., consistent withan increasing convex function), suggesting that thesehigh centrality subsystems receive extensive coordina-tion attention. These observations support our reason-ing behind Hypothesis 1 that intermediate centralitysubsystems may not be receiving coordination effortcommensurate with their complexity.From the results of the models in Table 2, it is clear

that warranty claims in the previous year has a largeimpact on the warranty claims this year. To checkthe robustness of coefficients of the other variablesin the models, we removed this lagged variable fromthe models and and reran the statistical analysis. Wedid not observe a significant change in the directionor impact of the coefficients, but as expected, the over-all explanatory power of the models were reduced


(e.g., adjusted R-squared of Model 2 was reduced toaround 11% from 69%) by excluding of the laggedvariable.

5.4. Economic Significance ofCoordination Deficit

In addition to the statistical analyses, we conductedan analysis of economic significance to get a senseof the magnitude of the association between coor-dination deficit and warranty claims. Following theconvention used in other studies (see Nerkar andParuchuri 2005, Song et al. 2003), we computed thepercentage of change in the dependent variable (i.e.,warranty claims) associated with a one standard devi-ation change in the independent variable (i.e., coor-dination deficit), evaluated at the mean of the data.Reducing coordination deficit by one standard devi-ation (which is equal to 0�047) from the mean in ourmodel predicts a 2�6975×0�047= 0�127 unit reductionin warranty claims, which represents a 0�127÷4�115=3�08% reduction. For our client, this would translateinto millions of dollars annually in direct savings,plus an important reputational benefit (i.e., becauseConsumer Reports and other rating services considerwarranty claims in their evaluation and recommenda-tion of vehicles).Although a 3�08% reduction in warranty claims

is economically important, the percentage of qual-ity issues that are related to coordination deficit mayactually be substantially larger than this. The reasonis that in our model, the variable representing war-ranty claims from the previous year may also containclaims that are associated with coordination deficit(i.e., design flaws that were introduced in previousyears and carried over to this year’s model throughpart reuse). So, if the firm were to reduce coordina-tion deficit by one standard deviation in each designcycle, some of the quality improvements would carryover to future vehicles (through components basedon previous designs). Although one cannot calculatethis carryover effect of the reduction in coordinationdeficit with precision, we can get an approximate fig-ure by using our models.One way to estimate the potential magnitude of an

ongoing reduction in coordination deficit is to makeuse of Model 3 as follows: First, we note that theregression 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 = 0�70. Then, wesuppose that 1 remains constant over time and thatwe reduce coordination deficit by x% in each design

cycle. (Recall that redesigns occur every five to sixyears.) If we let

yn = total percentage of reduction (both direct andindirect) in warranty claims in the nth redesign�

and we assume that warranty claims are constantbetween redesigns, then the total reduction in agiven cycle will be equal to the direct reduction pluscarryover, which is

y1 = x

y2 = x+ 0�70xy3 = x+ 0�70x+ �0�70�2x

· · ·yn = x+ 0�70x+ �0�70�2x+ �0�70�3x+ · · ·+ �0�70�n−1x�

(5)

As n → , this geometric series converges to x +x�0�70/1− 0�70 = 3�33x. So, if coordination deficit isreduced by one standard deviation (i.e., x = 3�08%),and if there is no redesign at all, then the percentageof warranty reduction in the nth year converges to3�33× 3�08%= 10�26% when n is large. If, instead, weassume that because of technology change and modelretirement, the dependence on old designs extendsback only five design cycles, then an ongoing onestandard deviation reduction in coordination deficitultimately results in a y5× x (=2�94× 3�08%= 9�06%)reduction in warranty claims.To get a sense of the total number of warranty

claims that are associated with mismatches betweenthe organizational coordination and product architec-ture networks (as opposed to the improvement pre-dicted by our model for a realistically achievable onestandard deviation reduction in coordination deficit),we consider the predicted impact eliminating coor-dination deficit entirely. According to Model 3, thiswould result in a 5�3% direct reduction in warrantyclaims, which would yield a reduction of 3�33 ×5�3%= 17�65% in the limit, and a reduction of 2�94×5�3%= 15�58% if the carryover effect is limited to fivedesign cycles.Although these calculations give us a general sense

of the impact of ongoing reduction in coordinationdeficit, we should be cautious in interpreting the indi-vidual coefficients and carryover effects. In the abovecalculations, warranty claims from the previous year isa proxy variable, which may include many causaleffects, of which coordination deficit is only one.Using our regression model to estimate the amount ofthis variable that is attributable to coordination deficitis reasonable, but far from precise. Moreover, becausewe would expect design flaws from prior years to getcorrected over time, our estimate is probably an upperbound on the overall economic impact of coordinationdeficit reductions on warranty claims.


According to a 2006 J.D. Power and Associatesquality survey, an average of 52 of the 124 (i.e.,42%) of quality problems observed in automobileswere design defects (Jensen 2006). If that is true,then our analysis suggests that almost half of thesedesign related warranty claims are due to organiza-tional coordination problems. The rest, presumably,are associated with individual errors.Along with the cost savings, these numbers also

indicate a major potential reputational benefit. Forexample, in a recent J.D. Power and Associates ini-tial quality survey, Toyota observed 104 complaintsper 100 vehicles (fourth in ranking), while Hondaobserved 110 (seventh in ranking), Ford observed 112(eighth in ranking), and Chevrolet observed 113 prob-lems (tenth in ranking) in the first 90 days of own-ership (Bennett and Boudette 2008). Although thesecomplaints are not the same as our warranty inci-dents, they are certainly related. Because these num-bers are very close for the brands, a 10% reductioncould move a brand from tenth place to third place.So, relatively small improvements in warranty inci-dents could make a significant improvement in afirm’s quality rankings and hence its reputation.

6. DiscussionOur analysis shows that warranty claims in the pre-vious year have significant power for predicting war-ranty claims this year. Indeed, when we use only theprevious year’s warranty claims in a simple regres-sion, it explains about 55% of the variation in thisyear’s warranty claims. Though intuitive, this result isnot of great managerial use, because it merely impliesthat trouble spots in a vehicle tend to persist overtime. In this sense, using last year’s warranty claimdata to predict this year’s warranty claims is a bit likeusing yesterday’s weather to predict today’s weather.There is a substantial correlation, but the model isobvious. Only by going beyond this level of predic-tion can we derive useful forecasts.We also observed a positive correlation between the

fraction of problematic change orders and the numberof warranty claims. Subsystems for which we observea high percentage of problems during design are thevery subsystems that result in a higher number ofwarranty incidents. The implication is that engineersfix some of the design problems by issuing and resolv-ing ECOs, but not all of them. Because some designproblems reach the marketplace and lead to warrantyclaims, management efforts to reduce the problematicchange orders will both speed the vehicle develop-ment process and improve vehicle quality.Another factor shown by our analysis to be cor-

related with warranty claim incidents is the percent-age of new parts in a subsystem. This is intuitivegiven the learning involved in the design of a newpart. From a management perspective, this implies

that design organizations should devote extra atten-tion and resources to subsystems with higher frac-tions of new parts. Although our client clearly knewthis already, the fact that warranty claims are stillpositively correlated with the fraction of new con-tent suggests that current levels of attention and/orresources may not be enough.A somewhat counterintuitive implication of our

results is that tardiness of engineering change ordersand quality problems are negatively correlated.Although one might expect tardiness to compromisequality (e.g., by causing haste or chaos in the designprocess), we observed that subsystems with moreECO tardiness tend to have fewer quality problemsin the field. This may be due to a simple time versusquality trade-off; more time on a component results ina lower probability of a problem, even at the expenseof missing due dates. Of course, while missing a duedate in order to spend more time on a given com-ponent may improve that component, it may alsobe detrimental to other components or the vehiclelaunch as a whole. So, although this result may sug-gest that management should be careful about com-pressing design times too much, it certainly shouldnot be taken as support for missing due dates estab-lished by the ECO system.The inverted-U association we observed between

subsystem centrality and warranty incidents suggeststhat subsystems of intermediate centrality are moreprone to quality problems. We conjecture that this isbecause intermediate centrality parts are more diffi-cult to evaluate with regard to their complexity thanare high centrality parts (which are obviously com-plex) or low centrality parts (which are obviouslysimple). As such, it is more difficult to determinethe appropriate amount of resources and coordinationeffort for intermediate centrality parts than for eitherhigh or low centrality parts. Though intriguing froma research perspective, this result does not identifyspecific subsystems in need of greater attention andhence is of limited managerial use.Our most important contributions are (1) introduc-

ing the coordination deficit metric for quantifyingmismatches between the product architecture and theorganizational structure, and (2) showing that thismetric is positively correlated with warranty claimincidents. This result is significant to the literatureon network analysis of product development sys-tems because (a) it is the first effort to formally mea-sure misalignment between an organization and itsassigned work, and (b) it provides support for thecommon conjecture that misalignment of the designorganization with the product architecture is detri-mental to performance.From a management perspective, this work sug-

gests some potentially appealing insights. First, our


analysis highlights a means for mining ECO systemdata to monitor the alignment of the organization withthe product being designed. Our coordination deficitmetric provides a simple quantitative measure of thedegree of mismatch and points out specific pairs ofsubsystems where the level of formal coordination isless than the extent of connectivity in the productarchitecture. Such pairs of subsystems may be can-didates for additional coordination attention. Becausethe coordination deficit metric also identifies pairs ofsubsystems where the level of coordination activityexceeds the amount of connectivity in the productarchitecture, it may also suggest places where coor-dination efforts can be reduced with minimal impacton performance. This suggests that it is possible toimprove the match between organizational coordina-tion and the product architecture without increasingthe total amount of coordination activity.Although the statistical correlations we have iden-

tified in this study only suggest, rather than prove,causality, the existence of a positive associationbetween coordination deficit and quality problems isof managerial interest. Because of the large cost ofdesign quality problems (e.g., recalls), managers ofproduct development organizations must be sensitiveto any factor that may have an impact on design qual-ity problems. No statistical study (e.g., of the typeused in Six Sigma programs) can ever provide proofof causality, so managers can only pursue improve-ments by addressing factors shown to be associatedwith quality problems. Our paper introduces andquantifies coordination deficit as one such factor. Fur-thermore, because researchers and practitioners havebeen arguing (indeed assuming) that alignment of theorganization with the product is desirable, our find-ings are consistent with current management theory.Our results support this theory and provide concreteguidance on how to act upon it in practice.

7. ConclusionsIn this paper, we have presented an empirical modelthat characterizes the misalignment of the prod-uct architecture and organizational interactions andhave investigated the impact of this misalignment onquality (measured by warranty claims) in a vehicledevelopment process. Our results suggest that orga-nizational factors and product architecture have asignificant impact on quality.Our analysis made use of data from an ECO

system like that used in most product developmentprocesses. These data enabled us to specify bothproduct architecture and organizational coordinationnetworks. As such, our study is the first, of whichwe are aware, that bases a network analysis of theproduct development process entirely on standard

data from a firm’s information system. Because wedo not rely on cumbersome and time consuming sur-veys, our methodology is more likely to find use inpractice than survey based methods.Our work, along with the other studies that have

made use of emerging tools of complex networks tocharacterize both product architecture (a network ofcomponents) and organizational structure (a networkof people), highlight the potential importance of suchnetwork tools to the science and practice of NPD.Our results suggest that misalignment of the designorganization with the product architecture negativelyaffects product quality and uses network tools tohighlight the specific areas of misalignment. Sosaet al. (2004, 2007) suggest that such misalignments areinfluenced by various features of the organizationalstructure and use network tools to characterize thesefeatures. Because of the power and flexibility of thesenetwork tools, they are already becoming a standardpart of the NPD research tool kit. We expect them tobecome similarly prevalent as practical managementtools in the future.To further the science and practice of NPD pro-

cesses, this work could be extended in several direc-tions. First, our research exclusively relied on archival(e.g., ECO, warranty) data. Although this is of sub-stantial practical use, because it captures formal con-nections, it leaves out informal connections, suchas communication outside the channels indicated bythe distribution lists. Hence, a complementary studycould make use of surveys or e-mail and phonerecords to characterize informal communication foruse as an additional predictor of quality performance.A second dimension along which our model could

be refined is the granularity of the product data. Wehave performed our analysis at the subsystem level.This was largely because our client only had warrantyclaims data that could be appropriately aggregated atthis level. However, we could obtain warranty claimsat the part level, we could perform a much moredetailed analysis of the impact of coordination deficiton product quality. Our expectation is that this wouldfacilitate more precise matching of the organizationalstructure to product architecture. It would also enablea more accurate prediction of potential quality troublespots.Finally, we note that the ultimate managerial pur-

pose of this type of analysis is to better adapt thedesign organization to the products being developed.Our results provide an approach for identifying gapsbetween organizational structure and product archi-tecture. However, we have only analyzed vehicleprograms for one model year. To get a deeper under-standing of how vehicle architectures evolve overtime and where the organizational coordination prac-tices lag behind product changes, it would be useful


to perform a longitudinal study over multiple modelyears. Although getting data extending back acrossmultiple design cycles would be a huge challenge,such an analysis would represent an important stepin the use of complex network methods to further thescience of product development.

8. Electronic CompanionAn electronic companion to this paper is available aspart of the online version that can be found at http://mansci.journal.informs.org/.

AcknowledgmentsThe authors gratefully acknowledge the support of thisresearch by the National Science Foundation underGrants DMI-0423048 and DMI-024377. The authors thankChristoph Loch (the department editor), the associate edi-tor, and the reviewers for their excellent feedback and wisesuggestions, which have improved this paper substantially.

ReferencesAdler, P. S., A. Mandelbaum, V. Nguyen, E. Schwerer. 1995. From

project to process management: An empirically-based frame-work for analyzing product development time. ManagementSci. 41(3) 458–484.

Baldwin, C. Y., K. B. Clark. 2000. Design Rules: The Power of Modu-larity. MIT Press, Cambridge, MA.

Baltagi, B. 2001. Econometric Analysis of Panel Data, 2nd ed. JohnWiley & Sons, New York.

Bennett, J., N. E. Boudette. 2008. Ford and GM show gains inquality 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 Win-dows: Software for Social Network Analysis. Analytic Technolo-gies, Harvard, MA.

Brown, S. L., K. M. Eisenhardt. 1995. Product development: Pastresearch, present findings, and future directions. Acad. Manage-ment 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 bound-aries: Boundary objects in new product development. Organ.Sci. 13(4) 442–455.

Clark, K. B., T. Fujimoto. 1990. The power of product integrity. Har-vard Bus. Rev. 68(November–December) 107–118.

Clark, K. B., T. Fujimoto. 1991. Product Development Performance:Strategy, Organization, and Management in the World Auto Indus-try. 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. BrookingsPapers Econom. Activity 1987(3) 729–781.

Contractor, N. S., P. R. Monge. 2002. Managing knowledge net-works. 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 incomplex systems. Management Sci. 50(2) 159–173.

Fleming, L., O. Sorenson. 2004. Science as a map in technologicalsearch. 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 effectiveknowledge sharing in multiunit companies. Organ. Sci. 13(3)232–248.

Hausman, J. A. 1978. Specification tests in econometrics. Economet-rica 46(6) 1251–1271.

Heckman, J. J., G. J. Borjas. 1980. Does unemployment cause futureunemployment? Definitions, questions and answers from acontinuous-time model of heterogeneity and state dependence.Economica 47(187) 247–283.

Henderson, R. M., K. B. Clark. 1990. Architectural innovation: Thereconfiguration of existing product technologies and the failureof 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 engineeringchange management in UK manufacturing industries. Internat.J. Oper. Production Management 19(1) 21–37.

Ibarra, H. 1993. Network centrality, power, and innovation involve-ment: Determinants of technical and administrative roles. Acad.Management J. 36(3) 471–501.

Jensen, C. 2006. Design quirks erode ’06 quality ratings. NewYork Times (June 11), http://www.nytimes.com/2006/06/11/automobiles/11QUALITY.html.

Kraatz, M. S., E. J. Zajac. 2001. How organizational resources affectstrategic change and performance in turbulent environments:Theory and evidence. Organ. Sci. 12(5) 632–657.

Krackhardt, D., J. R. Hanson. 1993. Informal networks: The com-pany behind the chart. Harvard Bus. Rev. 71(4) 104–111.

Krishnan, V., K. T. Ulrich. 2001. Product development decisions: Areview of the literature. Management Sci. 47(1) 1–21.

Loch, C. H., C. Terwiesch. 1999. Accelerating the process of engi-neering change orders: Capacity and congestion effects. J. Prod-uct 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 oscilla-tions in complex engineering projects. Management Sci. 49(6)733–750.

Nerkar, A., S. Paruchuri. 2005. Evolution of R&D capabilities: Therole 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 organi-zational mortality: Legitimacy or unobserved heterogeneity?Amer. Sociol. Rev. 56(3) 399–409.

Sanchez, R., J. T. Mahoney. 1996. Modularity, flexibility, and knowl-edge management in product and organization design. Strate-gic Management J. 17 63–76.

Schilling, M. A. 2002. Toward a general modular systems theoryand its application to interfirm product modularity. Acad. Man-agement Rev. 25(2) 312–334.

Song, J., P. Almeida, G. Wu. 2003. Learning-by-hiring: When ismobility more likely to facilitate interfirm knowledge transfer?Management Sci. 49(4) 351–365.

Sosa, M. E., S. D. Eppinger, C. M. Rowles. 2004. The misalignmentof product architecture and organizational structure in complexproduct 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 interfacesin complex product development. Working paper, INSEAD,France.

Terwiesch, C., C. H. Loch. 1999. Managing the process of engi-neering change orders: The case of the climate control systemin automobile development. J. Product Innovation Management16(2) 160–172.

Ulrich, K. 1995. The role of product architecture in the manufactur-ing firm. Res. Policy 24(3) 419–441.

Wasserman, S., K. Faust. 1994. Social Network Analysis. CambridgeUniversity Press, London.

Wooldridge, J. M. 2002. Econometric Analysis of Cross Section andPanel Data. MIT Press, Cambridge, MA.

S

Ya

b

a

ARRAA

KSSSSNSN

1

baacbicbstne

vaif(Bii

0d

Journal of Operations Management 29 (2011) 194–211

Contents lists available at ScienceDirect

Journal of Operations Management

journa l homepage: www.e lsev ier .com/ locate / jom

tructural investigation of supply networks: A social network analysis approach

usoon Kima,∗, Thomas Y. Choib, Tingting Yanb, Kevin Dooleyb

Department of Management, Marketing, and Logistics, College of Business Administration, Georgia Southern University, Statesboro, GA 30460, United StatesDepartment of Supply Chain Management, W. P. Carey School of Business, Arizona State University, Tempe, AZ 85287, United States

r t i c l e i n f o

rticle history:eceived 26 October 2009eceived in revised form 5 November 2010ccepted 10 November 2010vailable online 18 November 2010

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. Inthis paper we demonstrate how to use social network analysis to investigate the structural characteristicsof supply networks. Our theoretical framework relates key social network analysis metrics to supplynetwork constructs. We apply this framework to the three automotive supply networks reported in Choiand Hong (2002). Each of the supply networks is analyzed in terms of both materials flow and contractual

eywords:upply networksupply chain managementecond-tier suppliersocial network analysisetwork structure

relationships. We compare the social network analysis results with the case-based interpretations in Choiand Hong (2002) and conclude that our framework can both supplement and complement case-basedanalysis of supply networks.

© 2010 Elsevier B.V. All rights reserved.

tructural analysisetwork indices

. Introduction

Supply chain management has focused on linear relationships ofuyers and suppliers (Cox et al., 2006; Zhu and Sarkis, 2004). Whilelinear perspective may be useful for planning certain mechanicalspects of transactions between buyers and suppliers, it fails toapture the complexity needed to understand a firm’s strategy orehavior, as both depend on a larger supply network that the firm

s embedded in (Choi and Kim, 2008). A firm’s “supply network”onsists of ties to its immediate suppliers and customers, and tiesetween them and their immediate suppliers and customers, ando on (Cooper et al., 1997; Croxton et al., 2001). In the past decadehere has been increased discussion of the benefits of adopting aetwork perspective in supply chain management research (Choit al., 2001; Lazzarini et al., 2001; Lee, 2004; Wilding, 1998).

From a supply network perspective, the relative position of indi-idual firms with respect to one another influences both strategynd behavior (Borgatti and Li, 2009). In this context, it becomesmperative to study each firm’s role and importance as derivedrom its embedded position in the broader relationship structure
Borgatti and Li, 2009; DiMaggio and Louch, 1998). For example,urkhardt and Brass (1990) and Ibarra (1993) claim that power and
nfluence derive from a firm’s structural position in its surround-ng 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).

272-6963/$ – see front matter © 2010 Elsevier B.V. All rights reserved.oi: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 net-works that have been done have relied on qualitative methods toderive theoretical and practical insights (e.g., Harland et al., 2001;Jarillo and Stevenson, 1991). While qualitative interpretations havetheir merits, their validity is threatened by a researcher’s boundedrationality, which includes the difficulty to conceptualize complexphenomena such as networks. Thus in this paper we propose toanalyze the structural characteristics of supply networks using aformal, 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 supple-ment and compliment more traditional, qualitative interpretationmethods when analyzing cases involving supply networks.

Social network analysis (SNA) has recently gained acceptanceamong scholars for its potential to integrate the operations andsupply management field with other branches of managementscience (Autry and Griffis, 2008; Borgatti and Li, 2009; Carteret al., 2007). According to Borgatti and Li (2009), SNA conceptsare particularly suitable for studying how patterns of inter-firmrelationships in a supply network translate to competitive advan-
tages through management of materials movement and diffusion ofinformation.
To date, SNA has not been applied in an empirical study of realsupply networks; in fact there is a general paucity of SNA appli-

dx.doi.org/10.1016/j.jom.2010.11.001

http://www.sciencedirect.com/science/journal/02726963

http://www.elsevier.com/locate/jom

mailto:[email protected]

dx.doi.org/10.1016/j.jom.2010.11.001

ons M

celtidimsaaa

f(tHIHUioo

2

2

ivTpson2w

tsntahsCwmtt(pdp

hPwNm(1pa

Y. Kim et al. / Journal of Operati

ations in operations and supply management, with only a fewxceptions (e.g., Carter et al., 2007; Choi and Liker, 1995). This isargely because there is lack of conceptual clarification as to howhe key SNA metrics (e.g. centrality) can be theoretically interpretedn the context of supply networks. Therefore in this study we linkifferent SNA metrics at the node- or firm-level to specific roles

n a supply network. We consider supply networks based on bothaterials flow and contractual relationships. The metrics yield six

upply network related constructs: supply load, demand load, oper-tional criticality, influential scope, informational independence,nd relational mediation. Different network-level SNA metrics arelso linked to their implications for supply network performance.

We apply our framework to real supply network data derivedrom three published case studies of automotive supply networksChoi and Hong, 2002). In that study the authors created empiricallyhree complete network maps of the center console assembly foronda Accord, Acura CL/TL, and DaimlerChrysler Grand Cherokee.

n the present paper, we convert the network data from Choi andong (2002) into matrix forms and analyze them using the softwareCINET 6 (Borgatti et al., 2002). These quantitative results are then

nterpreted using our theoretical framework. Finally, we discussur quantitative SNA results comparing to the qualitative findingsf Choi and Hong (2002) and consider the implications.

. Literature review

.1. Supply networks

Supply networks consist of inter-connected firms that engagen procurement, use, and transformation of raw materials to pro-ide goods and services (Lamming et al., 2000; Harland et al., 2001).he relatively recent incorporation of the term “network” into sup-ly chain management research represents a pressing need to viewupply chains as a network for firms to gain improved performance,perational efficiencies, and ultimately sustainable competitive-ess (Corbett et al., 1999; Dyer and Nobeoka, 2000; Kotabe et al.,002). Therefore, it is increasingly important to analyze the net-ork structure of supply relationships.

In the operations and supply management field, a complex sys-em perspective has been used as a theoretical lens for describingupply networks. Wilding (1998) studied dynamic events in supplyetworks through what he referred to as “supply chain complexityriangle” (p. 599). Choi et al. (2001) conceptualized supply networkss a complex adaptive system (CAS). Surana et al. (2005) proposedow various complex systems concepts can be harnessed to modelupply networks. Pathak et al. (2007) discussed the usefulness ofAS principles in identifying complex phenomena in supply net-orks. Others have examined supply networks from a strategicanagement perspective. Greve (2009), using supply networks in

he maritime shipping industry, studied whether technology adop-ion is more rapid in centrally located network positions. Mills et al.2004) suggested different strategic approaches to managing sup-ly networks depending on whether a firm is facing upstream orownstream and whether it is seeking its long-term or short-termosition in the supply network.

Methodologically, simulation models have been used to studyypothetical supply networks (Kim, 2009; North and Macal, 2007;athak et al., 2007). Others have studied real-world supply net-orks using the case study approach (Jarillo and Stevenson, 1991;ishiguchi, 1994; Choi and Hong, 2002). Scholars in the industrial
arketing have developed descriptive models of supply networks
Ford, 1990; Håkansson, 1982, 1987; Håkansson and Snehota,995). Descriptive case studies in this genre illustrate how com-anies such as Benetton, Toyota, or Nissan attained competitivedvantage through their supply networks (Jarillo and Stevenson,

anagement 29 (2011) 194–211 195

1991; Nishiguchi, 1994). Other studies focused on developingtaxonomies of supply networks (Harland et al., 2001; Lamminget al., 2000; Samaddar et al., 2006).

More recently, Borgatti and Li (2009) have highlighted thesalience of SNA to study supply networks. In fact, there have beena few studies in the operations and supply management field thatused or promoted the use of SNA. Choi and Liker (1995) used SNA toinvestigate the implementation of continuous improvement activ-ities in automotive supplier firms. Carter et al. (2007) providedan example of the application of SNA in a logistics context. Autryand Griffis (2008) applied the concept of social capital, framed aspart of social network theory, to supply chain context. However,still lacking in such studies is a theoretical framework that relatessocial network theory to supply network dynamics and the com-prehensive application of SNA to studying supply networks. In thefollowing section, we provide a brief overview of SNA, focusing onthe key metrics useful for investigating and explaining phenomenawithin 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 inthat they have an ability to make choices. With its computationalfoundation in graph theory (Cook et al., 1998; Kircherr, 1992; Liand Vitányi, 1991), SNA analyzes the patterns of ties in a network.Naturally, SNA has been used to study community or friendshipstructure (Kumar et al., 2006; Wallman, 1984) and communicationpatterns (Koehly et al., 2003; Zack and McKenney, 1995). It has beenadopted 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 direc-torships (Robins and Alexander, 2004; Scott, 1986) and networkeffects on individual firms’ performance (e.g., Ahuja et al., 2009;Burkhardt and Brass, 1990; Gulati, 1999; Jensen, 2003; Rowleyet al., 2005; Stam and Elfring, 2008; Uzzi, 1997).

Operations and supply management scholars have also notedthe methodological potential of SNA. For instance, Choi et al. (2001)stated that one could approach the study of supply networks fromthe social network perspective. Ellram et al. (2006) acknowledgedsocial network theory as a useful tool to study influence in sup-ply chains. Carter et al. (2007) identified SNA as a key researchmethod to advance the fields of logistics and supply chain man-agement. More recently, Borgatti and Li (2009) and Ketchen andHult (2007) echoed such sentiments. They have also recognized thedifficulty of collecting network-level data in supply networks butargued its imperativeness for operations and supply managementto be integrated with other management disciplines.

According to Borgatti and Li (2009), a more systematic adoptionof SNA will be instrumental in exploring behavioral mechanisms ofentire supply networks. A SNA approach allows us to better under-stand the operations of supply networks, both at the individual firmlevel and network level—how important the individual firms are,given their positions in the network and how the network structureaffects 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 ofnetwork metrics at the node- or network-level to characterize thedynamics inside a social network.

2.3. Key network metrics

Network metrics can be calculated at two levels—the node leveland network level. Node-level metrics measure how an individ-ual node is embedded in a network from that individual node’s

1 ons M

pmlftd

2

mTmtnc111

nd(tbcnnwM

snAwunnco

smnofdstwo

2

nwioiu

c(t(tn

96 Y. Kim et al. / Journal of Operati

erspective. In this study, we focus on three types of node-leveletrics—degree, closeness, and betweenness centrality. Network-

evel metrics compute how the overall network ties are organizedrom the perspective of an observer that has the bird’s eye view ofhe network. The network-level metrics we consider are networkensity, centralization, and complexity.

.3.1. Node-level metricsIdentifying the key actors in a social network is one of the pri-

ary uses of SNA (Tichy et al., 1979; Wasserman and Faust, 1994).he concept of centrality is fundamental to node-level networketrics (Borgatti and Everett, 2006; Borgatti and Li, 2009). Cen-

rality reflects the relative importance of individual nodes in aetwork. A node’s central position in a social network has a signifi-ant impact on its and others’ behaviors and well-beings (Mizruchi,994). Centrality has been associated with social status (Bonacich,972; Freeman, 1979), power (Coleman, 1973), and prestige (Burt,982).

There are different types of centrality metrics and they identifyodes that are important, in different aspects. Most prominent areegree centrality, closeness centrality, and betweenness centralityEverett and Borgatti, 1999; Krackhardt, 1990; Marsden, 2002). Ofhese, the most straightforward is degree centrality. This conceptuilds on an observation that the more links a node has the moreentral it is—when a node is connected to a large number of otherodes, the node has high degree centrality. Due to its greater con-ectedness with other nodes, a node with high degree centralityould necessarily be more visible in the network (Freeman, 1979;arsden, 2002).Another centrality concept is closeness centrality. As the term

uggests, this metric focuses on how close a node is to all the otherodes in the network beyond ones that it is directly connected to.node is central if it can quickly reach all the others, and that ishy closeness centrality includes indirect ties. This centrality issually associated with node’s autonomy or independence in socialetworks (Freeman, 1979; Marsden, 2002)—a node with high close-ess centrality has more freedom from others’ influence and higherapacity for independent actions. Such nodes become less reliantn other nodes.

Betweenness centrality measures how often a node lies on thehortest path between all combinations of pairs of other nodes. Theore a given node connects nodes that would otherwise be discon-

ected, the more central that node is—other nodes are dependentn this node to reach out to the rest of the network. This metricocuses on the role of a node as an intermediary and posits that thisependence of others makes the node central in the network. Asuch, the betweenness centrality usually denotes a node’s poten-ial control or influence in the network (Marsden, 2002). A nodeith high betweenness centrality has a great capacity to facilitate

r constrain interactions between other nodes (Freeman, 1979).

.3.2. Network-level metricsSNA also yields metrics concerning the structure of the overall

etwork, such as network density, network centralization, and net-ork complexity. Network density refers to the number of total ties

n a network relative to the number of potential ties. It is a measuref the overall connectedness of a network (Scott, 2000)—a networkn which all nodes are connected with all other nodes would gives a network density of one.

Network centralization captures the extent to which the overallonnectedness is organized around particular nodes in a network
Provan and Milward, 1995). Conceptually, network centraliza-ion can be viewed as an extension of the node-level centralityFreeman, 1979)—if a network had such a highly centralized struc-ure that all connections go through few central nodes, then thatetwork would be high on network centralization. The network
anagement 29 (2011) 194–211

with highest possible centralization is one with a star structure,wherein a single node at the center is connected to all other nodesand these other nodes are not connected to each other. Likewise, thelowest centralization occurs when all nodes have the same numberof connections to others.

Network centralization and network density are complemen-tary. Whereas centralization is concerned with the distribution ofpower or control across the network, density reflects network cohe-siveness. A network that has every node connected with everyoneelse would have a highest possible density (i.e., density of one).This network would be a highly cohesive network but would havea diffuse and distributed control structure.

Network complexity is defined as “the number of dependencyrelations within a network” (Frenken, 2000, p. 260) and thus woulddepend on both the number of nodes in the network and the degreeto which they are interlinked (Frenken, 2000; Kauffman, 1993). Inthe context of a supply network, complexity relates to the collec-tive operational burden born by the members in the network (Choiand Krause, 2006). For instance, a large number of units in a systemis likely to entail high coordination cost (Kim et al., 2006; Provan,1983). Further, if these units are highly interdependent, then thecollective operational burden would be high and thus more com-plex at the system level.

Network complexity is related to network density and networkcentralization. First, more complex networks require higher opera-tional burden (Lokam, 2003; Pudlák and Rödl, 1992, 1994). Second,network density is conceptually linked with network complexitybecause a denser network requires more effort to build and main-tain (Marczyk, 2006). Finally, network centralization is associatedwith network complexity because the highest coordination costsrequire when every node is connected to all other nodes (i.e., anetwork 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 estab-lished between firms in the supply network. For example, a tiemight be established between two firms if they were collaboratingon a new product development or if they had overlapping boardmembership or belonged to the same trade organization. In thispaper 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 mate-rials, or they can be linked through a contractual relationship (Choiand Hong, 2002). In a tree-like structure of materials flow (Berryet al., 1994; Chopra and Sodhi, 2004; Hwarng et al., 2005), thenetwork describes which supplier delivers to which customer. Theother type of network is based on contractual relationships. Often,when a buying company wants to control the bill of materials, itengages in directed sourcing, wherein it establishes a contract witha second- or third-tier supplier and directs the top-tier supplierto receive materials from them (Choi and Krause, 2006; Chopraand Sodhi, 2004; Park and Hartley, 2002). In this context, materi-als flow occurs between two firms who do not have a contract andvice versa. These two types of supply networks, although based onthe 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 constructsWe now consider the key node-level SNA metrics and discuss

how they can be used to interpret different roles in supply net-

Y. Kim et al. / Journal of Operations Management 29 (2011) 194–211 197

Table 1Node-level centrality metrics and their implications for supply networks.

Network type Centralitymetrics

Supply networkconstructs

Conceptual definitions Implication for central nodes

Rolea Description Key capabilities

Materials flow Indegreecentrality

Supply load The degree of difficulty faced by afirm in managing incomingmaterial flows from the upstreamfirms

Integrator To put together or transformdifferent parts into a value-addedproduct and ensure it functionswell

System integrationDesign/developmentArchitecturalinnovation

Outdegreecentrality

Demand load The degree of difficulty faced by afirm in dealing with demands fromthe downstream firms

Allocator To distribute limited resourcesacross multiple customers,focusing on scale economies

Process/manufacturingQuality managementComponent innovationOut-bound logistics

Betweennesscentrality

Operationalcriticality

The extent to which a firm impactsthe final assembler’s operationalperformance in terms of productquality, coordination cost andoverall lead-time.

Pivot To facilitate or control the flows ofsupply across the whole network

Risk managementIn- and out-boundlogisticsCross-functional integ.

Contractualrelationship

Degreecentrality

Influential scope The extent to which a firm has animpact on operational decisions orstrategic behavior of other firms inthe supply network

Coordinator To reconcile differences of networkmembers and align their opinionswith the greater supply networkgoals

Contract managementSRM/CRM

Closenesscentrality

Informationalindependence

The extent to which a firm hasfreedom from the controllingactions of others in terms ofaccessing information in thesupply network

Navigator To explore, access, and collectvarious information with greaterautonomy in the supply network

Information acquisitionStrategic alignmentwith OEM

Betweenness Relational The extent to which a firm canover

firms

Broker To mediate dealings between Information processing

wcfpr

oissdta

3mC

C

wnG1i

C

d

ndsid

centrality mediation intervene or has controlinteractions among otherthe supply network

a Network role given high centrality.

orks. Table 1 offers an overview of key centrality metrics, theorresponding supply network constructs, and their implicationsor network roles in the context of modeling supply networks. Weropose this new framework for the interpretation of the SNA met-ics in the supply network context.

To illustrate these constructs, we first discuss the calculationf key SNA metrics and the essential properties of each. Then, wentegrate each key SNA metric separately with the two types ofupply networks (i.e., materials flow and contractual relation). Wehould note that the supply network based on materials flow isirectional, whereas the supply network based on contractual rela-ionship is non-directional as legal obligations are mutually agreednd enacted.

.2.1.1. Degree centrality in supply network. Degree centrality iseasured by the number of direct ties to a node. Degree centrality

D(ni) for node i(ni) in a non-directional network is defined as:

D(ni) =∑

j

xij =∑

j

xji

here xij is the binary variable equal to 1 if there is a link betweeni and nj but equal to 0 otherwise (Freeman, 1979; Glanzer andlaser, 1959; Nieminen, 1973; Proctor and Loomis, 1951; Shaw,954). To account for the impact of network size g, degree centrality

s normalized as the proportion of nodes directly adjacent to ni:

′D(ni) = CD(ni)

g − 1.

For comparison purposes, in this study, we convert normalizedegree centrality to a 0–100 scale by multiplying by 100.

A high degree centrality points to “where the action is” in a
etwork (Wasserman and Faust, 1994, p. 179). Freeman (1979)escribes it as reflecting the amount of relational activities, anduch activities make the nodes with high degree more visible. Fornstance, in a non-directional contractual relationship network, theegree centrality refers to the extent to which the firm influences
innetwork members and turn theminto its own advantage

Strategic alignmentwith OEM

other firms on their operations or decisions as the firm has moredirect contacts with others (Cachon, 2003; Cachon and Lariviere,2005; Ferguson et al., 2005). In contrast, nodes with low degreecentrality are considered peripheral in the same network. If a nodeis completely isolated (i.e., zero degree), then removing this nodefrom the network has virtually no effect on the network. There-fore, a firm who has more contractual ties in the network garners abroad range of influence on others, and at the same time such a firmwould often be required to reconcile conflicting schedules or inter-ests between others. For the final assembler, for instance, it wouldmake sense to align with suppliers with high degree centrality.

In a directional network of materials flow, the focus is eitheron the flow initiated (out-degree) or flow received (in-degree). Forinstance, out-degree centrality of a node is defined as:

C ′D(ni) = xi+

g − 1.

In-degree centrality and out-degree centrality indicate the sizeof the adjacent upstream tier and downstream tier, respectively.A high in-degree or out-degree can capture transactional intensityor related risks for a firm (Powell et al., 1996). In a materials flownetwork, in-degree centrality for a firm can reflect the degree ofdifficulty faced by the firm when managing the incoming materialflows. In other words, this metric measures the firm’s operationalload coming from the upstream suppliers. A firm with high in-degree centrality may serve the role of an integrator, as they aretasked with organizing and incorporating a range of parts from var-ious suppliers to maintain the overall integrity of the product orservice (Parker and Anderson, 2002; Violino and Caldwell, 1998).Such members in a supply network are instrumental and vital incarrying 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 difficultyin managing the needs of customers. The more direct customersthere are in downstream, the more challenging it is for the firmto ensure on-time delivery, cost-effective inventory, and order

1 ons M

mtrWhpooocn(

3cm1eIttpn

C

w

nwnrbat

rBcarstatScr2

3cbadnafe

C

wa


anagement for their customers. The number of direct cus-omers is thus positively associated with the operational loadelated to demand integration and resource allocation (Frohlich and

estbrook, 2002). In a materials flow supply network, a firm withigh out-degree centrality tends to be a common supplier to multi-le downstream firms. Such supplier can economize and capitalizen its own internal resources as it aggregates demands from a rangef customers (Nobeoka, 1996). Further, this firm is more likely thanthers to gain access to proprietary assets or information of itsustomer firms. This firm is in the best position to allocate or chan-el production or technical information to others in the networkCassiman and Veugelers, 2002).

.2.1.2. Closeness centrality in supply network. The calculation ofloseness centrality is based on geodesic distance d(ni, nj) —theinimal length of a path between two nodes ni and nj (Hakimi,

965; Sabidussi, 1966). In this study, closeness centrality is consid-red only in contractual relationship networks, as shown in Table 1.n a directed network (e.g., materials flow), the geodesic(s) from nio nj may not be the same as the one(s) from nj to ni, or there can bewo geodesics between two non-adjacent nodes. In the case of sup-ly networks, this does not make physical sense. Therefore, typicalode closeness is defined as:

C (ni) =

⎡⎣

g∑j=1

d(ni, nj)

⎤⎦

−1

here∑g

j=1d(ni, nj) is the total distance between ni and all other

odes. At a maximum, the index equals (g − 1)−1, which happenshen the node is adjacent to all other nodes. When all the otherodes are not reachable from the node in question, the indexeaches its minimum value of zero. The index can be normalizedy multiplying CC(ni) by g − 1. The value then ranges between 0nd 1 regardless of network size (Beauchamp, 1965). In this study,he normalized index is converted to a 0–100 scale.

Nodes with high closeness need not much rely on others forelaying information or initiating communications (Bavelas, 1950;eauchamp, 1965; Leavitt, 1951). This metric, in a supply networkontext, thus can represent the extent to which a firm can actutonomously and navigate freely across the network to accessesources in a timely manner. Such a firm has comparatively shorterupply chains, both upstream and downstream. Shorter chainsranslate into less distortion of information and better ability toccess reliable information (e.g., demand forecasts, supply disrup-ion) in a timelier manner (Lee et al., 1997; Chen et al., 2000).uch accessibility to high-quality information increases the firm’sapability to match supply and demand (Cachon and Fisher, 2000),esulting in less inventory and lower operational costs (Lee et al.,000).

.2.1.3. Betweenness centrality in supply network. Betweennessentrality appears under both types of networks. A firm can lieetween a pair of non-adjacent firms either along their materi-ls flow or contractual relationship. The intermediary will haveifferent effects on the firms it links, whether directionally oron-directionally. Measuring betweenness centrality begins withn assumption that a connection between two nodes, nj and nk,ollows their geodesics. Therefore, betweenness centrality can bexpressed as (Freeman, 1977):

∑
B(ni) =
j<k

gjk(ni)gjk

here gjk is the total number of geodesics linking the two nodes,nd gjk(ni) is the number of those geodesics that contain ni. The

anagement 29 (2011) 194–211

ni’s betweenness is then simply the sum of the probabilities thatthe node lies between other nodes. The betweenness reaches themaximum when ni falls on all geodesics and has a minimum of zerowhen ni falls on no geodesics. We normalize it to a value between0 and 100:

C ′B(ni) = CB(ni)

[(g − 1)(g − 2)/2]× 100.

The betweenness can be viewed as indicating how much “gate-keeping” ni does for the other nodes (Borgatti and Everett, 2006;Freeman, 1980; Spencer, 2003). Gatekeeping occurs because a nodeon geodesic can control the flows of materials or communication(Marsden, 2002). When applying to materials flow networks, firmswith high betweenness act as a hub or pivot that transmits materi-als along the supply chains, and betweenness centrality relates tothe 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 afirm with high betweenness transmits materials to a wrong placeor 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 caneasily infect the broader supply network, interfering with normalproduct flows (Kleindorfer and Saad, 2005). Therefore, operationalhiccups caused by such firms can surely hamper the functioning ofthe entire supply network (Hendricks and Singhal, 2005). Consid-ering 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 andSinghal, 2003).

In a contractual relationship network, the metric can denote theextent to which a firm can affect the interactions among others inthe same supply network. A firm with high betweenness centralitymediates many pathways and thus can either facilitate or interferewith the network communications. The social network literaturesuggests that a node linking dense regions of relationships enjoysthe benefits of non-redundant information to increase its controlover others (Burt, 1992, 1998). Supply network research also postu-lates that a buyer can enjoy the increased sourcing leverage whenit lies between two disconnected, competiting suppliers (Choi andWu, 2009; Wu and Choi, 2005). For instance, the buyer can playtwo rival suppliers off each other to drive down the purchasingprice.

3.2.2. Network-level constructsWe now discuss the key network-level metrics. Table 2 sum-

marizes the theoretical interpretation of the metrics and theirimplications for network performance in the context of supply net-works.

3.2.2.1. Supply network centralization. Recall that CD(ni) is node-level 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

i=1[CD(n∗) − CD(ni)]

max∑g

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 whenone node is connected with all other g − 1 nodes, and the others
interact only with this node. Its minimum value of 0 occurs when alldegree centrality values are equal. In supply networks, centraliza-tion can refer to how much power or control the core firms exerciseover other network members (Choi and Hong, 2002). In this study,besides centralization based on degree, two other centralization


Table 2Network-level metrics and their implications for supply networks.

Network type Network-levelmetrics

Conceptual definition in supplynetworks

Implication of overall network structurea

Characteristics Performance implications

Materials flow Centralization The extent to which particularfocal firms control and managethe movement of materials in asupply network

Operational authority (e.g., power to makedecisions on materials flow) concentratedin few central firmsCentralized decision implementationprocess

High level of controllability in productionplanningLow level of operational effectiveness atthe network-level (i.e., more time taken toreach a decision and take actions on issuesat a local level)

Complexity The amount of collectiveoperational burden born by themember firms in a supplynetwork

More firms engaged in the delivering andreceiving of materialsMore steps required to move the materialsalong

Low level of operational efficiency at thenetwork level (i.e., longer lead time fromthe most upstream to the final assembleror more parts for the same productfunction)

Contractualrelationship

Centralization The extent to which particularfocal firms exercise bargainingpower or relationshipmanagement control overother firms in a supply network

Lack of interactions between central andperipheral firms in a supply networkDecoupled relationships between firms atdifferent tiers

High level of controllability in productdesign, product quality, and/or costmanagementLow level of responsiveness to or moretime for resolution on issues occurring at alocal level

Complexity The amount of load on thesupply network as a whole that

More firms involved in transferringinformation

Low level of robustness or high degree ofvulnerability to supply disruptions (i.e.,

Active interactions at a local levelrelay

nstrea

ic

icaaib1(r

3rtretpontm

sptlictbtbili

requires relationshipcoordination Slow

dow

a Implications given high metric score.

ndices are also used—ones based on closeness and betweennessentrality.

Further, there are other proxy measures of centralization usedn this study. They are multiple indices of density that involve theore and periphery sub-groups in a network (see Table 6). Whennetwork is partitioned into two clusters, a core cluster appears

mong nodes that are densely connected together and a peripherys formed among nodes that are more connected to core mem-ers than to each other (Borgatti and Everett, 2000; Luce and Perry,949). For instance, in Fig. 1, there are 19 firms in the core groupsee Table 6) around Honda and CVT who appear at the center. Theest appears in the periphery.

.2.2.2. Supply network complexity. Supply network complexityefers to the load on the network as a whole that requires coordina-ion (Choi and Hong, 2002). While the general state of the literatureegarding the property of complexity at the network level is stillmerging (Butts, 2001a,b; Everett, 1985; Freeman, 1983), we adopthe idea put forth by Kauffman (1993) and Frenken (2000). Theyropose that network complexity can be indicated by the numberf nodes and degree of interdependency among nodes in a givenetwork. Therefore, we use two types of SNA output metrics—size-ype and density-type—to represent the number of supply network

embers and the level of connectedness among them, respectively.The size-type outputs are shown in network size and core

ize, and the density-types include network density, core density,eriphery density, core-to-periphery (CTP) density, and periphery-o-core (PTC) density. Network size relates to the average pathength among nodes in the network (Ebel et al., 2002). More firmsn a network translate into more steps and more time needed toomplete the same task, whereby creating a higher likelihood ofhe supply being interrupted en route and higher collective burdenorn at the system level (Frenken, 2000). Likewise, between the
wo networks of identical size, more links imply a higher proba-ility that the functioning of the individual nodes in the network
s likely to be impeded by others, leading to a greater coordinationoad on the whole network (Choi and Krause, 2006). For instance,f an OEM has two top-tier suppliers, the firm would necessarily

more time to channel information and ahigher likelihood of information distortionacross a supply network)

ing communications fromm to the final assembler

incur a greater amount of coordination load, compared to a situ-ation where there is only one top-tier firm. Therefore, a complexsupply networks would be associated with large network size, largecore size, high network density, high core density, high peripherydensity, high CTP density, and high PTC density. Note that in thecontractual relation supply networks, the PTC and CTP densitiesare identical, since every link in the network is non-directional andthe adjacency matrix representing this network is symmetric.

4. Research methodology

4.1. Data source

Choi and Hong (2002) (hereafter, denoted as C&H) reportedthree supply networks from raw materials suppliers to a finalassembler involved in the production of an automobile center con-sole assembly. The three product lines represented were HondaAccord, Acura CL/TL, and DaimlerChrysler (DCX) Grand Cherokee.Using an inductive case study approach, the authors derived propo-sitions 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 networkas 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 differenttypes of network information—one pertaining to materials flowand another based on contractual relationships. These two differenttypes of network data yield a total of six supply networks—threebased on directional materials flow and three based on non-directional 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 rep-resenting 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 bymaterials flow or contractual relationship, and would be “0” oth-erwise. Supply networks may yield adjacency matrices that are

200 Y. Kim et al. / Journal of Operations Management 29 (2011) 194–211

ow ne

sdrcmU2

sSz1dsaUoF

5

5

wTsti

sT

Fig. 1. Materials fl

ymmetric (i.e., non-directional) or asymmetric (i.e., directional),epending on the nature of the linkages. As noted earlier, a mate-ials flow network is directional and thus asymmetric, while aontractual relationship network is non-directional and thus sym-etric. Once generated, the adjacency matrices are imported intoCINET 6 and are used as inputs for network analysis (Borgatti et al.,002).

UCINET is a comprehensive software package for the analysis ofocial network data. It has been one of the most widely acceptedNA tools for conducting the structural analysis of interorgani-ational networks (e.g., Gulati, 1995, 1999; Human and Provan,997; Rowley et al., 2005; Ahuja et al., 2009). The program containsozens of network analytic methods such as centrality measures,ubgroup identification, role analysis, elementary graph analysis,nd permutation-based statistical analysis. While performing SNA,CINET can create network visualizations. A visualization of eachf the six supply networks, also known as a sociogram, is shown inigs. 1–6.

. Results

.1. Node-level results

Tables 4 and 5 list key firms in the two types of supply net-orks. We identify key firms based on their centrality values.

ables 4 and 5 build on Table 1. The supply network constructshown on the top row come from Table 1, and centrality computa-
ions are conducted on the corresponding centrality metrics shownn Table 1.
As indicated below each table, there is a cut-off point for eachupply network construct (e.g., 10 for in-degree, 6 for out-degree).he cut-off point is determined based on one rule: when there is

twork for Accord.

a noticeable drop-off in the score, the previous score constitutesthe threshold. In all cases except one, there are multiple key firms.The exception is out-degree centrality for the materials flow typeof DCX’s supply network. Every node in the network, except forthe OEM, has only one customer, showing the same value on out-degree centrality; consequently, there was no threshold value. InTables 4 and 5, the number shown in parenthesis next to a firmname represents the centrality score.

5.2. Network-level results

Tables 6–8 show SNA results at the network level. Tables 6 and 7focus on centralization metrics, respectively, for directional mate-rials flow and non-directional contractual relationships. Table 8summarizes all complexity metrics for both types of networks.

In Table 6, various network-level indicators are shown acrossthree different supply networks. Beginning with network sizeand density, individual node-level centrality scores are averagedfor each supply network. Then, the three network centralizationscores are listed. Up to this point, all values reflect network-levelattributes. Below the network-level values, Table 6 lists values atthe 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 discus-sion on core and peripheral groups). It then moves on to listingother group-level measures.

Addressing contractual relationship supply networks, Table 7 isconstructed much the same way. Since this supply network type
is non-directional, network measures shown on the left-side col-umn are slightly different from those of Table 6, as discussed underTable 1. Also note that most of the network-level metrics are in nor-malized form, which allows us to compare them across the threedifferent supply networks.


Table 3Summary of case data from Choi and Hong (2002).

Networkmeasures

Product type

Honda Accord Acura CL/TL DaimlerChrysler (DCX) Grand Cherokee

Centralization Two firms, CVT and JFC, are top-tier suppliersto Honda

One top-tier supplier, Intek, a completeintegrator of this supply network

Textron as the sole top-tier supplier thatintegrates parts and subassemblies

Several second- or third-tier suppliers (e.g.,Emhart, Garden State, and Miliken) directlyselected by HondaSome third-tier suppliers directly selected byCVT, based on Honda’s core supplier listHonda’s penchant for centralized control whenit comes to the product design and supplierselection

Honda engaging in directed sourcing at thesecond, third, and even fourth tiersIntek likewise engages in directed sourcing byselecting its own suppliers and even theirsupplier’s suppliers, based on Honda’s coresupplier listDirected sourcing generally for high-priced orstrategic itemsHonda’s centralized control of the productdesign activities

Textron-Farmington and Leon Plastics appearas two key second-tier suppliersTextron assumes the leading role in designingconsoleDirected sourcing occurs only on a limitedbasis

Complexity All together, 50 network entities: 2 first-tier,21 second-tier, 18 third-tier, 7 fourth-tier, and2 fifth-tier suppliersMajority of the suppliers at the second-tierlevelFour different nature of businesses in thenetwork—manufacturing companies, rawmaterials suppliers (e.g., GE Plastics),distribution centers (e.g., Iwata Bolt), andtrading houses (e.g., Honda Trading)Reciprocal relationship between CVT and JFC,two top-tier suppliers, contributing to eitherreduction or increase of network complexitydepending on the relational nature

76 entities in the network: 1 first-tier, 20second-tier, 28 third-tier, 17 fourth-tier, 9fifth-tier, and 1 sixth-tier suppliersThe coupling between Honda and Intek basedon their shared history may reduce the level ofcomplexityThe decoupling between Intek and JFC, asecond-tier supplier of the criticalsubassembly, may further the complexityHonda’s effort to centralize second-tiersuppliers may increase complexity of thenetwork as a whole

41 entities: 2 first-tier, 10 second-tier, 22third-tier, and 7 fourth-tier suppliersAt the top-tier level, Textron is engaged inassembly work and also acts as a conduit for apart from Leon as it ships the front console matdirectly to the DCX plant with Textron’s labelNo reciprocal relations among suppliersAs per DCX’s recommendation, Textron hasconsolidated the second-tier suppliers, leadingto reduced number of second-tier suppliersand subsequently reduced complexity

Fig. 2. Materials flow n
etwork for Acura.


Fig. 3. Materials flow network for Grand Cherokee.

Fig. 4. Contractual relationship network for Accord.


Fig. 5. Contractual relationsh

Fig. 6. Contractual relationship ne

ip network for Acura.

twork for Grand Cherokee.


Table 4List of key firms based on materials flow network.

Supply loada Demand loadb Operational criticalityc

Accord CVT (59d), JFC (15), HFI (11) CVT (15), C&C (7.4), JFC (7.4), GE (7.4), Yamamoru (7.4),Industry Products (7.4)

CVT (13), Emhart (2), Yamamoru (1.7), Fitzerald (1.7),JFC (1.3)

Acura Intek (58), Arkay (21), Select Ind. (12) Iwata Bolt (9.1), Tobutsu (6.1), Arkay (6.1), Twist (6.1),Milliken (6.1), Garden State (6.1), Select Ind. (6.1)

Intek (3), Arkay (1.7)

DCX Textron (65), Leon Plastics (31) None Textron (3.8), Leon Plastics (2.5)

a Firms with in-degree > 10.b Firms with out-degree > 6.c Firms with betweenness > 1.0.d Centrality score.

Table 5List of key firms based on contractual relationship network.

Influential scopea Informational independenceb Relational mediationc

Accord CVT (52), Honda (30),Yamamoru (15)

CVT (57), Honda (53), Yamamoru (40) CVT (79), Honda (64), Emhart (21) Yamamoru(15), Fitzerald (14)

Acura Intek (45), Honda (36), Arkay(18), Select Ind. (15)

Intek (62), Honda (56), Arkay (44), Select Ind.(43), Tobutsu (41), HFI (40)

Intek (77), Honda (63), Select Ind. (14), IwataBolt (12), Arkay (10)

DCX Textron (62), Leon Plastics (35) Textron (72), Leon Plastic

a Firms with degree > 15.b Firms with closeness > 40.c Firms with betweenness > 10.

Table 6Network-level results for materials flowa supply networks.

Network measures Product type

Accord Acura DCX

Network size (firms) 28 34 27Network density 0.046 0.037 0.037Average in-degree 4.630 3.654 3.704Average out-degree 4.630 3.654 3.704Average betweenness 0.809 0.231 0.234Centralization (in-degree) 0.567 0.556 0.641Centralization (out-degree) 0.106 0.056 0.001Centralization (betweenness) 0.128 0.029 0.038

Core group size (firms) 19 23 4Core density 0.067 0.059 0.250Core to periphery (CTP) density 0.006 0.000 0.000

lrsbg

TN

Periphery to core (PTC) density 0.064 0.043 0.250Periphery density 0.000 0.000 0.000

a Represented by asymmetric matrix.

Table 8 re-organizes some information from Tables 6 and 7. Itists values for the select indicators of network complexity—they
epresent the degree of interdependency among firms. Networkize is listed as the first indicator. Network density is then listed inoth materials flow and contractual relationships networks. Then,roup-level indicators are listed in both types of supply networks.
able 7etwork-level results for contractual relationshipa supply networks.

Network measures Product type

Accord Acura DCX

Network size (firms) 28 34 27Network density 0.074 0.066 0.074Average degree 7.407 6.595 7.407Average closeness 35.716 37.747 41.959Average betweenness 7.407 5.375 5.778Centralization (degree) 0.479 0.413 0.585Centralization (closeness) 0.459 0.513 0.641Centralization (betweenness) 0.748 0.738 0.854

Core group size (firms) 17 6 3Core density 0.125 0.467 0.667CTP or PTC density 0.048 0.179 0.333Periphery density 0.036 0.000 0.000

a Represented by symmetric matrix.

s (58) Daimler (46) Textron (88), Leon Plastics (53) Daimler (15)

6. Interpretation of results

In this section, we recapitulate the SNA results shown inTables 4–8 with reference to the supply network constructs devel-oped in this study (see Tables 1 and 2). We provide networkdynamics implications of the node-level results first and then thoseof the network-level results. A summary of the SNA results at thenode- 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 networksTable 4 compares groups of firms across supply load, demand

load, and operational criticality (see Table 1 for definitions). CVT, afirst-tier supplier in Accord supply network, appears highly cen-tral, showing the highest scores on all three columns. In otherwords, CVT assumes the most operational burden on both thesupply side and demand side. This firm is tasked with integrat-ing multiple parts into a product, which also means the firm canmake the most of its resources by pooling customer demands andthe related risks. CVT is also the pivotal player in the movementof materials. Without this firm, the entire supply chain would bedisrupted. In contrast, we observe that another top-tier supplierof Accord, JFC, is not as central. Its centrality scores are markedlylower 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 suppli-ers 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 underboth supply load and operational criticality. The bulk of networkresources flow into and through this firm. However, unlike CVT inAccord network, Intek does not appear central under demand load.This is because the firm primarily receives materials (see Fig. 2). Infact, Iwata Bolt, a second-tier supplier for Acura, comes first underdemand load. This simply means that this firm delivers to a rela-tively large number of buying firms, which implies that this supplier
has leverage in allocating its internal resources across multiple cus-tomers. Another noteworthy finding is that Arkay, a second-tiersupplier, is the only firm that ranks high on all the three centralitymetrics. Without conducting SNA, Arkay’s central role in the Acurasupply network may very well be overlooked.


Table 8Key indicators for network complexity.

Network size(firms)

Materials flow network Contractual relationship network

Networkdensity

Core size Coredensity

CTP density PTC density Networkdensity

Core size Coredensity

Peripherydensity

PTC density

snoeTtait

6

AtaaTHTmJimC

tmbIsHk

TN

Accord 28 0.046 19 0.067 0.006Acura 34 0.037 23 0.059 0.000DCX 27 0.037 4 0.250 0.000

There are a comparatively less number of central firms in DCX’supply network. The implication is that the structure of the DCXetwork is simpler (see Fig. 3) than those of Honda and Acura. Forne, there are no firms listed under demand load. This is becausevery supplier in this network has only one customer, includingextron and Leon, a top-tier and a key second-tier supplier, respec-ively. These two suppliers appear under both supply load and oper-tional criticality. Both firms engage in value-adding activities byntegrating parts and facilitating their flows. The supply streams inhis supply network take place primarily through Textron or Leon.

.1.2. Key firms in contractual relationship supply networksIn Table 5, CVT is again prominent on all centrality metrics in

ccord supply network. This firm appears as most influential onhe operation of the contractual relationship supply network, justs it does in the materials flow network. Nonetheless, there arefew notable differences. First, Honda does not appear at all in

able 4, but in this network based on contractual relationships,onda emerges quite visibly (second to CVT) on all three columns.his is because Honda maintains a contractual relationship withany of its second- and third-tier suppliers (see Fig. 4). Second,

FC, a top-tier supplier who appears in all three centrality metricsn Table 4, is gone in Table 5. In other words, when it comes to

anaging contracts, Honda emerges as central and JFC disappears.learly, JFC is more isolated in the contractual relationship network.

For Acura supply network, Intek appears yet again as most cen-ral, while Honda emerges as central also. Thus, Intek looks like

ost influential in the contractual relation network and none could
ypass Intek to connect with Honda. The network position allowsntek to take control of information and communication flows. Oneupplier for Acura that appears in Table 5 but did not in Table 4 isFI. HFI is a lone third-tier supplier that SNA picked up as being aey firm under Informational Independence. This is largely because
able 9ode-level overview.

Materials flow network

Accord CVT, a 1st-tier suppliers, is most central, and assumes the most operationalburden on both supply and demand sidesJFC, another 1st-tier supplier, is not as much central as CVTHonda appears not central in this networkHFI and C&C, two 2nd-tier suppliers, need to handle high degrees of supplyload and demand load, respectivelyTwo 2nd-tier suppliers (Emhart and Yamamoru), and one 3rd-tier (Fitzerald)are also central as a go-between along the materials flow

Acura Intek, the sole 1st-tier supplier, is most central under both supply load andoperational criticalityArkay, a 2nd-tier supplier, is central on every centrality metricIwata Bolt, a 2nd-tier supplier, is most central underdemand loadHonda is virtually out of sight in this network

DCX Textron, the sole 1st-tier supplier, and Leon, a 2nd-tier supplier, are mostcentral under both supply load and operational criticalityNo central firm under demand loadDaimler is rather central only under supply load

0.064 0.074 17 0.125 0.036 0.0480.043 0.066 6 0.467 0.000 0.1790.250 0.074 3 0.667 0.000 0.333

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 5for DCX shows little change. There were two firms (Textron andLeon) in Table 4 and the same firms appear again in Table 5. Theonly 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 sup-pliers, Irwin and E.R. Wagner (see Fig. 6). Daimler thus has leverageover 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 net-works based on Tables 6 and 7.

6.2.1. Characteristics of the materials flow supply networksIn Table 6, Accord’s supply network shows a comparatively

high density compared to the other two networks of Acura andDCX. Accord’s supply network also features relatively high aver-age scores on the key centrality metrics. Particularly, on averagebetweenness, Accord’s lead is substantial. It implies that firmsin this supply network are more engaged in both delivering andreceiving materials than firms in other supply networks. It alsomeans that there are more steps required to move the materialsalong. From an operational standpoint, it might indicate that thisnetwork provides less efficiency (e.g., longer lead time, more parts
used for the same function) as it imposes more managerial atten-tion on the firms in a central position. Looking at centralizationscores for Accord, indegree score stands out, suggesting the inflowof materials is concentrated in a small group of firms in the sup-ply network. We also note a rather large discrepancy in the scores
Contractual relationship network

CVT is most central under all three measures—operational flexibility,managerial independence, and relational controlHonda emerges as the close second to CVT on all centrality metricsJFC is extinct and becomes isolated in this networkYamamoru, a 2nd-tier suppliers, emerges as central under managerialindependenceEmhart, a 2nd-tier supplier, is central under relational control

Intek is again most central on all threecentralityHonda is the close second to Intek on every centrality metricHFI, a 3rd-tier supplier, emerges as key under managerial independence due toits ties with other key suppliersTwo 2nd-tier suppliers, Arkay and Select Industries, rank consistently high onall three centrality metrics

Little change from materials flownetworkTextron and Leon are two most central on every centrality metricsDaimler comes next but by a large margin on all three metricsNo other firms, than the three firms, appear as central in this network


Table 10Network-level overview.

Materials flow network Contractual relationship network

Accord Comparatively high overall density Relatively high overall densityHighest average score on all three centrality metrics Highest average betweenness but lowest average on closenessRelatively high centralization across all three types, and substantial lead onaverage betweenness score

Largest core group with lowdensity

Much higher indegree centralization than the other two types Relatively high periphery densityNo connectivity among peripheral firms Comparatively low PTC densityLittle reciprocity between the core and peripheral firms (much higher PTCdensity than CTP density)

There are more interactions overall among moremembers

Peripheral firms engage solely in supplying to core firms Rather complex at the network levelRelatively less centralized

Acura Comparatively large overall membership but with low density Largest overall membership but with lowest overall densityRelatively low average scores on all the three centrality metrics Lowest average betweenness scoreComparatively low centralization indices More tightly coupled core groupVery large core group with very low density No interactions among peripheral firmsVirtually no materials flows among peripheral firms Network activities mostly concentrated around the core groupNo reciprocity between the core and the periphery firms Relatively high PTC densityNetwork activities concentrated around the core group Comparatively more centralized around the smaller core groupComparatively less complex at the network level Comparatively less complex at the network level

DCX Smallest membership with relatively low density Highest average scores on closeness centralityComparatively high indegree centralization, but quite low outdegreecentralization

Highest centralizationindices

Smallest and tightly knit core group Smallest core group with very high densityNo materials flows among peripheral firms No interactions among peripheral firms

obnafi

banaFagSpowo

ooAiosdmctSntsc

6

T

Largest discrepancy between PTC and CPT density among three SNsPeripheral firms engage exclusively in supplying to the core firmsRelatively more centralized and least complex at the network level

f CTP density and PTC density, which signifies little reciprocityetween the core and peripheral firms. Further, the far-off firms doot interact at all, as demonstrated in the periphery density of 0,nd this is true for all product types. In other words, the peripheralrms engage solely in supplying to the core firms.

Acura’s supply network has comparatively large membershiput low overall density. The three average centrality scores are rel-tively low. Acura’s supply network, compared to Accord’s, has lessumber of links and the overall steps required to get things donere not as many, which may indicate higher operational efficiency.urther, based on centralization scores, Acura’s supply networkppears as less centralized than Accord’s. At the local level, the coreroup has very large membership but with relatively low density.ince there are more firms in the core group, it suggests that theower in the network is more spread out; the more flat structure ofperational authority again may be an indication that this networkorks more efficiently (e.g., less time expended to make a decision

n issues at a local level).DCX’s supply network has the smallest membership, and the

verall density is also relatively low. The centralization index basedn in-degree is comparatively high, as is the case with Accord andcura. However, notable difference occurs with out-degree central-

zation. It is quite low, indicating that most of the materials flowut to few common dominant firms, and this observation is alsoupported by the small size of the core group. There is also a hugeiscrepancy between CTP density and PTC density, which simplyeans that the majority of materials flow links in the network is

oncentrated on a small number of firms. As expected, these firms inhe core group are tightly knit, as evidenced by a high core density.uch simple structure can provide high operational efficiency at theetwork-level (e.g., shorter lead time from upstream suppliers tohe final assembler); however, if multiple issues were to happenimultaneously they could overwhelm the few central players and
ould require much more time for resolution.
.2.2. Characteristics of contractual relationship supply networksThe density for Accord is much higher in Table 7 than it was in

able 6. This is because contracts can jump across several tiers. As

By far higher PTC densityMajority of network activities centers around the core groupPeripheral firms engage only in supplying to the core firmsMost centralized and least complex at the network level

expected, the same thing happens for Acura and DCX as well. Interms of centrality metrics, Accord’s supply network shows rela-tively low average closeness but high average betweenness scores.Such a structure may be less responsive or more susceptible tosupply disruptions. It would possibly take more time channelinginformation and there is a higher chance that information becomesdistorted on its way along the chains as more firms get involved intransferring it. Therefore, such structure is likely to be less robustor less effective when it comes to coping with supply disruptions.By the same token, the structure would provide greater complexityat the network level for Accord, as also evidenced by Accord’s rel-atively large core group size (see Table 7). Further, it has relativelyhigh periphery density, which further indicates that the network iscomplex because there are more interactions going on even amongperipheral members. Still, more contacts among members at thelocal level might facilitate identifying, if any, supply issues occur-ring locally.

Acura’s supply network shows relatively low overall density butwith large membership, which correspond with less number ofcontractual links overall. Regarding average betweenness, this sup-ply network shows the lowest score, indicating that this networkneeds a smaller number of channels to get things done. Compara-tively, therefore, this supply network appears as more efficient, forinstance, in managing such issues as supply disruptions becausecommunications at the network level can be comparatively fasterand more organized than those of Accord’s, which is also supportedby Acura’s comparatively more tightly knit core group and zeroperiphery density.

Interestingly, DCX’s supply network shows the highest averagecloseness score. In other words, the firms in the DCX’s networkare more readily reachable from each other, indicating that infor-mation can travel faster across the network. To put it differently,the network structure is more conducive to the centralized con-
trol by dominant actors. As might be expected, this supply networkfeatures the highest centralization indices among all three supplynetworks. There is additional evidence for DCX’s high centralizationat the network level—the majority of the activities in the supplynetwork seem to center around a very small group of firms (i.e.,

ons M

tenntotsas

7

7

7

drdwcwasAodIscr

npcTaDc(hSnnadCibatpSccts

acwptr


hree core firms) that are highly interwoven together (i.e., the high-st core density of 0.667). Further, the firms in the periphery, witho interactions among them, focus on catering to the core firms’eeds, evidenced by high PTC density. Because network informa-ion tends to spread relatively fast and converge at a small groupf dominant actors, the network as a whole would be compara-ively more effective and robust when it comes to dealing withupply disruptions. Particularly, active interactions between corend periphery firms would further enhance such capability of theupply network.

. Discussion

.1. Comparisons between SNA results and C&H study

.1.1. Overlapping and divergent resultsOne of the main findings of C&H was the three OEMs’ varying

egrees of centralized control over their supply networks. The SNAesults confirm this. In particular, the final assemblers’ practice ofirected sourcing is captured in the contractual relationship net-ork structure. For instance, the high values in Honda’s various

entralities and overall density in the contractual relationship net-ork, compared to those in the materials flow network, is clearly

ttributable to the added links that represent Honda’s directedourcing practice involving its second- and third-tier suppliers.nother finding shared by both studies is the relational saliencef those tertiary-level suppliers in the network that are sourcedirectly by OEMs. All of such suppliers (e.g., Emhart for Accord and

wata Bolt for Acura) emerge as visible in the contractual relation-hip network, through their exhibiting high scores on the variousentrality metrics or becoming a member of the core group in theirespective supply networks.

Divergent results between the two studies relate largely toetwork-level properties such as network centralization and com-lexity. First, C&H describe Honda’s two supply networks as moreentralized than DCX’s. However, SNA suggests the opposite (seeables 6 and 7). In evaluating network centralization, C&H actu-lly take the perspective of the final assemblers (i.e., Honda andCX). They present the argument that Honda is more centralizedompared to DCX because it has more direct ties with its suppliersi.e., top-tier as well as second- and third-tier suppliers)—Hondaas more centralized control of its supply networks. However,NA, in contrast, looks at how central all firms are in the supplyetwork, not just the final assembler. SNA evaluates the relativeode-level centrality scores of all the network members to arrivet the indicators of network centralization. The two studies alsoiverge when considering which supply network is most complex.&H suggest that Acura’s network is most complex. This judgment

s based on the network-level physical attributes (e.g., total num-er of entities, average geographical distance between companies)nd qualitative evidence regarding the lack of shared history andhe perceived level of decoupling among members. Contrarily, SNAoints to Accord’s network as being most complex. This is becauseNA focuses on how individual firms and their relationships areonnected to one another at the network level. For instance, SNAonsiders various aspects of interdependence among members inhe network, such as network density, core density, periphery den-ity, and PTC density.

The two studies, as such, draw different conclusions on somespects of supply network properties. Nonetheless, we want to
aution that this does not mean one is more accurate; rather, weant to say that they just focus on different aspects of the samehenomenon—the case approach focuses on contextual informa-ion, whereas SNA operates on numerical breakdown of data onelative positions of members.
anagement 29 (2011) 194–211 207

7.1.2. What C&H offer but SNA does notC&H’s qualitative approach offers a contextually rich picture of

network dynamics. For instance, they make statements about thenetwork structure by drawing on such observations as Honda’sstrong penchant toward centralized policy with respect to sup-plier selection and product design and DCX’s practice of delegatingauthority to the first-tier supplier as to who will be second-tiersuppliers and how to design the console. Further, the case methodcan provide more detailed accounts of how the supply networksoperate and behave. For instance, in the Honda’s supply networks,the second-tier suppliers selected directly by Honda tend to beless cooperative with the top-tier supplier, which contributes tofurthering complexity at the network level; in the DCX’s net-work, Daimler commissions the top-tier supplier to consolidatethe second-tier suppliers to reduce operational complexity. Suchfindings are context-specific and would be very difficult for SNA tocapture.

Also, C&H offer some propositions representing the overarch-ing principles of the supply networks, derived from the qualitativedata. For instance, the study observes, “Formalized rules, norms,and policies lead to the varying degrees of centralization in thesupply network . . .” (p. 488); “The cost consideration representsthe most salient force that shapes the emergence of the supply-network structure” (p.488); and “A centralized approach to supplynetwork involves a common list of core suppliers and the designactivities are tightly controlled by the final assembler” (p. 489).Only from case-based qualitative studies could such propositionsbe compiled. SNA would be unable to capture such contextuallyrich information.

7.1.3. What SNA offers but C&H do notSimply, SNA offers many quantitative metrics that qualitative

approaches cannot. By analyzing the structural characteristics ofsupply networks, SNA brings us new intriguing results that wouldlikely be overlooked by qualitative methods. First, by producingvarious network metrics, from node- to group- and to network-level, SNA facilitated a comprehensive analysis of supply networks.For instance, SNA evaluated differing roles of the individual nodesand their relative importance with respect to others in the samenetwork (see Tables 4 and 5).

Second, SNA allowed for a comparative analysis of two differentnetwork structures—materials flow and contractual relationship.Between the two different network structures, we have observedsome divergent results even on the same network metrics (e.g.,density, betweenness centrality). Those discrepancies, as notedearlier, come from the fact that the two structures are constructedbased on different types of relational connection. Thus, it is notproper to say that one type of link is a more accurate depiction of agiven network than the other; but rather the two different types ofnetwork information should be considered jointly to fully under-stand a supply network. Further, SNA enabled a group-level analysisby partitioning each supply network into two structurally distinctclusters—core and periphery sub-groups. The core-periphery anal-ysis in fact facilitated assessing network-level properties acrossdifferent supply networks (i.e., network centralization and networkcomplexity).

7.2. Academic contributions

Our goal in this paper has been to introduce SNA as a meansto analyze the structure of supply networks and draw theoretical
conclusions from such analysis. Our framework translates key SNAmetrics into the context of supply networks, and discusses howroles of individual supply network members vary depending ontheir relative structural position in the network. Subsequently, wesuggested a guideline as to how to identify central nodes and eval-

2 ons M

usntoViTshma

sSdtiefnlctopaq

7

ttanswTtsakiwcstita

csklmswclkda

b


ate them differently. Central firms require possessing a particularet of capabilities corresponding to the roles they assume in theetwork (see Table 1). For instance, firms with high in-degree cen-rality should focus on developing a capability in system integrationr product architectural innovation (Parker and Anderson, 2002;iolino and Caldwell, 1998); firms with high betweenness central-

ty may be in a better position to engage in supply risk management.hus, it would be prudent for a buying company (e.g., OEMs), whenelecting or developing a supplier, to consider these issues. Weope that the theoretical framework of this study would be instru-ental in facilitating future supply network research adopting SNA

pproach.The paper’s methodological contribution is two-fold. First, this

tudy demonstrates the value of SNA in studying supply networks.NA considers all member firms in a given supply network toetermine which firms are most important, in what aspect, tohe operation of the whole network. Capitalizing on computat-ng power, SNA can generate various analytic outputs reflectingither individual- or group-level behavioral dynamics, which inact facilitate gaining a more comprehensive and systematic view ofetwork dynamics. Second, applying the widely accepted network-

evel analytical concepts (i.e., network density, centralization, andore-periphery), SNA can complement qualitative methods in cap-uring the structural intricacy of the whole network in a morebjective way. As has been demonstrated, SNA has considerableotential for enhancing our studies of supply networks (Borgattind Li, 2009; Carter et al., 2007) and can effectively complementualitative methods.

.3. Managerial contributions

Based on C&H’s data, our study brings to the fore the salience ofwo types of supply networks—materials flow and contractual rela-ionship. We propose that managers consider these two types forny given supply network, as we have demonstrated how the twoetworks organize and behave differently. For instance, in Acura’supply networks, the size of the core group becomes much largerhen based on materials flow than the contractual relationship (see

ables 6 and 7 for comparison). Also, managers should note thathere can be different sets of key firms between the two types ofupply network (see Tables 4 and 5). One firm that does not appears central in one type (e.g., HFI in the Acura network) may be aey player in another. Depending on which type of link to focus on,ndividual suppliers’ position of importance and the strategic roles

ill vary. For instance, the key firms in a materials flow networkan have a considerable effect on the operational quality of overallupply network, affecting lead time, product quality, OEM’s inven-ory level, or stockout costs (Bourland et al., 1996). Key suppliersn a contractual relation network could facilitate the timely inden-ification or resolution of those system-level operational problemsnd other supply disruption risks (Lee, 2002).

Further, it may be prudent for a manufacturing firm to identifyentral second- or third-tier suppliers using SNA. Some of theseuppliers become a key player by being linked to more visible otherey firms in the supply network. In other words, some tertiary-evel suppliers emerge as important because they are vital to other

ore prominent suppliers in supply networks. We anticipate theseecond- and third-tier suppliers that previously went unnoticedill play a more significant role in future. As the issue of supply

hain scalability takes the center stage for safety and sustainability,arge final assemblers are moving toward identifying and managing
ey tertiary-level suppliers. Collecting complete supply networkata and applying SNA, as we have done in this paper, may serves a useful approach.
In general, having a pictorial rendition of a supply network wille useful to managers. SNA can help generate network sociograms

anagement 29 (2011) 194–211

(see Figs. 1–6). As a visual embodiment of relationship patterns insupply networks, these sociograms can be instrumental in attaininga realistic picture of networking patterns and the dynamics. Just asall graphs, network drawings can help save search efforts, facilitaterecognition, and provide interesting new perspectives and insightsinto supply networks. Also, SNA provides a methodological framefor collecting and organizing data, which will be useful for plan-ning and monitoring changes in the operation of supply networks.The position of a node in the network affects the opportunities andconstraints 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 empir-ically investigating supply networks using SNA concepts. Weacknowledge that our study is limited in ways that suggest oppor-tunities for future research. First, our analysis is confined to aspecific automobile module (i.e., center console assembly). Any onesupplier in the supply network might be involved in several over-lapping supply networks across different product lines. A supplier’srole based on one supply network will look quite different from thatderived by considering the multiple supply networks together it isa member of. Therefore, the central roles a supplier plays in ouranalysis should be qualified to the single product line. It would notbe reasonable to consider the results of our analysis as a generalstatement 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 networksstudied here were also mapped based on information obtainedfrom the final assemblers. Therefore, any possible effect eachsupplier’s extended network can have on the firm’s strategicimportance to the OEM could not be captured in our analysis. Forinstance, one second tier supplier to Honda may have a tie to otherOEMs. If such extended ties were also counted, certain centralitymetrics (e.g., betweenness) for the supplier might have shown dif-ferent scores from those based on the egocentric network, wherebyplacing the supplier in a different strategic position with respect toHonda. Such egocentric network approach, albeit considered a reli-able substitute for complete (sociocentric) network data (Marsden,2002), may not be enough to provide a full understanding or poten-tial 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 thevariances in strength. All the links considered in our analysis weretreated as having the same weight, while the link an OEM has withthe first-tier suppliers should involve more intensive informationexchanges (i.e., kanban system) or a greater amount of materials(i.e., larger contract size) than those with the second-tier firms, forinstance. Also, we viewed supply networks based on the materi-als flow and contract connections. However, certainly there aremany other relational connection types that can be consideredin supply networks, such as ownership, technology dependence,intellectual property, and risk sharing. Network ties could be rep-resented by the number of joint programs or of shared patents, levelof trust, or perceived transactional risks. Future studies thereforecan incorporate the relative strength of supply ties using SNA as themethod can effectively illustrate networks with “weighted” links(Borgatti and Li, 2009; Battini et al., 2007). Exchange ties involv-ing a multi-level interface will have differential impact compared
to other comparable supply ties based only on a single type oftransaction.
We note that most supply networks are considered a scale-free network, whose degree distribution closely follows a powerlaw (Albert and Barabási, 2002; Pathak et al., 2007). That is, most

ons M

n(tactiacnIcmGhclwd

roncsowaoutun

R

A

A

A

A

B

B

B

B

B

B

B

B

B

B

B

B


odes have very few links and only a small number of nodese.g., core firms) have many connections. Future studies may applyhe scale-free network metrics to studying supply networks, suchs clustering coefficient and characteristic path length. Clusteringoefficient measures the degree to which nodes in a network tendo cluster together around a given node (Barabási et al., 2002), andt can inform us of how suppliers behave with respect to the finalssembler at both the local and the global level. For instance, itan tell us how suppliers would come together for better coordi-ation, based on some governance mechanism involving an OEM.

ndicating the system-level “closeness,” characteristic path lengthan assist in evaluating whether a given supply network is opti-ally designed (Braha and Bar-Yam, 2004; Lovejoy and Loch, 2003).iven a supply network, it can be of considerable interest to knowow the path length compares to the “best” or “worst” possibleonfiguration for networks with the same number of nodes andines. This can provide implications for how effectively the net-

ork is designed and how robust it can be to possible supplyisruptions.

Finally, SNA could be applied to advancing existing theoriesegarding the structure or topology of supply networks. A rangef SNA metrics can serve as a useful means in this effort. Suchetwork variables as density and various centralities could be appli-able to characterizing typological archetypes of supply networktructures, eventually leading to the development of a portfoliof contingent approaches to supply management. In conclusion,e hope that this paper can serve as a call to other operations

nd supply management researchers regarding the importancef framing supply chains as networks and continuing to developseful supply network indices. We hope to see more researchersaking advantage of the usefulness of SNA for untangling andnderstanding the complex phenomena embedded in supplyetworks.

eferences

brahamson, E., Rosenkopf, L., 1997. Social network effects on the extent of innova-tion diffusion: a computer simulation. Organization Science 8 (3), 289–309.

huja, G., Polidoro Jr., F., Mitchell, W., 2009. Structural homophily or socialasymmetry? The formation of alliances by poorly embedded firms. StrategicManagement Journal 30 (9), 941–958.

lbert, R., Barabási, A.L., 2002. Statistical mechanics of complex networks. Reviewsof Modern Physics 74, 47–97.

utry, C.W., Griffis, S.E., 2008. Supply chain capital: the impact of structural and rela-tional linkages on firm execution and innovation. Journal of Business Logistics29 (1), 157–174.

arabási, A.L., Jeong, H., Néda, Z., Ravasz, E., Schubert, A., Vicsek, T., 2002. Evolution ofthe social network of scientific collaborations. Physica A: Statistical Mechanicsand its Applications 311 (3–4), 590–614.

attini, D., Persona, A., Allesina, S., 2007. Towards a use of network analysis:quantifying the complexity of Supply Chain Networks. International Journal ofElectronic Customer Relationship Management 1 (1), 75–90.

avelas, A., 1950. Communication patterns in task-oriented groups. Journal of theAcoustical Society of America 22, 271–282.

eauchamp, M.A., 1965. An improved index of centrality. Behavioral Science 10,161–163.

erry, D., Towill, D.R., Wadsley, N., 1994. Supply chain management in the electronicsproduct industry. International Journal of Physical Distribution and LogisticsManagement 24 (10), 20–32.

onacich, P., 1972. Factoring and weighting approaches to status scores and cliqueidentification. Journal of Mathematical Sociology 2, 113–120.

orgatti, S.P., Everett, M.G., 2000. Models of core/periphery structures. Social net-works 21 (4), 375–395.

orgatti, S.P., Everett, M.G., 2006. A graph-theoretic perspective on centrality. SocialNetworks 28 (4), 466–484.

orgatti, S., Everett, M., Freeman, L., 2002. UCINET 6 for Windows: Software for SocialNetwork Analysis. Analytic Technologies, Inc., Natick, MA.

orgatti, S.P., Li, X., 2009. On social network analysis in a supply chain context.Journal of Supply Chain Management 45 (2), 5–21.

ourland, K.E., Powell, S.G., Pyke, D.F., 1996. Exploiting timely demand informa-tion to reduce inventories. European Journal of Operational Research 92 (2),239–253.

raha, D., Bar-Yam, Y., 2004. Information flow structure in large-scale productdevelopment organizational networks. Journal of Information Technology 19,244–253.

anagement 29 (2011) 194–211 209

Burkhardt, M.E., Brass, D.J., 1990. Changing patterns or patterns of change: the effectsof a change in technology on social network structure and power. AdministrativeScience Quarterly 35, 104–127.

Burt, R.S., 1980. Innovation as a structural interest: rethinking the impact of networkposition 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 Uni-

versity 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 Math-

ematical 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 ofshared information. Management Science 46 (8), 1032–1048.

Cachon, G.P., Lariviere, M.A., 2005. Supply chain coordination with revenue-sharingcontracts: 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 logisticsresearch. Journal of Business Logistics 28 (1), 137–169.

Cassiman, B., Veugelers, R., 2002. Complementarity in the innovation strategy: Inter-nal R&D, external technology acquisition and cooperation. Discussion paperseries No. 3284. Centre for Economic Policy Research, Cambridge.

Chen, F., Dresner, Z., Ryan, J.K., Simchi-Levi, D., 2000. Quantifying the bullwhip effectin 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 complexadaptive systems: control versus emergence. Journal of Operations Management19 (3), 351–366.

Choi, T.Y., Hong, Y., 2002. Unveiling the structure of supply networks: case studiesin Honda, Acura, and DaimlerChrysler. Journal of Operations Management 20(5), 469–493.

Choi, T.Y., Liker, J.K., 1995. Bringing Japanese continuous improvement approachesto 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 fortransaction costs, risks, responsiveness, and innovation. Journal of OperationsManagement 24 (5), 637–652.

Choi, T.Y., Kim, Y., 2008. Structural embeddedness and supplier management: anetwork perspective. Journal of Supply Chain Management 44 (4), 5–13.

Choi, T.Y., Wu, Z., 2009. Triads in supply networks: theorizingbuyer–supplier–supplier relationships. Journal of Supply Chain Management45 (1), 8–25.

Chopra, S., Sodhi, M.S., 2004. Managing risk to avoid supply-chain breakdown. MITSloan 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 improvesupply chains. Sloan Management Review 40 (4), 71–82.

Cox, A., Sanderson, J., Watson, G., 2006. Supply chains and power regimes: towardan analytic framework for managing extended networks of buyer and supplierrelationships. Journal of Supply Chain Management 37 (2), 28–35.

Croxton, K.L., García-Dastugue, S.J., Lambert, D.M., Rogers, D.S., 2001. The supplychain management processes. The International Journal of Logistics Manage-ment 12 (2), 13–36.

DiMaggio, P., Louch, H., 1998. Socially embedded consumer transactions: for whatkinds of purchases do people most often use networks? American SociologicalReview 63, 619–637.

Dyer, J.H., Nobeoka, K., 2000. Creating and managing a high-performanceknowledge-sharing network: the Toyota case. Strategic Management Journal21 (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 inte-grative approach to three-dimensional concurrent engineering. InternationalJournal of Physical Distribution & Logistics Management 37 (4), 305–330.

Everett, M.G., 1985. Role similarity and complexity in social networks. Social Net-works 7, 353–359.

Everett, M.G., Borgatti, S.P., 1999. The centrality of groups and classes. Journal ofMathematical Sociology 23 (3), 181–201.

Ferguson, R.J., Paulin, M., Bergeron, J., 2005. Contractual governance, relational gov-ernance, and the performance of interfirm service exchanges: the influence ofboundary-spanner closeness. Journal of the Academy of Marketing Science 33(2), 217–234.

Ford, I.D., 1990. Understanding Business Markets: Interactions, Relationships andNetworks (Ed.). Academic Press, San Diego, CA.

Freeman, L.C., 1977. A set of measures of centrality based on betweenness. Sociom-etry 40, 35–41.

Freeman, L.C., 1979. Centrality in social networks: conceptual clarification. SocialNetworks 1, 215–239.

2 ons M

F

F

F

F

G

G

G

G

G

G

H

H

H

H

H

H

H

H

H

H

H

H

I

I

J

J

K

K

K

K

K

K

K

K

K


reeman, L.C., 1980. The gatekeeper, pair-dependency and structural centrality.Quality and Quantity 14 (4), 585–592.

reeman, L.C., 1983. Spheres, cubes, and boxes: graph dimensionality and networkstructure. Social Networks 5, 139–156.

renken, K., 2000. A complexity approach to innovation networks. The case of theaircraft industry (1909–1997). Research Policy 29 (2), 257–272.

rohlich, M.T., Westbrook, R., 2002. Demand chain management in manufactur-ing and services: web-based integration, drivers and performance. Journal ofOperations Management 20, 729–745.

lanzer, M., Glaser, R., 1959. Techniques for the study of group structure and behav-ior. I. Analysis of structure. Psychological Bulletin 56, 317–331.

reve, H.R., 2009. Bigger and safer: the diffusion of competitive advantage. StrategicManagement Journal 30 (1), 1–23.

rover, V., Malhotra, M.D., 2003. Transaction cost framework in operations andsupply chain management research: theory and measurement. Journal of Oper-ations Management 21 (4), 457–473.

ulati, R., 1995. Social structure and alliance formation patterns: a longitudinalanalysis. Administrative Science Quarterly 40, 619–652.

ulati, R., 1999. Network location and learning: the influence of network resourcesand firm capabilities on alliance formation. Strategic Management Journal 20(5), 397–420.

ulati, R., Nohria, N., Zaheer, A., 2000. Strategic networks. Strategic ManagementJournal 21 (special issue), 203–215.

åkansson, H. (Ed.), 1982. International Marketing and Purchasing of IndustrialGoods: An Interaction Approach. Wiley, Chichester.

åkansson, H. (Ed.), 1987. Industrial Technological Development: A NetworkApproach. Croom Helm, London, UK.

åkansson, H., Ford, D., 2002. How should companies interact in business networks?Journal of Business Research 55 (2), 133–139.

åkansson, H., Snehota, I., 1995. Developing Relationships in Business Networks.Routledge, London, England.

akimi, S.L., 1965. Optimum locations of switching centers and the absolute centersand medians of a graph. Operations Research 12, 450–459.

arland, C.M., Lamming, R.C., Cousins, P.D., 1999. Developing the concept of supplystrategy. International Journal of Operations & Production Management 19 (7),650–673.

arland, C.M., Lamming, R.C., Zheng, J., Johnsen, T.E., 2001. A taxonomy of supplynetworks. The Journal of Supply Chain Management 37 (4), 21–27.

enderson, R., Clark, K., 1990. Architectural Innovation: the reconfiguration of exist-ing product technologies and the failure of established firms. AdministrativeScience Quarterly 35, 9–30.

endricks, K.B., Singhal, V.R., 2003. The effect of supply chain glitches on shareholdervalue. Journal of Operations Management 21 (5), 501–522.

endricks, K.B., Singhal, V.R., 2005. An empirical analysis of the effect of supplychain disruptions on long-run stock price performance and equity risk of thefirm. Production and Operations Management 14 (1), 35–52.

uman, S., Provan, K., 1997. An emergent theory of structure and outcomes in small-firm strategic manufacturing networks. Academy of Management Journal 40 (2),369–403.

warng, H.B., Chong, C.S.P., Xie, N., Burgess, T.F., 2005. Modelling a complex supplychain: understanding the effect of simplified assumptions. International Journalof Production Research 43 (13), 2829–2872.

ansiti, M., 2000. How the incumbent can win: managing technologicaltransitions in the semiconductor industry. Management Science 46 (2),169–185.

barra, H., 1993. Network centrality, power, and innovation involvement: determi-nants of technical and administrative roles. Academy of Management Journal36, 471–501.

arillo, J.C., Stevenson, H.H., 1991. Co-operative strategies: the payoffs and the pit-falls. Long Range Planning 24 (1), 64–70.

ensen, M., 2003. The role of network resources in market entry: commercial banks’entry into investment banking, 1991–1997. Administrative Science Quarterly48 (3), 466–497.

auffman, S.A., 1993. The Origins of Order: Self-Organization and Selection in Evo-lution. Oxford University Press, New York.

etchen Jr., D.J., Hult, G.T.M., 2007. Bridging organization theory and supply chainmanagement: the case of best value supply chains. Journal of Operations Man-agement 25 (2), 573–580.

im, W.S., 2009. Effects of a trust mechanism on complex adaptive supply networks:an agent-based social simulation study. Journal of Artificial Societies and SocialSimulation 12 (3), 4.

im, T.Y., Oh, H., Swaminathan, A., 2006. Framing interorganizational networkchange: a network inertia perspective. Academy of Management Review 31,704–720.

ircherr, W., 1992. Kolmogorov complexity and random graphs. Information Pro-cessing Letters 41 (3).

leindorfer, P.R., Saad, G.H., 2005. Disruption risk management in supply chains.Production and Operations Management 14 (1), 53–68.

lovdahl, A.S., 1985. Social networks and the spread of infectious diseases: the AIDSexample. Social Science & Medicine 21 (11), 1203–1216.

oehly, 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 non-polyposis colorectal cancer genetic testing and family functioning. CancerEpidemiology, Biomarkers & Prevention 12, 304–313.

otabe, M., Martin, X., Domoto, H., 2002. Gaining from vertical partnerships: knowl-edge transfer, relationship duration and supplier performance improvement in

anagement 29 (2011) 194–211

the US and Japanese automotive industries. Strategic Management Journal 24(4), 293–316.

Krackhardt, D., 1990. Assessing the political landscape: structure, cognition, andpower in organizations. Administrative Science Quarterly 92, 142–154.

Kumar, R., Novak, J., Tomkins, A., 2006. Structure and evolution of online socialnetworks. 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 ofsupply networks. International Journal of Operations & Production Management20 (6), 675–691.

Lazzarini, S.G., Chaddad, F.R., Cook, M.L., 2001. Integrating supply chain and networkanalyses: 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. Califor-nia 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: Proceed-

ings 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 Sys-tems 36, 71–88.

Lovejoy, W.S., Loch, C.H., 2003. Minimal and maximal characteristic path lengths inconnected sociomatrices. Social Networks 25, 333–347.

Luce, R.D., Perry, A.D., 1949. A method of matrix analysis of group structure. Psy-chometrika 14, 95–116.

Marczyk, J., 2006. Measuring and tracking complexity. In: Proceedings of the SixthInternational Conference on Complex Systems , June 25–30. New England Com-plex 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”. Inter-national Journal of Operations & Production Management 24 (9/10), 1012–1036.

Mizruchi, S.M., 1994. Social network analysis: recent achievements and currentcontroversies. Acta Sociologica 37 (4), 329–343.

Mizruchi, S.M., Marquis, C., 2006. Egocentric, sociocentric, or dyadic? Identifyingthe appropriate level of analysis in the study of organizational networks. SocialNetworks 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 themanufacturer-supplier network: benefits of quasi-market strategy in theJapanese automobile industry. Kobe Economic and Business Review 41,69–99.

North, M.J., Macal, C.M., 2007. Managing Business Complexity: Discovering StrategicSolutions with Agent-based Modeling and Simulation. Oxford University Press,New York.

Park, S., Hartley, J.L., 2002. Exploring the effect of supplier management on per-formance in the Korean automotive supply chain. Journal of Supply ChainManagement 38 (2), 46–53.

Parker, G.G., Anderson, E.G., 2002. From buyer to integrator: the transformation ofthe supply-chain manager in the vertically disintegrating firm. Production andOperations Management 11 (1), 75–91.

Pathak, S.D., Day, J.M., Nair, A., Sawaya, W.J., Kristal, M.M., 2007. Complexity andadaptivity in supply networks: building supply network theory using a complexadaptive systems perspective. Decision Sciences 38 (4), 547–580.

Pollock, T.G., Porac, J.F., Wade, J.B., 2004. Constructing deal networks: brokers asnetwork “architects” in the US IPO market and other examples. The Academy ofManagement Review 29 (1), 50–72.

Powell, W.W., Koput, K.W., Smith-Doerr, L., 1996. Interorganizational collaborationand the locus of innovation: networks of learning in biotechnology. Administra-tive 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, NewYork, pp. 561–586.

Provan, K.G., 1983. The federation as an interorganizational network. Academy ofManagement Review 8 (1), 79–89.

Provan, K.G., Milward, H.B., 1995. A preliminary theory of network effectiveness: acomparative study of four community mental health systems. AdministrativeScience Quarterly 40 (1), 1–33.

Pudlák, P., Rödl, V., 1992. A combinatorial approach to complexity. Combinatorica14, 221–226.

Pudlák, P., Rödl, V., 1994. Some combinatorial-algebraic problems from complexity
theory. Discrete Mathematics 136, 253–279.
Pudlák, P., Rödl, V., Savicky, P., 1988. Graph complexity. Acta Informatica 25,515–535.

Robins, G., Alexander, M., 2004. Small worlds among interlocking directors: net-work structure and distance in bipartite graphs. Computational & MathematicalOrganization Theory 10, 69–94.

ons M

R

R

SS

SSS

S

S

S

T


owley, T.J., 1997. Moving beyond dyadic ties: a network theory of stakeholderinfluences. The Academy of Management Review 22 (4), 887–910.

owley, T.J., Baum, J.A.C., Greve, H.R., Rao, H., Shipilov, A.V., 2005. Time to breakup: social and instrumental antecedents of firm exits from exchange cliques.Academy of Management Journal 48 (3), 499–520.

abidussi, G., 1966. The centrality index of a graph. Psychometrika 31, 581–603.amaddar, S., Nargundkar, S., Daley, M., 2006. Inter-organizational information shar-

ing: the role of supply network configuration and partner goal congruence.European Journal of Operational Research 174 (2), 744–765.

cott, J.P., 1986. Capitalist Property and Financial Power. Brighton, Wheatsheaf.cott, J., 2000. Social Network Analysis: A Handbook. Sage Publications, London.haw, M.E., 1954. Group structure and the behavior of individuals in small groups.

Journal of Psychology 38, 139–149.pencer, 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.

tam, W., Elfring, T., 2008. Entrepreneurial orientation and new venture perfor-mance: the moderating role of intra- and extra-industry social capital. The
Academy of Management Journal 51 (1), 97–111.
urana, A., Kumara, S., Greaves, M., Raghavan, U.N., 2005. Supply chain network:a complex adaptive systems perspective. International Journal of ProductionResearch 43, 4235–4265.

ichy, N.M., Tushman, M.L., Fombrun, C., 1979. Social network analysis for organi-zations. Academy of Management Review 4 (4), 507–519.

anagement 29 (2011) 194–211 211

Uzzi, B., 1997. Social structure and competition in inter-firm networks: the paradoxof embeddedness. Administrative Science Quarterly 42, 35–67.

Valente, T.W., 1996. Social network thresholds in the diffusion of innovations. SocialNetworks 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 ongoingcomputer-supported management groups. Organization Science 6 (4), 394–422.

Zaheer, A., Bell, G.G., 2005. Benefiting from network position: firm capabili-
ties, structural holes, and performance. Strategic Management Journal 26 (9),809–825.
Zhu, Q., Sarkis, J., 2004. Relationships between operational practices and perfor-mance among early adopters of green supply chain management practices inChinese manufacturing enterprises. Journal of Operations Management 22 (3),265–289.

MANAGEMENT SCIENCEVol. 52, No. 11, November 2006, pp. 1737–1750issn 0025-1909 �eissn 1526-5501 �06 �5211 �1737

informs ®

doi 10.1287/mnsc.1060.0582©2006 INFORMS

A Typology of Plants in GlobalManufacturing NetworksAnn 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 MeyerJudge Business School, Cambridge University, Trumpington Street, Cambridge CB2 1AG, United Kingdom,

[email protected]

The purpose of this paper is to propose a new, empirically derived typology of plants in the internationalmanufacturing 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 positionof plants in international manufacturing networks. The focus has been primarily on the intangible knowledgenetwork, and secondarily on the physical, logistic network. Our analysis leads to four types of plants withdifferent network roles: the isolated plants, the receivers, the hosting network players, and the active networkplayers. Our analysis shows that the different types of plants play a different strategic role in the company, havea different focus, and differ in age, autonomy, and level of resources and investments. Also, the analysis suggeststhat the evolution of the plant depends to some extent on the network role of the plant. Finally, two scenariosfor the development of a strong network role are identified. The research is useful for the scholar studying thearchitecture of knowledge networks, as well as for the practitioner who is in charge of an international networkof manufacturing units.

Key words : manufacturing strategy; knowledge management; international manufacturing; plant networksHistory : 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. IntroductionIn 1964, Skinner warned, “the time has come whenwe must begin to sharpen the management of inter-national manufacturing operations” (Skinner 1964,p. 126). As competition is globalizing and the com-plexity of the environment in which companies oper-ate is increasing, managing an integrated interna-tional network has become an increasingly impor-tant task for manufacturing managers (Bartlett andGhoshal 1989, Ferdows 1997a). However, despite theimportance attached to it by both academics and prac-titioners, the field of international operations manage-ment is still at a relatively early stage of theory devel-opment (Roth et al. 1997) and could be enriched byinsights from empirical research (Chakravarty et al.1997).In the field of international operations manage-

ment, at least two categories of research can be distin-guished (Chakravarty et al. 1997). The first category ofresearch consists mainly of international comparisons.The basic question here is to what extent models andconcepts in production and operations managementare applicable in different countries or regions. Thesecond category studies the management of interna-tional networks of facilities, suppliers, and markets.

The basic question here is how to design and man-age the flows of goods, people, technology, and infor-mation in international networks (Chakravarty et al.1997). Our research contributes to this second cate-gory of international operations research.Competitiveness today is not solely based on the

application of state-of-the-art management techniquesin each of the individual plants, but also on the imple-mentation of an integrative strategy on the networkof plants (Ferdows 1997a). From a logistics perspec-tive, this requires the optimization of the company’ssupply chain. From an organizational perspective, itrequires managing the creation and transfer of knowl-edge in the network. Plants adopt a different role inthese networks. As plants differ in product allocationand in focus, they play different roles in the supplychain (Hayes and Schmenner 1978). As they differin the level of creation, sharing, and absorption ofinnovations, they play different roles in the intangibleknowledge network in the company (Ferdows 1997b).The purpose of our research has been to understandthe different roles of plants in this knowledge net-work. Based on rigorous and in-depth case research,a new typology of plants has been derived. The planttypes differ in the extent to which they share inno-

1737

Vereecke et al.: Typology of Plants in Global Manufacturing Networks1738 Management Science 52(11), pp. 1737–1750, © 2006 INFORMS

vations with the other plants, in the level of visits toand from the other plants, and in the level of com-munication with the other plants. The analysis alsoshows that different roles in the knowledge networkcoincide with different roles in the supply chain.

2. Literature Review2.1. Operations in a Multinational: A Network

PerspectiveOver the last two decades, research on the structureand organization of multinationals has shifted froma focus on the one-to-one headquarters-subsidiariesrelationships toward a focus on managing a networkof units (Kogut 1989). Ghoshal and Bartlett (1990,p. 620) claim that the network approach “is particu-larly suited for the investigation of such differencesin internal roles, relations, and tasks of different affil-iated units �� and of how internal co-ordinationmechanisms might be differentiated to match the vari-ety of sub-unit contexts.”In the management of these networks, the focus

has often been on the flow of information. Doz andPrahalad (1991, p. 160), for example, state that differ-ences in the mission of subsidiaries are reflected inthe “pattern and intensity of information flows.” Intheir more recent work Doz et al. (2001) argue that thesuccess of some multinational companies lays in theirability to “sense” information and knowledge and todistribute it rapidly throughout the network.The information flow is only one type of network

relationship between the subsidiaries and headquar-ters, and among the subsidiaries. The physical flowof components, semifinished goods or end products,financial flows, and “flows” of people moving aroundin the network are other types of network relation-ships (Bartlett and Ghoshal 1989).This trend toward describing the multinational com-

pany as a network of units can also be observed inthe manufacturing strategy literature. Work has beendone, for example, in the description of the benefitsand methods of the transfer of best practices acrossthe manufacturing network. Chew et al. (1990) showthat the improvement of the overall performance ofmultisite companies depends on the local innovative-ness of the plants, as well as on the interplant transferof these local innovations. Flaherty (1986, 1996) addsto this the importance of coordination. She argues thatthe coordination of international operations in a net-work can improve cost and delivery performance andenhance the learning from the experiences of units inthe network.However, the systematic analysis of the relationship

between the plants in the manufacturing networkrequires an appropriate methodology. Nohria (1992,

p. 8) claims that, “if we are to take a network per-spective seriously, it means adopting a different intel-lectual lens and discipline, gathering different kindsof data, learning new analytical and methodologicaltechniques, and seeking explanations that are quitedifferent from conventional ones.” Network analy-sis is a particularly powerful methodology for thedescription and analysis of the structure of networksand the position of the units in the network (Knokeand Kuklinski 1982). The next section describes thenetwork relationships between the units in the manu-facturing network from a conceptual perspective. Theoperationalization of these network relationships andthe application of network analysis techniques aredescribed in §3.

2.2. Network Position of PlantsThe purpose of our research is to understand the posi-tion of manufacturing units in international manu-facturing networks. Our hypothesis is that distinctplants play different roles in these networks by hav-ing relationships of different type and intensity withthe other plants and with headquarters. Bartlett andGhoshal (1989) recognize four types of relationshipsbetween subsidiaries: physical goods, information,people, and financial resources. The flow of financialresources in the strict sense of providing capital tosubsidiaries is of lesser importance in our study ofnetwork relationships between plants, and will there-fore not be discussed here. The three other typesof relationships—goods, information, and people—differ in their degree of tangibility. Our interest liesprimarily in the intangible knowledge network of themultinational, which is explained in the next two sec-tions because we are exploring how the network ofproduction facilities of the multinational may enhancethe creation of strategic capabilities. The logisticsorganization of the multinational, which is reflectedin the focus of the plants and in the tangible trans-fer of components on semifinished goods through thenetwork, is discussed in §4.4.

2.2.1. The Information Network. Two types ofinformation flow can be distinguished: the admin-istrative information flow and the knowledge flow(Gupta and Govindarajan 1991). In a manufactur-ing context, the administrative information flows con-sist of information on inventory levels, purchasingrequirements, forecasts, production plans, etc. Theseinformation flows depend to a large extent on thedegree of centralization of manufacturing tasks, suchas planning, inventory management, and procure-ment. From a manufacturing strategy perspective, theknowledge flows are the more interesting ones. It iscommonly accepted that one of the main reasons forthe existence of multinationals is the possibility toacquire, create, and use technological assets across

Vereecke et al.: Typology of Plants in Global Manufacturing NetworksManagement Science 52(11), pp. 1737–1750, © 2006 INFORMS 1739

national boundaries (Dunning 1993, p. 290). Conse-quently, the ability to transfer innovations throughthe multinational’s network is crucial for attaininga competitive advantage. Three categories of inno-vation flows have been studied: the developmentand introduction of a new product, the developmentand introduction of a new production process, andthe implementation of a new management system(Ghoshal and Bartlett 1988).

2.2.2. The People Network. The flow of peoplein the manufacturing network may take differentshapes. A typical example is the position of a man-ager having line or staff responsibility in two or moreplants. This can be at the level of the plant man-ager, as well as the functional levels reporting tothe plant manager. This type of relationship can becalled “interlocking management” by analogy withthe interlocking directorship; i.e., one person being amember of the board of directors of two or more com-panies (Gerlach 1992). Of equal importance are the“dispatched managers,” i.e., the managers who havebeen transferred from one operating unit to another,on a permanent or a temporary basis, by analogywith the dispatched director. A third shape of theflow of people refers to the day-to-day operationsof the network. These relations between units arerealized through “coordinators”—managers travelingfrequently between operating units to share informa-tion and to accomplish cooperation between the units.The role of such coordinators has received a lot ofattention in the organization literature. They are spe-cific examples of what Galbraith (1977) and Mintzberg(1979) have defined as the “liaison devices” of anorganization.A major advantage of these coordinators is the op-

portunity they create for personal contact betweenpeople in the organization. Ghoshal et al. (1994) haveshown that the relationship among subsidiary man-agers and the relationship between managers of sub-sidiaries and managers of headquarters have a signif-icant influence on the frequency of the intersubsidiarycommunication and on the frequency of communica-tion between the subsidiaries and headquarters. Com-munication plays an important role as a facilitator ofthe transfer of innovations in multinationals (Ghoshaland Bartlett 1988, Gupta and Govindarajan 1991).We retain from this short discussion three variables

that are particularly relevant for our study: (1) theflow of innovations between the units in the network;(2) the extent to which coordination exists in the net-work through managers traveling between the units;and (3) the frequency of communication between theunits in the network.Interlocking management has not been retained as

such in the research because it can be regarded as aspecial reason for frequent travels between the two

plants involved. Dispatching has not been retainedeither because we assume that this creates a tight rela-tionship between the dispatching and the receivingunit only if the dispatched manager keeps in touchwith his original unit. Measuring the communicationbetween the two units then captures this.

3. Research Methodology3.1. Case ResearchThe research reported here is part of a larger researchstudy on the international plant configuration. Theresearch was exploratory, i.e., we wanted to under-stand the “how” and “why” of the international plantnetwork. Thus, case study research has been preferredover 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 ex-haustive, we mention that a strict research protocolhas been designed, a questionnaire with both closed-and open-ended questions has been developed asguidance for the interviews, accommodations havebeen made to avoid interview fatigue, and both qual-itative and quantitative data have been collected in arigorous and structured way and have been analyzedin a systematic way. Several variables have been mea-sured through multiple item measures. The reliabilityof these variables has been assessed by calculating theCronbach alpha, and factor analysis has been used toreject or confirm the assumption that some theoreti-cal constructs underlie the items (Carmines and Zeller1979, DeVellis 1991).To enhance construct validity, multiple raters have

been used. This tactic avoids the risk that data comesfrom a single respondent with a biased view or withlimited access to information (Speier and Swink 1995,Boyer and Verma 1996). The intraclass correlation(ICC) method has been used to assess the interraterreliability of the variables. The ICC index measuresthe variance of the scores of the raters within a plant,relative to the between-plant variance. Data on theICC for all variables used in the analyses can be foundin Appendix 1.

3.2. Data CollectionThe case research has been carried out in eight man-ufacturing companies headquartered in Europe, indifferent industries: food products (two companies),textile goods, plastic products, leather products, pri-mary metal, fabricated metal, and electrical goods.Thus, no single industry dominates the sample. Thecompanies had between four and 10 manufacturingplants. The primary selection criterion for the caseshas been diversity, at the level of the company aswell as the plant. At the company level, it is impor-tant to have diversity in terms of the international


environment in which the company operates becauseone of the research objectives was to explore the linkbetween the characteristics of the company’s interna-tional environment and the plant configuration in thecompany. Consequently, the cases are distributed overthe integration/responsiveness grid, as defined byBartlett and Ghoshal (1989). Two of the cases are clas-sified as “global,” two as “transnational,” and four as“multinational” (Vereecke and Van Dierdonck 1999c).Diversity at the plant level has been obtained byselecting companies with a minimum of four plants,spread over a broad geographical region—the ratio-nale being that with three plants or less, companieshave few opportunities for differentiating the role andfocus of their plants. A geographical spread of theplants (pan-European or even global) was expectedto result in a broad range of drivers for establish-ing the plant, and therefore also in a broad rangeof plant roles (Ferdows 1997b). The sample was lim-ited to companies with their headquarters in WesternEurope.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 head-quarters. In total, data has been collected on 59 man-ufacturing plants, through 37 interviews (with a totalduration of approximately 120 hours). The numberof interviews varied between two and six per case.A structured questionnaire with closed- and open-ended questions has been used as a guide through theinterviews.• A second questionnaire has been sent to the plant

managers and/or the manufacturing managers in thedistinct production plants. One hundred fourty fourquestionnaires have been sent to 54 out of the 59plants. For five of the plants, headquarters asked usnot to send a questionnaire to the plant managers.Eighty three percent of the questionnaires have beenreturned from 50 plants. This implies that in total wehave received data from the plant managers on 50 outof the 59 plants (85%). The number of questionnairesreturned from the plants varied between one and fiveper plant.• Information has also been obtained from desk re-

search on company brochures, publications, and com-pany archives.Fourty-two plants were located in Europe, spread

over 14 different countries. The other 17 plants werespread over 10 different countries in East Asia andthe Middle East, the United States and Canada, andSouth Africa and Australia. We thus have a trulyinternational sample. The number of years the planthad been part of the company ranges between 0 (thisplant was starting up at the moment of the research)

and 50 years, with an average of 17 years. The num-ber of employees in the plants ranges between 77 and1,100 with an average of 340.

3.3. Operationalization of the Network Position ofthe Plants

In describing the manufacturing network of a multi-national company as an information and people net-work, the network units considered are all the plantsand the group of managers in headquarters respon-sible for manufacturing (in this paper, referred toas “headquarters”). As discussed earlier, the networkrelationships considered in this research are the flowsof innovation, the use of coordinators, and the com-munication between the units in the network.The innovation transfers have been measured by

asking managers in the plants (through the mail ques-tionnaires) and in headquarters (through the inter-views) to enumerate and describe the transfers ofproduct, process, and managerial innovations theyknow of over the past three years. A similar opera-tionalization has been used by Ghoshal and Bartlett(1988). The information that has been gathered fromthese different sources has been checked, comple-mented, and corrected by at least one manager inheadquarters, in the course of the in-depth interviews.The presence of coordinators has been operational-

ized as the extent to which people are traveling fromone unit to another. This information on people flowshas been collected through the mail questionnaire tothe plants. The measurement is based on the tool usedin the research by Ghoshal (1986). The respondentshad to report the number of days they had spent, overthe previous year, in headquarters and in each of theplants in the company’s network.One of the questionnaire items measures the com-

munication between the managers in the plants andin headquarters. However, such self-reported answersmay suffer from recollection problems. This problemis severe if the data collection method consists ofan interview or questionnaire asking the respondentto name the persons he/she communicates with fre-quently. This approach has been used in early stud-ies of communication networks in R&D laboratories(Allen 1977). An alternative approach is to provide alist of people, and to ask the respondent with whomon this list he/she has communicated, rather than let-ting the respondent name the people he communi-cated with (Knoke and Kuklinski 1982). This approachhas been followed in our research. A score of 3, 2,and 1 has been given to daily, weekly, and monthlycommunication, 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 net-work relations are mutual (as is the case for the com-munication network), we measure centrality of the


unit through its degree. The degree of a unit is definedas the proportion of other units with which a unithas a direct relationship (Knoke and Kuklinski 1982).If network relations are not mutual (as is the casefor the flows of people and innovations), two degreemeasures are used: the unit’s indegree and outdegree(Knoke and Kuklinski 1982). The indegree of a unit isdefined as the proportion of relations received by theunit from all other units. The outdegree of a unit isdefined as the proportion of relations from that unitto all other units.Based on these definitions of centrality, the follow-

ing network variables have been defined:• The communication centrality of plant i captures

the frequency of communication of the manufacturingstaff of plant i with the manufacturing staff of theother units in the network.• The innovation indegree of plant i captures the

intensity of the innovation flow transferred (and im-plemented) 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 num-

ber of days plant i has received visitors from the man-ufacturing staff team of the other plants.• The people outdegree of plant i captures the num-

ber of days manufacturing staff people of plant i havebeen visiting other plants in the plant configurationIn network analysis, the consequences of missing

data are severe because the lack of data from a singleunit implies the lack of data on the N − 1 possiblerelationships of this unit with the other units in thenetwork. Estimates such as centrality can therefore bedistorted if data are missing. Consequently, great carehas been taken so as to maximize the response rate(Vereecke and Van Dierdonck 1999b).

3.4. Clustering of the DataTo ensure the validity of the network typology, a two-stage procedure has been followed to cluster the data(Ketchen and Shook 1996). We had sufficient data on49 of the plants to involve them in the cluster anal-ysis. Ward’s hierarchical clustering method has beenused to define the number of clusters. This num-ber of clusters has then been used as the parame-ter in the nonhierarchical K-means clustering methodwith Euclidian distance measure. K-means clusteringis preferred over the hierarchical cluster methods forthe development of the typology because it is an iter-ative partitioning method and thus is compensatingfor a poor initial partitioning of the cases. Because theunits of measurement for the network relationshipsdiffer substantially and Euclidian distance is used asthe distance measure in the cluster analysis, the vari-ables have been standardized prior to the clustering(Aldenderfer and Blashfield 1984, p. 21).

As suggested by Ketchen and Shook (1996), thenumber of clusters has been determined through theuse 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 vari-ance explained at each step. After seven clusters, theincreases in R2 are low (below 3%). This observationpoints at a classification into four or seven clusters.• The cubic clustering criterion (CCC) points at

nine clusters. However, tests have indicated that theCCC may suggest too many clusters (Milligan andCooper 1985).• We have used the analytics software SAS to per-

form a number of the tests that have been put forwardby Milligan and Cooper as most effective (Milligan1996). The pseudo F statistic, developed by Calinskiand Harabasz (1974), has local peaks at two and sevenclusters. The pseudo t2 statistic, based on Duda andHart (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. Becausethere is partial agreement among the test results,Milligan (1996) suggests opting for the larger number,that is, seven. However, when going from the four tothe seven-cluster solution, we see that the pattern ofthree clusters is roughly maintained, while the fourthcluster falls into four smaller clusters (including acluster of one unit), which are difficult to distinguish.Consequently, the seven-cluster solution merely addscomplexity without providing revealing insights. Wehave therefore opted for a classification into fourclusters.

4. Empirical Results4.1. A Network Typology of PlantsThe four clusters represent different positions of plantsin the plant network of information and people. Theaverage of the network variables in each of the clus-ters is represented graphically in Figure 1.The typology of plants resulting from this clus-

ter analysis is summarized in Table 1. We distin-guish three levels for each of the variables: “low”for average value below 0; “medium” for averagelevel between 0 and 1; and “high” for average valueabove 1. These cut-off values are defined on the stan-dardized 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, fewmanufacturing staff people come to visit such a plant,few manufacturing staff people from this plant go


Figure 1 Network Clusters: Graphical Representation

Plot of means for each cluster

Variables

–2

–1

0

1

2

3

Communication centra

Innovation indegree

Innovation outdegree

People outdegree

People indegree

Cluster BCluster A

Cluster DCluster C

visit other plants, and there is little communicationbetween the manufacturing staff people of this plantand the other manufacturing managers in the net-work.A plant in Cluster B is comparable to the isolated

plant on all but one variable: it receives more inno-vations from the other units in the network. We willtherefore label these plants as “receivers.” Clusters Aand B thus consist of plants that are only weaklyembedded in the manufacturing network. They rep-resent 37 out of the 49 plants in the sample.Clusters C and D consist of plants that are true net-

work players. A type C plant frequently exchangesinnovations, both ways, with the other units and itsmanufacturing staff communicates extensively withthe other manufacturing managers in the network.A C plant is also frequently hosting visitors fromother units in the network. In the network, the C plantthus 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 communicationcentrality and the outflow of innovations are evenhigher in the type D than in the type C plants (signif-icantly different at p = 10% for communication cen-trality and at p= 5% for innovation outflow). Second,the major flow of visitors is in the opposite direction.

Table 1 Network Typology of Plants

Cluster C Cluster DHosting Active

Cluster A Cluster B network networkNetwork variable Isolated Receiver player player

Number of plants in cluster 11 26 8 4Communication centrality Low Low Medium HighInnovation indegree Low Medium Medium HighInnovation outdegree Low Low Medium HighPeople indegree Low Low High MediumPeople outdegree Low Low Medium High

Whereas in type C plants the inflow of visitors is sig-nificantly higher than the outflow �p < 1%�, in type Dplants 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 labelthem as the “active network players.”

4.2. Cluster ValidationAnalysis of variance on the variables used to generatethe cluster solution is frequently used to test the valid-ity of the cluster analysis solution. The test results aresummarized in Table 2.However, we do not want to overemphasize the

value of this analysis of variance. Because the clus-tering method attempts to minimize variance withinthe clusters, it is logical that the F -test is significant(Aldenderfer and Blashfield 1984, p. 65). External cri-teria analysis is more appropriate. Such analysis isbased on statistical tests on variables that have notbeen used to generate the cluster solution, and yet arerelevant (Aldenderfer and Blashfield 1984, Milliganand 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 roleof the plant as the extent to which the plant con-tributes to the other units in the manufacturing net-work (Vereecke and Van Dierdonck 1999a). We havemeasured the importance of the strategic role of theplant on a nine-point Likert scale, describing plantswhich have as their main goal “to get the productsproduced” at the lowest extreme, to plants that are a“center of excellence, and serve as a partner of head-quarters in building strategic capabilities in the manu-facturing function” at the highest extreme. Given ourdefinition, the importance of the strategic role of theplants in Cluster D should be high. The importance ofthe strategic role of the plants in Clusters A and B, onthe other hand, should be low because these plantsmake little contributions to the plant network. Theplants in Cluster C are expected to play a strategicrole of medium importance.The average and median of the importance of

strategic roles are shown in Table 3. We should notehere that for the importance of the strategic role, as

Table 2 Analysis of Variance on Four-Means ClusterSolution

Network variable F p-level

Communication centrality 12, 18 0.000006Innovation indegree 17, 38 0.000000Innovation outdegree 21, 69 0.000000People indegree 47, 81 0.000000People outdegree 14, 76 0.000001


Table 3 Importance of Strategic Role of the Plants

Median test: obs-expValid N Mean Median below median∗

Cluster A 11 4.80 4.67 0�39Cluster B 26 4.52 4.69 1�73Cluster C 8 5.76 6.44 −0�08Cluster D 4 7.97 8.10 −2�04Overall 49 5.07 4.80

∗Number of cases observed minus number of cases expected below theoverall median level of strategic role, that is, below 4.80. A positive numbershows the number of cases observed below the overall median, and conse-quently indicates a relatively low level of strategic role in the cluster.

well as for most of the plant characteristics that willbe discussed later, the assumption of normality is vio-lated. For those variables, the nonparametric alterna-tives to the ANOVA, the Kruskall-Wallis and MedianTests, have been used.The Kruskal-Wallis test indicates a significant dif-

ference in the level of the strategic role between theclusters �p < 10%�. The Median Test confirms that thedifference in strategic role follows the hypothesizedpattern, as can be seen in Table 3. Cluster B containsslightly more cases below the median level of strategicrole than could be expected if the strategic role wereevenly distributed over the four clusters, indicating arelatively low level of strategic role. Cluster D con-tains more cases above the median level of strategicrole than could be expected if the strategic role wereevenly distributed over the four clusters, indicatinga relatively high level of strategic role. The Mann-Whitney U-Test confirms that the level of strategicrole in Clusters A and B is significantly lower �p < 5%�than in Cluster D.

4.3. Future Strategic Role of the PlantWe have discussed the relationship between the net-work position of the plant and the importance ofthe strategic role played by the plant. Our researchalso provides information on the expected changes inthe strategic role of the plant. The interviewees wereasked to estimate the importance of the strategic roleof the plant as they expect it to be in five years on thenine-point Likert scale described above. The data sug-gests that in Clusters C and D, only a few marginalincreases and decreases in strategic role are expected.This suggests that the plants which occupy an inte-grated position in the network (Clusters C and D) arefairly stable in terms of the importance of the strategicrole they play in the company. Several of the A and Bplants, on the other hand, are expected to experiencean increase in strategic role. For some, the expectedincrease is quite substantial. Given the relationshipthat we observed between the role of the plant and itsnetwork position, it is fair to expect that these plantswill probably be moving from Clusters A or B toward

Clusters C or D. Several of the other plants in Clus-ters A and B are expected to experience a decrease instrategic role. Again, for some, the expected decreaseis quite substantial. It is clear that these two clustersof nonintegrated plants are less stable than the twoclusters of integrated plants.An example illustrates our point. Two of the “re-

ceiver” plants in the sample have been closed sincewe started the case research. We do not want to inferhere that the plants in the “isolated” or “receiver”clusters are on the waiting list for closure. The exam-ples of plants with a positive expectation in strate-gic role would certainly contradict this point. Ourhypothesis is that the plants in these two clusters arein a variable position, and that this variability maylead toward an increase as well as a decrease in termsof the importance of the role the plant plays in tomor-row’s network. These plants seem to provide strategicflexibility in the network.It is interesting to mention that the decrease in

strategic role that is predicted by headquarters forsome of the isolated plants and the receivers is notexpected by the managers in the plants. The lack ofnetwork relationships for the isolated plants and thereceivers seems to create a gap between the expecta-tions of plant management and the considerations inheadquarters. It may also suggest that the managersin A and B plants are less involved in strategic deci-sion making and, thus, are less well informed.

4.4. Characteristics of the Plant TypesTo better understand the network typology of plants,the four types of plants have been compared on a setof 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 productrange, and

Market focus: the extent to which the plantfocuses on a narrow portion of the geographical mar-ket served by the company.• The supplier/user relationship with other plants in

the network: the extent to which a plant suppliescomponents or semifinished goods to or uses com-ponents or semifinished goods from another plant inthe network. It has been measured as the centrality(outdegree and indegree) of the plant in the physicalnetwork of goods. The outdegree of plant i capturesthe portion of plants in the plant configuration, towhich plant i supplies components or semifinishedgoods. The indegree of plant i, (analogously) captures


the portion of plants in the plant configuration, fromwhich plant i receives components or semifinishedgoods.• The level of investment: A list of 14 potential in-

vestments has been included in the questionnaires.From this list of 14 items, four types of investmenthave been identified through factor analysis:

(1) Investments in the production process, that is,in setup time reduction, plant automation, processanalysis, productivity improvement, and through-put time reduction (Cronbach alpha of the resultingfactor= 0�77).

(2) Investments in planning, that is, in materialand/or capacity planning and just-in-time systems(Cronbach alpha of the resulting factor= 0�79).

(3) Investments in managerial improvement pro-grams, that is, in statistical process control, supplierpartnerships, total quality management, and em-ployee participation programs (Cronbach alpha of theresulting factor= 0�73).

(4) Investments in new product development.• The autonomy of the plant. Both strategic auton-

omy and operational autonomy have been measuredthrough 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 havebeen identified, through factor analysis:

(1) Strategic autonomy in decisions concerningthe operations of the plant, that is, the decision to de-velop a new product or to introduce a new planningsystem and the selection of a new supplier (Cronbachalpha of the resulting factor= 0�81).

(2) Strategic autonomy in decisions concerningthe design of the plant, that is, the decision to de-velop a new production process and the choice ofa new technology (Cronbach alpha of the resultingfactor= 0�85).Two dimensions of operational autonomy have

been identified, through factor analysis:(1) Operational logistics autonomy, that is, in

developing a production plan, placing purchasingorders, managing inventories (Cronbach alpha of theresulting factor= 0�84).

(2) Operational autonomy in design and engi-neering, that is, in developing new products and pro-cesses (Cronbach alpha of the resulting factor= 0�88).• The level of capabilities in the plant. Two types

of capabilities are distinguished: the capabilitiesto develop new products and managerial capabili-ties. They have been measured in the headquartersinterviews through a 1–9 Likert scale. The Cronbachalpha 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 questionnairesent to the plant management teams. Because this per-formance data is self-reported, it is important to havedata from multiple respondents per plant, and to eval-uate the interrater reliability. Two dimensions of per-formance have been identified through factor analysis(see Appendix 1):

(1) Performance on time measures, that is, per-formance relative to the target set for manufactur-ing throughput time, delivery lead time, and on-timedelivery to customers (Cronbach alpha of the result-ing factor= 0�85).

(2) Performance on cost and quality measures,that is, performance relative to the target set for unitproduction cost, productivity of direct workers, defectrates, and overall product quality (Cronbach alpha ofthe resulting factor= 0�83).The results of the (mostly nonparametric) compar-

isons of the four clusters on these variables are listedin Table 4. For those variables that showed a sig-nificant difference across the four clusters (with sig-nificance level p < 10%), pairwise comparison of themean 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 mar-

ket focused than plants in Clusters C and D; andplants in Cluster B are significantly more marketfocused than plants in Cluster C.(3) The outflow of components and semifinished

goods is significantly lower for plants in Clusters Aand B than for plants in Cluster D.(4) The inflow of components and semifinished

goods is significantly lower for plants in Cluster Athan for plants in Cluster B; and is significantly lowerfor plants in Cluster B than for plants in ClustersC and D.(5) The level of strategic autonomy in plant design

for plants in Cluster A is significantly lower than forplants in Clusters B, C, and D. Plants in Cluster Bhave a significantly lower level of strategic autonomyin plant design than plants in Cluster D.(6) The level of process investment in plants in

Cluster D is significantly higher than in plants inClusters A, B, and C.(7) Plants in Cluster A invest significantly more

in managerial improvement programs than plants inClusters 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 clus-

ters that result from these comparisons. The com-ments made throughout the interviews provide someadditional insights in the profile of the clusters. Thesecomments are listed in Appendix 2.


Table 4 Statistics on Plant Characteristics by Cluster

Mean/median

Plantcharacteristic Variable A B C D Difference between clusters

Age Number of years plant is 11�1 16�8 30�6 19�7 Anova �p < 1%�

part of company A< Bn�s�/A< C∗∗/A< Dn�s�/B< C∗∗/B< Dn�s�/C> Dn�s�

Size Number of employees 154 240 362 533 Not significantNumber of workers 111 165 251 308 Not significantNumber of salaried workers 43 43 126 226 Not significantNumber of manufacturing 13 21 41 40 Not significant

staff people

Market focus Proportion of market range 0 �18 0 �63 0 �90 0 �89 Kruskal-Wallis Anova with p < 5%supplied by the plant Mann Whitney U-test

A< Bn�s�/A< C∗∗/A< D†/B< C∗/B< Dn�s�/C≈ Dn�s�

Product focus Proportion of product range 0 �15 0 �22 0 �30 0 �38 Not significant

Supplier/user Outdegree 0 0 0 0 �47 Kruskal-Wallis Anova with p < 5%relationship Mann Whitney U-test

A≈ B/A≈ C/A< D∗∗/B≈ C/B< D†/C< Dn�s�

Indegree 0 0 �11 0 �22 0 �42 Kruskal-Wallis Anova with p < 5%Mann Whitney U-test

A< B†/A< C∗∗/A< D∗/B< C∗/B< D†/C< Dn�s�

Operational Logistics 6�2 6�9 6�4 5�8 Not significantautonomy Development and engineering 4�4 4�8 5�8 6�2 Not significant

Strategic Operations of the plant 4�1 5�2 5�1 5�4 Not significantautonomy Design of the plant 3�7 4�8 5�7 6�3 Anova �p < 5%�

A< B∗/A< C∗∗/A< D∗∗/B< Cn�s�/B< D†/C< Dn�s�

Investment Process investment 5�5 5�3 5�1 6�8 Anova �p < 10%�

A> Bn�s�/A> Cn�s�/A< D†/B> Cn�s�/B< D∗/C< D∗

Investment in planning 4�4 4�9 4�6 6�3 Not significantManagerial investment 6�5 4�9 4�9 5�7 Anova �p < 5%�

A> B∗∗/A> C∗/A> Dn�s�/B≈ C/B< Dn�s�/C< Dn�s�

New product investment 4�9 5�2 5�7 7�0 Not significant

Plant capabilities Level of resources 6�4 5�3 6�4 7�5 Anova �p < 5%�

A> B†/A≈ C/A< Dn�s�/B< C†/B< D∗∗/C< Dn�s�

Performance Time performance 1�0 0�72 0�84 0�82 Not significantrelative to target Cost and quality performance 1�0 0 �63 0 �02 0 �69 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 othervariables, 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. DiscussionSome general lessons can be drawn from the planttypology and the characteristics of the four types ofplants.First, the plants providing innovations to the manu-

facturing network, the “hosting network players” andthe “active network players,” are at the same time re-ceivers of innovations from other units in the network.Apparently, transferring knowledge is beneficial, notonly for the receiver, but also for the provider. Anexplanation may be that the quality of the relationshipbetween two units is a major factor in the exchangeof innovations, or as Szulanski (1996, p. 36) has putit, “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 net-work of ideas and in the tangible network of goods.This is in line with Nonaka and Takeuchi (1995),who argue that codified and noncodified knowledgecomplement and reinforce each other. The “isolated”plant, which is not actively taking part in the networkof ideas, is also isolated in the physical sense: weobserved very little flows of components or semifin-ished goods from these plants to the other plantsin the network, and vice versa. The network players(type C and D), on the other hand, are typically sup-


Table 5 Summary of Plant Characteristics by Cluster

Plant characteristics

A Relatively young; market focused; little inflow and outflow ofcomponents and semifinished goods; relatively low level of strategicautonomy in plant design; relatively high level of managerialinvestment

B Relatively young; little outflow of components and semifinished goods;relatively low level of managerial investment; relatively low level ofcapabilities

C Relatively old; broad market; high inflow of components andsemifinished 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; relativelyhigh level of process investment

pliers to the other plants (in the case of Cluster D) orcustomers 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 intrafirmflows are products and technology, and the latter isoften embodied in the former,” and also observedthis link between knowledge and physical flows. Ourresearch suggests that the product is not only a car-rier 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 Cand D) is 28 years, whereas the average age of the twomore isolated types of plants (type A and B) is only 15years. The difference in age between these two groupsis significant �p < 1%�. Networks apparently developover a long period of time.A fourth conclusion is that the four different net-

work roles reflect very different plant characteristics.The “isolated” plant in Cluster A is very indepen-dent. In its isolated position, it does not contribute tothe network, but on the other hand, it also does notdepend on the other network units for its componentsor for maintaining or improving its manufacturingcapabilities. Plant management has the capabilities torun the plant independently. The receivers in Clus-ter B typically are local players that need support—technical and/or managerial—of headquarters or theother plants in the network for their survival. Theyneed this support either because of the negative atti-tude and lack of skills in the plant, or because of thestrategic decision of headquarters to keep investmentsin the plant relatively low. The hosting network play-ers (Cluster C) are typically fairly old, they supplya broad market, and they are characterized by a lowlevel of managerial investment. The hosting networkplayer has been observed in seven of the eight com-panies studied. With one exception, this role is playedby only one of the plants per company. It is interest-

ing that half of the eight C plants are the “motherplant,” the earliest plant in the network, located closeto headquarters. We hypothesize that because of itsage, the broad market it supplies, and its easy accessto headquarters, the plant has gained a lot of expe-rience, which explains why the plant is seen as acompetence center by other plants. The other fourC plants are located close to another plant with whichthey have established tight relationships. The inflowof people in these C plants dominantly comes fromthis neighboring plant, which also has a higher levelof strategic role. In two cases, the neighboring planthappens to be a D plant. The profile we see here isone of a satellite plant that is heavily influenced bythe presence of another network unit. We concludethat the scenario which leads to a C-type plant seemsto build on heritage: the network relationships existbecause the plant has been in the network for a verylong time and is located close to headquarters or to anactive network player. The C plant seems to undergothis scenario in a passive way, rather than to play anactive 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 investmentsunder a relatively high level of autonomy. Such plantsare actively building network relationships by send-ing manufacturing staff to other plants and throughextensive communication. It is their enthusiasm andtheir technical specialization that makes them animportant network player.From interaction with managers about the typology,

we have noticed that the D cluster is an intriguingcategory for plant managers. The D plants are typ-ically plants that act as a center of excellence or asa pilot plant for new products, they are regarded asthe “think tank” or “engine” in the network, and areknown as the technical “specialist” plant in the net-work (see Appendix 2). This intriguing profile raisesquestions as to the further evolution of these plants,which makes it an issue for future research.Fifth, there is no significant difference in perfor-

mance between the clusters. Reaching the targets oncost, quality, or time measures does not appear moreor less difficult in the distinct clusters. This suggeststhat there is not a unique optimal network positionfor a plant. Rather, the network position of the plantshould be regarded from a contingency perspective.Finally, the analyses suggest that the future per-

spectives of the plant depend on the plant’s net-work position. Plants that are strongly embedded inthe production network are expected to maintain thehigh level of strategic role they are already play-ing in the network. The future of plants in ratherisolated positions has been predicted to be in twoopposite directions: some plants are expected to grow


in strategic importance and are assumed to developnetwork relationships; others are expected to becomeless important and may even disappear from the man-ufacturing network. A possible explanation may bethat in case of overcapacity and cost cutting, an “iso-lated” or “receiver” plant is a welcome candidate fordisinvestment or closure. Closing such a plant impliesa reduction of overall capacity, which is exactly whatis aimed at. It does not imply, however, an importantreduction in knowledge transfers because these plantsdo not contribute considerably to the other plants inthe network. This is apparently a headquarters’ deci-sion plant managers are not aware of.

6. Contributions to Researchers andPractitioners

Previous classifications of plants have focused onthe tangible characteristics of plants: the productsthe plant produces, the processes it has in place, themarkets it serves, and the parts it supplies to otherplants in the network (Hayes and Schmenner 1978).The typology developed in our research differs, asit classifies plants on the basis of their position inthe intangible knowledge network. We focus primar-ily on flows of knowledge, rather than flows of goods.The conclusions of our work are therefore useful toany scholar who wants to study the architecture ofknowledge networks in manufacturing, and eventu-ally in other environments such as R&D or serviceoperations.The research has allowed us to identify, among the

plants we studied, 12 network players, i.e., plants thatshowed a strong interaction with other units in thenetwork. This interaction between plants is a fairlynew trend, or at least, a trend not previously welldocumented. Moreover, our research offers a method-ology for identifying network relationships in manu-facturing networks.To the manager in charge of a multinational net-

work of manufacturing plants, the typology serves asa “toolbox” for drawing a map of the plant network.In our multiple discussions with managers about thetypology, we learned that the typology has high facevalidity to them, and allows them to classify theirplants, even without actually measuring the in- andout-flows of the plants. An evaluation of this mapmay help them in identifying possible gaps or unbal-ances. Because the position of the plant in the net-work does not impact the plant’s performance, anyof the types of plants can be effectively present inthe network. If managers believe that their networkwould benefit from plants spreading best practices,they should identify which plants have active andhosting capabilities, and foster these plants in their

network. However, the hosting network players seemto be a result of the past, while the existence ofthe active network players can be stimulated. Ourresearch indicates what it takes to develop an activenetwork player. On the other hand, the manager mayfind it wise to have some isolated plants or receivers(types A and B). These are quite mobile buildingblocks of the network. Reducing the number of iso-lated plants or receivers does not impact the potentialfor transferring knowledge. As such, the presence ofisolated plants and receivers gives the manager somestructural flexibility in managing his network.To the plant manager, the research shows the dan-

ger of a protective attitude toward the exchangeof knowledge. The isolated position taken by theseplants may well result in a difference in view betweenplant managers and company managers about thestrategic future of the plant.

7. Limitations and Future ResearchAn important limitation of the research is the focuson the intracompany network relationships. Whilewe acknowledge that intercompany network relation-ships are important in creating sustainable competi-tive advantage, we have limited our research to thenetwork relationships between units of the same com-pany. Whether the hosting and active network playersare also tightly embedded in the external, intercom-pany network with suppliers, customers, and othernetwork partners remains to be studied.Second, our research describes the strategic role

played by plants in international plant networks. Itidentifies those plants that develop knowledge andcapabilities and that transfer this knowledge to theother plants in the network. The research does notexplain how this knowledge is developed, nor does itdescribe the mechanisms used for the diffusion of thisknowledge and their effectiveness. Also, as stated ear-lier, the research is static and raises questions as to thefurther evolution of the plants and of their position inthe network. This is an area of future research.The absence of significance in the difference in per-

formance in the network typology may point at a lackof difference in performance. However, it may as wellbe a consequence of the performance measure thathas been used, which is static and rather restrictive(that is, performance relative to the target set for theplant) and operationalized as a perceptual measure.There is definitely still a need to study the relation-ship between plant performance and the position theplant plays in the manufacturing network.Also, we did not make any assertions about the

relationship between the portfolio of plants in termsof their network type and the performance of thecompany. We hypothesize that the optimal portfolio


of plants is contingent on the company’s competi-tive environment. However, this needs to be studied.As mentioned in the Methodology section (§3), thispaper is based on case research. While one of themajor advantages of case research is the depth of theinformation that can be collected, its major disadvan-tage is the limitation in sample size, and therefore thepotential limitation in external validity. However, weare convinced that the careful selection of the casesfrom a diversity of industries improves the externalvalidity of the work.The cases have been limited to companies head-

quartered in Europe to avoid cultural differences be-tween the cases. Whether the conclusions still holdin multinationals headquartered in other continents isunexplored and can be subject to future research.Finally, the research focuses on manufacturing com-

panies only. Whether a similar typology can be devel-oped for service companies is an open question.

8. ConclusionIn the research, network analysis has been used asa methodology for understanding the position ofplants in international manufacturing networks. Thefocus has been primarily on the intangible knowledge

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 Logistics Developing a master production schedule 0.81Developing material and capacity plans 0.78Developing the shop floor schedule 0.70Developing sales forecasts 0.89Placing purchasing orders 0.80Managing inventories 0.70

Development and engineering Developing new products 0.74Making changes to existing products 0.77Developing new production processes 0.78Making changes to existing production processes 0.79

Strategic autonomy Operations of the plant Decision to develop a new product 0.69Decision to make changes to an existing 0.76

product designSelection of a new supplier 0.77Decision to introduce a new planning and 0.80

control systemChoice of standards, goals, and performance 0.70

measures for quality management

Design of the plant Decision to develop a new production process 0.76Decision to make changes to an existing 0.78

production processChoice of technology 0.73

network, and secondarily on the physical, logistic net-work. A typology of plants in a manufacturing net-work has resulted from the research. Four types ofplants, with a different strategic role, different charac-teristics, and different perspectives for the future haveemerged. The typology indicates that flows of knowl-edge between plants seem to be reciprocal, and thatthere is a clear correlation between tangible and intan-gible flows in the network. The driver behind inten-sive network relations may be either heritage or adeliberate investment in capabilities inside the plant.Anyhow, building network relationships takes time.We have also observed that the future of plants thatare tightly embedded in the network is more stableand secure.Overall, this leads us to believe that in manag-

ing international networks of plants, managers canbalance long-term knowledge development and me-dium-term flexibility. In approving investments in thenetwork relationships, they allow some of the plantsto play an active role in the creation and diffusionof knowledge in the network, thus creating long-termcompetitive advantage. The other plants provide themanager with strategic flexibility. Their role in the net-work can be adapted in the medium term, accordingto the changing needs of the business.


Appendix 1. Continued.

Construct Factor Item ICC

Investment Process investment Setup time reduction 0.67Plant automation 0.73Process analysis 0.73Productivity improvement 0.75Throughput time reduction 0.85

Investment in planning Material and/or capacity planning 0.57Just-in-time systems 0.72

Managerial investment Statistical process control 0.87Supplier partnerships 0.81Total quality management 0.89Employee participation programs 0.76

New product investment New product development 0.77

Plant capabilities Level of resources Capabilities in developing new products 0.66Managerial capabilities 0.62

Not included in the analyses Level of technical resources 0.34

Performance relative to target Time performance Manufacturing throughput time 0.61(from start until finish of production)

Service level (on-time delivery to customers) 0.75Delivery lead time (from customer’s order 0.60

until delivery)Cost and quality performance Average defect rates at the end of manufacturing 0.75

Average unit production costs for a typical product 0.80Productivity of direct production workers 0.78Overall product quality as perceived by the customers 0.69

Not included in the analyses Rate of new product introduction 0.47Equipment setup time 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 materialsand/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 ICCcutoff 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 CommentsCluster AIndependent �2×�Local �2×�Improved/learning �4×�; problem solvers �2×�Manufacturing capabilities �7×�Motivated �2×�; creativeDevelopment for their ownProduct focused

Cluster BOnly executes �2×�Some development �9×�; a lot of development �1×�Expert (in quality, service, material handling, CIM,

energy savings)Limited; simpleComplex; difficult; below expectationLocal market �7×�; Local improvements; local cultureNeeds help �2×�; receives (technical) support �3×�Lives its own life; goes its own way; distanceLack of motivation; inflexible management;

management problems �2×�; lack of skills �4×�;Insufficient experience; negative mentality;

counterproductive mentality; mentality is to acceptProblems as they come; social climate has improvedCrew of Belgian managers; no own management;

group of expatriates; satellite plant �2×�;

Management input from HQ; strong liaison managerin other plant

Cluster CPilot plant �1×�; test site �1×�; development site �4×�

Center of excellence �3×�; center of competenceProduct know-how �3×�; process know-howTraining centerSupports other plants �2×�; motor for all products �1×�

Close to HQ; home player; mother plant �1×�

Quite motivated to experimentLack of focusLack of investmentSatellite �1×�

Product specialistCluster DCenter of excellence �2×�; development center; pilot

plant �3×�

Think tank; generator of ideasAtmosphere of activity; do-spirit; enthusiasm; happy to

experimentMotor of the other plants; engineGives technical assistance; high tech; specialist; process

know-howClose to HQ �1×�


ReferencesAldenderfer, M. S., R. K. Blashfield. 1984. Cluster Analysis. SAGE

Publications, Beverly Hills, CA.

Allen, T. J. 1977. Managing the Flow of Technology. MIT Press, Cam-bridge, MA.

Bartlett, C. A., S. Ghoshal. 1989. Managing Across Borders: The Trans-national Solution. Hutchinson Business Books, Boston, MA.

Boyer, K. K., R. Verma. 1996. A note on the use of multiple ratersin 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 oper-ations management research: A review and tutorial. ProductionOper. Management 9(2) 128–140.

Calinski, T., J. Harabasz. 1974. A dendrite method for cluster anal-ysis. 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 inter-national operations versus internationalizing. Production Oper.Management 6 100–101.

Chew, B. W., T. F. Bresnahan, K. B. Clark. 1990. Measurement, coor-dination and learning in a multiplant network. R. S. Kaplan,ed. Measures for Manufacturing Excellence. Harvard BusinessSchool Press, Boston, MA, 129–162.

Collins, R. S., R. W. Schmenner, D. C. Whybark. 1989. Pan-Europeanmanufacturing: The road to 1992. Eur. Bus. J. 1 43–51.

De Bodinat, H. 1980. Influence in the Multinational Corporation: TheCase 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 anew 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. Fer-dons, ed. Managing International Factory Network. Elsevier Sci-ence Publishers, Amsterdam, The Netherlands, 3–21.

Ferdows, K. 1997a. Made in the world: The global spread of pro-duction. Production Oper. Management 6 102–109.

Ferdows, K. 1997b. Making the most of foreign factories. HarvardBus. Rev. 75(2) 73–88.

Flaherty, M. T. 1986. Coordinating international manufacturing andtechnology. 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 blockmodelanalysis. Admin. Sci. Quart. 37 105–139.

Ghoshal, S. 1986. The innovative multinational: A differen-tiated network of organizational roles and managementprocesses. Doctoral dissertation, Harvard Business School,Cambridge, MA.

Ghoshal, S., C. A. Bartlett. 1988. Creation, adoption, and diffusionof innovations by subsidiaries of multinational corporations.J. Internat. Bus. Stud. 19(3) 365–388.

Ghoshal, S., C. A. Bartlett. 1990. The multinational corporationas an interorganizational network. Acad. Management Rev. 15603–625.

Ghoshal, S., H. Korine, G. Szulanski. 1994. Interunit communicationin multinational corporations. Management Sci. 40 96–110.

Gupta, A. K., V. Govindarajan. 1991. Knowledge flows and thestructure of control within multinational corporations. Acad.Management Rev. 16 768–792.

Hayes, R. H., R. W. Schmenner. 1978. How should you organizemanufacturing? Harvard Bus. Rev. 56(1) 105–119.

Ketchen, D. J., C. L. Shook. 1996. The application of cluster analysisin strategic management research: An analysis and critique.Strategic Management J. 17 441–458.

Knoke, D., J. H. Kuklinski. 1982. Network Analysis. Sage Publica-tions, Beverly Hills, CA.

Kobrin, S. J. 1991. An empirical analysis of the determinants ofglobal integration. Strategic Management J. 12 17–31.

Kogut, B. 1989. Research notes and communications: A note onglobal strategies. Strategic Management J. 10 383–389.

Miles, M. B., A. M. Huberman. 1994. Qualitative Data Analysis. SagePublications, Thousand Oaks, CA.

Milligan, G. W. 1996. Clustering validation: Results and implica-tions 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 proceduresfor determining the number of clusters in a data set. Psychome-trika 50 159–179.

Mintzberg, H. 1979. The Structuring of Organizations: A Synthesis ofthe Research. Prentice Hall, Engelwood Cliffs, NJ.

Nohria, N. 1992. Is a network perspective a useful way of study-ing organizations? N. Nohria, R. G. Eccles, eds. Networks andOrganizations. 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 technologyand operations management: Resource toolkit for research andteaching. 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: Anexamination of research methods and analytical techniques.Working paper, College of Business Administration, Universityof Oklahoma, Norman, OK.

Szulanski, G. 1996. Exploring internal stickiness: Impediments tothe transfer of best practice within the firm. Strategic Manage-ment J. 17 27–43.

Vereecke, A., R. Van Dierdonck. 1999a. Design and Management ofInternational Plant Networks: Research Report. Academia Press,Gent, Belgium.

Vereecke, A., R. Van Dierdonck. 1999b. Network analysis: A power-ful tool in global manufacturing research. Proc. Decision SciencesAnnual Meeting, New Orleans, LA, 1250–1252.

Vereecke, A., R. Van Dierdonck. 1999c. A typology of interna-tional plant networks: Empirical study. E. Bartezzaghi, R. Fil-ippini, G. Spina, A. Vinelli, eds. Managing Operations Network.SGEditoriali, Padua, Italy, 653–660.

Yin, R. K. 1984. Case Study Research: Design and Methods. Sage Pub-lications, Beverly Hills, CA.

strategy in industrial networks - welcome to...

Documents