Decision Support Systems 83 (2016) 22–31

Contents lists available at ScienceDirect

Decision Support Systems

j ourna l homepage: www.e lsev ie r .com/ locate /dss

Topological analysis and visualization of interfirm collaborationnetworks in the electronics industry

Rahul C. Basole ⁎School of Interactive Computing & Tennenbaum Institute,Georgia Institute of Technology,Technology Square Research Building,85 Fifth Street NW,Atlanta, Georgia 30332, United States

⁎ Corresponding author. Tel.: +1 404 385 6269; fax: +E-mail address: basole@gatech.edu.

http://dx.doi.org/10.1016/j.dss.2015.12.0050167-9236/© 2015 Elsevier B.V. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 17 October 2014Received in revised form 20 December 2015Accepted 20 December 2015Available online 29 December 2015

This study examines the topological characteristics of interfirm collaboration networks (CNs) in the global elec-tronics industry. Our results show that high-performing firms exhibit significant relational CN power, manageCNs that follow a power-law shape degree distribution, are predominantly horizontally integratedwith low geo-graphic complexity, and maintain a balanced exploration-exploitation collaboration relationship portfolio. Wecomplement our topological analysis with graphical visualizations of each of these CNs over three timeframes(2004-06; 2007-09; 2010-12). Theoretically, we demonstrate the association of topological CN characteristicswith high-performance of firms. Methodologically, our study defines and implements a data-driven analysesand visualization of CNs in high clockspeed industries. Our study makes important managerial contributions tothe systemic design, engineering, and management of CNs.

© 2015 Elsevier B.V. All rights reserved.

Keywords:Interfirm collaboration networksTopologyNetwork analysisVisualizationElectronics industry

1. Introduction

Interfirm collaboration networks (CNs) are increasingly importantin today’s complex, global business environment [50,26]. Driven byadvances in information and communication technologies [47], CNsenable participating firms to share and distribute risks [29], enhancecommunication and trust [59], reduce transaction costs [48], and gainaccess to complementary assets, skills, and knowledge [3].

Despite the economic importance, little is known about variation inthe structural shape – or topology – of CNs [19]. There is a general un-derstanding that firms must align their CNs to the market, customer,firm strategy and capabilities [13]. The ”one size fits all” configuration,however, is recognized as inadequate; the ideal CN is firm-, industry-and context-specific [27,1]. This leads to the following research issues:What topological characteristics do high-performing CNs exhibit? Andhow do you best visualize the topological shape of CNs for sensemakingand decision support?

We pursue these questions by grounding our study in theories ofcomplex enterprise systems, interfirm collaboration and network anal-ysis and drawing on multiple carefully curated and integrated second-ary datasets. We complement our empirical analysis with defining amethodology for developing time-based visualizations that enable usto graphically compare interfirm CN structures and provide system-level insights. In doing so, we answer the call for rigorous data-drivenstudies of complex socio-technical systems [46] andmacroscopic inves-tigations of complex strategic issues [2], further our understanding of

1 404 385 6127.

the systemic design, engineering, andmanagement of CNs, and contrib-ute to the emerging interfirm decision support literature [22,9,7].

The remainder of this paper is organized as follows. Section 2 pre-sents the theoretical foundation. Section 3 describes our methodology.Section 4 presents the analysis, visualizations, and a discussion of re-sults. Section 5 concludes the paperwith implications and opportunitiesfor future research.

2. Theoretical foundation

2.1. Interfirm collaboration networks (as) systems

There has been a long-standing recognition that CNs are complexsystems [37,46]. Building on Porter’s linear value chain framework,[54], for instance, describes supply chains as a systemwhose constituentparts include material suppliers, production facilities, distribution ser-vices, and customers linked together via a feed forward flow of mate-rials and the feedback flow of information. Today, industrial CNs arecomposed of a diverse set of vertical and horizontal interactions be-tween suppliers, manufacturers, distributors, retailers, and customers,which have transformed the traditional linear value chain into a com-plex network of interactions between system members [31,32]. At thesame time, globalization has led to geographically dispersed CNs withhigh levels of interfirm dependency [25]. Management of such complexnetworked systems requires a significant level of coordination, collabo-ration, delegation, and monitoring [46,13,55,36].

Traditional modeling and analysis approaches focus on individualfirms or employ a dyadic lens. This approach, however, fails to accountfor the systemic effects resulting from the complex topological and

1 www.gartner.com/technology/supply-chain/top25.jsp2 Supply chain leaders are determined by an assessment of three weighted compo-

nents: financial performance (50%), analyst opinion (25%) and peer opinion by supplychain professionals (25%). Financial data is taken from each firm’s annual report.

23R.C. Basole / Decision Support Systems 83 (2016) 22–31

behavioral aspects inherent in CNs [16]. It has been argued that effec-tive value chain management must emphasize the importance andconsideration of behavior and performance of the entire CN [45]. Inparticular, research has shown the value and applicability in model-ing CNs as complex networked systems comprised of autonomous,self-organizing, interdependent, and adaptive members involved inthe manufacturing, integration, and delivery of products and ser-vices [24,56].

2.2. Network analytic perspective

Bellamy and Basole [16] argue that there are three distinct but relat-ed research foci of prior network analytic studies of complex systems:architecture (i.e. CN structure), behavior (i.e. CN dynamics), and control(i.e. CN strategy). A network analytic perspective enables researchers toincorporate both technical and social issues and thereby offers a moreholistic picture of CNs [20]. The network lens draws on the well-established field of graph theory. In the interfirm CN context, nodestend to represent firms (or other organizational entities, such as facto-ries) and edges represent relationships between firms, such as buyer-supplier relationships, material flow, and information exchange [14].The resulting topology and structural properties describe the positionand (inter)connectedness of firms within the CN [16].

It has been argued that real-world CNs assume one of three commonnetwork topologies (random, small-world and scale-free), each havingstrengths and weaknesses [42], and differing impact on performance,dynamics, and governance (e.g. [42,38]). Basole et al. [12], for example,empirically showusing a network analytic lens that centrally positionedfirms tend to accrue substantial benefits, helping them reduce transac-tional costs and improve operational efficiency, ultimately leading tobetter operating and business performance. Correspondingly, studieshave shown that the type and nature of CN relationships matter aswell. For instance, researchers identified that the strength of a CN rela-tionship – assessed in terms such as frequency, age, or intensity – canpositively facilitate knowledge exchange [61] and new product devel-opment outcomes [41].

2.3. Collaboration

CNs are characterized by two ormore participating firms agreeing toinvest resources, share information, resources, rewards, and responsi-bilities, as well as often make decisions and solve problems jointly[50]. Collaboration thus involves some cooperative behavior. There areplentiful motivations for firms to collaborate, including capturing in-creased economies of scale, operational cost-effectiveness and efficien-cy in todays global markets, access to resources, core competencies,and innovative skills, better financial performance, and greater innova-tion [52,28].

Two well-established theoretical lenses explain these motivations.First, the transaction cost economics (TCE) perspective suggests thatfirms will establish collaborative relationships when the costs incurredfor particular activities is perceived to be lower than when performedwithin existing organizational boundaries [60]. [33], for instance, exam-ined the motivation of a firms knowledge transfer behavior from a TCEperspective and found that firms will engage in relationships whenknowledge transfers are more efficient than market means. TCE hasproven particularly applicable in explaining the shift towards verticalintegration, which has been shown to occur in interfirm contextswhere there is a high frequency of interaction and a great deal of assetspecificity [60]. A second perspective is the resource based view (RBV)of the firm [5]. The RBV suggests that firms pursue collaborative rela-tionships not necessarily to reduce transaction costs, but because higherlevels of integration of resources, assets and capabilities is often difficultto imitate and can thus lead to greater growth and performance of thefirm [33].

Collaboration is particularly prevalent in high-growth, technology-intensive industries where technology and knowledge necessary forsustained innovation often lie outside a firms traditional core compe-tence [23]. Through a series of case studies, [53] found that CNs notonly enabled firms to integrate and link operations for increased effec-tiveness but also enabled radical and incremental innovation. Collabora-tion allows sharing of knowledge and enhances knowledge creation andinnovation spillovers from the supplier [30]. Collaboration in the supplychain also enhances innovation as evidenced in various logisticsactivities such as new product development, process improvements,service delivery, inventory management, technology transfer and ca-pacity planning [30,53]. These findings are corroborated by [52] in anempirical study that found that CN members who had higher levels ofcollaboration achieved better operational performance and innovationactivities.

The organizational learning literature also distinguishes collabora-tive relationships in terms of their motivation to either exploit existingcapabilities or to explore new opportunities [35]. Exploitation placesemphasis on the development of existing products, processes, or re-sources with incremental improvements, efficiency and risk reductionas primary objectives. Exploration on the other hand typically relatesto the exploration of knowledge as well as the search and discovery ofinnovation, with radical improvements, experimentation and risk-taking as central objectives.

3. Methodology

3.1. Data

Our study utilizesmultiple data sources to create the topology of CNsin the global electronics industry. We focus on the electronics industryfor several reasons. First, prior work has shown that the electronics in-dustry is characterized by high levels of collaboration and partnering[51]. Second, the electronics industry operates under a high clock-speed, with new products and services emerging rapidly. New formsof supply chain IT solutions and practices are thus more likely to beadopted. Lastly, the electronics industry is arguably one of the mostglobal with the majority of firms coming from Asia, Europe, and NorthAmerica, enabling us to capture the geographic footprint of CNs.

In order to understand the state of practice,we limit our study on theCN structure of high-performing supply chain firms. We identify rele-vant focal firms using the Gartner Top 25 Supply Chain list.1 This list,first launched by AMR Research in 2005, identifies global supply chainleaders drawn from the Fortune Global 500 and Forbes Global 2000rankings.2 The list is widely used in the supply chain management andstrategy literature (e.g. [21]). An examination of the annual rankingsfrom 2007-2012, shown in Table 1, reveals 12 well-known, highlyreputable and very innovative electronics companies. We chose 2007 asthe starting year of our study as it marked the era of transformativechange in the electronics industrywith the emergence of the smartphone.

The CN structure for each of these companies was then built usingtwo data sources: Thomson Reuters SDC Platinum Alliance & Joint-Venture database (from hereon SDC) and Connexiti. SDC is a commonlyused data source for the study of strategic alliances and industry net-works and is regarded as one of the most comprehensive databases ofits kind [49]. SDC includes information on many different types of col-laborative relationships, including strategic alliances, supply, researchand development (R&D), marketing, licensing and manufacturing. Weinclude all active relationships between 2004 and 2012 in which atleast one of the companies described below has participated. We

Table 1Top Supply Chain Firms (2007-2012). Firms in bold are focal firms in our analysis.

Rank 2007 2008 2009 2010 2011 2012

1 Nokia Apple Apple Apple Apple Apple2 Apple Nokia Dell Procter & Gamble Dell Amazon3 Procter & Gamble Dell Procter & Gamble Cisco Systems Procter & Gamble McDonald’s4 IBM Procter & Gamble IBM Wal-Mart Stores Research in Motion Dell5 Toyota Motor IBM Cisco Systems Dell Amazon P&G6 Wal-Mart Stores Wal-Mart Stores Nokia PepsiCo Cisco Systems Coca-Cola7 Anheuser-Busch Toyota Motor Wal-Mart Stores Samsung Electronics Wal-Mart Intel8 Tesco Cisco Systems Samsung Electronics IBM McDonalds Cisco Systems9 Best Buy Samsung Electronics PepsiCo Research in Motion PepsiCo Wal-Mart Stores10 Samsung Electronics Anheuser-Busch Toyota Motor Amazon.com Samsung Electronics Unilever11 Cisco Systems PepsiCo Schlumberger McDonalds Coca-Cola Colgate-Palmolive12 Motorola Tesco Johnson & Johnson Microsoft Microsoft PepsiCo13 Coca-Cola Coca-Cola Coca-Cola Coca-Cola Colgate-Palmolive Samsung Electronics14 Johnson & Johnson Best Buy Nike Johnson &Johnson IBM Nike15 PepsiCo Nike Tesco Hewlett-Packard Unilever Inditex16 Johnson Controls SonyEricsson Walt Disney Nike Intel Starbucks17 Texas Instruments Walt Disney Hewlett-Packard Colgate-Palmolive Hewlett-Packard H&M18 Nike Hewlett-Packard Texas Instruments Intel Nestle Nestle19 Lowe’s Johnson & Johnson Lockheed Martin Nokia Inditex Research in Motion20 GlaxoSmithKline Schlumberger Colgate-Palmolive Tesco Nike Caterpillar21 Hewlett-Packard Texas Instruments Best Buy Unilever Johnson & Johnson 3M22 Lockheed Martin Lockheed Martin Unilever Lockheed Martin Starbucks Johnson & Johnson23 Publix Supermarkets Johnson Controls Publix Super Markets Inditex Tesco Cummins24 Paccar Royal Ahold SonyEricsson Best Buy 3M Hewlett-Packard25 AstraZeneca Publix Super Markets Intel Schlumberger Kraft Foods Kimberly-Clark

24 R.C. Basole / Decision Support Systems 83 (2016) 22–31

exclude alliances that were terminated during the period. We cross-validated and augmented the SDC dataset with information fromConnexiti. Connexiti is a comprehensive supply chain intelligence data-base that captures both supply and customer relationships for nearly20,000 global companies. Several previous studies have used Connexitito study alliance networks in the high-technology industry (e.g. [6,11,17]). Operational performance data of CNs are taken from theCompuStat database and cross-examined using theMergent Online da-tabase. Innovation performance data are scraped from theUSPTOpatentdatabase.

Table 2Definition and Computation of Topological Metrics.

Metric Definition Computa

BetweennessCentrality

The extent to which a firm can intervene or has controlover interactions among other firms in the supplynetwork.

BCi ¼ ∑

ClosenessCentrality

The extent to which a firm has freedom from thecontrolling actions of others in terms of accessinginformation in the supply network.

CCi ¼ ∑

EigenvectorCentrality

The extent to which a firm has influential power on theactions of other firms in the supply network, based onits direct ties to firms with high relational power.

ECi ¼ α∑

ClusteringCoefficient

The extent to which a firms supply network partnersform sub networks. The forming of these sub networkscan help facilitate collaboration among partners.

Ci ¼ nni ðni−

Embeddedness The extent to which a firm is central and interconnectedin the network.

EMBi=B

Efficiency The extent to which a firms relationships among itssupply network partners are non-redundant.

EFi ¼�∑

Constraint The extent to which a firm is constrained by one ormore of its supply network partners, based on theirdirect ties shared among each other.

CSi ¼ ðpij

DegreeCentrality

The extent of activity in the entire network. D=ni/n

WeightedCluster Coefficient

The extent to which nodes in the network clustertogether.

CW ¼ 1n∑

Density The ratio of actual to potential network connections DN ¼ rn�ðn−

2

Small-WorldCoefficient

The extent to which the nodes in the network can bereached from any other node in a small number of steps.

SW ¼ LC

3.2. Network construction

Each firm’s CNwas constructed as a binary adjacencymatrix, with cellentries marked as 1 if there is any relationship between two firms and 0otherwise. Sincewewere primarily concernedwhether a relationship be-tween two firms exists and not withmultiplex relationships, multiple re-lationships between the same pair of firms were treated as a single link.As collaborative relationships are considered to be bi-directional, our net-work resulted in an undirected unipartite graph [39]. Following [34], wecoded R&D and technology transfer relationships as exploratory; supply,

tion Parameter Description

:q; j≠ipq;i; j pq,i,j represents the proportion of shortest paths between q and

j that run through i

ni¼1dðpi; pjÞ d(pi,pj) represents the number of edges in the shortest path

linking pi and pj

:jAijc j Aij represents the adjacency matrix of supply network

relations, α is a constant and cj represents the centrality ofsupply network entity j

p

1Þ=2 t represents the number of existing ties among all ni directpartners p of focal firm i

Ci×Ci BCi and Ci represents the betweenness centrality and clusteringcoefficient of firm i respectively

:j

�1−∑:

qpiqmjq��=ni piq is the proportion of focal firm is ties invested in the

relationship with q,mjq is the marginal strength of the tiebetween entities j and q (who are both directly connected to i)and ni is the total number of direct partners of focal firm i.

−∑:qpiqpqjÞ2 pij is the proportional strength of firms is relationship with j

and pqj is the proportional strength of qs relationship with j.

ni represents direct partners of focal firm i and n represents thetotal number of firms in the network

:iCi n is the total number of firms in the network and Ci is the

clustering coefficient of firm i.1ÞÞ r is the number of actual connections and n is the number of

firms in the network.L is the average path length between two firms in the networkand C is the average clustering coefficient.

Fig. 1. Conceptual Model of our CN Visualization Approach.

25R.C. Basole / Decision Support Systems 83 (2016) 22–31

marketing, and licensing as exploitative relationships. Finally, we con-structed each firm’s CN structure at k=3 system boundary depth,allowing us to identify how firms leverage local (i.e. direct) and global(i.e. indirect) network connections. We created CNs for three timeframes(2004-06; 2007-09; 2010-12) to reveal how these firms cultivate andmanage important relationships over time.

3.3. Metrics

3.3.1. TopologyWe used UCINET 6.365, a social network analysis package, to

compute topological CN measures at both the firm and network-level [18]. Firm-level network measures include three centralitymeasures (betweenness centrality, closeness centrality, and eigen-vector centrality), clustering coefficient, embeddedness, efficiency,and constraint. Network-level measures include average degree cen-trality, weighted cluster coefficient, density, and small-world coeffi-cient. Table 2 summarizes these metrics and provides a descriptionas they pertain to the study of our CNs.

3.3.2. PerformanceWe use both well-stablished operational and innovation measures to

understand the performance of CNs. The three operational measures

Fig. 2. Degree Distribution of CNs.

include return on asset (ROA), computed by the net income divided bytotal assets; inventory turns (IVT), computed by the cost of goods sold di-vided by inventory; and revenue growth (RV), computed by the percent-age change in revenue from the previous year. We report a five-yearmoving average from 2008-2012. We use patent yield (PY) computedby patent counts divided by R&D expenditures, to measure a firm’s inno-vation efficiency.

3.4. Visualization

Visualizations can provide important novel and complementary in-sights into the structure, dynamics, and strategy of CNs [16,8,10]. Visu-alizations can be used to explore, interpret and communicate data andaid decision makers overcoming cognitive limitations. With the newtsunami of available data, visualization is increasingly recognized as anintegral part of the scientific approach and considered a fundamentalmethod of transforming data to knowledge [58].

There are many examples of network visualizations including bio-logical and ecological networks, social networks, the Internet and cita-tion networks [39]. Visualizations of industry networks are alsoemerging and used as complementary analyses to traditional statisticalsummaries (e.g. [6,11]). Rosenkopf and Schilling [44] argue that visual-izations are effective when trying to explain substantive differences be-tween network structures. It has also been shown that visualizations areparticularly valuable for understanding and analyzing business issues,including competitive intelligence, strategy, scenario planning andproblem-solving [57,11].

Most prior work has focused primarily on the visualization of specif-ic industries, but not explicitly on CNs of individual firms. Unquestion-ably, CN visualization is challenging and resource-intensive. Completeor even comprehensive CN data is generally not available. At the sametime, even if the data is collected and appropriately curated, the amountof information can often be overwhelming to the analyst if not present-ed appropriately [43,22]. Effective visualizations must therefore ensurea careful balance between detail, abstraction, accuracy, efficiency, andaesthetics.

We use Gephi 0.8.2., an open-source software for visualizing and an-alyzing large network graphs to create graphical representations of CNsin the electronics industry [15]. We use a concentric layout approachwith a NoOverlap algorithm to create visually appealing and insightfulCN representations. The focal firm is placed at the center of the visuali-zation while firms k steps away from the focal firm are placed on the kth

circle. Node size is proportional to the firm’s importance asmeasured by

Table 3Topological, Portfolio, and Performance Characteristics of CNs (2007-2012).

Firm-Level Network-Level Portfolio Performance

Firm n e BC CC EC C EMB EF CS D CW DN SW XPR XPT ROA IVT RV PY

APL 286 2,372 47.30 564.00 0.02 0.33 562.85 0.72 0.44 16.59 0.18 0.06 14.97 0.49 0.51 0.17 51.68 0.42 0.30CSCO 866 7,239 8,579.66 1,646.00 0.11 0.15 2,892.28 0.86 0.05 16.72 0.12 0.02 41.96 0.54 0.46 0.11 9.92 0.09 0.05DELL 467 5,696 149.23 916.00 0.03 0.26 1,042.92 0.76 0.20 24.39 0.16 0.05 13.27 0.67 0.33 0.08 42.03 0.02 0.18HP 700 6,825 893.63 1,345.00 0.08 0.26 3,367.22 0.74 0.07 19.50 0.13 0.03 27.52 0.44 0.56 0.07 12.25 0.07 0.10IBM 633 6,461 1,229.00 1,229.00 0.05 0.18 1,323.05 0.83 0.10 20.41 0.14 0.03 22.96 0.44 0.56 0.12 19.26 0.03 0.98INTC 697 6,555 537.39 1,367.00 0.05 0.27 2,111.09 0.74 0.15 18.81 0.13 0.03 28.22 0.53 0.47 0.13 3.40 0.09 0.08MSFT 396 4,685 553.65 771.00 0.03 0.16 533.42 0.85 0.16 23.66 0.18 0.06 13.46 0.54 0.46 0.22 10.95 0.10 0.09NOK 360 3,828 100.73 709.00 0.02 0.14 268.91 0.88 0.24 21.27 0.17 0.04 13.93 0.40 0.60 0.07 12.48 0.00 0.03RIM 577 6,256 374.78 1,125.00 0.04 0.28 1,506.31 0.77 0.13 21.68 0.14 0.04 20.50 0.35 0.65 0.21 11.30 0.49 0.41SE 550 5,396 169.00 1,084.00 0.03 0.39 1,562.14 0.64 0.25 19.62 0.15 0.04 22.67 0.34 0.66 0.09 9.26 0.10 0.64SONY 647 6,309 41.90 1,267.00 0.04 0.28 901.59 0.73 0.14 19.50 0.13 0.03 23.71 0.41 0.59 0.01 9.08 0.03 0.19TI 701 6,849 2,988.50 1,336.00 0.09 0.19 3,017.51 0.81 0.06 19.54 0.13 0.03 26.75 0.49 0.51 0.17 3.65 0.00 0.21min 286 2,372 41.90 564.00 0.02 0.14 268.91 0.64 0.05 16.59 0.12 0.02 13.27 0.34 0.33 0.01 3.40 0.00 0.03max 866 7,239 8,579.66 1,646.00 0.11 0.39 3,367.22 0.88 0.44 24.39 0.18 0.06 41.96 0.67 0.66 0.21 51.68 0.49 0.98avg 589.42 5,860.25 1305.40 1,113.25 0.05 0.24 1,590.77 0.78 0.17 20.14 0.15 0.04 22.49 0.47 0.53 0.11 16.27 0.12 0.27med 640 6,385 456.09 1,177.00 0.04 0.25 1,414.68 0.76 0.14 19.58 0.15 0.04 22.81 0.47 0.53 0.11 11.12 0.08 0.18stdev 175.50 1,476.21 2,434.65 317.76 0.03 0.08 1,041.53 0.07 0.11 2.37 0.02 0.01 8.30 0.09 0.09 0.06 15.00 0.16 0.28ind avg 911 7,311 789.12 2,488.24 0.02 0.12 1,137.51 0.86 0.45 16.04 0.12 0.02 45.73 0.2 0.8 0.04 6.58 0.12 0.12

26 R.C. Basole / Decision Support Systems 83 (2016) 22–31

eigenvector centrality.3 Node color indicates a firm’s modularity class.Modular communities are defined as groups of densely interconnectednodes that are only sparsely connected with the rest of the network[40]. We used a consistent categorical color scheme to indicate the larg-est to smallest community. A conceptual representation of a CN visual-ization is presented in Fig. 1.

4. Results & discussion

4.1. Topological characteristics

The distribution of the number of connections per firm, or degreedistribution, is widely used as a primary summary of the topology ofcomplex networks [39]. The shape of a degree distribution is typicallyindicative of the processes structuring the network. Degree distribu-tions have been studied in a range of different networks, including bio-logical, ecological, engineered and social networks. Our result, shown inFig. 2, reveals that degree distributions of high-performing CNs are notrandom but rather follow a strong power-law distribution shape.4 Inother words, high-performing CNs are comprised of few firms withmany and amajority offirmswith only a few relationships, respectively.This is in line with previous findings which have shown thatmany real-world networks have long-tail connections. Barabasi et al. [4] argue thatthis phenomenon can be explained by the fact that newnodes (orfirms)preferentially attach with an existing node (or firm) based on the num-ber of connections that the existing node already has.

Table 3 summarizes the topological characteristics for each firm’s CNas well as the industry average. The first observation is that high-performing CNs are substantially smaller in size (n) and relationships(e) than the industry average, suggesting a potentially more judiciousCN partner selection approach.

High-performingCNs have nearly twice asmanydirect partners (de-gree centrality), on average, than the industry average, indicating an in-fluential advantage over other firms in the electronics network.

3 We could have used any of the firm-level topological metrics to denote node size.Wechose Eigenvector centrality as it presents a comprehensive measure of network promi-nence of the firm.

4 A power law (PL), also known as a scaling law, is the form taken by a remarkable num-ber of phenomena in the natural, social, and engineered systems. It is a relation of the typeY=kXα, where Y and X are variables of interest, α is called the power law exponent, and kis a typically unremarkable constant. A power-law implies that small occurrences are ex-tremely common, whereas large instances are extremely rare.

High-performing CNs have a substantially higher betweenness cen-trality (BC) score than the industry average (65.4% higher), suggestingthat they gain significant benefits of more timely access to resources.Closeness centrality (CC) measures the shortest path connecting a focalfirm to any other firm in the CN, with a lower score signifying fewersteps needed to reach other partners. Thus, a 44.7% lower average close-ness centrality score than the industry average indicates similarly thathigh-performing CNs have more timely access to resources. Togetherthis provides strong evidence that firms with high-performing CNs pos-sess high relational power.

High-performing CNs are also characterized by higher clustering(C) than the industry average. This result strongly suggests high collab-oration intensity and greater levels of knowledge sharing and learningamong CN members, with performance benefits spilling over to thefocal firm. Among the focal firms, Samsung Electronics (0.39) andApple (0.33) have the highest clustering level. Firms with highembeddedness (EMB) have high relational power and collaboration in-tensity. Our results show that high-performing CNs have an averageembeddedness score that is 39.1% higher than the industry average.One interpretation of this result is that high-performing firms are em-bedded in CNs that facilitate the knowledge sharing and transfer, as de-rived from the greater number of opportunities for collaboration tooccur between network entities. High-performing firms also are, on av-erage, 9.3% less structurally efficient (EF) than the industry at large. Thissuggests that these focal firms allow for greater redundancy in their CNconfiguration, in part through higher levels of interfirm collaborationbetween partners, enabling greater adaptability and agility to changingmarket requirements. Higher levels of constraint (CS) translate toweaker brokerage ability. Our results show that the industry averageconstraint is more than double the average constraint of the high-performingCNs. In fact, all high-performingfirmshave better brokeragepower compared with the industry average. High-performing firmsshare a considerably high percentage (86%) of relationships withfirms outside of their segment, having to manage a much larger (small-er) amount of horizontal (vertical) complexity. The results also showthat the majority of high-performing firms share roughly 60% of theirconnections with firms in the same home country as them.

Overall, all high-performing CNs exhibit greater levels of averagedegree (D) than the industry average. Interestingly, Apple (16.6) andCisco (16.7) have the two lowest average degrees within their net-work. While the average (weighted) clustering coefficient (CW)score is nearly the same across any network, high-performing CNshave as much or higher clustering levels than the industry at large.We also observe that the high-performing CNs have significantly

27R.C. Basole / Decision Support Systems 83 (2016) 22–31

higher amounts of centralization compared with the electronics in-dustry. This result further supports the notion that power and influ-ence is concentrated in a small number of firms and not as dispersedas in the case for any firm in the industry as a whole. This result alsoimplies that power of individual firms varies rather substantially andthat overall positional advantages are unevenly distributed. We alsofind that high-performing CNs are, on average, twice as dense (DN)as the average firm in the electronics industry, indicating greater in-teraction and collaboration between firms. The small-world quotient(SW) reveals that while electronics firms in general portray smallworld characteristic, high-performing CNs do not. One exception isCisco. This is not surprising as Cisco’s CN covers over 95% of the en-tire electronics industry. The CNs of the other firms tend to follow atopology similar to a scale-free network.5

4.2. Portfolio characteristics

The majority of high-performing firms maintain a relatively bal-anced exploration (XPR=47%) versus exploitation (XPT=53%) rela-tionship strategy. Three notable exceptions can be observed. BothResearch inMotion and Samsung relationship portfolio is tipped heavilytowards exploitation (65% and 65%, respectively). Dell’s portfolio isflipped, with a higher proportion of exploration-based relationships(67%). These findings are particularly interesting when compared tothe relationship portfolio of the industry. The results show that therest of the industryis heavily skewed towards exploitation (80%).

4.3. Performance characteristics

The results in Table 3 confirm that all three operational performancemeasures (ROA, IVT, RV) of high-performing CNs are higher than the in-dustry average. This holds also true for individual years. This is not sur-prising as this is a key criteria of the Gartner list. Our data also revealsthat nearly half of the high-performing firms had R&D expendituresbelow the industry average across our time frame. This is a somewhatcontradictory finding, but suggests that high-performing firms may le-verage their CNs for innovation, instead of relying solely on internalR&D. This results supports the growing recognition that innovationmore often occurs outside the firm [23]. When computing patent yield(PY), or innovation efficiency, our results show that high-performingCNs have higher levels of patent-yield than the industry average.

4.4. Visual representations

The three timeframes for the 12 firms are shown as color-encodedsmall multiples ordered alphabetically in Fig. 3a-aj. This approach al-lows for fast visual comparison of topological commonalities and differ-ences over time [58]. By visualizing the topological structure, we canalso leverage our pre-attentative cognitive abilities to discern evolution-ary patterns for each firm and across different firms.

Let’swalk through two illustrative cases in further detail: a between-company comparison (e.g. Apple vs. Dell from 2010-12 shown in Fig. 3(c) and (i), respectively) and a within-company comparison (e.g. IBM,shown in Fig. 3 (m)-(0)). Apple’s CN consists primarily of first andsecond-tier partners, while Dell has a very rich third-tier. The intercon-nectivity between partners is relatively sparse for Apple as compared toDell, which exhibits a relatively dense pattern. Both Apple and Dell haveprominent direct partners, as indicated by the large node size of tier-1partners. When examining IBM’s CN over time, we can see an overallgrowth in partners at each tier as well as a significant growth in inter-connectivity among partners, evidenced by the dense network pattern.

Extending these sample interpretative steps to the other visualiza-tions, several general observations can be made. First, most companies

5 A scale-free network is a connected network with the property that the number oflinks k originating from a given node exhibits a power law distribution.

have a rich set of first and second-tier, but much sparser third tier ofCN partners. This tier-based distribution does not change significantlyover time. Some companies in fact have very sparse third-tier partnerCNs (e.g. Apple, Cisco, Sony). While we see some differences across net-work size and tier distributions, there appears to be a fairly consistentnumber of modular communities across all firms as indicated by thecolor distribution (4-6). We also notice that the density of interconnec-tions between the focal firmand itsfirst tier partner increases over time.In fact some CNs tend to increase density across multiple tiers, as is thecase with Samsung Electronics, Microsoft, and IBM. Apple, on the otherhand,maintains a very leanCN,withmost prominent partners (high be-tweenness score) at the first tier level. Cisco appears to maintain a sim-ilar CN pattern. We also observe that many companies had the greatestincrease in relationship and interconnectivity from the first (2004-06)to the second period (2006-09), suggesting higher relationship activitytimeframe. We observe less differences between the second and thirdtime periods across most firms. Interestingly, we do see that somemore prominent firms (depicted by node size) tend to move from thesecond to the first tier (e.g. Apple, Cisco Systems, HP, Microsoft) overtime, suggesting that firms create closer ties with influential partnersin their CN.

5. Conclusions

Interfirm CNs have unquestionably become an important topic insupply chain and strategic management research. A network per-spective provides a powerful way to systemically understand thecomplex interdependencies and flows that create performance dif-ferences in CNs. In this paper, we show – both statistically and visu-ally – that high-performing global CNs tend to exhibit many commonfundamental topoplogical characteristics. In particular, we show thathigh-performing firms exhibit significant relational power, manageCNs that follow a power-law shape degree distribution, are predom-inantly horizontally integrated with low geographic complexity, andmaintain a balanced exploration-exploitation interfirm relationshipportfolio.

5.1. Implications

Our study has several important implications for supply networkresearchers and practitioners and decision support in general.

5.1.1. Building and sustaining relational powerFirms pursuing high-performing CNs must continuously build,

cultivate, and manage their relationships. There is an increasing rec-ognition that the shift from simple transactional and contractual-based relationships to more long-term relational forms of collabora-tion between CN partners can lead to relational power and manybeneficial outcomes. However, building and sustaining relationalpower requires well-defined partner strategies. Our data-drivenstudy shows that firms with high-performing CNs maintain only amodest level of direct relationships. Relational power can thus beachieved, in part, by carefully selecting and designing the right setof network relationships. However, sustaining relational power re-quires the right combination of ”sticks and carrots” that encourage,incentivize, and empower partners to meet the focal firms desiredhigh operational standards. In order to build and sustain ability tolearn and innovate, managers should cultivate a collaborative philos-ophy with a focus on leveraging the strength of the CN andmust con-tinually scan their networks for value creation potential from thecapacities and capabilities of their partners.

5.1.2. Leveraging innovation capabilities of the CNFirms with high-performing CNs are often innovation leaders as

well. However, depending on the nature of their business and strategicorientation, they are not necessarily prolific patentees. Exceptions

28 R.C. Basole / Decision Support Systems 83 (2016) 22–31

include IBM, Samsung, and Sony, who in fact run internal research orga-nizations. Instead what appears to be the case is that high-performingfirms leverage the innovation capabilities of their CN partners. Theyform strategic relationships with partners that provide them access tonovel resources and knowledge. This is particularly reflected in theirpronounced balance in exploration and exploitation portfolios.

5.1.3. Lowering geographic complexity of the CN baseOur study also contributes to emerging research in the economic im-

plications of the geography of global CNs. Contrary to the perceptionthat innovativefirms source globally, our resultsfind that the total num-ber of international (cross-border) collaborations are actually relativelysmall compared to the number of domestic collaborations. This findingmay be questioned in particular due to the increasing trend of globaliza-tion and digitization of business. However, the importance of CN

(a) Apple (2004-06) (b) Apple (2

(d) Cisco (2004-06) (e) Cisco (2

(g) Dell (2004-06) (h) Dell (20

(j) HP (2004-06) (k) HP (20

Fig. 3. Concentric Visualizations of CNs. Color ∈ M

partner proximity to innovation is not that surprising, as it has beenshown that face-to-face communication is quite common in innovativebuyer-supplier relationships. This finding is also in line with researchexamining the economic implications of regional clustering of firms,which has shown that firms, through strong local ties, have competitiveadvantages over their isolated rivals.

5.1.4. Developing CN intelligence capabilitiesCNs are becoming increasingly global and complex. Identification

and management of risks and opportunities is increasingly challeng-ing. Most firms have only very little visibility into their CN beyondthe first tier. Our study shows that visualization can provide systemicinsight into the underlying structure. Visualization combined withanalytics leads to an important CN capability that allows mappingof firms, flow, information, and risk. Integrating these diagrammatic

007-09) (c) Apple (2010-12)

007-09) (f) Cisco (2010-12)

07-09) (i) Dell (2010-12)

07-09) (l) HP (2010-12)

odularity Class; Size ∝ Eigenvector Centrality.

(m) IBM (2004-06) (n) IBM (2007-09) (o) IBM (2010-12)

(p) Intel (2004-06) (q) Intel (2007-09) (r) Intel (2010-12)

(s) Microsoft (2004-06) (t) Microsoft (2007-09) (u) Microsoft (2010-12)

(v) Nokia (2004-06) (w) Nokia (2007-09) (x) Nokia (2010-12)

Fig. 3 (continued).

29R.C. Basole / Decision Support Systems 83 (2016) 22–31

representations and developing custom views–such as heatmaps ofcollaboration intensity–with enterprise information systems, andthe creation of new metrics (e.g. structural velocity, complexity, orburst) will enable timely identification of peripheral activities andrisks in the CN. Further, they will help shed light on the criticalroles and linkages that previously would have been left unnoticed.At the same time, it provides insight into the competitive strategiesother firms are pursuing. We believe the approach presented inthis paper is a first step towards the development of novel and criti-cal CN intelligence capabilities.

5.2. Limitations and future research

Our study does have some limitations, each of which presentsexciting future research opportunities. We are cognizant that thereare many other high-performing electronics firms that were not

included in the Gartner list. Future studies may wish expand thelist of high-performing firms, or conversely, perform a more in-depth analysis comparing only a few of high-performing firmscompeting for the same market segment. Secondly, we limited ourcomparative analysis to specific firms in a single industry. Future re-search may include other dynamic industries, such as biotechnology,automotive, or consumer goods. We also did not explicitly addressother important aspects of supply network relationships, such asmultiplexity, strength, scope, and governance. Similarly, supply net-works are not static forms of organization. They change and evolveover time. While we have covered a six year span of alliance, opera-tional, and innovation data, future research could incorporate a lon-gitudinal lens for building and portraying CNs. Lastly, successful CNmanagement requires new governance and orchestration mecha-nisms. Empirical studies and in-depth case studies of specific CNsmay reveal important managerial, operational, and technological

(y) RIM (2004-06) (z) RIM (2007-09) (aa) RIM (2010-12)

(ab) Samsung (2004-06) (ac) Samsung (2007-09) (ad) Samsung (2010-12)

(ae) Sony (2004-06) (af) Sony (2007-09) (ag) Sony (2010-12)

(ah) TI (2004-06) (ai) TI (2007-09) (aj) TI (2010-12)

Fig. 3 (continued).

30 R.C. Basole / Decision Support Systems 83 (2016) 22–31

capabilities associated with high-performance and complement ouranalysis.

Acknowledgments

The author would like to thank Marcus Bellamy for help with datacollection and analysis as well as Hyunwoo Park and Jagannath Putrevufor valuable feedback on earlier versions. This study was generouslysupported in part by the Tennenbaum Institute, the Institute for Peopleand Technology, and Intel Corporation.

References

[1] R. Adner, Match your innovation strategy to your innovation ecosystem, Harv. Bus.Rev. 84 (2006) 98–107.

[2] R. Adner, The Wide Lens: A New Strategy for Innovation, Portfolio/Penguin, NewYork, 2012.

[3] G. Ahuja, Collaboration networks, structural holes, and innovation: A longitudinalstudy, Adm. Sci. Q. 45 (2000) 425–455.

[4] A.L. Barabási, R. Albert, Emergence of scaling in random networks, Science 286(1999) 509–512.

[5] J. Barney, Firm resources and sustained competitive advantage, J. Manag. 17 (1991)99–120.

[6] R.C. Basole, Visualization of interfirm relations in a converging mobile ecosystem, J.Inf. Technol. 24 (2009) 144–159.

[7] R.C. Basole, Visual business ecosystem intelligence: Lessons from the field, IEEEComput. Graph. Appl. (2014) 26–34.

[8] R.C. Basole, M.A. Bellamy, Supply network structure, visibility, and risk diffusion: Acomputational approach, Decis. Sci. 45 (2014) 753–789.

[9] R.C. Basole, M.A. Bellamy, Visual analysis of supply network risks: Insights from theelectronics industry, Decis. Support. Syst. 67 (2014) 109–120.

[10] R.C. Basole, M.A. Bellamy, H. Park, Visualization of innovation in global supply chainnetworks, Decis. Sci. (2016) 1–40.

[11] R.C. Basole, T. Clear, M. Hu, H. Mehrotra, J. Stasko, Understanding interfirm relation-ships in business ecosystems with interactive visualization, IEEE Trans. Vis. Comput.Graph. 19 (2013) 2526–2535.

31R.C. Basole / Decision Support Systems 83 (2016) 22–31

[12] R.C. Basole, S. Ghosh, M. Hora, Association between supply networks properties andperformance: Evidence from the electronics industry, Working Paper, 2015.

[13] R.C. Basole, W.B. Rouse, Complexity of service value networks: Conceptualizationand empirical investigation, IBM Syst. J. 47 (2008) 53–70.

[14] R.C. Basole, W.B. Rouse, L.F. McGinnis, D.A. Bodner, W.C. Kessler, Models of complexenterprise networks, J. Enterp. Transform. 1 (2011) 208–230.

[15] M. Bastian, S. Heymann, M. Jacomy, Gephi: An open source software for exploringand manipulating networks, International AAAI conference on Weblogs and SocialMedia, 2009.

[16] M.A. Bellamy, R.C. Basole, Network analysis of supply chain systems: A systematicreview and future research, Syst. Eng. 16 (2013) 235–249.

[17] M.A. Bellamy, S. Ghosh, M. Hora, The influence of supply network structure on firminnovation, J. Oper. Manag. 32 (2014) 357–373.

[18] S.P. Borgatti, M.G. Everett, L.C. Freeman, UCINET for Windows: Software for socialnetwork analysis, 2002.

[19] S.P. Borgatti, P.C. Foster, The network paradigm in organizational research: A reviewand typology, J. Manag. 29 (2003) 991–1013.

[20] S.P. Borgatti, X. Li, On social network analysis in a supply chain context, J. SupplyChain Manag. 45 (2009) 5–22.

[21] K. Butner, The smarter supply chain of the future, Strateg. Leadersh. 38 (2010)22–31.

[22] H. Chen, R.H. Chiang, V.C. Storey, Business intelligence and analytics: From big datato big impact, MIS Q. 36 (2012) 1165–1188.

[23] H.W. Chesbrough, Open innovation: The new imperative for creating and profitingfrom technology, Harvard Business Press, 2003.

[24] T.Y. Choi, K.J. Dooley, M. Rungtusanatham, Supply networks and complex adaptivesystems: control versus emergence, J. Oper. Manag. 19 (2001) 351–366.

[25] N.M. Coe, P. Dicken, M. Hess, Global production networks: realizing the potential, J.Econ. Geogr. 8 (2008) 271–295.

[26] J.H. Dyer, H. Singh, The relational view: Cooperative strategy and sources of interor-ganizational competitive advantage, Acad. Manag. Rev. 23 (1998) 660–679.

[27] M.L. Fisher, What is the right supply chain for your product? Harv. Bus. Rev. 75(1997) 105–117.

[28] M.T. Frohlich, R. Westbrook, Arcs of integration: an international study of supplychain strategies, J. Oper. Manag. 19 (2001) 185–200.

[29] A. Grandori, An organizational assessment of interfirm coordination modes, Organ.Stud. 18 (1997) 897–925.

[30] A.C. Inkpen, Creating knowledge through collaboration. Calif. Manag. Rev. 39(1996).

[31] Y. Kim, T.Y. Choi, T. Yan, K. Dooley, Structural investigation of supply networks: A so-cial network analysis approach, J. Oper. Manag. 29 (2011) 194–211.

[32] R. Klein, A. Rai, Interfirm strategic information flows in logistics supply chain rela-tionships, MIS Q. (2009) 735–762.

[33] B. Kogut, U. Zander, Knowledge of the firm, combinative capabilities, and the repli-cation of technology, Organ. Sci. 3 (1992) 383–397.

[34] D. Lavie, L. Rosenkopf, Balancing exploration and exploitation in alliance formation,Acad. Manag. J. 49 (2006) 797–818.

[35] J.G. March, Exploration and exploitation in organizational learning, Organ. Sci. 2(1991) 71–87.

[36] M. Marra, W. Ho, J.S. Edwards, Supply chain knowledge management: A literaturereview, Expert Syst. Appl. 39 (2012) 6103–6110.

[37] T.P. Moliterno, D.M. Mahony, Network theory of organization: A multilevel ap-proach, J. Manag. 37 (2011) 443–467.

[38] A. Nair, J.M. Vidal, Supply network topology and robustness against disruptions - aninvestigation using multi-agent model, Int. J. Prod. Res. 49 (2010) 1391–1404.

[39] M. Newman, S. Strogatz, D. Watts, Random graphs with arbitrary degree distribu-tions and their applications, Phys. Rev. E 64 (2000).

[40] M.E. Newman, Modularity and community structure in networks, Proc. Natl. Acad.Sci. 103 (2006) 8577–8582.

[41] A. Oke, M. Idiagbon-Oke, F. Walumbwa, The relationship between brokers’ influ-ence, strength of ties and npd project outcomes in innovation-driven horizontalnetworks, J. Oper. Manag. 26 (2008) 571–589.

[42] S.D. Pathak, D.M. Dilts, S. Mahadevan, Investigating population and topological evo-lution in a complex adaptive supply network, J. Supply Chain Manag. 45 (2009)54–57.

[43] T. Ramakrishnan, M.C. Jones, A. Sidorova, Factors influencing business intelligence(BI) data collection strategies: An empirical investigation, Decis. Support. Syst. 52(2012) 486–496.

[44] L. Rosenkopf, M.A. Schilling, Comparing alliance network structure across industries:observations and explanations, Strateg. Entrep. J. 1 (2007) 191–209.

[45] W.B. Rouse, Enterprises as systems: Essential challenges and approaches to transfor-mation, Syst. Eng. 8 (2005) 138–150.

[46] W.B. Rouse, Complex engineered, organizational and natural systems, Syst. Eng. 10(2007) 260–271.

[47] V. Sambamurthy, A. Bharadwaj, V. Grover, Shaping agility through digital options:Reconceptualizing the role of information technology in contemporary firms, MISQ. (2003) 237–263.

[48] R.C. Sampson, Organizational choice in R&D alliances: Knowledge-based and trans-action cost perspectives, Manag. Decis. Econ. 25 (2004) 421–436.

[49] M.A. Schilling, Understanding the alliance data, Strateg. Manag. J. 30 (2009)233–260.

[50] M.A. Schilling, C.C. Phelps, Interfirm collaboration networks: The impact of large-scale network structure on firm innovation, Manag. Sci. 53 (2007) 1113–1126.

[51] N. Shin, K. Kraemer, J. Dedrick, R&D, value chain location and firm performance inthe global electronics industry, Ind. Innov. 16 (2009) 315–330.

[52] T.M. Simatupang, R. Sridharan, The collaborative supply chain, Int. J. Logist. Manag.13 (2002) 15–30.

[53] C.A. Soosay, P.W. Hyland, Driving innovation in logistics: case studies in distributioncentres, Creat. Innov. Manag. 13 (2004) 41–51.

[54] G. Stevens, Integrating the supply chain, Int. J. Phys. Distrib. Logist. Manag. 19 (1989)3–8.

[55] M. Subramani, How do suppliers benefit from information technology use in supplychain relationships? MIS Q. (2004) 45–73.

[56] A. Surana, S. Kumara, M. Greaves, U.N. Raghavan, Supply-chain networks: a complexadaptive systems perspective, Int. J. Prod. Res. 43 (2005) 4235–4265.

[57] E. Tufte, The Visual Display of Quantitative Information, Graphics Press, Cheshire,1983.

[58] E. Tufte, Envisioning Information, Graphics Press, Cheshire, 1990.[59] B. Uzzi, Social structure and competition in interfirm networks: The paradox of

embeddedness, Adm. Sci. Q. 42 (1997) 35–67.[60] O.E. Williamson, Markets and hierarchies: Some elementary considerations, Am.

Econ. Rev. 63 (1973) 316–325.[61] H. Yli-Renko, E. Autio, H.J. Sapienza, Social capital, knowledge acquisition, and

knowledge exploitation in young technology-based firms, Strateg. Manag. J. 22(2001) 587–613.

Rahul C. Basole is an Associate Professor in the School of Interactive Computing, the Asso-ciate Director for Enterprise Transformation in the Tennenbaum Institute/ IPaT, and an a_liated facultymember in the GVUCenter at theGeorgia Institute of Technology. He is also aVisiting Scholar in HSTAR at Stanford University and a Fellow of the Batten Institute at theDarden School of Business. He is also the Editor-in-Chief of the Journal of Enterprise Trans-formation. His research and teaching focuses on computational enterprise science, infor-mation visualization, and strategic decision support. His work has received numerousbest paper awards andhehas extensively published in leading computer science,manage-ment, and engineering journals. He holds a Ph.D. in industrial and systems engineeringfrom the Georgia Institute of Technology.