Contents lists available at ScienceDirect

Journal of Operations Management

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

Impact of supply base structural complexity on financial performance: Rolesof visible and not-so-visible characteristics

Guanyi Lua,∗, Guangzhi Shangb

a Oregon State University, USAb Florida State University, USA

A R T I C L E I N F O

Accepted by: Mikko Ketokivi

Keywords:Supply chain structureSupply chain complexitySupply baseEliminative and cooperative structural linksROATobin's Q

A B S T R A C T

Supply chains have become increasingly complex in the last decade, which makes their structural characteristicsimportant determinants of firm performance. Prior studies on supply chain structure have largely emphasizednetwork-level attributes but ignored supply-base level characteristics. However, in many cases it is the 1st tiersuppliers, not those “deep in the network,” that have most immediate influence on the buyer. In addition, somestructural characteristics, such as direct links between the buyer's suppliers and its customers, are not-so-visibleto the buyer, yet can impact its financial performance dramatically. The existing literature has overlooked thesenot-so-visible structural links. Using objective supply chain data collected fromMergent Online and Compustat, wemap the supply base structure of 867 public firms. We construct three visible (horizontal, vertical and spatial)and two not-so-visible (eliminative and cooperative) structural complexity metrics, and examine their impactson buyer firms' financial performance as measured by Return on Assets and Tobin's Q. Our empirical analysisshows that the five dimensions have differential effects: some have negligible impacts while others appear tostrongly affect financial performance. Contrary to the common belief that complexity hurts performance, we findthat an individual complexity dimension may have both positive and negative effects, and the overall effect maybe non-linear.

1. Introduction

“Supply chains are growing increasingly complex, making themharder to manage, operate, and change in response to customer,competitive, and financial shifts.” – Wilson Perumal and Company,2015.

Steinhilper et al. (2012) find that costs caused by supply chaincomplexity account for up to 25% of manufacturing firms’ total ex-penditure. Supply chain complexity impedes decision-making (Manujand Sahin, 2011), fertilizes disruptions (Chopra and Sodhi, 2014) anderodes plant level operational efficiency (Bozarth et al., 2009). Despitethese disadvantages, a general consensus among practitioners andacademics is that supply chains have become increasingly complex overthe last decades—with little relief in sight—owing to increasingly so-phisticated customer requirements (Bode and Wagner, 2015; KPMG,2011). As a result, the structural complexity characteristics of supplychains have become important determinants of firm performance (Kim,2014). A study by A.T. Kearney (2007) indicates that firms can increaseearnings by 3%–5% if they can make improvements based on supplychain structure. Supplier management now involves more than just

building mutually beneficial, long-term relationships. It also requires anin-depth understanding of the structural complexity of globally inter-connected supply chains (Kim et al., 2015).

Supply chain structure has garnered much interest from OperationsManagement (OM) scholars. Prior studies have largely adopted a socialnetwork perspective to understand the network-level attributes of in-terconnected firms and their influences (Bellamy et al., 2014; Kim et al.,2011). While overall supply network structure is important (Kim et al.,2015), a more nuanced understanding of the supply base structure is alsoimperative. A supply base largely consists of 1st tier suppliers directlyconnected to the focal buyer. Overall network structure emerges withno single firm deliberately orchestrating its exact shape (Choi andHong, 2002). But while a supplier “deep in the network” may affect thebuyer, in many cases it is the supply base that more directly andstrongly influences performance (Wilhelm et al., 2016). As Sivadasanet al. (2002, p.80) observed, “in a dynamic environment such as asupply chain, even basic supplier-customer systems, with structurallysimple information and material flow formation, have a tendency toexhibit operational complexity” and eventually impact buyers' financialperformance (Manuj and Sahin, 2011). A central challenge for

http://dx.doi.org/10.1016/j.jom.2017.10.001Received 21 March 2016; Received in revised form 4 August 2017; Accepted 6 October 2017

∗ Corresponding author.E-mail address: guanyi.lu@oregonstate.edu (G. Lu).

Journal of Operations Management 53–56 (2017) 23–44

Available online 29 October 20170272-6963/ © 2017 Elsevier B.V. All rights reserved.

T

advancing supply chain structure research, therefore, is to show howand why supply base structural characteristics influence buyers’ fi-nancial performance.

In addition, some structural links are not-so-visible to the buyer, yetimpact its financial performance dramatically. For example, a suppliercan sell directly to the buyer's customers, with the potential to replace it(Rossetti and Choi, 2005). Prior empirical studies have investigated thevisible structural links (i.e., the links that connect to the buyer). Yet theneglect of not-so-visible structural links masks information critical tosupply chain management decisions. Furthermore, the visible and not-so-visible links may be related. For instance, the potential eliminationthreat posed by supplier-customer links (a not-so-visible factor) is likelyto be minimized when suppliers reside in geographically dispersed lo-cations (a visible complexity measure). Thus, our motivation is toconstruct a comprehensive set of supply base structural metrics andanswer the research question: how do the characteristics of supply basestructural complexity affect the buyer's financial performance?

Utilizing a proprietary objective dataset compiled from two datasources—Mergent Online and Compustat, we map the supply basestructure of 867 public firms and construct two sets of structuralcomplexity metrics (specifically, visible and not-so-visible, see details inSection 2). We empirically examine the individual effects of thesecomplexity dimensions on the buyer's financial performance. Contraryto the literature, which states that supply chain complexity hurts firmperformance (Bozarth et al., 2009; Bode and Wagner, 2015), we pro-pose that the effects of complexity dimensions at the supply base levelare complicated and mixed. An individual dimension may have bothpositive and negative effects and the overall effect is contingent on themagnitude of the complexity dimension itself. We find that somecomplexity dimensions reveal a nonlinear (U-shaped or inverted-U)relation with the buyer's financial performance. In addition, thesecomplexity dimensions exhibit differential effects; some wield con-siderably stronger impacts than others.

This study makes two major theoretical contributions. First, it expandsour understanding of supply chain structure by channeling focus from thebroad network level to the more nuanced supply base level. The supplybase has stronger and more immediate performance impacts than the restof the supply network due to its “proximity” to the buyer (Wilhelm et al.,2016). Second, by also emphasizing the not-so-visible structural linkswhere the buyer is generally not directly involved, our study extends theconceptualization of supply base complexity and provides a more com-prehensive set of structural dimensions. Understanding the impacts ofthese dimensions is critical because the buyer is likely to influence only itsdirect links (Bode and Wagner, 2015). As a result, this study addresses aresonant theme within the supply chain structure research: showing how afirm should manage structural characteristics separately, with the poten-tial to mitigate the negative impacts of complexity dimensions it cannotdirectly control. Our study also carries a methodological implication. Al-most all studies on supply chain structure rely on information from thebuyer—primarily survey and qualitative data—to measure structuralcomplexity. However, the use of data solely from the buyer risks over-looking the impact of structural links of which the focal firm is unaware.We overcome this problem by constructing objective measures frombuyer-supplier links identified by independent third parties.

The rest of this article is organized as follows: Section 2 reviews therelated literature. Section 3 proposes the theoretical framework and de-velops research hypotheses. Section 4 discusses data source and variableconstruction. Section 5 depicts methods and reports results. Section 6concludes the paper with a discussion on contributions and limitations.

2. Literature review

2.1. Complexity in supply chains

The concept of complexity has triggered research in multiple aca-demic disciplines. It generally pertains to system-level attributes about

connections among system constituents. In social science, Simon (1962,p.468) offers an influential definition that a system is complex if it is“made up of a large number of parts that interact in a non-simple way.”This definition highlights two critical traits of complexity: structure andbehavior (Perrow, 1984; Senge, 2006). According to Bode and Wagner(2015, p.216), the former “is often termed structural complexity (alsostatic or detail complexity) and refers to the number and variety ofelements defining the system.” The latter is often labeled “dynamiccomplexity,” referring to the interactions of those elements. The twotraits are usually interrelated in practice. A large number of elementsimplies a great number of possible interactions, which is especially truewhen connected firms jointly assemble a final product (Bozarth et al.,2009; Manuj and Sahin, 2011).

Prior studies on supply chain complexity have capitalized on bothtraits and viewed complexity as a multi-dimensional concept. For in-stance, Vachon and Klassen (2002) propose two dimensions: un-certainty (which is associated with structure, i.e., the number of con-stituents), and complicatedness (associated with behavior, i.e.,interaction among constituents). Choi and Krause (2006) identify threedimensions: the number of direct suppliers (structure), differentiationamong direct suppliers (structure), and the relationships among thesuppliers (behavior). Bozarth et al. (2009) also propose three: internalmanufacturing complexity, downstream complexity and upstreamcomplexity. They explicitly state that each of their complexity dimen-sions can be characterized as both structural and behavioral. Whileearly studies provide insights into supply chain complexity, researchershave failed to achieve a consensus about which dimensions best de-scribe supply chain complexity, partly due to their different foci (Jacobsand Swink, 2011; Manuj and Sahin, 2011). Our study focuses onstructural complexity, because it is explicitly measured by buyer-sup-plier relationships. These relationships also reflect the interactions be-tween firms. However, we note that interactions are difficult, if notimpossible, to capture fully and objectively. The scope of this study isthus decidedly restricted to the structural complexity of a firm's supplybase with a particular focus on 1st tier suppliers.

Table 1 summarizes the most relevant studies. As we noted earlier,the small number of studies examining supply chain structural attri-butes largely emphasize network-level measures that link relational tiesto performance metrics such as firm innovation, social capital and re-source access. For example, Bellamy et al. (2014) demonstrate thatsupply network accessibility is significantly associated with innovation.Kim et al. (2011) investigate the supply networks of three automobileproduct lines (Honda Accord, Acura CL/TL and DaimlerChrysler GrandCherokee) and show how network centrality and density affect materialflow and contractual relationships. In network studies, the network“position” matters. According to the social network theory, firms oc-cupying a “central” network position (as manifested by measures suchas in-bound centrality, out-bound centrality, and network accessibility)will outperform competitors due to superior access to resources. Notethat in a network, a buyer does not necessarily have “position” ad-vantage over its suppliers because a supplier can have the same or evenhigher level of centrality (or other network measures) than its buyer. Incontrast, at the supply base level, the buyer naturally occupies thecentral position. In a supply network, position is a firm attribute suchthat a number of firms may share similar position advantages. In asupply base, what matters more is the link attribute, i.e., which parties(e.g., two suppliers, or a supplier and a customer) are linked. Thus, howthe focal firm connects to its suppliers and customers and how theyconnect with each other have strong performance implications. Com-pared with existing network measures which reflect a firm's positionrelative to others, our supply base measures capture how links are“distributed” within a supply base. While the kernels of social networktheory can still be used in supply base research, its measures likelycannot. Among the studies we reviewed, only Bode and Wagner (2015)use “lower-than-network” level measures to study upstream supplychain disruptions. The lack of research on complexity at the supply base

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

24

level (Wilhelm et al., 2016) is surprising given that related literaturehas repeatedly argued that supply base characteristics affect costs, in-novation, and supply chain risks and responsiveness (Choi and Krause,2006).

2.2. Supply base structural complexity metrics and their impacts

The conceptual literature suggests that five dimensions describe thestructural characteristics at the supply base level: three visible (hor-izontal, vertical, and spatial complexity) and two not-so-visible (elim-inative and cooperative complexity) (Bode and Wagner, 2015; Pathaket al., 2014; Rossetti and Choi, 2005). We note that the level of visibilityof the two sets of measures is relative rather than absolute. Table 2summarizes the definitions of these key concepts and their supportingliterature.

The visible dimensions refer to the attributes of structural links thatdirectly involve the focal firm. Early studies propose that the “width” ofthe supply base is one of its important attributes (Perrow, 1984; Senge,2006). This metric has been generally operationalized via the numberof suppliers directly connected to the focal buyer (Choi and Krause,2006) and labeled as horizontal complexity (Bode and Wagner, 2015).We improve this measure by incorporating product groups (see Section4.3).

Also relevant is the “depth” of the supply base, which captures thehierarchical level and is referred to as vertical complexity (Tolbert andHall, 2009). Conceptually, hierarchy can be defined as the number oftiers in a supply network (Blackhurst et al., 2005), but its meaning isunclear in a largely single-tier supply base (Choi and Hong, 2002).Additionally, identifying the tier in which a firm resides is difficult inmany circumstances. For example, Samsung is highly vertically

Table 1Relevant literature.

Study Data Structure related metrics Key findings or propositions

EmpiricalBellamy et al.,

2014Objective data from: the Electronics Business 300listings, the Connexiti database and the ThomsonReuters SDC Platinum Joint Ventures/Alliancesdatabase (final n = 390)

Network level:

• supply network accessibility

• supply networkinterconnectedness

Supply network accessibility is significantly associated withinnovation performance. Interconnectedness of supplynetworks strengthens the association between networkaccessibility and innovation performance. The influence of thetwo characteristics on innovation is enhanced by a firm'sabsorptive capacity as well as the supply network partners'innovativeness.

Bode andWagner,2015

Survey data of 396 firms from Germany, Austria, andSwitzerland

Supply chain level:

• horizontal complexity

• vertical complexity•spatial complexity

All three complexity drivers are positively associated with thefrequency of disruptions. The drivers also interact and intensifyeach other's impacts in a synergistic fashion.

Choi and Hong,2002

Case study (supply networks of three auto productlines)

Network level:

• formalization

• centralization

• complexity

The three structural metrics significantly affect one another.Cost consideration appears to be the predominant force thatshapes the structure of supply network.

Gao et al., 2015 Survey data of 202 Chinese firms Network level:

• buyer-supplier relationalstrength

• supply network density

Novel information sharing partially mediates the effect oftechnological diversity in supplier network on buyer firms' newproduct creativity. The effect of technological diversity isenhanced by buyer–supplier relational strength but inhibitedby supplier network density.

Kim et al., 2011 The same data used in Choi and Hong (2002) butconverted into social network analysis metrics

Network level:

• network degree centrality

• network closeness centrality

• network betweenness centrality

Applying social network analysis yields results thatcomplement with the findings reported in Choi and Hong(2002). For instance, Choi and Hong (2002) report the twosupply networks of Honda are more centralized than DCX's.However, social network analysis advises the opposite.

ConceptualKim et al., 2015 N/A Network level:

• block-diagonal

• Scale-free

• centralized

• diagonal

Defines supply network disruption and differentiate betweendisruptions at the node/arc level vis-a-vis network level. Thestudy shows that node/arc-level disruptions do not necessarilycause network-level disruptions. Different structuralrelationships among firms have varying levels of resilience.

Choi and Krause,2006

N/A Supply base level:

• the number of suppliers

• the degree of differentiationamong these suppliers

• the level of inter-relationshipsamong the suppliers

Formulates four propositions about how supply basecomplexity dimensions impact the four major SCM researchareas—“transaction costs, supply risk, supplier responsiveness,and supplier innovation.”

Pathak et al.,2014

N/A Network level:

• community supply network

• federal supply network

• consortium supply network

• hierarchical supply network

Employs the interrelated dimensions of ties between firms andnetwork-level entity, and governance to postulate four networkarchetypes and then describe how co-opetitive relationshipsmay evolve in these network archetypes.

Vachon andKlassen, 2002

N/A Network level:

• technological dimension of thesupply chain

• information processingdimension of complexity

The two dimensions furnish a two-by-two matrix that definessupply chain complexity and provides a theoretical foundationfor relating different dimensions of supply chain complexity todelivery performance.

Yan et al., 2014 N/A Network level:

• operational nexus supplier

• monopolistic nexus supplier

• informational nexus supplier

Offers a theory of a concept dubbed “the nexus supplier”. Thetheory is furnished in the form of typologies, which describethree ideal types of nexus suppliers and how they affect abuyer's operational performance.

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

25

integrated. It manufactures LCD touch screens, hard drives, and CPUsfor its smartphones. If we use the number of tiers as a measure forhierarchy, Samsung will have fewer tiers in its supply chain than its lessvertically-integrated competitors (such as Apple) who purchase mate-rials from outside suppliers. However, conceptually Samsung and Appleshould share the same hierarchy level (i.e., both are 1st tier supplier toretailers such as Best Buy). One possible remedy is discussed in Choiand Krause (2006), who suggest that the average number of lower-tiersuppliers a firm's 1st tier suppliers have can be used as a proxy for itssupply base's hierarchy.

The third visible structural dimension is spatial complexity, whichmeasures the geographical expanse of the supply base (Vachon andKlassen, 2002). Spatial complexity is generally measured by the numberof countries or regions that a firm's suppliers span (von Corswant andFredriksson, 2002). On one hand, geographic reach allows access toglobal knowledge as well as cheaper materials and labor (Gilley andRasheed, 2000). On the other hand, wide-spread supply bases face moredifficulties in coordinating production due to higher complicatedness(Vachon and Klassen, 2002).

The not-so-visible dimensions capture the attributes of structurallinks in which the focal firm is generally not directly involved, but atleast one of its 1st tier suppliers is. Despite the importance and re-levance of these not-so-visible links (Rossetti and Choi, 2005; Pathaket al., 2014), the few empirical studies have largely focused on visiblelinks.

A 1st tier supplier can link itself to the focal buyer's customer(s) orother 1st tier supplier(s). The “not-so-visible” (rather than “invisible”)label comes from the fact that such links may be initiated by the buyer(e.g., buyer asks the supplier to offer maintenance services or partsdirectly to its customer or asks one supplier to purchase from another).Rossetti and Choi (2005) refer to a direct link between a supplier and itsbuyers' customers as “supply chain disintermediation” as the linkweakens the buyer's competitive position (Li and Choi, 2009). Since anextreme case might see a supplier replacing the buyer and usurping itscustomers, we dub this dimension eliminative complexity. One well-known example of this transformation is the Taiwan-based HTC. As acell phone OEM in the 1990s, HTC launched its own brand in 2002 andbecame one of the top five smartphone brands in the world (Wan andWu, 2016).

A supplier can also link itself to peers in the supply base. Wilhelm(2011) shows that links between suppliers can improve cooperation andsubsequently affect the buyer's performance. Pathak et al. (2014, p.255) concur and explicitly state “a cooperative relationship betweentwo companies refers to a direct link between two firms.” Thus, we callthis structural dimension cooperative complexity. Since cooperative

complexity can also have negative impacts, such as those arise fromsupplier collusion, we consider both pros and cons of the connectionsbetween suppliers.

The literature seldom accounts for the differential effects of struc-tural complexity dimensions. For instance, eliminative complexity cantrigger a crisis of survival, whereas increased horizontal complexitygenerally brings an increase in communication costs. Managers ought totailor their efforts to address the specific challenges different dimen-sions impose. Furthermore, the literature focuses on operational per-formance and implicitly assumes linear relationships (e.g. Bode andWagner, 2015). However, the effects of complexity dimensions on fi-nancial performance are more complicated. For example, as the numberof suppliers (captured by horizontal complexity) increases, deliverycoordination is likely to increase. However, more number of suppliermay also indicate higher redundancy in the supply base which wouldresult in lower risk of supply disruptions. Our review also reveals thescarcity of empirical inquiries in the area. A large fraction of priorstudies are conceptual in nature (see Table 1). This reflects the difficultyof collecting supply chain structure information and the challenge ofoperationalizing structural measures (Kim et al., 2011). More im-portantly, almost all data employed in the few existing empirical in-vestigations are collected solely from the buyer (i.e., Bode and Wagner,2015; Gao et al., 2015; Choi and Hong, 2002; Kim et al., 2011; Kim,2014). While Bellamy et al. (2014) also use objective third-party data,the focus and insights of our study are substantially different due to thelevel of analysis (network vs. supply base).

To summarize, very few studies have examined a comprehensive setof structural complexity dimensions at the supply base level. Our studyof supply base structural complexity dimensions based on objectivedata fills these gaps.

3. Theoretical framework and hypothesis development

3.1. Theoretical framework

Concepts from three theoretical perspectives provide the foundationto link changes in supply base complexity to financial performance:transaction cost economics (TCE), knowledge-based view (KBV) andsocial network theory (SNT). We briefly elaborate below.

TCE proposes that difficulties emerging from market-based ex-changes generate transaction costs (e.g., negotiation and contracting,monitoring and enforcement, and resolution when dissent occurs) andrelates cost efficiency to organization choices (Williamson, 1975;Grover and Malhotra, 2003). It has been widely applied to explore re-lated topics such as strategic alliance formation (Parkhe, 1993),

Table 2Supply base structural complexity dimensions.

Definition Supporting literature

Supply base 1st tier suppliers that are actively managed by the focal firm Choi and Hong, 2002; Choi and Krause, 2006

Structural link A buyer-supplier relation between two firms (a link indicates product flowbetween the two firms)

Pathak et al., 2014

Supply base structural link A buyer-supplier relation involving at least one of the 1st tier suppliers or the focalfirm

Pathak et al., 2014; Wilhelm, 2011

Supply base structural complexityHorizontal complexity The breadth level (width) of the supply base Perrow, 1984; Choi and Hong, 2002; Tolbert and Hall, 2009;

Senge, 2006Vertical complexity The hierarchical level (depth) of the supply base Choi and Hong, 2002; Choi and Krause, 2006; Bode and Wagner,

2015Spatial complexity The level of geographical spread of the supply base Choi and Hong, 2002; Bode and Wagner, 2015; Vachon and

Klassen, 2002.Eliminative complexity The level of connection between the 1st tier suppliers and the focal buyer's

customersRossetti and Choi, 2005

Cooperative complexity The level of connection between the 1st tier suppliers within the supply base Pathak et al., 2014; Wilhelm, 2011

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

26

offshore outsourcing (Ellram et al., 2008) and vertical integration(Rabinovich et al., 2007). Two elements of TCE are particularly re-levant to our study: bounded rationality and uncertainty. Bounded ra-tionality states that firms have limited memories and cognitive proces-sing power. They cannot assimilate all the information at their disposal.Despite knowing all the “rules” by which competition operates, firmsstill cannot reliably foresee all the possible results. Bounded rationalityhas serious cost implications, as the actions of suppliers are sometimesunpredictable, due in part to their self-interest and in part to the in-herent dissimilarity between firms in “internal culture, resources andmotivation for supply chain membership” (Flynn et al., 2016, p. 5). Thestructure of the supply base itself can be inherently complex (Sivadasanet al., 2002), and depending on those characteristics and the resultantinformation processing needs, some firms may manage transactioncosts effectively while others simply pursue a “good enough” (probablysuboptimal) solution.

Uncertainty reflects the difficulty of predicting transaction out-comes. Flynn et al. (2016) propose three types of uncertainty in asupply base context: micro-level variation, meso-level informationshortage, and macro-level equivocality. Micro-level variation focuseson the “the mathematical aspects of uncertainty” (Duncan, 1972, p.317) and is primarily rooted in the unpredictability of task execution(Tushman and Nadler, 1978), due to variation in the flow of goods andinformation within a supply base (Germain et al., 2008). Variation isassociated with cost (Flynn et al., 2016). For instance, high variation indelivery lead-time can increase operational costs, as the buyer is forcedto adjust its production plan. Meso-level information shortage arisesfrom individual supply chain members, who might withhold informa-tion (Rabinovich et al., 2007; Shrivastava and Mitroff, 1984) in theirown interest. Meso-level uncertainty is common in supply bases. Forexample, assessing suppliers is routinely done with incomplete in-formation (Wu and Barnes, 2012). The bullwhip effect (Lee et al., 1997)illustrates how information shortage has caused firms to hold extrainventory to compensate for meso-level uncertainty. Macro-level equi-vocality is associated with ambiguous and ill-structured situations. Thebuyer may possess a wealth of information, but it is difficult to knowwhich information is needed or how to interpret it. Economic varia-bility, market fluctuations, regulatory inconsistency and technologicalturbulence are among the primary sources of macro-level uncertainty(Beckman et al., 2004; Germain et al., 2008). Macro-level uncertainty isdifficult to understand, yet it impacts supply base operations pro-foundly. Facing vague information, managers replace “maximum effi-ciency” criteria with a call for “satisfactory” or “sufficient” performance(Tiwana et al., 2007), due to their bounded rationality (Simon, 1979;Thompson, 1967). Increases in any type of uncertainty likely result inhigher transaction costs; nonetheless macro- and meso-level un-certainties tend to have greater impact on firm performance becausethey are often intertwined with strategic decisions while micro-leveluncertainty is associated with tactical operations and is more “pre-dictable” (Flynn et al., 2016).

Despite its comprehensiveness, TCE alone may not fully account forthe effects of all complexity dimensions. Also relevant is the knowledge-based view, which places knowledge among a firm's most importantstrategic resources (Kogut and Zander, 1992; Felin and Hesterly, 2007).KBV suggests that because knowledge is embedded within firms and isusually difficult to imitate, access to heterogeneous knowledge bases is amajor determinant of sustained performance. Supply base complexity isrelated to knowledge-based resources in two ways. First, it affects thechannels through which these knowledge-based resources can be ac-quired and shared. Links among the buyer, its suppliers, and its cus-tomers influence the access to knowledge. Second, it affects the het-erogeneity of knowledge. This is particularly related to spatial andvertical complexity because knowledge asymmetries likely exist amonggeographically or hierarchically distant firms. For example, the litera-ture on supply chain collaboration shows that idiosyncratic knowledgefrom upstream suppliers enhances buyer performance (see a review by

Flynn et al., 2010). Relying on suppliers' knowledge allows a buyer tofocus on “what it does best” (a.k.a. “core competence”, Prahalad andHamel, 1990). Relatedly, the mainstream economic perspective sug-gests that the hierarchical level of a supply chain can be viewed as aproxy for the level of specialization at each tier (Williamson, 1975). Themore tiers, the more likely suppliers at each tier are highly specialized.According to Prahalad and Hamel (1990), specialized suppliers oftenown unique knowledge that enables them to provide good qualityproducts at low prices.

Social network theory also adds insights to our discussion (Ahuja,2000). SNT asserts that firms occupying a central network position arelikely to achieve better performance due to better access to informa-tion/resources. More importantly, unlike TCE and KBV, SNT highlightsthe effect of lack of connection (i.e., structural holes) on buyer per-formance (Burt, 1992). A buyer who bridges otherwise disconnectedparties (holding a structural hole position) enjoys brokerage opportu-nities. A hole suggests that firms on either side have access to distinctinformation (Hargadon and Sutton, 1997). Thus, a structural hole re-duces information redundancy and maximizes resource-sharing benefitsfor the buyer. In addition, such benefits to the buyer vary noticeablywith the level of connection between its suppliers and customers(eliminative complexity) or between suppliers within the supply base(cooperative complexity).

Incorporating the three theoretical perspectives and the broad lit-erature on supply base management, we next articulate how supplybase structural dimensions affect buyer financial performance as man-ifested by return on assets (ROA) and Tobin's Q (TQ). ROA measures afirm's short-run, tangible financial performance and has been widelystudied (Mackelprang et al., 2015; Wagner et al., 2012). In addition, weare also interested in investigating whether a firm's supply base com-plexity might attract the attention of its investors and the stock market,and hence affect the long-term, intangible aspects of financial perfor-mance, which is captured by TQ (Chung and Pruitt, 1994; Villalonga,2004). The inclusion of both allows us to observe if there are differ-ential impacts of supply base complexity on short-term and long-termfinancial performance.

3.2. Hypotheses

Horizontal complexity refers to the “width” of the supply base (Bodeand Wagner, 2015; Perrow, 1984; Senge, 2006). Modern manufacturersusually contract a number of suppliers for the same product group inorder to mitigate risk. Such a strategy reduces supplier dependence. Italso allows the buyer to transfer part of the risk of task variability(Harrigan, 1985) to its immediate suppliers in a technologically tur-bulent environment (Holcomb and Hitt, 2007). Aside from the benefitsof risk mitigation, multi-sourcing often leads to cost advantages. Spe-cialized suppliers may be better at achieving cost efficiencies that aredifficult for the buyer to attain (Holcomb and Hitt, 2007). Outsourcingtasks to specialized suppliers is positively associated with financialperformance for firms adopting a cost-leader strategy (Gilley andRasheed, 2000). The literature also shows that order splitting (betweensuppliers) can significantly reduce the buyer's safety stock and in-ventory-related costs (Kelle and Silver, 1990). Although communicationand monitoring costs will increase with the number of suppliers, thebenefits are likely to outweigh these costs as long as the number ofsuppliers is not too large to manage (Holcomb and Hitt, 2007). Thebuyer's financial performance will improve as the horizontal complexityincreases due to risk mitigation and supplier specialization.

However, at high levels of horizontal complexity, bounded man-agerial efforts constrain the buyer's ability to maintain exchange re-lationships and monitor supplier behaviors (Pilling et al., 1994). Studiessuggest that an increase in the number of suppliers is significantly as-sociated with the difficulty of communicating requirements and ob-taining consistent inputs (Vachon and Klassen, 2002). Indeed, as thenumber of suppliers continues to increase, the buyer tends to encounter

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

27

more uncertainties. Bode and Wagner (2015) show that micro-leveluncertainty (e.g., delivery failure) increases with the number of sup-pliers because no supplier is perfectly reliable (Babich, 2006;Chaturvedi and Martínez-de-Albéniz, 2011). Meso-level uncertaintymay also grow. A large number of suppliers makes it difficult to monitorsupplier behavior. Lack of accurate and timely information prevents thebuyer from making appropriate production decisions (Flynn et al.,2016). Macro-level uncertainties rise as well. The buyer's informationcomes from suppliers with conflicting interests (competing for a greatershare of orders), and is thereby potentially ill-structured. The buyingfirm may receive conflicting information, and its bounded rationalitymay curb its ability to tell fact from fiction. All these issues have ad-verse effects on cost efficiency and financial performance. Krause andHandfield (1999) observe that one firm in the electronics industry onceoutsourced aggressively but then reduced its supply base to only about40 suppliers in order to maximize financial performance. Thus, beyonda certain level of horizontal complexity, we expect this relationship tobe negative. Specifically, we propose:

H1a. Horizontal complexity will have a non-linear (inverted-U) effecton the focal buyer's ROA, with the slope positive at low levels ofhorizontal complexity but negative at high levels of horizontalcomplexity.

H1b. Horizontal complexity will have a non-linear (inverted-U) effecton the focal buyer's Tobin's Q, with the slope positive at low levels ofhorizontal complexity but negative at high levels of horizontalcomplexity.

Vertical complexity reflects the hierarchical level (the “depth”) ofthe supply base (Blackhurst et al., 2005; Tolbert and Hall, 2009). TheOM literature suggests that supply chain partners at different tiers carryout interdependent tasks in order to accomplish the manufacturingprocess (Khurana, 1999). The existence of hierarchy may benefit thebuyer in three ways. First, it leads to cost benefits. As discussed earlier,vertical complexity reflects the level of specialization of a manu-facturing process from an economic perspective. Specialized firms maybe able to pool uncorrelated customer demands to achieve smootherproduction and greater economies of scale. Buyers can benefit bysourcing from these firms rather than producing in house. Second,supply chain hierarchy protects firms from technological obsolescence(Balakrishnan and Wernerfelt, 1986). Irreversible investment in tech-nology is less valuable in a technologically volatile environment(Balakrishnan and Wernerfelt, 1986; Williamson, 1975), inasmuch asevolution quickly renders “new” technologies obsolete (Barney, 1991).Buyers with a shallow supply base likely have to invest in multipletechnologies that are necessary to succeed in the industry and thereforebear high financial risks. Far-reaching buyers, on the other hand, canshare the burden, since suppliers at each tier may possess uniquetechnology and knowledge that reduce the buyer's need to own it(Prahalad and Hamel, 1990). Third, supply chain collaboration litera-ture suggests that unique knowledge from upstream suppliers enhancesbusiness performance (Flynn et al., 2010). A far-reaching buyer canbetter incorporate upstream knowledge into product development andreduce production cost. As such, we expect that vertical complexity ispositively associated with the buyer's financial performance.

Conversely, an excessively hierarchical process could have a nega-tive impact on performance. Greater vertical complexity carries with itthe struggle to specify product attributes and integrate process stageswith supply chain partners. Take the pharmaceutical industry, for ex-ample. An upstream supplier's error can dramatically hurt end-productquality. The deeper the hierarchy, the stronger the negative impact ofsuch an event (Khurana, 1999). The buyer suffers more failures when itssupply base's vertical complexity is too high (Bode and Wagner, 2015).Moreover, information asymmetry is likely to increase as verticalcomplexity increases (Wang et al., 2017). Consequently, communica-tion and monitoring costs can offset and even exceed the savings due tosupplier specialization. Finally, from the perspective of complicatedness

(Vachon and Klassen, 2002), more tiers in a supply chain imply aprofusion of constituents interacting in even more ways. It follows thatthe information ambiguity (macro-level uncertainty) will be muchhigher in such a supply base. Firms will have to spend resources bar-gaining over issues such as profit and joint decisions, dropping thevalue of vertical complexity noticeably. Taken together, we expect firmperformance to increase with vertical complexity up to a point and thendecrease past that point.

H2a. Vertical complexity will have a non-linear (inverted-U) effect onthe focal buyer's ROA, with the slope positive at low levels of verticalcomplexity but negative at high levels of vertical complexity.

H2b. Vertical complexity will have a non-linear (inverted-U) effect onthe focal buyer's Tobin's Q, with the slope positive at low levels ofvertical complexity but negative at high levels of vertical complexity.

Spatial complexity gauges the geographical spread of the supplybase (Vachon and Klassen, 2002) and is mainly related to offshoresourcing. There are three major benefits of a widely-spread supply base.First, firms with geographically disperse supply bases get access tolower-priced materials and labor and world-class manufacturing cap-abilities (Bode and Wagner, 2015; Gilley and Rasheed, 2000; KPMG,2011). OEMs in relatively under-developed countries help buyers re-duce fixed investment in production and thus lower the breakevenpoint, which subsequently boosts financial performance. Second,knowledge asymmetries between locations (Lahiri, 2010) allow buyersto leverage a global knowledge base. Each location contributes idio-syncratic manufacturing know-how as a result of collaboration betweenfirms in different locations (Almeida and Kogut, 1999). Last but notleast, from the core competency perspective (Prahalad and Hamel,1990; Kotabe and Murray, 2004), offshore sourcing frees up organiza-tional resources for emerging needs or to develop important firm cap-abilities (Hamel et al., 1989). The continuous rise of global trade inrecent decades seems to attest to the positive contribution of a geo-graphically extended supply base.

Geographical dispersion of a supply base also has its drawbacks.First, geographically proximate suppliers are likely to be more homo-genous than spread-out suppliers in areas with different degrees ofeconomic development (Handfield and Nichols, 1999). Such homo-geneity leads to consistent inputs and thus reduces the buyer's manu-facturing costs in dealing with quality problems (Vachon and Klassen,2002). On the opposite end of the spectrum, it is hard for geo-graphically scattered suppliers to supply homogenous inputs, owing totheir dissimilarities in technology, culture (western vis-à-vis eastern)and local environments. Second, firms with a large number of foreignsuppliers have to comply with a myriad of foreign trade regulations,ranging from anti-corruption laws to import-export controls. The coststo adapt to these business rules can greatly affect the buyer's operationsand hurt its financial performance. Third, a geographically-stretchedsupply base may pose logistics challenges due to unpredictable eventssuch as the 2011 Japanese tsunami or more recent Middle East politicalturmoil. Dealing with these low-probability, high-impact accidents mayweaken the buyer's ability to reap the benefits of offshore sourcing.

Thus, geographical dispersion presents both opportunities andchallenges. Based on the KBV and prior supply chain studies, we arguethat the benefits of spatial complexity will outweigh costs at high levels.First, unlike horizontal and vertical complexity, spatial complexity af-fects financial performance mainly by exposing the buyer to differentresource locations (Ang and Inkpen, 2008). Geographic dispersion en-ables interactions between experts from different regions, which allowstransfer of tacit knowledge and local manufacturing know-how(Audretsch and Feldman, 1996). In other words, geographic colla-boration facilitates knowledge spillover (moderation effect). The greaterthe distribution of supplier locations, the greater the likelihood of ac-cruing knowledge (Lahiri, 2010). The supply chain literature concursand suggests that firm competitiveness is partly determined by how thefirm can effectively utilize global resources. Although newly acquired

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

28

knowledge may not have value in that local unit, it may find applicationin another location. A firm with a widely-spread supply base is morelikely to understand and apply new knowledge for problem solving.Second, the benefits of a global knowledge base and access to cheaplabor and materials are likely to increase exponentially with spatialcomplexity owing to potential optimal orchestration of global re-sources. For example, Procter & Gamble once limited its offshore sour-cing, but has outsourced about half of its R & D activities to globalpartners. This resulted in a 60% increase in innovation productivity andover $10 billion in revenue (Forbes, 2010). As Kotabe and Murray ex-plicitly state, “many companies, regardless of their nationality, thathave a limited scope of global sourcing are at a disadvantage over thosethat exploit it to the fullest extent in a globally competitive market-place” (2004, p.9). Taking into account both the challenges and bene-fits, we conjecture that firms probably would not be able to fully re-cognize the value of geographical dispersion at low levels of spatialcomplexity. Thus, the costs may outweigh benefits up to a certain pointbut the positive impact of effective use of global resources will overridethe costs beyond that point.

H3a. Spatial complexity will have a non-linear (U-shaped) effect on thefocal buyer's ROA, with the slope negative at low levels of spatialcomplexity but positive at high levels of spatial complexity.

H3b. Spatial complexity will have a non-linear (U-shaped) effect on thefocal buyer's Tobin's Q, with the slope negative at low levels of spatialcomplexity but positive at high levels of spatial complexity.

Eliminative complexity refers to the phenomenon of “supply chaindisintermediation” (Rossetti and Choi, 2005) and measures the degreeof connection between the buyer's 1st tier suppliers and its customers.Note that we label this dimension “eliminative” because in the extremecase, a supplier may replace its buyer. Rossetti and Choi (2005) reportthat many buyers and their former suppliers have become competitorsin the aerospace industry. Bleeke and Ernst (1995) illustrate howThomson Consumer Electronics (a former supplier to JVC) eventuallybecame one of JVC's major competitors.

While not all 1st tier suppliers can and will replace the buyer (e.g., asupplier who provides screws to both the buyer and the buyer's custo-mers may not have the ability to make the buyer's products), there areat least two ways that the presence of eliminative complexity can hurtthe buyer's financial performance. First, customers are more likely tosqueeze the buyer's profit when they have direct links to upstreamsuppliers. Information from these suppliers provides the customers abetter understanding of the buyer's operational capabilities and coststructure. Consequently, the customers could negotiate a better pricewith the buyer or impose otherwise nonexistent clauses on the contact,reducing the buyer's profit margin. Indeed, with sufficient informationabout its suppliers, customers may be able to force the focal firm's profitto the bare survival level (Choi and Hong, 2002). Second, the structuralhole perspective suggests that a supplier's direct access to the customersweakens the buyer's role as a “broker” (Ahuja, 2000; Burt, 1992). If thesupplier has the technological and manufacturing potential to offercompeting products, its reliance on the buyer would greatly decline.The supplier will be less likely to coordinate its operations to supportthe buyer, thus weakening the buyer's competitiveness in the market-place (Rossetti and Choi, 2005). Thus, eliminative complexity generatesadverse effects on the buyer's financial performance.

Worth noting is that the link between the supplier and the customermay be initiated by the buyer. For example, a buyer may ask the sup-plier to supply maintenance parts directly to the customer (Li and Choi,2009). These links may reduce buyer's coordination costs but the twoaforementioned drawbacks still apply; customers can learn more aboutthe buyer while suppliers get direct access to customers. In addition,customers could blame the buyer for any bad experience with thesupplier, regardless of the fact that the buyer may have no control overthe product/service the supplier provides. In service operations litera-ture, scholars have showed that this type of relation can significantly

hurt the buyer, despite its intent (Zhang et al., 2015). In other words,while some links between suppliers and customers may be beneficial,the overall effect of eliminative complexity at the supply base level islikely to be detrimental. This is in line with the implications of struc-tural holes. Connections between customers and suppliers remove thefocal buyer's structural hole position and thereby erode buyer perfor-mance.

Nonetheless, the rule of diminishing returns applies. While the ab-solute amount of information about the buyer will increase as moresuppliers offer products directly to its customers, the net growth ofunique, non-redundant information is likely to decrease (Galbraith,1973). Thus, the customers' ability to exploit the buyer will increasemore slowly and is likely to be capped after certain point due to in-formation saturation (Galbraith, 1973). From the SNT perspective, thebuyer's structural hole advantages (Burt, 1992) vanish at high levels ofeliminative complexity and consequently any further increase in elim-inative complexity is unlikely to generate additional harm. Taken to-gether, eliminative complexity should have its strongest negative mar-ginal impact at lower levels (i.e., from nonexistence to presence).

H4a. Eliminative complexity of the supply base will have a negative yetdiminishing impact on the focal buyer's ROA.

H4b. Eliminative complexity of the supply base will have a negative yetdiminishing impact on the focal buyer's Tobin's Q.

Cooperative complexity gauges the degree of connection amongsuppliers within the buyer's supply base (Pathak et al., 2014; Wilhelm,2011). A link between two suppliers shows direct communication andexchange of physical goods between the two firms, and implies thatthey do not produce an equivalent part or product (Pathak et al., 2014).Links between suppliers can affect buyer performance positively byproviding three substantive benefits: information acquisition, co-ordination spillover, and innovation. First, links between suppliers leadto information acquisition. While the buyer is not involved in this re-lation, it may indirectly gain a better understanding of a supplier,providing opportunities to better coordinate with that supplier. Thisview is consistent with the strategic management literature, which as-serts that suppliers are a major source of “corporate intelligence” (Fuld,1988; Porter, 1980).

Second, links between suppliers reduce supply base uncertainty,thanks to the spillover effects of coordination. These links serve as in-formation conduits such that technology improvements, insights toproblems and failed approaches can travel from one firm to another(Choi and Hong, 2002; Wilhelm, 2011). This lowers variation in sup-pliers' manufacturing planning, which in turn can reduce lead timevariance for their mutual customer (i.e., the buyer). The spillover ef-fects also include standardization of transactions. One supplier is likelyto accept or adapt to the standards of another supplier when they are ina buyer-supplier relation. The resultant “jointly developed” standardsare likely to lower the buyer's purchasing cost as the same procedurescan be used for both suppliers.

Third, the literature finds that direct links among suppliers posi-tively affect a buyer's innovation through knowledge exchange in thesupply base (Ahuja, 2000), which curtails the buyer's macro-leveltechnological uncertainty. In addition, a link exists when there is abuyer-supplier relation. Supplier-supplier links thus imply the use ofcommon parts across partners (e.g., the same ball bearing used byToyota and its suppliers) within the supply base. Literature suggeststhat the use of common parts helps the buyer reduce its products' timeto market (Choi and Krause, 2006).

Links among suppliers, however, also provide opportunities forsuppliers to exploit the buyer. Collusion can be a concern, but suchcollusion may seldom exist, given the buyer's ability to closely monitor1st tier suppliers and the threat of reputation loss (Ahuja, 2000).Overall, links among suppliers at the supply base level would lead tobenefits. But the benefits likely have limits. In a supply base wheremany suppliers connect with each other, information is likely to be

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

29

transparent. Adding such connections would have marginal impact onthe buyer's understanding of suppliers. The rents of coordination spil-lover may diminish as well, because firms can neither completelyeliminate operational variations through coordination nor standardizetransaction procedures for all partners. Rather, the coordination costmay increase with number of connected suppliers, suggesting dimin-ishing returns. Furthermore, the need for information processing willincrease with cooperative complexity. However, bounded rationalityconstrains the buyer's ability to reap innovation rents beyond a certainpoint. After that, further increase in cooperative complexity would havelittle marginal impact on buyer's financial performance. Taken to-gether, we propose:

H5a. Cooperative complexity of the supply base will have a positive yetdiminishing impact on the focal buyer's ROA.

H5b. Cooperative complexity of the supply base will have a positive yetdiminishing impact on the focal buyer's Tobin’ Q.

4. Data and variables

4.1. Data source and collection process

We start our data collection process by identifying the buyer firms ofinterest. We focus on public companies in five two-digit SIC codes (28,35, 36, 37, and 38) that together define chemical (including pharma-ceutical) and electronics manufacturers. All five SICs are under thegeneral category of Manufacturing, which is in line with our focus onthe supply base of manufacturing buyers. SIC 36 (Electronic & OtherElectric Equipment) has been widely studied in the literature in the lastdecades owing to rapid technological evolution. In addition, SIC 35(Industrial Machinery & Equipment), 37 (Transportation Equipment),and 38 (Instruments & Related Products) tend to have similar char-acteristics—all are related to high-tech. Indeed, they are collectivelylabeled “complex technology” sectors (Coad and Rao, 2008), wheresuppliers play important roles in buyers’ manufacturing processes(Cohen et al., 2000). We also included SIC 28 (Chemical & Allied Pro-ducts) as we argue that supply base structural characteristics are criticalin this industry (Hoffman, 1999).

Our empirical sample merged data from two sources, Compustat andMergent Online (Mergent hereafter). The former is a well-respectedsource for company-level financial data of publicly-traded U.S. com-panies. The latter is a unique secondary data source for supplier/cus-tomer relationship and product portfolio information, which we discussfurther in the following. Mergent uses information sources such asSecurities and Exchange Commission (SEC) required filings (includingannual reports), news articles, trade publications, and company web-sites to identify a buyer's suppliers and customers, as well as to compileits product information.1 For each buyer firm, we collect supplier/customer lists, product tree, and other company characteristics (e.g.,address and board members). Fig. 1 illustrates the exposition of re-lationship and product tree data on Mergent. While the supplier andcustomer lists are straightforward, we elaborate on how the producttree (Fig. 1, Panel C) is used to define a company's product groups. Theillustrative company in the figure, AMAG Pharmaceuticals Inc., has aproduct tree that includes four product groups, each defined by a spe-cific sector-to-product path. For example, Product Group 1 is specifiedby the "Healthcare→ Biopharmaceuticals→Digestive Systems" path. Note

that measuring certain dimensions of supply base structural complexityinvolves relationship and product information not only for the buyer,but also for its 1st tier suppliers. Thus, we repeat the data collectionprocess for each of the 1st tier suppliers identified above. Since Mergentupdates its data on a continuous basis, we make the conservative as-sumption that our data represent the reality on or before the data col-lection date, August 2016.

Compustat's data, on the contrary, exhibit a significant time lag.Fortunately, it has a clearly documented “as of” date, which variesacross companies, according to their different annual report releasedates. Thus, if a company's “as of” date from Compustat is after August2016, we obtain a sufficient condition that our dependent variables(ROA and TQ, constructed from Compustat data) are measured after thefocal independent variables of interest (five complexity dimensions,constructed from Mergent data) for that company. We collect data inMay 2017 from Compustat's Fundamental Annual database, whichyields 78.2% satisfying the above condition. To further improve thispercentage, we collect additional data from Compustat's FundamentalQuarterly database to obtain the most recent quarterly ROA and TQmeasures, 98.7% of whose “as of” dates are after the Mergent data.Therefore, we use the quarterly ROA and TQ in our main analysis, whilereplicating the results with the annual measures in the robustnesschecks.

Merging data between Compustat and Mergent is a painstaking pro-cess, since company identifiers in the former carry no meaning in thelatter. A direct name match does not work well because the samecompany's name is often formatted differently in the two data sources.De Leeuw and Keijl (2015) report that different matching approachescould result in drastically different “yield rates” (i.e. match-able ob-servations) and hence affect empirical results. To minimize the poten-tial impact of both human and algorithmic errors, Roth et al. (2015)recommend a combination of automatic and manual steps. We employsuch a matching process in this study. First, we input each companyname provided by Compustat into Mergent's search engine and checkwhether exactly one company is identified. We automated this stepwith an R script and identified around 90% perfect matches. Second, forthose with no match, we manually experiment with other companyname formats. For those with multiple matches, we manually identifythe correct one. The above process produces a 97% yield.

Next, we discuss the data cleaning process, as summarized inTable 3. We start with the 1891 total firms under the five SIC codesidentified by Compustat. After constructing the dependent and controlvariables such as sales and resource efficiencies, we remove observa-tions with missing values in one or more variables—a practice con-sistent with the literature (Modi and Mishra, 2011; Mishra et al., 2013).Then, we remove a small group of firms with extreme values (Iglewiczand Hoaglin, 1993) of either ROA (outside the range from −2 to 2) orTQ (greater than 10).2 We obtain a sample of 1150 companies beforecombining data from Mergent. Data collected from Mergent allows us toconstruct supply base complexity measures for 867 firms (the finalsample size as shown in Table 3), providing a coverage of 75%. On onehand, this imperfect coverage underscores the difficulty of identifyingcomplete supplier and customer relationships. Coverage rates reportedin the literature using other data sources are comparable. For example,Wu and Birge (2014) show that in the manufacturing and logisticssector, about 65% of firms listed on the Center for Research in SecurityPrices (CRSP) database have supplier/customer data in the BloombergSPLC database. Bellamy et al. (2014) report that among the companieson the Electronics Business 300 list, 151 (or 50%) have supplier/cus-tomer data in the Connexiti database, which they use to construct1 Compustat Segment Customer data, another source for customer lists, relies primarily

on SEC filings (Ellis et al., 2012). SEC requirements state that “if 10% or more of therevenue of an enterprise is derived from sales to any single customer, that fact and theamount of revenue from each such customer shall be disclosed.” Communication withMergent suggests that the Compustat Segment Customer data is one of their sources forcompiling supplier and customer lists. Our own comparison between Mergent and Com-pustat data suggests that the former is indeed more comprehensive than the latter. Thiscomparison is available from the authors upon request.

2 The magnitude of sample reduction due to these steps of data cleaning is comparablewith prior research using Compustat data. For example, Mishra et al. (2013, pp. 301–302)collected 656 firms from Compustat and retained 357 (or 54.4%) firms with non-missingvalues. As shown in Table 3, we collected 1891 firms from Compustat and retained 1150(or 60.8%) firms with non-missing values.

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

30

supplier network measures. On the other hand, the imperfect coveragealso cautions us against the potential impact of sample selection bias onour empirical results. Section 5.4 assesses the extent and implications ofthe supply base data coverage issue.

4.2. Dependent variables

In the next three subsections, we discuss the construction of ourvariables, which is summarized in Table 4. Consistent with prior studies(Mackelprang et al., 2015; Wagner et al., 2012; Jacobs et al., 2016), weoperationalize our dependent variables by Return on Assets and Tobin'sQ. ROA measures a firm's short-run profitability while Tobin's Q cap-tures long-run market valuation and intangible assets (Modi andMishra, 2011; Bharadwaj et al., 1999; Chung and Pruitt, 1994;Villalonga, 2004). Fig. 2 shows the histograms of the two main de-pendent variables—quarterly ROA (ROA_Q) and TQ (TQ_Q). Histogramsfor the annual measures show similar shapes. As discussed previously,we remove outlier firms experiencing abnormal growth or decline(Table 3, Steps 6 and 7).

4.3. Supply base structural complexity variables

When operationalizing our five supply base complexity dimensions,we attempt to create measures that are continuous and easily inter-preted. Continuous measurements are crucial, given that their hy-pothesized impacts are nonlinear. While it might be possible to de-monstrate a nonlinear effect through categorical variables, it is difficultto compute an “optimal” level of complexity, which might be of

practical interest to supply chain managers. To facilitate discussion, weillustrate with an example buyer in Fig. 3 (i.e. the same firm shown inFig. 1).

Horizontal complexity is conceptualized in the literature as thenumber of 1st tier suppliers (Choi and Hong, 2002; Tolbert and Hall,2009; Senge, 2006). However, while a diversified firm with a broadproduct portfolio may have the same number of 1st tier suppliers as aless diversified firm, the complexity levels of the two firms' supply basesare obviously different. For example, the horizontal complexity of a 2-product-group, 2-supplier firm should be lower than a 1-product-group,2-supplier firm because the former has only one supplier to manage ineach product group (and each product group usually has its ownmanagement team). In addition, just as more product groups mightrequire more suppliers, a large product group might call for moresuppliers than a small product group. That is, the size of each productgroup also matters. To capture the heterogeneity of product portfolioacross buyer firms, we operationalize Horizontal by a firm's number of1st tier suppliers divided by a weighted sum of its product groups,where the weighting mechanism captures the varying sizes of productgroups.3 Since the buyer in Fig. 3, AMAG Pharmaceuticals Inc., has five1st tier suppliers and four product groups, its Horizontal =

∑ = Weight5

i i14 .To

calculate Weightis for each product group, we count its frequency ofappearance among all companies in the same two-digit SIC code (28) asAMAG. It is reasonable to expect that a large product group is likely tohave a high frequency. To convert this frequency into weights that arecomparable across industries, we calculate the percentage of companiesin SIC 28 having each of these product groups (15.6%, 6.8%, 12%, and2.4%). Plugging these weights into the above formula, we obtain Hor-izontal = =+ + + 13.65

0.156 0.068 0.12 0.024 for buyer AMAG.Vertical complexity is measured by the average number of 2nd tier

suppliers per 1st tier supplier (Vertical = + + + +5 1 1 42 1345 = 36.6 in

Fig. 3). The logic for this is that the more low-tier suppliers a 1st tiersupplier has, the more likely its supply chain reaches the deepest up-stream manufacturers, as we discussed in the literature review (Choiand Krause, 2006). This proxy is consistent with our focus on 1st tiersuppliers.

Spatial complexity is measured by the number of countries re-presented in the supply base. Similar measures have been used in stu-dies on geographical dispersion of multinational corporations (e.g., vonCorswant and Fredriksson, 2002). In Fig. 3, the five suppliers are lo-cated across three countries, U.S., Netherlands, and Germany. Thus,Spatial = 3. We also consider two alternative measures in the robustnesschecks (see Section 5.4).

Eliminative complexity first computes the product of two factors for

Fig. 1. Data exposition on mergent online.

Table 3Data construction and cleaning process.

Steps Reduction # of Firms

Compustat onlyStart: raw data from Compustat 1891Step 1: quarter or annual Return on Assets (ROA) missing 45 1846Step 2: quarter or annual Tobin's Q (TQ) missing 82 1764Step 3: Market Efficiency (MktEff) missing 82 1682Step 4: Production Efficiency (PdtEff) missing 189 1493Step 5: Inventory Efficiency (InvEff) missing 249 1244Step 6: remove extreme TQ (> 10) 73 1171Step 7: remove extreme ROA (< -2 or> 2) 21 1150

After merging with Mergent OnlineStep 8: no matched company 37 1113Step 9: no relationship or product tree data 246 867

Supply base data coverage 75%

3We thank the review team for the suggestion of incorporating both the number andthe size of product groups in the Horizontal measure.

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

31

Table4

Variableco

nstruc

tion

.

Variable

Description

Source

Calcu

lation

Referen

ce

ROA_Q

/AReturnon

Assetscalculated

from

quarter/an

nual

data

Com

pustat

IB/A

TMacke

lprang

etal.,20

15;Wag

neret

al.,20

12TQ

_Q/A

Tobin'sQ

calculated

from

quarter/an

nual

data

Com

pustat

(CSH

O*P

RCC_F

+PS

TKL+

DLT

T+

INVT+

LCT-ACT)/A

TMod

ian

dMishra,

2011

;Bha

radw

ajet

al.,19

99Horizon

tal

Horizon

talco

mplexity

Merge

nt#

of1s

ttier

supp

liers

divide

dby

weigh

tedprod

uctgrou

psAda

pted

from

:Bod

ean

dWag

ner,

2015

Vertical

Verticalco

mplexity

Merge

ntAve

rage

#of

2ndtier

supp

lierpe

r1s

ttier

supp

lier

Ada

pted

from

:Bod

ean

dWag

ner,

2015

Spatial

Spatialco

mplexity

Merge

nt#

ofsupp

lierco

untries

vonCorsw

antan

dFred

riksson,

2002

Elim

inative

Elim

inativeco

mplexity

Merge

nt%

of1s

ttier

supp

liers

directly

selling

tobu

yer's

custom

ersweigh

tedby

prod

uctov

erlap

New

,not

empirically

exam

ined

Coo

perative

Coo

perative

complexity

Merge

ntActua

llin

ksov

erpo

ssible

links

amon

g1s

ttier

supp

liers

New

,not

empirically

exam

ined

Num

Prod

ucts

Num

berof

prod

uctgrou

psMerge

ntCou

ntof

prod

uctgrou

psextractedfrom

prod

ucttree

MacDuffi

eet

al.,19

96Num

Customers

Num

berof

custom

ers

Merge

nt#

of1s

ttier

custom

ers

Kim

,201

7InvE

ffEffi

cien

cyof

utilizing

inve

ntoryresources

Com

pustat

SALE

/(IN

VT+

LIFR

),then

stan

dardizeat

4-digitSICleve

lMod

ian

dMishra,

2011

,equ

ation(4)&

(5)

PdtEff

Efficien

cyof

utilizing

prod

uction

resources

Com

pustat

SALE

/PPE

GT,

then

stan

dardizeat

4-digitSICleve

lMod

ian

dMishra,

2011

,equ

ation(6)

MktEff

Efficien

cyof

utilizing

marke

ting

resources

Com

pustat

SALE

/XSG

A,then

stan

dardizeat

4-digitSICleve

lMod

ian

dMishra,

2011

,equ

ation(7)

SalesLog

Logg

edsalesin

millions

Com

pustat

Logg

edSA

LEHen

dricks

andSing

hal,20

09Sa

lesG

rowthLo

gLo

gged

salesch

ange

from

prev

ious

year

inmillions

Com

pustat

Logg

eddifferen

cein

SALE

betw

eencu

rren

tan

dprev

ious

years

Hen

dricks

andSing

hal,20

09Bo

okTo

Mkt

Book

tomarke

tratio

Com

pustat

BKVLP

S/MKVALT

Hen

dricks

andSing

hal,20

09Deb

tToE

quity

Deb

tto

equity

ratio

Com

pustat

(AT-LT

)/MKVALT

Hen

dricks

andSing

hal,20

09MktSh

are

Marke

tsharein

percen

tage

Com

pustat

SALE

divide

dby

totalSA

LEacross

firm

swithinthesame4-digitSIC

Hen

dricks

andSing

hal,20

09Sh

areh

olde

rLog

Logg

ednu

mbe

rof

shareh

olde

rsin

thou

sand

sMerge

ntLo

gged

numbe

rof

shareh

olde

rsin

thou

sand

sAnd

res,

2008

BoardS

ize

Num

berof

boardmem

bers

Merge

ntCou

ntof

boardmem

bers

And

ersonet

al.,20

04Bo

ardT

enure

Ave

rage

boardmem

bertenu

reMerge

ntAve

rage

#of

yearsbo

ardmem

bers

worke

dfortheco

mpa

nyAnd

ersonet

al.,20

04Bo

ardA

geAve

rage

boardmem

berag

eMerge

ntAve

rage

ageof

boardmem

bers

And

ersonet

al.,20

04Bo

ardC

OO

Yes/N

oChief

Ope

rating

Officeron

board

Merge

nt1/

0=

existenc

e/ab

senc

eof

COO

onbo

ard

Roh

etal.,20

16

Note:

Foreach

Com

pustat-based

variab

le,the

capitaliz

edterm

sin

thecalculationco

lumnrepresen

ttheoriginal

variab

lena

mes

used

inCom

pustat.

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

32

each 1st tier supplier, existence of direct selling to the buyer's customers(a 1/0 dummy) and product group overlap with the buyer (a percen-tage), and then sums this product across the supply base and divides thesum by the total number of 1st tier suppliers. For example, AMAG's first1st tier supplier, Pfizer Inc., engages in direct selling. Using the buyer'sproduct portfolio as the denominator for the overlap measure4 itsproduct overlap with the buyer is = 75%3

4 since among AMAG's fourproduct groups, only “Healthcare → Healthcare Services → HealthcareSupport Services” is absent from Pfizer's product group portfolio. As aresult, the multiplication of existence of direct selling and productgroup overlap is × =1 75% 0.75. Since none of its other four supplierssell directly to AMAG's customers, this procedure yields zero. Last, wesum the five results and divide the total by five, obtaining Eliminative= = 0.150.75

5 . This metric has an intuitive interpretation: one can view itas the actual threat over the maximal threat (i.e., if all suppliers engagein direct selling and produce items in all product groups the buyer of-fers, the threat would be × × =(1 100%) 5 5).

Cooperative complexity (Pathak et al., 2014; Wilhelm, 2011) ismeasured by the number of links among 1st tier suppliers over the totalpossible links. In Fig. 3, the five suppliers have + + + =4 3 2 1 10possible links and 1 actual link (between Merck KGaA and Pfizer Inc.),which yields Cooperative = = 0.11

10 .

4.4. Control variables

We consider several control variables to mitigate potential omittedvariable bias in our focal effects (Angrist and Pischke, 2009). Firms witha diverse product portfolio might have more suppliers and experience ahigher financial performance variation as shown in Jain et al. (2014).We thus control for the number of product groups in a buyer firm'sportfolio, NumProducts. Similar to the upstream suppliers, the size ofthe downstream customer base (NumCustomer: number of customers)might also influence the buyer's financial performance (Kim, 2017).

Apart from supply base configuration, the buyer's profitability isalso driven by internal and operational decisions, which are manifestedby its efficiency in utilizing various firm resources. Low efficiency

implies sub-optimal resource utilization and higher waste (non-valueadding resources), which hurt financial performance (Chase et al.,2006). We employ three efficiency measures proposed in Modi andMishra (2011): inventory efficiency (InvEff), production efficiency(PdtEff), and marketing efficiency (MktEff). A firm's efficiency in uti-lizing internal resources signals its operational strength (Modi andMishra, 2011), which might also explain why some firms are better atsupply base management than others. As a result, controlling for in-ternal operational factors not only explains variation in financial per-formance, but also helps alleviate omitted variable bias due to the op-erational strength of a firm.

Another common driver of financial performance and supply basecomplexity is firm size. The literature shows evidence for the associa-tion between firm size and financial performance (Koufteros et al.,2014). It is conceivable that larger firms tend to have more suppliersand more complex supply bases. As a result, we use logged salesamount, SalesLog, to control for firm size (Hendricks and Singhal,2009). Since all buyers in our sample are publicly traded U.S. firms, wealso include the logged number of shareholders, ShareholderLog, as anadditional firm size control.5

Since a firm's decision to add or replace suppliers is largely de-pendent on its current financial standing, we control for its growthtrajectory and debt level. SalesGrowthLog captures the (logged) changein sales from last year, while the book to market ratio, BookToMkt,controls for growth potential in the future (Hendricks and Singhal,2009). We adopt the commonly used debt to equity ratio, DebtToEquity,to measure the debt level (Hendricks et al., 2009). Investing in thesupply base might also be dictated by the buyer firm's market share(i.e., whether it is a big or small player in the industry). FollowingHendricks and Singhal (2009), we calculate a percentage measure ofmarket share at the four-digit SIC level, MktShare.

Given that supply base management is generally considered a high-level strategic decision, we control for a variety of board characteristics.BoardSize, BoardAge, and BoardTenure are the count, the average ageand the average years of experience of the board members. Hendricks

Fig. 2. Histograms and scatter plots for severalkey variables.

4 We use the buyer's product portfolio as the denominator for the overlap percentagebecause Eliminative attempts to measure the threat of “elimination” that the 1st tiersuppliers impose on the buyer.

5 Other frequently used firm size controls include earnings, cashflow, assets, andnumber of employees (Hendricks and Singhal, 2009; Modi and Mishra, 2011; Mishraet al., 2013). However, these variables are highly correlated with SalesLog in our sampleand thus are not used as additional controls.

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

33

et al. (2014) demonstrate the financial impact of the appointment of aChief Operations Officer (COO). Given that the role of COOs6 has ex-panded over time from internal functions such as managing inventoryturnovers to external functions such as facilitating supply chain in-tegration (Wagner and Kemmerling, 2014), it is reasonable to expectthe presence of a COO might also motivate supplier management in-itiatives such as increasing the visibility of 1st tier suppliers. Thus, weconstruct a dummy variable, BoardCOO, to control for the presence/absence of a COO on the executive board.

Last, we create dummy variables to account for systematic differ-ences across industries at the four-digit SIC level. A breakdown of firmsinto their respective industries is discussed in Section 5.4 when weaddress the potential sample selection bias caused by incomplete supplybase data coverage. Table 5 presents the descriptive statistics, as well asthe correlation matrix of our variables.

5. Empirical analysis

We first consider potential endogeneity problems because supplybase configuration and financial performance are both managerialoutcomes. We test our hypotheses next, followed by presenting thesignificant nonlinear effects graphically. Last, we conduct three sets ofrobustness checks: (1) employing annual measures, ROA_A and TQ_A, asalternative dependent variables, (2) examining two alternative mea-sures of Spatial complexity, and (3) assessing the extent and implica-tions of supply base data coverage bias.

5.1. Potential endogeneity concerns

We carefully consider endogeneity issues that might arise throughomitted variable bias, and justify our strategy with theoretical, con-textual, and empirical arguments (Guide and Ketokivi, 2015). Note thatsimultaneity is unlikely to be a source of endogeneity in our sample,since we have explicitly verified that our complexity variables’ timestamps precede those of the outcome variables. Further, if financialperformance triggers supply base changes, it takes a relatively long timeas Jain et al. (2014) argued when studying global sourcing and

inventory investment.Instrumental variables and panel datasets are commonly used to

address the omitted variable bias (Ketokivi and McIntosh, 2017).Nonetheless, identification through within-firm variation via a panelapproach appears to be non-tenable, given data availability. To ourknowledge, secondary data on supply chain structure did not becomeavailable until lately. Major data vendors, including Bloomberg(Osadchiy et al., 2015), Connexiti (Bellamy et al., 2014), and Mergent,do not provide archival or longitudinal information. Even if one collectsdata repetitively, the variation has been reported to be minimal(Osadchiy et al., 2015). We have also verified this through our own datacollection on Mergent. Therefore, the panel data approach would re-quire a long time-series to generate enough within-firm variation of thecomplexity variables. The instrumental variable approach would alsobe problematic given the extreme difficulty of arguing for exogeneity inour context. A good instrument needs to have strong correlation withsupply base configuration (relevance condition) but not with financialperformance of the focal firm (exogeneity condition) (Rossi, 2014).Since both supply base configuration and financial performance arelargely affected by managerial inputs, a quality instrument is unlikelyto come from the buyer. Rather, link-breaking or link-forming activitiesinitiated by suppliers are more reasonable candidates for instruments.Unfortunately, this information is hard to collect and quantify. AsKetokivi and McIntosh (2017, p.7) state: “Applying instrumental vari-ables amounts to trading one set of untestable assumptions for another,and using a bad instrument may well make things worse than stickingto OLS … This observation offers a segue to the next candidate solutionto tackling endogeneity: instead of trying to come up with bettermodels, perhaps an actionable answer lies in getting better data.”

As a result, our empirical strategy is to collect and construct controlvariables that are purported in the literature to affect both financialperformance and supply base configuration. In other words, we attemptto include variables in our regression that if omitted, could cause en-dogeneity concerns. Shang et al. (2017) is a recent example that usesthe control variable approach instead of instruments. We recognize thateven with our extensive set of controls as shown in Table 4, completelyeliminating endogeneity is unlikely, which we acknowledge as a lim-itation of this study.

Fig. 3. Illustrating the operationalization of supply chain complexity measures.

6 Also called Chief Supply Chain Officers, CSCO, in some companies.

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

34

Table5

Variablesummarystatistics

andco

rrelationmatrix( N

=86

7).

Variable

Mean

SDMin

Max

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(1)

ROA_Q

−0.02

0.1

−0.93

0.26

(2)

TQ_Q

1.7

1.31

0.01

9.33

0.23

**(3)

ROA_A

−0.07

0.27

−1.87

0.59

0.81

**0.23

**(4)

TQ_A

1.61

1.21

0.01

9.12

0.22

**0.95

**0.25

**(5)

Horizon

tal

58.99

197.03

0.59

1415

0.06

0.04

0.02

0.02

(6)

Vertical

15.19

29.5

025

3−0.05

−0.01

−0.06

−0.01

0.32

**(7)

Spatial

3.64

5.13

026

0.14

**0.05

0.12

**0.06

0.56

**0.29

**(8)

Elim

inative

0.1

0.22

01

0.05

0.01

0.04

−0.03

0.06

0.08

*0.21

**(9)

Coo

perative

0.06

0.14

01

0.09

**0.06

0.06

0.02

0.13

**0.22

**0.30

**0.66

**(10)

Num

Prod

ucts

4.98

4.96

034

0.13

**0.05

0.15

**0.05

−0.13

**−

0.02

0.17

**0.11

**0.08

*(11)

Num

Customers

18.57

23.7

125

10.07

*−0.06

0.03

−0.09

*0.09

*−

0.02

0.22

**0.30

**0.27

**0.30

**(12)

InvE

ff−

0.01

0.82

−1.97

6.5

0.20

**0.18

**0.24

**0.18

**0.05

00.04

0.04

0.06

0.06

0.03

(13)

PdtEff

00.88

−1.61

7.94

0.08

*0.19

**0.12

**0.19

**−

0.05

0.08

*−

0.01

0.03

0.05

0−0.05

(14)

MktEff

0.22

0.91

−2.04

6.91

0.36

**−0.06

0.41

**−0.03

0.02

−0.04

0.10

**0

−0.01

0.12

**0.04

(15)

SalesLog

6.14

2.43

012

.28

0.54

**0.09

**0.57

**0.08

*0.16

**−

0.01

0.38

**0.18

**0.18

**0.25

**0.27

**(16)

SalesG

rowthLo

g10

.90.37

011

.14

0.15

**0.23

**0.20

**0.24

**0.01

00.07

*0.02

0.03

0.03

−0.01

(17)

Book

ToMkt

0.31

1.86

−42

.84.11

−0.11

**−0.67

**−

0.11

**−0.69

**−

0.14

**−

0.04

−0.16

**0.04

−0.04

−0.02

0.01

(18)

Deb

tToE

quity

0.21

5.14

−23

.43

147.66

0.14

**−0.10

**0.13

**−0.10

**0.06

−0.04

0.11

**−0.02

00.07

*0.16

**(19)

MktSh

are

0.09

0.18

01

0.47

**0.04

0.50

**0.02

0.09

**−

0.05

0.22

**0.14

**0.10

**0.20

**0.18

**(20)

Shareh

olde

rLog

5.86

2.61

014

.88

0.30

**0.02

0.28

**0.01

0.11

**−

0.03

0.21

**0.12

**0.07

0.15

**0.11

**(21)

BoardS

ize

15.59

8.53

166

0.30

**0.07

*0.26

**0.06

0.12

**−

0.06

0.16

**0.04

0.06

0.20

**0.11

**(22)

BoardT

enure

13.4

5.57

135

.33

0.30

**−0.03

0.31

**−0.03

−0.01

−0.06

0.03

0.04

0.04

0.15

**0.09

**(23)

BoardA

ge58

.39

4.33

41.4

760.10

**−0.10

**0.10

**−0.10

**−

0.04

−0.03

00.01

0.01

0.09

*−0.01

(24)

BoardC

OO

0.18

0.39

01

−0.03

−0.05

−0.06

−0.04

0.04

0.02

−0.04

−0.04

−0.05

0.01

−0.02

(12)

(13)

(14)

(15)

(16)

(17)

(18)

(19)

(20)

(21)

(22)

(23)

(13)

PdtEff

0.13

**(14)

MktEff

0.12

**0.08

*(15)

SalesLog

0.24

**0.03

0.36

**(16)

SalesG

rowthLo

g0.03

0.12

**0.15

**0.16

**(17)

Book

ToMkt

−0.19

**−0.12

**0.03

−0.16

**−0.16

**(18)

Deb

tToE

quity

0.07

*0

0.11

**0.29

**0.01

0.04

(19)

MktSh

are

0.21

**−0.01

0.31

**0.81

**0.04

−0.04

0.23

**(20)

Shareh

olde

rLog

0.09

*−0.04

0.12

**0.43

**−0.04

−0.03

0.10

**0.43

**(21)

BoardS

ize

0.18

**−0.03

0.14

**0.49

**0

−0.13

**0.15

**0.39

**0.34

**(22)

BoardT

enure

−0.03

−0.06

0.16

**0.15

**−0.05

0.15

**0.07

*0.19

**0.25

**0.15

**(23)

BoardA

ge−

0.05

−0.09

**0.08

*−

0.03

−0.06

0.14

**−

0.04

0.02

0.20

**0.10

**0.53

**(24)

BoardC

OO

−0.03

−0.02

−0.04

−0.09

**−0.01

−0.02

0.02

−0.12

**−

0.11

**0.07

−0.03

−0.04

Note:

*:p

<0.05

,**p

<0.01

.

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

35

5.2. Regression results

Tables 6 and 7 contain the results of our main analysis using OLSregressions, with four-digit SIC dummies as fixed effects. We enter thecomplexity variables sequentially. In both tables, Models (1) through(5) enter one complexity dimension (first and second order terms) at atime, while Model (6) combines all five dimensions. We discuss thecontrol variable effects first. Among the three resource efficiencyvariables, we observe both inventory (InvEff) and marketing (MktEff)efficiencies show a positive sign, albeit the statistical significance ismore salient for TQ than for ROA. This result confirms prior literature(e.g., Modi and Mishra, 2011). The effect of production efficiency(PdtEff), however, shows opposite signs for ROA (positive) and TQ(negative). This may imply PdtEff has different impacts on short-termand long-term performance. It may be more difficult to improve pro-duction efficiency than inventory and marketing efficiencies. HighPdtEff thus suggests little potential improvement can be made in thefuture from market investors' perspective (negative impact on TQ).Consistent with Hendricks et al. (2009), we find firm size, measured bySalesLog, is associated with superior financial performance, while ahigher debt level (DebtToMarket) reduces financial performance. Last,we find that an executive board consisting of members with longertenure is associated with stronger financial performance, confirmingprior results in the literature (e.g., Anderson et al., 2004).

We discuss the visible structural complexity dimensions next. Theinverted U-shaped relationship posited in H1 and H2 implies the second-order term is negative for Horizontal and Vertical. We find empiricalsupport for the former, but not the latter. Along with the positive andstatistically significant first-order effect for Horizontal, we conclude itsimpacts on ROA and TQ are positive at lower levels and its marginal

impact becomes smaller as it increases.7 In terms of Vertical, we con-clude from the evidence gathered in Model (6) in Tables 6 and 7 that itsimpact on financial performance is insignificant and hence H2 is notsupported. For Spatial, the U-shaped relationship posited in H3 implies apositive second order effect, which is observed for both ROA and TQ.Together with a negative and significant first order effect, we concludeits impact is negative at lower levels and its marginal impact becomessmaller as it increases. For both Horizontal and Spatial, the questionsremain whether and when the second-order effect flips the marginaleffect from one direction to the other (i.e., from positive to negative forHorizontal, negative to positive for Spatial). We answer this question byvisualizing their nonlinear effects through confidence band plots in thenext section.

Turning to the two not-so-visible dimensions, both H4 and H5 posit adiminishing impact, implying positive and negative second-order ef-fects for Eliminative and Cooperative, respectively. Model (6) showssupportive evidence for both. Taken together with their significant first-order term coefficients, we conclude that Eliminative has a negativediminishing effect on financial performance, while Cooperative's impactis positive diminishing. These nonlinear effects are visualized in thenext section as well. We note that when Cooperative is entered into theROA regression on its own in Model (5) of Table 6, its effects are smallerin size and statistically insignificant. This difference between Model (5)and (6) could be explained by the correlation between Eliminative andCooperative. Both dimensions of supply base complexity are based onthe structural links the 1st tier suppliers have. It is conceivable thatsuppliers with ties to the other suppliers are more likely to have tieswith the buyer's customers, since they are more connected. This leads to

Table 6Regressions results (DV: ROA).

(1) (2) (3) (4) (5) (6)

Horizontal (×10−2) 0.010*(0.005) 0.012∗(0.005)Horizonta2 (×10−4) −0.001*(0.001) −0.001∗∗(0.001)Vertical (×10−2) −0.028 (0.033) −0.022 (0.031)Vertical2 (×10−4) 0.013 (0.015) 0.008 (0.014)Spatial −0.005∗(0.002) −0.005∗(0.002)Spatial2 0.0002∗ (0.0001) 0.0002∗ (0.0001)Eliminative −0.122∗∗ (0.045) −0.163∗∗∗ (0.049)Eliminative2 0.115∗ (0.049) 0.138∗∗ (0.049)Cooperative 0.065 (0.050) 0.192∗∗(0.061)Cooperative2 −0.072 (0.063) −0.185∗∗ (0.069)NumProducts −0.001(0.001) −0.0004(0.001) −0.0005(0.001) −0.0004(0.0005) −0.0004(0.001) −0.001(0.001)NumCustomers −0.0002(0.0001) −0.0002(0.0001) −0.0002(0.0001) −0.0001(0.0001) −0.0002(0.0001) −0.0001(0.0001)InvEff 0.005(0.004) 0.005(0.004) 0.006(0.004) 0.005(0.004) 0.005(0.004) 0.004(0.004)MktEff 0.003(0.004) 0.002(0.004) 0.002(0.004) 0.002(0.004) 0.002(0.004) 0.003(0.004)PdtEff 0.014∗∗∗(0.003) 0.015∗∗∗(0.003) 0.015∗∗∗(0.003) 0.015∗∗∗(0.003) 0.015∗∗∗(0.003) 0.014∗∗∗(0.003)SalesLog 0.020∗∗∗(0.002) 0.020∗∗∗(0.002) 0.022∗∗∗(0.003) 0.021∗∗∗(0.003) 0.020∗∗∗(0.003) 0.021∗∗∗(0.003)SalesGrowthLog −0.002(0.002) −0.002(0.002) −0.0001(0.002) −0.001(0.002) −0.002(0.002) 0.0002(0.002)BookToMkt −0.001(0.001) −0.001(0.002) −0.001(0.001) −0.001(0.001) −0.001(0.002) −0.0002(0.001)DebtToEquity 0.001(0.001) 0.001(0.001) 0.001(0.001) 0.001(0.001) 0.001(0.001) 0.001(0.001)MktShare −0.095∗∗∗(0.020) −0.092∗∗∗(0.020) −0.086∗∗∗(0.019) −0.096∗∗∗(0.020) −0.090∗∗∗(0.019) −0.092∗∗∗(0.020)ShareholderLog −0.0005(0.001) −0.0005(0.001) −0.0002(0.001) −0.0002(0.001) −0.0004(0.001) 0.0003(0.001)BoardSize −0.0004(0.0003) −0.0005(0.0003) −0.0005(0.0003) −0.001(0.0003) −0.0004(0.0003) −0.001†(0.0003)BoardTenure 0.003∗∗∗(0.001) 0.003∗∗∗(0.001) 0.003∗∗∗(0.001) 0.003∗∗∗(0.001) 0.003∗∗∗(0.001) 0.003∗∗∗(0.001)BoardAge −0.001(0.001) −0.001(0.001) −0.001(0.001) −0.001(0.001) −0.001(0.001) −0.001(0.001)BoardCOO −0.018(0.011) −0.017(0.011) −0.017(0.011) −0.017(0.011) −0.017(0.011) −0.016(0.011)Constant −0.065(0.060) −0.046(0.069) −0.068(0.064) −0.066(0.065) −0.050(0.065) −0.102†(0.059)4-digit SIC Dummies Yes Yes Yes Yes Yes Yes

Observations 867 867 867 867 867 867R2 0.392 0.366 0.370 0.373 0.366 0.413

Note: †p < 0.1; *p < 0.05; **p < 0.01; ***p < 0.001. Robust standard errors in parentheses.

7 Given our linear regression framework, the marginal effect of Horizontal is given by= +∂

∂β β Horizontal2E ROAi

Horizontalii

( )1 2 , where β1 and β2 are coefficients on Horizontali and

Horizontali2. Thus, β1 is essentially the marginal effect when =Horizontal 0i .

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

36

the expectation that buyers with higher Cooperative score are also likelyto have higher Eliminative scores. Indeed, the correlation between thesecomplexity dimensions is 0.66 in our sample. Thus, when Cooperative isentered alone, it will capture part of Eliminative's effect. Since the effectsof Eliminative and Cooperative are in the opposite direction, Cooperative'seffects are attenuated in Model (5). The same line of logic also explainswhy the Eliminative effects are smaller and insignificant in Model (4) ofTable 7. These observations further attest to the importance of in-vestigating different complexity dimensions in an integrated model.8

5.3. Plots for significant nonlinear effects

Our regression results show that among the five complexity di-mensions, four demonstrate significant nonlinear effects. To gain ad-ditional insights into whether and when the marginal effect transitionsfrom one “significance zone” into another, such as from positive sig-nificant to insignificant and from insignificant to negative significant,we construct eight confidence band plots (Fig. 4), each correspondingto the impact of one complexity dimension on one financial outcome(i.e. ROA or TQ). We illustrate the construction process using the impactof Horizontal on ROA. First, we quantify the effect size, = ∂

∂θiE ROA QHorizontal

( )ii,

and its standard error, σθi, for each firm in our sample through the DeltaMethod (Oehlert, 1992). Assuming θi is normally distributed, its 95%confidence interval is simply [ −θ σ1.96i θi , +θ σ1.96i θi]. Within the 867firms in the sample, 806 have a positive and significant effect (i.e.,

+ <θ σ1.96 0)i θi , 44 register an insignificant effect( + > > −θ σ θ σ1.96 0 1.96 ),i θ i θi i and 17 show a negative and significant

effect ( − <θ σ1.96 0)i θi . These statistics are marked on Panel A1 ofFig. 4.

Second, we provide more general insights on when the effect “shiftszones.” Bauer and Curran (2005) advocates the use of confidence bandplots (a.k.a. Johnson-Neyman technique) for this purpose.9 As shown inPanel A1, the effect of Horizontal on ROA switches from positive toinsignificant when the lower bound of its 95% confidence interval hitszero at =Horizontal 139. In addition, it switches from insignificant tonegative when the upper bound crosses zero at =Horizontal 717. Notethat since Eliminative is marginally significant in the TQ regression (seeModel (6) of Table 7), its confidence band plot, Panel C2, uses the 90%confidence interval (all other plots use 95%). We discuss the implica-tions from these analyses in Section 6.

5.4. Robustness checks

We conduct three sets of robustness checks. First, we confirm thatthe robustness of our results extends to the case when annual measuresof ROA and TQ are used as alternative dependent variables. Second, weconsider two alternative measures for Spatial complexity. Last, we as-sess the potential impact of incomplete supply base data coverage onour empirical results.

To ensure that our dependent variables’ time stamp (i.e., “as of”dates on Compustat) is after that of the independent variables (i.e., datacollection date for Mergent), we have used the quarterly measures forROA and TQ. Although annual measures are less optimal in terms oftemporally preceding the independent variables (78.2% firms satisfyingthe temporal condition; recall the discussion in Section 4.1), they ap-pear to be more common in the literature (e.g., Mackelprang et al.,

Table 7Regressions results (DV: Tobin's Q).

(1) (2) (3) (4) (5) (6)

Horizontal (×10−2) 0.283∗(0.133) 0.404∗∗(0.147)Horizonta2 (×10−4) −0.019†(0.010) −0.027∗(0.011)Vertical (×10−2) −0.008(0.390) −0.185(0.361)Vertical2 (×10−4) 0.082(0.200) 0.128(0.184)Spatial −0.069∗(0.030) −0.097∗∗(0.033)Spatial2 0.003∗(0.001) 0.003∗(0.001)Eliminative −0.478(0.528) −1.179†(0.641)Eliminative2 0.588(0.623) 1.246†(0.719)Cooperative 1.957∗(0.878) 3.067∗∗(1.092)Cooperative2 −2.267∗(0.948) −3.371∗∗(1.108)NumProducts −0.024∗(0.010) −0.028∗∗(0.010) −0.028∗∗(0.009) −0.028∗∗(0.010) −0.028∗∗(0.010) −0.019∗(0.010)NumCustomers −0.002(0.002) −0.001(0.002) −0.001(0.002) −0.001(0.002) −0.001(0.002) −0.001(0.002)InvEff 0.148∗∗(0.053) 0.155∗∗(0.056) 0.155∗∗(0.056) 0.152∗∗(0.056) 0.152∗∗(0.056) 0.129∗(0.052)MktEff 0.167∗(0.069) 0.157∗(0.068) 0.156∗(0.066) 0.161∗(0.068) 0.151∗(0.068) 0.164∗(0.066)PdtEff −0.269∗∗∗(0.047) −0.260∗∗∗(0.049) −0.267∗∗∗(0.049) −0.263∗∗∗(0.049) −0.264∗∗∗(0.049) −0.290∗∗∗(0.048)SalesLog 0.098∗∗(0.032) 0.092∗∗(0.034) 0.107∗∗(0.035) 0.094∗∗(0.034) 0.087∗(0.034) 0.121∗∗∗(0.033)SalesGrowthLog −0.038(0.029) −0.037(0.026) −0.031(0.029) −0.036(0.027) −0.046†(0.025) −0.058†(0.032)BookToMkt −0.035(0.055) −0.041(0.058) −0.039(0.059) −0.041(0.058) −0.040(0.058) −0.035(0.056)DebtToEquity −0.009∗(0.004) −0.003(0.004) −0.004(0.004) −0.003(0.004) −0.003(0.004) −0.012∗∗(0.005)MktShare −0.450(0.326) −0.295(0.321) −0.232(0.317) −0.333(0.322) −0.263(0.316) −0.324(0.319)ShareholderLog 0.004(0.018) 0.011(0.018) 0.016(0.018) 0.012(0.018) 0.013(0.018) 0.015(0.018)BoardSize −0.002(0.006) −0.002(0.006) −0.002(0.006) −0.002(0.006) −0.002(0.006) −0.003(0.006)BoardTenure 0.014(0.009) 0.014(0.009) 0.014†(0.009) 0.014(0.009) 0.014†(0.009) 0.016†(0.009)BoardAge −0.008(0.011) −0.007(0.011) −0.007(0.011) −0.008(0.011) −0.008(0.011) −0.009(0.011)BoardCOO −0.127(0.124) −0.100(0.124) −0.103(0.123) −0.100(0.123) −0.091(0.123) −0.120(0.122)Constant 1.458∗(0.718) 1.460∗(0.729) 1.551∗(0.746) 1.549∗(0.728) 1.562∗(0.719) 1.468†(0.791)4-digit SIC Dummies Yes Yes Yes Yes Yes Yes

Observations 867 867 867 867 867 867R2 0.208 0.197 0.201 0.197 0.202 0.231

Note: †p < 0.1; *p < 0.05; **p < 0.01; ***p < 0.001. Robust standard errors in parentheses.

8 To check for multicollinearity of the complexity variables, we rerun Model (6)without the squared terms. This is necessary because a squared term creates a strongcorrelation with its first-order counterpart by definition and makes the multicollinearitydetection measures such as the Variance Inflation Factor (VIF) spuriously high(Wooldridge, 2015). VIF for the five complexity variables ranges between 1.2 and 1.8,well under the common rule-of-thumb thresholds.

9 This approach has a clear advantage over the conventional interaction plot in that itprovides information on the region of significance, which in turn spurs many applicationsin recent operations management studies (e.g. Malhotra et al., 2015; Tenhiäläa andHelkiö, 2015).

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

37

2015). Thus, in this robustness check, we rerun our analysis with theannual dependent variable measures, ROA_A and TQ_A. For succinctexposition, we present results from the full model specification in

Models (1) and (2) of Table 8. The observations are highly consistentwith those in the main analysis—all complexity dimensions but Verticalshow nonlinear effects on ROA and TQ.

Fig. 4. Visual illustration for nonlinear effects.

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

38

Next, we turn to alternative measures of Spatial. One might arguethat although our Spatial measure captures the number of suppliercountries, it does not capture to what extent the supply base consists offoreign suppliers. For example, let us change the country compositionof the supply base in Fig. 3 (US-Germany-Netherlands) from its current3-1-1 to a hypothetical 1-2-2. The number of supplier countries stays at3, but the extent of foreign suppliers grows from 40% to 80%. Such achange might make the supply base more complex to manage. Our firstalternative measure addresses this concern. Specifically, we use theproduct of the percentage of foreign suppliers and the number ofcountries represented in the supply base. Similar multiplicative mea-sures have been used in studies on geographical dispersion of multi-national corporations (e.g., Chon, 2004). The results from this alter-native measure are shown in Models (3) and (4) of Table 8. We observesimilar U-shaped effects as in the main analysis. The second alternativemeasure addresses the potential concern that a supplier's production isconducted by its subsidiaries in a country other than its headquarterscountry (supplier location in our Spatial measure is determined by itsheadquarters location).10 In this case, a common practice in the in-dustry is that the supplier will first purchase items from these sub-sidiaries (gain the ownership of items, an issue related to “transferprice” in the accounting literature) and then sell them to the buyer.Thus, the benefits and costs of having an “overseas supplier” as weargue in H3 are captured by the parent firm (i.e., the supplier at itsheadquarters location), not the buyer. To completely address thisconcern, however, requires knowledge of a supplier's actual productionlocation(s). Since we are unaware of any public database that containssuch detailed information, we consider a compromise based on the most

frequent subsidiary country of the supplier. We collect subsidiary lo-cation information for each supplier from Mergent. If the country withthe top frequency is different from the supplier's home country, wereplace the supplier's headquarters country with this “top subsidiarycountry.” Results from this subsidiary-based Spatial measure are re-ported in Models (5) and (6) of Table 8, which again show strongconsistency with our main analysis.

Last, we consider the coverage issue of supply base complexity data.As discussed in Section 4.1, among the 1150 firms for which Compustatprovides complete data, we are able to use Mergent as a data source toconstruct supply base complexity measures for 867 firms—an overallcoverage rate of 75%. This coverage rate is commensurate with pre-vious studies (e.g., Bellamy et al., 2014; Wu and Birge, 2014). Table 9provides a detailed look at the coverage rate across the five two-digitSIC industry categories. All five categories show similar coverage rates,with the lowest and the highest coverage rates at 66% and 82%, re-spectively. Thus, the availability of supply base data does not appear tobe particularly poor for industries in our sample.

We next consider whether the estimated effects of supply basecomplexity will be systematically biased using a selected sample offirms. In other words, if we have data for all 1150 firms, will the esti-mates change systematically? Based on the classic Heckman's sampleselection model (Heckman, 1979), the existence of such bias dependson whether the selection mechanism (why some firms are more likely tohave supplier data) is correlated with financial performance. Avail-ability of supplier data could be due to media coverage of either thebuyer or its suppliers. Since media coverage is more likely for re-markable events (McCarthy et al., 1996) such as considerable perfor-mance improvement or decline, we expect a correlation between theselection mechanism and financial performance, despite the fact thatthe direction of this correlation is a priori unclear. Motivated by this, we

Table 8Robustness check – different DV and alternative spatial complexity.

Annual DV Product Spatial Subsidiary Spatial

ROA TQ ROA TQ ROA TQ

(1) (2) (3) (4) (5) (6)

Horizontal (×10−2) 0.051∗∗ (0.016) 0.360∗ (0.143) 0.012∗∗ (0.005) 0.342∗ (0.135) 0.012∗∗ (0.005) 0.322∗ (0.134)Horizonta2 (×10−4) −0.004∗∗ (0.001) −0.027∗∗ (0.010) −0.001∗∗ (0.001) −0.023∗ (0.010) −0.001∗∗ (0.001) −0.022∗ (0.010)Vertical (×10−2) −0.051 (0.066) −0.168 (0.345) −0.027 (0.031) −0.237 (0.364) −0.025 (0.031) −0.212 (0.366)Vertical2 (×10−4) 0.030 (0.035) 0.075 (0.190) 0.011 (0.014) 0.166 (0.186) 0.010 (0.014) 0.158 (0.187)Spatial −0.017∗∗∗ (0.005) −0.086∗∗ (0.031) −0.007∗∗∗ (0.002) −0.106∗∗ (0.035) −0.005∗ (0.002) −0.085∗∗ (0.030)Spatial2 0.001∗∗ (0.0002) 0.003∗ (0.001) 0.0004∗∗∗ (0.0001) 0.004∗ (0.002) 0.0002∗ (0.0001) 0.003∗ (0.001)Eliminative −0.294∗∗ (0.097) −1.221∗ (0.610) −0.171∗∗∗ (0.050) −1.303∗ (0.646) −0.167∗∗ (0.051) −1.122† (0.644)Eliminative2 0.233∗ (0.107) 1.309+ (0.694) 0.147∗∗ (0.049) 1.359† (0.712) 0.141∗∗ (0.050) 1.150† (0.693)Cooperative 0.434∗∗ (0.148) 2.594∗ (1.083) 0.179∗∗ (0.060) 2.882∗∗ (1.091) 0.189∗∗ (0.062) 2.911∗∗ (1.100)Cooperative2 −0.434∗ (0.172) −2.914∗∗ (1.121) −0.168∗ (0.068) −3.131∗∗ (1.097) −0.177∗ (0.069) −3.148∗∗ (1.099)NumProducts 0.002† (0.001) −0.021∗ (0.009) −0.0001 (0.001) −0.013 (0.010) −0.0002 (0.001) −0.018† (0.010)NumCustomers −0.001∗ (0.0003) −0.002 (0.002) 0.00001 (0.0001) 0.001 (0.002) −0.00004 (0.0001) −0.001 (0.002)InvEff −0.0001 (0.013) 0.132∗ (0.052) 0.004 (0.004) 0.128∗ (0.052) 0.004 (0.004) 0.135∗ (0.053)MktEff 0.024∗∗ (0.008) 0.174∗∗ (0.064) 0.003 (0.004) 0.171∗∗ (0.065) 0.003 (0.004) 0.171∗ (0.067)PdtEff 0.054∗∗∗ (0.009) −0.261∗∗∗ (0.041) 0.013∗∗∗ (0.003) −0.288∗∗∗ (0.048) 0.014∗∗∗ (0.003) −0.278∗∗∗ (0.048)SalesLog 0.069∗∗∗ (0.006) 0.114∗∗∗ (0.031) 0.021∗∗∗ (0.002) 0.104∗∗ (0.032) 0.021∗∗∗ (0.002) 0.107∗∗∗ (0.032)SalesGrowthLog 0.001 (0.008) −0.044 (0.032) −0.002 (0.002) −0.057∗ (0.028) 0.001 (0.002) −0.014 (0.029)BookToMkt 0.018∗∗ (0.005) −0.037 (0.054) 0.00001 (0.001) −0.032 (0.055) −0.001 (0.001) −0.034 (0.055)DebtToEquity −0.001∗ (0.001) −0.010∗ (0.004) 0.001 (0.001) −0.009∗ (0.004) 0.001 (0.001) −0.008† (0.004)MktShare −0.281∗∗∗ (0.052) −0.336 (0.286) −0.092∗∗∗ (0.020) −0.363 (0.323) −0.093∗∗∗ (0.020) −0.356 (0.320)ShareholderLog 0.002 (0.003) 0.014 (0.017) −0.00002 (0.001) 0.007 (0.018) 0.0003 (0.001) 0.011 (0.018)BoardSize −0.003∗ (0.001) −0.005 (0.006) −0.001+ (0.0003) −0.003 (0.006) −0.001+ (0.0003) −0.003 (0.006)BoardTenure 0.008∗∗∗ (0.001) 0.014† (0.008) 0.003∗∗∗ (0.001) 0.014 (0.009) 0.003∗∗∗ (0.001) 0.015† (0.009)BoardAge −0.003 (0.002) −0.007 (0.011) −0.001 (0.001) −0.010 (0.011) −0.001 (0.001) −0.011 (0.011)BoardCOO −0.007 (0.021) −0.078 (0.117) −0.015 (0.010) −0.097 (0.122) −0.016 (0.011) −0.107 (0.122)Constant −0.395∗ (0.158) 1.268 (0.783) −0.069 (0.059) 1.617∗ (0.754) −0.091 (0.059) 1.169 (0.731)4-digit SIC Dummies Yes Yes Yes Yes Yes Yes

Observations 867 867 867 867 867 867R2 0.534 0.242 0.415 0.228 0.414 0.226

†p < 0.1; *p < 0.05; **p < 0.01; ***p < 0.001. Robust standard errors in parentheses.

10 We thank the review team for bringing this issue to our attention.

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

39

use the Heckman model to assess the magnitude and significance of thispotential bias. Adapting the standard two-equation setup to our con-text, we have:

= +ROA TQ X εOutcome Equation: { , } Bi i i i

= ⎧⎨⎩

+ >C Z τotherwise

Selection Equation: 1 if  Γ 00i

i i

where Ci is the coverage dummy that denotes whether a firm has supplybase data available and hence is covered in the final sample, Xi and Ziare variable vectors, and B and Γ are coefficient vectors. The error termsfollow a bivariate normal distribution such that ∼ε τ N, (0,Σ)i i , where

⎜ ⎟= ⎛⎝

⎞⎠

σ ρσρσ

Σ1

2.11 Note that specification of the outcome equation is

identical to Model (6) in Tables 6 and 7 When fitting the Heckmanmodel, it is recommended to have at least one variable in the selectionequation that does not belong to the outcome equation (i.e., exclusionrestriction). We believe firms existing for a longer time, as measured bythe number of years since incorporation (FirmAge), are more likely to bethe targets of business intelligence. As a result, FirmAge should sig-nificantly affect a firm's supplier data availability. However, there is noreason to believe older firms should have a higher or lower financialperformance (Anderson and Reeb, 2003). Therefore, we use FirmAge asthe exclusion restriction variable in the selection equation, which alsoincludes all Compustat-based variables in the outcome equation.Table 10 reports the two-step estimation results of the Heckman modelfor both ROA and TQ. To assess the bias caused by sample selection, wecompare the complexity effects in Table 10 with those in the mainanalysis. Strong consistency in statistical significance and effect sizesare observed in both cases. We further conduct a Hausman (1978) testto compare the systematic differences between the ten complexitycoefficients in Table 10 and those in Model (6) of Tables 6 and 7 All areinsignificant at 0.05 level. To conclude, although it is plausible to be-lieve the correlation between the selection mechanism and financialperformance might cause a bias in the estimated complexity effects, ouranalysis shows that the size of this bias is very small. More importantly,explicitly accounting for selectivity using a more complicated modeldoes not result in systematically different coefficient estimates for theeffects of supply base complexity on financial performance.

6. Discussion and conclusion

Managing the increasing complexity of global supply chains is aburden for many modern manufacturers (Bozarth et al., 2009; Chopraand Sodhi, 2014). To do this effectively, a firm has to go beyond justbuilding long-term relationships with key suppliers. It also needs tounderstand the structural links among its suppliers, its customers anditself. Prior studies of supply chain structure have largely focused onnetwork-level characteristics. The effects of structural attributes at thesupply-base level are not well understood. Additionally, not-so-visible

structural links are overlooked despite their potentially large impact onperformance.

6.1. Theoretical contributions

Our study offers evidence that supply base structure significantlyinfluences buyers’ financial performance and therefore reinforces theseldom-tested perspective that appropriate supply chain design enablesa firm to compete more effectively. Our findings provide more detailedinsights than those suggested by the literature: (1) we observe thatwhile not all supply base structural complexity dimensions affect

Table 9Supply base data coverage by SIC code.

Description SIC Compustat Cleaned Final Sample Coverage

Chemicals and Allied Products 28 333 250 75%Industrial and Commercial Machinery and Computer Equipment 35 180 148 82%Electronic and other Electrical Equipment and Components, except Computer Equipment 36 293 232 79%Transportation Equipment 37 100 76 76%Measuring, Analyzing, and Controlling Instruments; Photographic, Medical and Optical Goods; Watches and Clocks 38 244 161 66%Total 1150 867 75%

Note: SIC description accessed from https://www.osha.gov/pls/imis/sic_manual.html.

Table 10Robustness check – sample selection model for testing selection bias.

ROA (OLS) TQ (OLS) Selection (Probit)

(1) (2) (3)

Horizontal(×10−2)

0.015∗∗ (0.005) 0.370∗∗ (0.140)

Horizonta2

(×10−4)−0.002∗∗ (0.001) −0.025∗ (0.011)

Vertical(×10−2)

−0.028 (0.029) −0.149 (0.354)

Vertical2

(×10−4)0.013 (0.014) 0.114 (0.182)

Spatial −0.004∗ (0.002) −0.091∗∗ (0.032)Spatial2 0.0002∗ (0.0001) 0.003∗ (0.001)Eliminative −0.162∗∗∗ (0.046) −1.335∗ (0.625)Eliminative2 0.142∗∗ (0.046) 1.354† (0.694)Cooperative 0.173∗∗ (0.057) 2.982∗∗ (1.062)Cooperative2 −0.168∗ (0.065) −3.243∗∗ (1.078)NumProducts −0.0004 (0.001) −0.019∗ (0.010)NumCustomers −0.00005 (0.0001) −0.001 (0.002)ShareHolderLog 0.0003 (0.001) 0.017 (0.018)BoardSize −0.001∗ (0.0003) −0.003 (0.006)BoardTenure 0.002∗∗∗ (0.001) 0.014 (0.009)BoardAge −0.001 (0.001) −0.010 (0.010)BoardCOO −0.018† (0.010) −0.126 (0.120)InvEff 0.007∗ (0.004) 0.137∗∗ (0.052) −0.053 (0.051)MktEff 0.011∗∗ (0.004) 0.183∗∗ (0.069) −0.121∗∗ (0.046)PdtEff 0.018∗∗∗ (0.003) −0.275∗∗∗ (0.050) −0.073 (0.055)SalesLog 0.002 (0.004) 0.083 (0.052) 0.270∗∗∗ (0.028)SalesGrowthLog 0.010∗∗ (0.003) −0.035 (0.042) −0.575 (1.967)BookToMkt 0.002 (0.002) −0.032 (0.057) −0.026 (0.024)DebtToEquity −0.00000 (0.001) −0.013∗∗ (0.005) 0.011 (0.017)MktShare 0.021 (0.026) −0.203 (0.401) −1.687∗∗∗ (0.401)Inverse Mills

Ratio (IMR)−0.185∗∗∗ (0.042) −0.383 (0.471)

Age 0.005∗ (0.002)Constant −0.052 (0.049) 1.819∗ (0.724) 6.601 (21.480)4-digit SIC

DummiesYes Yes Yes

Observations 867 867 1150R2 0.433 0.224Log Likelihood −503.733

Note: †p < 0.1,*p < 0.05; **p < 0.01; ***p < 0.001. Estimation carried out using thestandard two-step procedure. Selection equation is the same for the outcome equations ofROA and TQ.

11 σ is the standard deviation of εi. Error variance of the selection equation is nor-malized to 1 for identification purpose. ρ is the correlation between the two error terms.

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

40

financial performance, some exert a strong influence; (2) their impactson ROA and TQ are similar but not identical; and (3) the effects ofcomplexity dimensions often are not straightforward; the overall effect(either positive or negative) of a complexity dimension is contingent onthe magnitude of the complexity dimension itself, despite the fact thatthe literature generally suggests a linear negative impact.

6.1.1. Impacts of horizontal, vertical, and spatial complexityAmong the three dimensions directly involving the focal firm, only

horizontal and spatial complexities exhibit significant impact on fi-nancial performance. The inverted U-shaped effect of horizontal com-plexity is consistent with the OM literature which suggests that a smallnumber of 1st tier suppliers (per weighted product group) makes ef-fective control possible (Inman et al., 2011; Schmenner and Swink,1998; Vastag, 2000) while a large number engenders uncertainties thatmay strain a firm's ability to cope. Most firms (806 firms) in our data areunder a horizontal complexity threshold that allows them to enjoy thebenefits of risk mitigation and supplier specialization. Fig. 4 Panel A1(we use ROA results as an example, but the results are quite qualita-tively consistent across the two dependent variables) shows that ben-efits dominate costs until horizontal complexity exceeds 139. Using fiveproduct groups (the average number of product groups in our sample is4.8) and 0.1 for moderate product group frequency, the weightedmeasure of 139 converts into 70 1st tier suppliers, implying that 70 isthe limit that a firm can effectively manage. Prior studies (e.g., Ogden,2006; Sarkar and Mohapatra, 2006) advocate supply base reductionbecause a smaller supply base leads to lower prices (by increasingpurchasing volume among fewer selected suppliers), fewer transactions,and better collaboration (with the chosen few). Anecdotal stories fromthe press suggest that many firms are aware of this and significantly cuttheir supply bases. For example, Ford reduced its stamping supply basefrom 150 suppliers to 11 (Wincel, 1999). However, our findings suggestthat firms should exercise caution when cutting their supply bases;many firms have the potential to improve both ROA and TQ by in-creasing multiple sourcing.

The U-shaped relation between spatial complexity and ROA/Tobin'sQ suggests that the initial costs of adopting a global sourcing modelmay hurt financial performance. However, as the level of spatial com-plexity increases, the benefits ultimately exceed the costs. The increasein global sourcing in the last few decades offers corroboration (Jainet al., 2014). Surprisingly, Fig. 4 Panel B1 shows a majority of firmsreside in the “disadvantaged zone” (Kotabe and Murray, 2004) wherespatial complexity has a negative marginal impact. The figure alsosuggests that no firm in our sample enjoys the rents of spatial com-plexity when ROA is considered. As we alluded in hypothesis devel-opment, the benefits of spatial complexity will increase exponentiallydue to facilitated knowledge sharing and resource orchestration. Theresult perhaps implies that most firms are still underexploring theglobal market. We also tested an alternative measure of spatial com-plexity—the product of the percentage of foreign suppliers and thenumber of foreign countries. We observe similar U-shaped effects as inthe main analysis.

While Bode and Wagner (2015) show that all the three visiblecomplexity dimensions are negatively associated with the frequency ofsupply disruption, our study suggests that horizontal and spatial com-plexity have curvilinear impacts when financial, rather than operational,performance is considered in the context of supply base management.Both dimensions may have positive and negative impacts and the re-lative strengths of the impacts are contingent upon the magnitude of thedimension itself. The negligible effect of vertical complexity appears tovalidate our focus on the supply base, suggesting that the impact of adeep upstream supplier on the focal firm's financial performance islimited.

6.1.2. Impacts of eliminative and cooperative complexityEliminative and cooperative complexities appear to have strong

financial impacts on the buyer. In practice, not-so-visible structuralcomplexity is relatively hard for the buyer to monitor. It is quite pos-sible that firms are less prepared for less-visible conditions and there-fore their impacts may be unexpectedly severe. Nonetheless, our find-ings add detailed insights to the literature. First, Rossetti and Choi(2005) conceptual work suggests that eliminative complexity nega-tively impacts the focal firm as it removes the buyer's structural holeadvantages. Our findings, on the other hand, suggest that this negativemarginal effect of eliminative complexity decreases. Customers' andsuppliers' ability to leverage their connections to exploit the focal firmdiminishes, owing to their bounded rationalities. They cannot explicitlyinterpret the large amount of information at high levels of eliminativecomplexity. Alternatively, it is also possible that buyers can proactivelyadapt to a changing environment (Choi et al., 2001) and mitigate thenegative impact of eliminative complexity. Fig. 4 Panel C1 shows thatthe marginal effect of eliminative complexity continues to decrease andapproaches zero when it reaches about 0.48. There are only very fewcases (27 or 3.1%) where eliminative complexity has a positive mar-ginal impact, which is largely consistent with our prediction that theeffect of eliminative complexity is non-positive.

Second, the conceptual advances by Pathak et al. (2014) proposethat cooperation among suppliers can be beneficial. However, theirconceptualization focuses on co-opetition, suggesting that the co-operating suppliers are also competitors. Our study relaxes such con-straint and empirically demonstrates the “net” positive effect of sup-plier cooperation on ROA and TQ. Fig. 4 Panel D1 suggests that 844firms benefit from the links between its suppliers, thanks to improvedinnovation performance (Ahuja, 2000) and spillover effect of suppliercoordination. Leveraging suppliers’ intellectual properties helps buyerscut costs and reduce the time to market (Choi and Krause, 2006). Whileprior studies on supplier development (Hartley and Choi, 1996) ad-vocate building relationships with “core” suppliers, our findings suggestthat the buyer should also encourage cooperation among its suppliers.Alternatively, our findings may suggest that the performance rents aremainly constrained by meso-level uncertainty (lack of information).The impact of macro-level uncertainty (equivocality) is negligible. In-creases in cooperative complexity promote information sharing but donot necessarily increase equivocality. Fig. 4 Panel D1 shows only 12firms with the highest cooperative complexity experienced negativeeffects. Combining information from different subunits of a firm leadsto new and useful knowledge (Huber, 1991). For example, a firm maylearn about a stock-out problem when comparing information from thesales department and the warehouse. Flynn et al. (2016) argue thatfirm-level insights are applicable at the supply-base level. Our resultsprovide empirical evidence to support this viewpoint. Multiple con-nections between suppliers not only reduce information asymmetry butalso allow connected firms to extract “useful” information that aidsdecision-making, especially when considering the fact that organiza-tions often “do not know what they know” (Brown and Duguid, 1991).

Collectively, our findings demonstrate that not all structural com-plexity dimensions have a significant impact on the buyer's financialperformance when other dimensions are controlled for. These sig-nificant effects are not always consistent with intuition; an increase in“complexity” may boost performance (cooperative complexity, for ex-ample). The curvilinear relations suggest that a complexity dimensioncan “transit” from facilitator to inhibitor (or vice versa), dependingupon its relative magnitude.

6.2. Managerial implications

Our study offers several managerial implications. First and fore-most, while recent industry reports by leading consulting firms (e.g.,KPMG, 2011) urge managing supply chains beyond a firm's immediatesuppliers, our results advise that the supplier base consisting of pri-marily 1st tier suppliers should not be overlooked. Structural links in-volving 1st tier suppliers have strong implications for financial

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

41

performance.Second, decision makers should not treat supply base structural

dimensions equally. Each dimension has its own value: some are large,some are small and they exert complicated effects. These curvilinearrelations suggest that there might be an optimal level of complexity.Our empirical results may provide guidance for firms to adjust theirsupply bases to improve financial performance. In addition, given theresource constraints they face today, many firms have to focus theirefforts on the complexity dimensions that contribute most to financialperformance. For example, our findings suggest that eliminative com-plexity exerts its strongest marginal impact at low levels. It is thusworth investing resources to track the structural links originating with1st tier suppliers continuously, especially in the early stages of change.

Third, there may be a trade-off between eliminative and cooperativeat low levels. A “well-connected” supplier may pose eliminative threatsbut at the same time generate cooperative rents. Buyers need to assesswhat type of suppliers best fit their particular needs. For example, if thebuyer wants to boost innovation performance, a well-connected sup-plier is preferred. However, if the goal is merely to find a low-costsource of parts, a less-connected supplier may be favored.

Last but not least, these complexity dimensions affect ROA and TQdifferently. Overall, the effect is smaller when TQ is considered. TQreflects a firm's long-term performance and intangible assets. The cur-rent supply network may not fit the firm's future needs due to productchanges and market dynamics. Thus, the value of the current supplystructure is far more uncertain and likely to be deferred in the future.

6.3. Limitations and future research

This study is subject to a few limitations, each offering opportunitiesfor future research. First, despite our efforts to mitigate endogeneity, weacknowledge that the snap-shot nature of the data prevents strongcausal interpretations. Future research can adopt a panel data approachas such data becomes available. Second, it is possible that the visibleand not-so-visible dimensions are related. We do not test the interactioneffects due to lack of theoretical support in the literature. Nonetheless,we think this is a promising area to advance supply base complexityresearch. Third, the current study emphasizes financial performance.Future research can implement a similar analysis strategy but in-vestigate a broader set of critical performance metrics such as cost,quality, delivery and flexibility.

References

Ahuja, G., 2000. Collaboration networks, structural holes, and innovation: a longitudinalstudy. Adm. Sci. Q. 45, 425–455.

Almeida, P., Kogut, B., 1999. Localization of knowledge and the mobility of engineers inregional networks. Manag. Sci. 45, 905–917.

Anderson, R.C., Reeb, D.M., 2003. Founding-family ownership and firm performance:evidence from the S & P 500. J. Finance 58 (3), 1301–1328.

Anderson, R.C., Mansi, S.A., Reeb, D.M., 2004. Board characteristics, accounting reportintegrity, and the cost of debt. J. Account. Econ. 37 (3), 315–342.

Andres, Christian, 2008. Large shareholders and firm performance—An empirical ex-amination of founding-family ownership. J. Corp. Finance 14 (4), 431–445.

Ang, S., Inkpen, A.C., 2008. Cultural intelligence and offshore outsourcing success: aframework of firm-level intercultural capability. Decis. Sci. 39 (3), 337–358.

Angrist, J.D., Pischke, J., 2009. Mostly Harmless Econometrics: an Empiricist'sCompanion. Princeton university press, Princeton 2009.

Audretsch, D.B., Feldman, M.P., 1996. R & D spillovers and the geography of innovationand production. Am. Econ. Rev. 86, 630–640.

Babich, V., 2006. Vulnerable options in supply chains: effects of supplier competition.Nav. Res. Logist. 53 (7), 656–673.

Balakrishnan, S., Wernerfelt, B., 1986. Technical change, competition and vertical in-tegration. Strategic Manag. J. 7, 347–359.

Barney, J., 1991. Firm resources and sustained competitive advantage. J. Manag. 17,99–120.

Bauer, D.J., Curran, P.J., 2005. Probing interactions in fixed and multilevel regression:inferential and graphical techniques. Multivar. Behav. Res. 40 (3), 373–400.

Beckman, C.M., Haunschild, P.R., Damon, J., 2004. Friends or strangers? Firm-specificuncertainty, market uncertainty and network partner selection. Organ. Sci. 15 (3),259–275.

Bellamy, M.A., Ghosh, S., Hora, M., 2014. The influence of supply network structure on

firm innovation. J. Operations Manag. 32 (6), 357–373.Bharadwaj, A.S., Bharadwaj, S.G., Konsynski, B.R., 1999. Information technology effects

on firm performance as measured by Tobin's q. Manag. Sci. 45 (7), 1008–1024.Blackhurst, J.V., Craighead, C.W., Elkins, D., Handfield, R.B., 2005. An empirically de-

rived agenda of critical research issues for managing supply chain disruptions. Int. J.Prod. Res. 43 (19), 4067–4081.

Bleeke, J., Ernst, D., 1995. Is your strategic alliance really a sale? Harv. Bus. Rev. 73 (1),97–105.

Bode, C., Wagner, S.M., 2015. Structural drivers of upstream supply chain complexity andthe frequency of supply chain disruptions. J. Operations Manag. 36, 215–228.

Bozarth, C.C., Warsing, D.P., Flynn, B.B., Flynn, E.J., 2009. The impact of supply chaincomplexity on manufacturing plant performance. J. Operations Manag. 27 (1),78–93.

Brown, J.S., Duguid, P., 1991. Organizational learning and communities-of-practice: to-ward a unified view of working, learning, and innovation. Organ. Sci. 2 (1), 40–57.

Burt, R.S., 1992. Structural Holes: the Social Structure of Competition. Harvard UniversityPress, Cambridge, MA.

Chase, R.B., Jacobs, R.F., Aquilano, N.J., 2006. Operations Management for CompetitiveAdvantage, eleventh ed. McGraw-Hill Irwin, New York, NY.

Chaturvedi, A., Martínez-de-Albéniz, V., 2011. Optimal procurement design in the pre-sence of supply risk. Manuf. Serv. Operations Manag. 13 (2), 227–243.

Choi, T.Y., Hong, Y., 2002. Unveiling the structure of supply networks: case studies inHonda, Acura, and Daimler Chrysler. J. Operations Manag. 20 (5), 469–493.

Choi, T.Y., Krause, D.R., 2006. The supply base and its complexity: implications fortransaction costs, risks, responsiveness, and innovation. J. Operations Manag. 24 (5),637–652.

Choi, T.Y., Dooley, K.J., Rungtusanatham, M., 2001. Supply networks and complexadaptive systems: control versus emergence. J. Operations Manag. 19 (3), 351–366.

Chon, B.S., 2004. The dual structure of global networks in the entertainment industry:interorganizational linkage and geographical dispersion. Int. J. Media Manag. 6(3 & 4), 194–206.

Chopra, S., Sodhi, M.S., 2014. Reducing the risk of supply chain disruptions. SloanManag. Rev. 55 (3), 73–80.

Chung, K.H., Pruitt, S.W., 1994. A simple approximation of Tobin's q. Financ. Manag. 23(3), 70–74.

Coad, A., Rao, R., 2008. Innovation and firm growth in high-tech sectors: a quantileregression approach. Res. Policy 37, 633–648.

Cohen, W.M., Nelson, R.R., Walsh, J.P., 2000. Protecting Their Intellectual Assets:Appropriability Conditions and Why US Manufacturing Firms Patent (Or Not). NBERWorking Paper 7552.

de Leeuw, T., Keijl, S., 2015. Research with secondary data: different matching methodsand is there a difference? In: Academy of Management Proceedings, pp. 14215.

Duncan, R.B., 1972. Characteristics of organizational environmental and perceived en-vironmental uncertainties. Adm. Sci. Q. 17 (3), 313–327.

Ellis, J.A., Fee, C.E., Thomas, S.E., 2012. Proprietary costs and the disclosure of in-formation about customers. J. Account. Res. 50 (3), 685–727.

Ellram, L.M., Tate, W.L., Billington, C., 2008. Offshore outsourcing of professional ser-vices: a transaction cost economics perspective. J. Operations Manag. 26, 148–163.

Felin, T., Hesterly, W.S., 2007. The knowledge-based view, nested heterogeneity, and newvalue creation: philosophical considerations on the locus of knowledge. Acad. Manag.Rev. 32 (1), 195–218.

Flynn, B.B., Huo, B., Zhao, X., 2010. Supply chain collaboration: capabilities for con-tinuous innovation: a contingency and configuration approach. J. Operations Manag.28 (1), 58–71.

Flynn, B.B., Koufteros, X., Lu, G., 2016. On theory in supply chain uncertainty and itsimplications for supply chain integration. J. Supply Chain Manag. 52, 3–27.

Forbes, 2010. Outsourcing Can Do Much More than Just Cut Costs. Available at: http://www.forbes.com/2010/06/15/outsourcing-capability-sourcing-leadership-managing-bain.html [Last assessed Sept. 29, 2016].

Fuld, L.M., 1988. Monitoring the Competition: Find Out What's Really Going on overThere. John Wiley, New York.

Galbraith, J., 1973. Designing Complex Organizations. Addison-Wesley, Reading,Massachusetts.

Gao, G.Y., Xie, E., Zhou, K.Z., 2015. How does technological diversity in supplier networkdrive buyer innovation? Relational process and contingencies. J. Operations Manag.36, 165–177.

Germain, R., Claycomb, C., Droge, C., 2008. Supply chain variability, organizationstructure and performance: the moderating effect of demand unpredictability. J.Operations Manag. 26, 557–570.

Gilley, K.M., Rasheed, A., 2000. Making more by doing less: an analysis of outsourcingand its effects on firm performance. J. Manag. 26 (4), 763–790.

Grover, V., Malhotra, M.K., 2003. Transaction cost framework in operations and supplychain management research: theory and measurement. J. Operations Manag. 21 (4),457–473.

Guide Jr., D.R., Ketokivi, M., 2015. Notes from the Editors: redefining some methodo-logical criteria for the journal. J. Operations Manag. 37 v-viii.

Hamel, G., Doz, Y.L., Prahalad, C.K., 1989. Collaborate with your competitors-and win.Harv. Bus. Rev. 67, 133–139.

Handfield, R.B., Nichols, E.L., 1999. Introduction to Supply Chain Management. Prentice-Hall, Englewood Cliffs, NJ.

Hargadon, A., Sutton, R.I., 1997. Technology brokering and innovation in a productdevelopment firm. Adm. Sci. Q. 42, 716–749.

Harrigan, K.R., 1985. Vertical integration and corporate strategy. Acad. Manag. J. 28 (2),397–425.

Hartley, J.L., Choi, T.Y., 1996. Supplier development: customers as a catalyst for processchange. Bus. Horizons 39, 37–44.

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

42

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

Heckman, J.J., 1979. Sample selection bias as a specification error. Econometrica 47, 153.Hendricks, K.B., Singhal, V.R., 2009. Demand-supply mismatches and stock market re-

action: evidence from excess inventory announcements. Manuf. Serv. OperationsManag. 11 (3), 509–524.

Hendricks, K.B., Singhal, V.R., Zhang, R., 2009. The effect of operational slack, diversi-fication, and vertical relatedness on the stock market reaction to supply chain dis-ruptions. J. Operations Manag. 27 (3), 233–246.

Hendricks, K.B., Hora, M., Singhal, V.R., 2014. An empirical investigation on the ap-pointments of supply chain and operations management executives. Manag. Sci. 61(7), 1562–1583.

Hoffman, A.J., 1999. Institutional evolution and change: environmentalism and the U.S.Chemical industry. Acad. Manag. J. 42 (4), 351–371.

Holcomb, T.R., Hitt, M.A., 2007. Toward a model of strategic outsourcing. J. OperationsManag. 25 (2), 464–481.

Huber, G.P., 1991. Organizational learning: the contributing processes and the literature.Organ. Sci. 2 (1), 88–115.

Iglewicz, B., Hoaglin, D.C., 1993. How to Detect and Handle Outliers. American Societyfor Quality Control, Milwaukee, WI.

Inman, R.A., Sale, R.S., Green Jr., K.W., Whitten, D., 2011. Agile manufacturing: relationto JIT, operational performance and firm performance. J. Operations Manag. 29 (4),343–355.

Jacobs, M.A., Swink, M., 2011. Product portfolio architectural complexity and opera-tional performance: incorporating the roles of learning and fixed assets. J. OperationsManag. 29 (7–8), 677–691.

Jacobs, B.W., Kraude, R., Narayanan, S., 2016. Operational productivity, corporate socialperformance, financial performance, and risk in manufacturing firms. Prod.Operations Manag. 25 (12), 2065–2085.

Jain, N., Girotra, K., Netessine, S., 2014. Managing global sourcing: inventory perfor-mance. Manag. Sci. 60 (5), 1202–1222.

Kearney, A.T., 2007. How Much Does Complexity Really Cost? Available at: https://www.atkearney.com/documents/10192/178350/complexity_secure.pdf/cd044988-91e9-4331-8afe-d72265c745e3 [Last assessed Dec. 28, 2015].

Kelle, P., Silver, E.A., 1990. Safety stock reduction by order splitting. Nav. Res. Logist. 37(5), 725–743.

Ketokivi, M., McIntosh, C.N., 2017. Addressing the endogeneity dilemma in operationsmanagement research: theoretical, empirical, and pragmatic considerations. J.Operations Manag. 52, 1–14.

Khurana, A., 1999. Managing complex production processes. Sloan Manag. Rev. 40 (2),85–97.

Kim, D., 2014. Understanding supplier structural embeddedness: a social network per-spective. J. Operations Manag. 32 (5), 219–231.

Kim, Y., 2017. The effects of major customer networks on supplier profitability. J. SupplyChain Manag. 53 (1), 26–40.

Kim, Y., Choi, T.Y., Yan, T., Dooley, K., 2011. Structural investigation of supply networks:a social network analysis approach. J. Operations Manag. 29 (3), 194–211.

Kim, Y., Chen, Y., Linderman, K., 2015. Supply network disruption and resilience: anetwork structural perspective. J. Operations Manag. 33–34, 43–59.

Kogut, B., Zander, U., 1992. Knowledge of the firm, combinative capabilities. ReplicationTechnol. Organ. Sci. 3 (3), 383–397.

Kotabe, M., Murray, J.Y., 2004. Global sourcing strategy and sustainable competitiveadvantage. Ind. Mark. Manag. 33, 7–14.

Koufteros, X., Lu, G., Peters, R., Lai, K., Wong, C.W.Y., Cheng, T.C.E., 2014. Productdevelopment practices, manufacturing practices, and performance: a mediationalperspective. Int. J. Prod. Econ. 156, 83–97.

KPMG, 2011. Supply chain Complexity: Managing Constant Change. Available at:https://www.kpmg.com/US/en/IssuesAndInsights/ArticlesPublications/Documents/supply-chain-complexity-managing-constant-change.pdf [Last assessed Dec. 28,2015].

Krause, D.R., Handfield, R.B., 1999. Developing a World-class Supply Base. Center forAdvanced Purchasing Studies, Tempe, AZ.

Lahiri, N., 2010. Geographic distribution of R & D activity: how does it affect innovationquality? Acad. Manag. J. 53 (5), 1194–1209.

Lee, H.L., Padmanabhan, V., Whang, S., 1997. Information distortion in a supply chain:the bullwhip effect. Manag. Sci. 43 (4), 546–558.

Li, M., Choi, T.Y., 2009. Triads in services outsourcing: bridge, bridge decay and bridgeTransfer. J. Supply Chain Manag. 45, 27–39.

MacDuffie, J.P., Sethuraman, K., Marshall, L.F., 1996. Product variety and manufacturingperformance: evidence from the international automotive assembly plant study.Manag. Sci. 42 (3), 350–369.

Mackelprang, A.W., Habermann, M., Swink, M., 2015. How firm innovativeness andunexpected product reliability failures affect profitability. J. Operations Manag. 38,71–86.

Malhotra, M.K., Ahire, S.L., Shang, G., 2015. Mitigating the impact of functional dom-inance in cross-functional process improvement teams. Decis. Sci. 48 (1), 39–70.

Manuj, I., Sahin, F., 2011. A model of supply chain and supply chain decision-makingcomplexity. Int. J. Phys. Distribution Logist. Manag. 41 (5), 511–549.

McCarthy, J.D., McPhail, C., Smith, J., 1996. Images of protest: dimensions of selectionbias in media coverage of Washington Demonstrations, 1982 and 1991. Am. Sociol.Rev. 61 (3), 478–499.

Mishra, S., Modi, S.B., Animesh, A., 2013. The relationship between information tech-nology capability, inventory efficiency, and shareholder wealth: a firm-level em-pirical analysis. J. Operations Manag. 31 (6), 298–312.

Modi, S.B., Mishra, S., 2011. What drives financial performance–resource efficiency orresource slack?: Evidence from US based manufacturing firms from 1991 to 2006. J.

Operations Manag. 29 (3), 254–273.Oehlert, G.W., 1992. A note on the delta method. Am. Statistician 46 (1), 27–29.Ogden, J.A., 2006. Supply base reduction: an empirical study of critical success factors. J.

Supply Chain Manag. 42, 29–39.Osadchiy, N., Gaur, V., Seshadri, S., 2015. Systematic risk in supply chain networks.

Management Science. Articles Adv. 1–23.Parkhe, A., 1993. Strategic alliance structuring: a game theoretic and transaction cost

examination of interfirm cooperation. Acad. Manag. J. 36 (4), 794–829.Pathak, S.D., Wu, Z., David, J., 2014. Toward a structural view of co-opetition in supply

networks. J. Operations Manag. 32 (5), 254–267.Perrow, C., 1984. Normal Accidents: Living with High-risk Technologies. Basic Books,

New York, NY.Pilling, B.K., Crosby, L.A., Jackson, D.W., 1994. Relational bonds in industrial exchange:

an experimental test of the transaction cost economic framework. J. Bus. Res. 30,237–251.

Porter, M.E., 1980. Competitive Strategy: Techniques for Analyzing Industries andCompetitors. Free Press, New York.

Prahalad, C.K., Hamel, G., 1990. The core competence of the corporation. Harv. Bus. Rev.68 (3), 79–91.

Rabinovich, E., Knemeyer, A.M., Mayer, C.M., 2007. Why do Internet commerce firmsincorporate logistics service providers in their distribution channels? The role oftransaction costs and network strength. J. Operations Manag. 25, 661–681.

Roh, J., Krause, R., Swink, M., 2016. The appointment of chief supply chain officers to topmanagement teams: a contingency model of firm-level antecedents and con-sequences. J. Operations Manag. 44, 48–61.

Rossetti, C., Choi, T.Y., 2005. On the dark side of strategic sourcing: experiences from theaerospace industry. Acad. Manag. Exec. 19 (1), 1–15.

Rossi, P.E., 2014. Even the rich can make themselves poor: a critical examination of IVmethods in marketing applications. Mark. Sci. 33 (5), 655–672.

Roth, A., Gray, J., Shockley, J., Weng, H.H.R., 2015. The Use of Secondary Source Datafor Measuring Performance in Operations Management Research. Available at: SSRN2271202.

Sarkar, A., Mohapatra, P.K.J., 2006. Evaluation of supplier capability and performance: amethod for supply base reduction. J. Purch. Supply Manag. 12, 148–163.

Schmenner, R.W., Swink, M., 1998. On theory in operations management. J. OperationsManag. 17 (1), 97–113.

Senge, P., 2006. The Fifth Discipline: the Art and Practice of the Learning Organization.Doubleday, New York, NY.

Shang, G., Pekgun, P., Ferguson, M., Galbreth, M., 2017. How much do online consumersreally value free product returns? Evidence from eBay. J. Operations Manag (inpress). https://doi.org/10.1016/j.jom.2017.07.001.

Shrivastava, P., Mitroff, I.I., 1984. Enhancing organizational research utilization: the roleof decision makers' assumptions. Acad. Manag. Rev. 9 (1), 18–26.

Simon, H.A., 1962. The architecture of complexity. Proc. Am. Philosophical Soc. 106 (6),467–482.

Simon, H.A., 1979. Rational decision making in business organizations. Am. Econ. Rev.69 (4), 483–513.

Sivadasan, S., Efstathiou, J., Frizelle, G., Shirazi, R., Calivescu, A., 2002. An information-theoretic methodology for measuring the operational complexity of supplier-cus-tomer systems. Int. J. Operations Prod. Manag. 22 (1), 80–102.

Steinhilper, R., Westermann, H.-H., Butzer, S., Haumann, M., Seifert, S., 2012.Komplexitätmessbarmachen. In: ZWF ZeitschriftfürwirtschaftlichenFabrikbetrieb,vol. 107. pp. 361.

Tenhiälä, Antti, Helkiö, Pekka, 2015. Performance effects of using an ERP system formanufacturing planning and control under dynamic market requirements. J.Operations Manag. 36, 147–164.

Thompson, J.D., 1967. Organizations in Action: Social Science Bases of AdministrativeTheory. Transaction Publishers, New Brunswick, NJ.

Tiwana, A., Wang, J., Keil, M., Ahluwalia, P., 2007. The bounded rationality bias inmanagerial valuation of real options: theory and evidence from information tech-nology projects. Decis. Sci. 38 (1), 157–181.

Tolbert, P.S., Hall, R.H., 2009. Organizations: Structures, Processes, and Outcomes, tenthed. Prentice Hall, Upper Saddle River, NJ.

Tushman, M.L., Nadler, D.A., 1978. Information processing as an integrative concept inorganizational design. Acad. Manag. Rev. 3 (3), 613–624.

Vachon, S., Klassen, R.D., 2002. An exploratory investigation of the effects of supplychain complexity on delivery performance. IEEE Trans. Eng. Manag. 49 (3), 218–230.

Vastag, G., 2000. The theory of performance frontiers. J. Operations Manag. 18, 353–360.Villalonga, B., 2004. Intangible resources, Tobin's Q, and sustainability of performance

differences. J. Econ. Behav. Organ. 54 (2), 205–230.von Corswant, F., Fredriksson, P., 2002. Sourcing trends in the car industry: a survey of

car manufacturers' and suppliers' strategies and relations. Int. J. Operations Prod.Manag. 22 (7), 741–747.

Wagner, S.M., Kemmerling, R., 2014. Supply chain management executives in corporateupper echelons. J. Purch. Supply Manag. 20 (3), 156–166.

Wagner, S.M., Grosse-Ruykena, P.T., Erhunb, F., 2012. The link between supply chain fitand financial performance of the firm. J. Operations Manag. 30, 340–353.

Wan, Z., Wu, B., 2016. When suppliers climb the value chain: a theory of value dis-tribution in vertical relationships. Forthcoming at Management Science. Articles Adv.1–20.

Wang, Yixin (Iris), Li, Jun, Anupindi, Ravi, Jun 11, 2017. Risky Suppliers or Risky SupplyChains? An Empirical Analysis of Sub-tier Supply Network Structure on Firm Risk inthe High-Tech Sector. Ross School of Business Paper No. 1297 Available at SSRN:https://ssrn.com/abstract=2705654 https://doi.org/10.2139/ssrn.2705654.

Wilhelm, M.M., 2011. Managing coopetition through horizontal supply chain relations:linking dyadic and network levels of analysis. J. Operations Manag. 29, 663–676.

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

43

Wilhelm, M.M., Blome, C., Bhakoo, V., Paulraj, A., 2016. Sustainability in multi-tiersupply chains: understanding the double agency role of the first-tier supplier. J.Operations Manag. 41, 42–60.

Williamson, O.E., 1975. Markets and Hierarchies: Analysis and Antitrust Implications.The Free Press, New York, NY.

Wilson Perumal & Company, 2015. Supply chain Complexity: the Growing Challenge forPerformance. Available at: http://www.wilsonperumal.com/expertise/index.php[Last assessed Dec. 28, 2015].

Wincel, J.P., 1999. A practitioner's view of strategic procurement. Supply Chain Manag.Rev. 2 (2), 56–64.

Wooldridge, J.M., 2015. Introductory econometrics: A modern approach. NelsonEducation.

Wu, C., Barnes, D., 2012. A dynamic feedback model for partner selection in agile supplychains. Int. J. Operations Prod. Manag. 32 (1), 79–103.

Wu, J., Birge, J.R., 2014. Supply chain Network Structure and Firm Returns. 2014.Available at: SSRN 2385217.

Yan, T., Choi, T.Y., Kim, Y., Yang, Y., 2014. A theory of the nexus supplier: a criticalsupplier from a network perspective. J. Supply Chain Manag. 51 (1), 52–66.

Zhang, J.J., Lawrence, B., Anderson, C.K., 2015. An agency perspective on service triads:linking operational and financial performance. J. Operations Manag. 35, 56–66.

G. Lu, G. Shang Journal of Operations Management 53–56 (2017) 23–44

44