complexity and adaptivity in supply networks

Upload: amimar

Post on 03-Apr-2018

229 views

Category:

Documents


1 download

TRANSCRIPT

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    1/34

    Decision Sciences

    Volume 38 Number 4

    November 2007

    C 2007, The Author

    Journal compilation C 2007, Decision Sciences Institute

    Complexity and Adaptivity in SupplyNetworks: Building Supply Network TheoryUsing a Complex AdaptiveSystems Perspective

    Surya D. Pathak

    Engineering Management Program, School of Engineering, Vanderbilt University, VU StationB 351831, 2301 Vanderbilt Place, Nashville, TN 37235, e-mail: [email protected]

    Jamison M. DayDepartment of Decision and Information Sciences, Bauer College of Business, University ofHouston, Melcher Hall 290D, Houston, TX 77204, e-mail: [email protected]

    Anand NairDepartment of Management Science, Moore School of Business, University of South Carolina,Columbia, SC 29208, e-mail: [email protected]

    William J. Sawaya

    Department of Civil and Environmental Engineering, Cornell University, 220 Hollister Hall,Ithaca, NY 14853, e-mail: [email protected]

    M. Murat KristalOperations Management and Information Systems Department, Schulich School of Business,York University, 4700 Keele Street Toronto, Ontario, Canada M3J 1P3,e-mail: [email protected]

    ABSTRACT

    Supply networks are composed of large numbers of firms from multiple interrelated

    industries. Such networks are subject to shifting strategies and objectives within adynamic environment. In recent years, when faced with a dynamic environment, several

    disciplines have adopted the Complex Adaptive System (CAS) perspective to gain in-

    sights into important issues within their domains of study. Research investigations in the

    field of supply networks have also begun examining the merits of complexity theory and

    the CAS perspective. In this article, we bring the applicability of complexity theory and

    CAS into sharper focus, highlighting its potential for integrating existing supply chain

    management (SCM) research into a structured body of knowledge while also providing

    a framework for generating, validating, and refining new theories relevant to real-world

    supply networks. We suggest several potential research questions to emphasize how a

    We sincerely thank Professors Thomas Choi (Arizona State University), David Dilts (Vanderbilt Uni-versity), and Kevin Dooley (Arizona State University) for their help, guidance, and support.

    Corresponding author.

    547

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    2/34

    548 Complexity and Adaptivity in Supply Networks

    CAS perspective can help in enriching the SCM discipline. We propose that the SCM

    research community adopt such a dynamic and systems-level orientation that brings to

    the fore the adaptivity of firms and the complexity of their interrelations that are often

    inherent in supply networks.

    Subject Areas: Adaptivity, Complex Adaptive System, Complexity, Complexity

    Theory, Decision Making, Supply Chain Management, and Supply Networks.

    INTRODUCTION

    Today, supply chain management (SCM) involves adapting to changes in a com-

    plicated global network of organizations. A typical supply network consists of

    interfirm relationships that may connect multiple industries. As a result, supply

    network decisions often require consideration of a large number of factors frommultiple dimensions and perspectives. Two emergent themes that managers fre-

    quently encounter when making these decisions are (i) the structural intricacies of

    their interconnected supply chains (Choi & Hong, 2002) and (ii) the need to learn

    and adapt their organization in a constantly changing environment to ensure its

    long-term survival (Brown & Eisenhardt, 1998).

    Complex interconnections between multiple suppliers, manufacturers, as-

    semblers, distributors, and retailers are the norm for industrial supply networks.

    When decision making in these networks is based on noncomplex assumptions

    (e.g., linearity, a buyersupplier dyad, sparse connectivity, static environment,

    fixed and nonadaptive individual firm behavior), problems are often hidden, leavingplenty of room for understanding and improving theunderlying processes. Consider

    the recent implementation of complexity-oriented decision making by American

    Air Liquide, a firm based in Houston, Texas. The following information was ac-

    quired through multiple employee interviews, associated document examinations,

    and observations of the Operations Control Center at American Air Liquide. The

    company produces industrial and medical gases such as nitrogen, oxygen, and hy-

    drogen at about 100 manufacturing locations in the United States and delivers to

    nearly 6,000 customer sites using a mix of pipelines, railcars, and more than 400

    trucks. In the past, its distribution routing was based on analytical optimization

    methods. However, this approach had a difficult time integrating environmentalvolatility, feedback from truck drivers, and dynamic sourcing opportunities. Af-

    ter working with NuTech Solutions (formerly Bios Group), they created a new

    complexity-based solution that leverages neural networks and agent-based mod-

    eling (with ant-foraging algorithms) to integrate decisions across their multinodal

    and multimodal supply network. Most important, the new solution method solves

    both sourcing and routing together in the optimization process. Charles Harper,

    director of National Supply & Pipeline and Supply Operations, summarizes the

    benefits of their complexity-based approach:

    After switching over, we drive less miles, we dont do stupid things, and wemove people to different jobs that didnt exist before. All those things add upto savings. Its been mind-blowing to see how much opportunity there was.The knowledge we gained from implementing the complexity-based solutionhelped us realize what the real-time incremental cost of the liquid going intocustomers tanks really was. Our supply network can now flexibly adapt to

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    3/34

    Pathak et al. 549

    volatility in the environment due to differentials in power prices or even hurri-canes. Complexity-based solutions are extremely applicable and people needto start using them or theyre going to lose out.

    American Air Liquide is far from being the only firm that is using the

    structural complexity (interconnectedness offirms) and adaptivity (dynamic learn-

    ing of individual firms) principles of Complex Adaptive Systems (CAS). Boe-

    ing has effectively used CAS principles to redesign their 787 Dreamliner sup-

    ply network, reducing the risk of expensive cascading supply network delays

    (Global Logistics and Supply Chain Strategies, 2007). Similarly, using CAS

    principles, Citibank Credit Risk uncovered $200 million in hidden expenses,

    Proctor and Gamble reduced supply network inventory by 25% and saved 22%

    on distribution expenses, and Southwest Airlines saved $2 million annually in

    their freight delivery operations (Kelly & Allison, 1999; Waldrop, 2003; GlobalLogistics and Supply Chain Strategies, 2007). As seen in these examples, a CAS-

    oriented approach can help firms reap benefits such as increased efficiency, rapid

    flexibility, better preparedness for external uncertainties, increased awareness of

    markets and competition, and improved decision making (Abell, Serra, & Wood,

    1999).

    Along with managing the complexity inherent in the interconnectivity of

    their supply networks, organizations have also started to learn the benefits of being

    adaptive in their behavior. Sheffi and Rice (2005) present an illustration of adaptive

    firm behavior in a cellular telephone supply network. They highlight the different

    approaches that Nokia and Ericsson took when a fire disrupted the supply from

    Philips, the sole supplier for a particular chip common to both manufacturers.

    While Ericsson suffered an estimated $2.34 billion loss, Nokia engaged directly

    with Philips to restore supply using alternate supply options. They modified designs

    of the handsets where possible and secured worldwide manufacturing capacity from

    Philips to ensure a steady supply of the chips. Meanwhile, the direct interaction

    between top management of Nokia and Philips further enhanced the ability of

    Nokia to adapt in the future. Wollin and Perry (2004) provide another example of

    how Honda adapted to the changing automotive sector environment by leveraging

    the notions of learning and path dependency of adaptive systems. They used theirAccord and Civic platforms as the basis of several of their most recent sport utility

    vehicles, and, as a result, they gained significant market share in that segment even

    though they were slow to enter the four-wheel-drive market.

    The pioneering article by Choi, Dooley, and Rungtusanatham (2001) exam-

    ined how properties of CAS are embodied by supply networks. Since this article,

    there have been only a handful of papers that use the CAS view of supply networks,

    signaling that the SCM discipline has yet to enthusiastically embrace the CAS per-

    spective. The intentof this position paper is to draw attention to recent developments

    in CAS theory from across multiple disciplines and articulate how this knowledge

    can be leveraged to enrich the operations management (OM) and SCM disciplines.We suggest leveraging the conceptualizations of Complex Adaptive Supply Net-

    works (CASN), such as those found in Choi et al. (2001) and Surana, Kumara,

    Greaves, and Raghavan (2005), to lay a foundation for both integrating existing

    work and developing new theories within the SCM body of knowledge. Specifically,

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    4/34

    550 Complexity and Adaptivity in Supply Networks

    we discuss how CAS principles can be useful for identification and organization

    of complex and adaptive phenomena in supply networks, such as individual firm

    adaptation, self-organization and emergence, buyersupplier relationships, supply

    network performance, environmental change, and feedback mechanisms. Finally,we examine the challenges associated with CASN theory development and provide

    suggestions for future research efforts and CASN theory development.

    A CAS VIEW OF SUPPLY NETWORKS

    Because organizations exhibit adaptivity and can exist in a complex environment

    with myriad relationships and interactions, it is a natural step to identify a supply

    network as a CAS. Choi et al. (2001) argue that supply networks should be recog-

    nized as CAS by providing a detailed mapping of each property of CAS to a supply

    network. In a similar way, subsequent research has recognized this same inherentcomplexity of supply networks (Surana et al., 2005). For brevity, we use Anderson

    (1999) and Choi et al. (2001) to offer an overview of CAS and its framing of SCM

    research.

    A CAS is an interconnected network of multiple entities (or agents) that

    exhibit adaptive action in response to changes in both the environment and the sys-

    tem of entities itself (Choi et al., 2001). Collective system performance or behavior

    emerges as a nonlinear and dynamic function of the large number of activities made

    in parallel by interacting entities. For example, the individual decisions made by

    firms facing imperfect information and variable demand lead to a globally observed

    phenomenon (i.e., the bullwhip effect) (Lee, Padmanabhan, & Whang, 1997). An-

    derson (1999) outlined four common properties of such systems.

    First, a CAS consists of entities that interact with other entities and with

    the environment by following a set of simple decision rules (i.e., schema). These

    entities may evolve over time as entities learn from their interactions. In contrast

    to relational modeling, which tries to use one set of variables to explain variation

    in another set of variables, CAS examines how changes in an individual entitys

    schema lead to different aggregate outcomes.

    Second, a CAS is self-organizing. Self-organization is a consequence of in-

    teractions between entities. Self-organization is defined as a process in which newstructures, patterns, and properties emerge without being externally imposed on

    the system. Because the behavior in complex systems comes from dynamic inter-

    actions among the agents and between the environment and the agents, the changes

    tend to be nonlinear with respect to the original changes in the system. Thus, there

    may be small changes that have a dramatic effect on the system, or, conversely,

    large changes that have relatively little effect. Choi et al. (2001, p. 357) state, the

    behavior of a complex system cannot be written down in closed form; it is not

    amenable to prediction via the formulation of a parametric model, such as a statis-

    tical forecasting model. Even though it may not be possible to predict the future

    in an exact manner, the future may exhibit some underlying regularity. While thechanges that are made to a system may be dramatic and unpredictable, there may

    be patterns of behavior that can be considered prototypical. Appropriate analyses

    may yield some knowledge of key patterns of behavior that are likely to develop

    in the system over time.

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    5/34

    Pathak et al. 551

    Third, a CAS coevolves to the edge of chaos. Choi et al. (2001) explain

    coevolution, positing that a CAS reacts to and creates its environment so that as

    the environment changes it may cause the agents within it to change, which, in

    turn, cause other changes to the environment. These actions and reactions can betriggered by external events such as natural disasters (e.g., Hurricane Katrina) or

    the actions of agents (e.g., a decision to implement an enterprise resource planning

    system). A CAS exhibits dynamism as changes occur in the environment; this

    dynamism affects the system. Environmental factors may cause changes to which

    the agents must adapt, influencing the way agents perceive their environment or the

    schema used by the agents themselves. Thus, the rules followed by the individual

    entities organize thesystem, because individual entities are notprivy to the objective

    function of the system as a whole. The coevolution of the system happens in the

    rugged fitness landscapes in which the CAS exists. The concept of landscape was

    first introduced by biologist Sewell Wright (1932). It refers to the mapping from anorganisms genetic structure to its fitness level. In management research, the idea of

    landscape is analogous to the domain of social and economic phenomena (Levinthal

    & Warglien, 1999). Specifically, these landscapes may be thought of in terms of

    an analogy of a range of mountains that represents an objective function (i.e.,

    performance function) that is filled with hills and valleys (Kauffman, 1995). The

    hills or peaks represent the desired optimal states, in which a rugged landscape has

    many peaks surrounded by deep valleys. For instance, in the Toyota supply network,

    the flow of goods between its Camry plant and the Johnson Controls seat-frame

    manufacturing plant controlled via a tightly coupled kanban system would reactdifferently to an external event than the flow of goods between Johnson Controls

    seat-frame manufacturing plants and their raw materials suppliers.

    Fourth, a CAS is recursive by nature, and it recombines and evolves over

    time. For example, going back to the bullwhip effect (Lee et al., 1997), the inter-

    firm orders could be characterized as orders from one organizational function to

    another organizational function, orders from an individual employee of one firm to

    an employee of another, or any combination of the involved individuals, functions,

    or firms. Furthermore, from a macroeconomic viewpoint, it can be posited that

    industry supply networks are interrelated within a national or international context

    and interact together as a CAS in a larger context (Arthur, Durlauf, & Lane, 1997).Thus, a CAS is often composed of entities that can themselves be characterized

    as CASs composed of smaller constituents (a nested hierarchy of smaller-scale

    complex systems). Changes in these smaller systems and even in individual entities

    can cause the entire system to change over time.

    Building on these properties, Choi et al. (2001) outline three key foci for

    supply chain research: internal mechanisms, the environment, and coevolution.

    For internal mechanisms, the key elements are agents (entities) and schema, self-

    organization and emergence, network connectivity, and network dimensionality.

    In the context of supply networks, an entity may be an organization, a division, a

    team, or an individual, or even a function of an individual s job. The key feature isthat agents have the ability to make decisions in response to the environment and to

    the action of other entities. In supply networks, schemas are the rules that the orga-

    nizations, or the decision makers within organizations, use to make the decisions

    for, and guide the actions of, the organization. Self-organization and emergence

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    6/34

    552 Complexity and Adaptivity in Supply Networks

    occur as a result of decisions that are made by the individual agents that cause the

    system to change and the collective system behavior to emerge over time. Network

    connectivity is the connection among the agents that determines the complexity

    of the network. As the connectivity among the agents increases, the interrelation-ships among the agents increase, in turn causing increases in the complexity of the

    network. In the case of supply-network relationships these connections are real,

    physical connections between organizations such as telephone lines, fax numbers,

    electronic data interchange systems, and so on. Dimensionality is the degree to

    which agents can act in an autonomous fashion without influencing other agents.

    Therefore, as the degree of connectivity increases, the dimensionality decreases

    as the actions of a given agent has a greater impact on those with which it is

    connected.

    As an example, Choi et al. (2001) present the interconnectivity of an aircraft

    engine manufacturer (Honeywell) with a university hospital (Metro UniversityHospital). Honeywell depends on mining companies for supplies of raw materi-

    als such as steel, copper, aluminum, and other composite materials. These mining

    companies source equipment that relies on the latest material extraction techniques

    developed by various firms and agencies. The material extraction techniques rely

    on pattern recognition technologies that aid in interpretations of X-ray scans of

    potential material vein and enable a firm to make appropriate decisions regarding

    extraction locations. It is conceivable that the required pattern recognition tech-

    nology is developed in a completely unrelated sector, such as health care. For

    example, a university hospital might develop a new pattern recognition techniquefor the purposes of medical treatment that could have potential application in mate-

    rial extraction. Over time, the knowledge gets passed on to the material extraction

    company via research conferences. This example illustrates complex interconnec-

    tivities among firms and the impact of decisions made by one firm on others in the

    network. We present the decisions and information flows among firms in Figure 1.

    Since the initial article on supply chains as CAS by Choi et al. (2001),

    there have been numerous developments in the CAS and network-related liter-

    ature across a wide range of disciplines, such as industrial engineering, computer

    science, physics, organizational science, new product development, and strategicmanagement. In the next section, we highlight these advancements and discuss

    how knowledge gained from these research studies can be beneficial for supply

    network research.

    NEW DEVELOPMENTS IN CAS AND THEIR APPLICABILITY

    TO SCM RESEARCH

    Research endeavors using the CAS perspective have been undertaken in diverse

    fields such as physics, biology, mathematics, computer science, engineering, psy-chology, political science, sociology, economics, and organizational behavior. To

    systematically approach this wide range of literature, we adopted the data trian-

    gulation approach. As a first step, we sought expert opinion regarding the state of

    recent research pertaining to CAS. This step provided an initial reference list and

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    7/34

    Pathak et al. 553

    Figure 1: Example of decision making in supply networks as complex adaptive

    systems (Based on the example in Choi et al., 2001).

    Honeywell(Aircraftengine

    manufacturer)

    Miningcompanies

    Miningequipment

    manufacturers

    Firms/agenciesengaged in

    development of newmaterial extraction

    techniques

    University HospitalPattern recognition

    technique developedin the medical field

    Information flow: Via research

    conferences and journal articles

    Decision: Purchase rawmaterials such as steel,copper, aluminum, and othercomposite materials

    Information flow:Shortages, costs,delivery schedules

    Information flow:Technology,capabilities ofequipments, cost

    Decision: Purchasemining equipment

    Information flow: Competingtechnological optionrequirementsDecision: Choice of

    material extractiontechnique

    guided our subsequent search process. In the next step, we undertook an extensive

    search of selected peer-reviewed journals (e.g., Academy of Management Journal;

    Management Science; Organizational Science;Non-Linear Dynamics, Psychology

    & Life Sciences; Emergence; and Complexity) by using the ABI/INFORMS and

    Business Source Premier databases. In the search process, we included keywords

    such as supply network, CAS, complexity theory, adaptation, adaptivity, chaos,

    SCM, and nonlinear time series analysis. From the results obtained, we selected

    more than 100 articles that were directly related to CAS and undertook an in-depth

    examination of these articles to identify significant theoretical, methodological,

    and technical developments related to all the major aspects of a CAS-based supply

    chain as described in Choi et al. (2001).

    Researchers across multiple disciplines have significantly advanced the theo-retical boundaries of CAS-based systems (Zhang, 2002; Fonseca & Zeidan, 2004;

    Richardson, 2004, 2005, 2007), especially focusing on organizational adaptation

    (Dooley, Corman, McPhee, & Kuhn, 2003), individual entity learning (Downs, Du-

    rant, & Carr, 2003), and network connectivity models (Barabaasi, 2002; Newman,

    2003). Methodological advancements such as sophisticated agent-based model-

    ing (Chatfield, Kim, Harrison, & Hayya, 2004; Sawaya, 2006; Pathak, Dilts, &

    Biswas, 2007), cellular automata (Wolfram, 2002; Mizraji, 2004), dynamical sys-

    tems theory (Surana et al., 2005), dynamic networks analysis (Carley, forthcoming),

    and empirical and case-study methods (Varga & Allen, 2006) have been applied

    to problems ranging from nursing and health care domains (Anderson, Issel, &McDaniel, 2003) to supply networks (Thadakamalla, Raghavan, Kumara, &

    Albert, 2004). Analysis techniques used within these articles include chaos theory

    (Strogatz, 1994), computational and statistical mechanics (Shalizi, 2001), and non-

    linear time series methods (Williams, 1997). Table 1 summarizes some of these

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    8/34

    554 Complexity and Adaptivity in Supply Networks

    Table1:Adv

    ancementsincomplexadaptivesystems(CAS)-basedrese

    arch.

    Research

    Contribution

    SignificantDevelopment

    ContributionRelatedto

    RepresentativePublications

    Theoretical

    Organizationaladaptation,in

    novation,

    intervention,andlearning

    Agents

    andschemas

    Barabaasi(2

    002),L

    issackandLetiche(2002

    ),

    Zhang(20

    02),AllenandStrathern(2003),

    Dooleyet

    al.(

    2003),Downsetal.(

    2003),

    Haslettan

    dOsborne(2003),Newman(2003),

    Anderson

    etal.(

    2003),Dagnino(2004),

    Fonsecaa

    ndZeidan(2004),R

    ichardson(2004),

    Aldunate,

    Pena-Mora,andRobinson(2005),

    Richardso

    n(2005),Burke,Fournier,andP

    rasad

    (2006),C

    hoiandKrause(2006),Peltoniemi

    (2006),Twomey(2006),R

    ichardson(2007)

    Communicationmechanisms

    inCAS

    Self-organization

    CASPerspectiveusedfortheorybuilding

    (evolutionaryeconomictheo

    ryandconsumer

    choicetheory)

    Connectivity

    Similaritiesbetweencomplexityandsystem

    theories

    Feedba

    ck

    Distributeddecisionmaking

    andentity

    coordinationinCAS

    Ruggedfitness

    landscape

    Networkemergence,scale-fr

    ee,andsmallworld

    networks

    Coevolution

    Entitylearningandemergentstrategy

    development

    Adapta

    tion

    CAS-basedmodelingofbusinessecosystems

    Emergence

    Designingemergence

    Learning

    Supplybasemanagement

    C

    ontinued

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    9/34

    Pathak et al. 555

    Table1:(Continued)

    Research

    Contribution

    SignificantDevelopment

    Contrib

    utionRelatedto

    RepresentativePublications

    Methodologica

    l

    Systemdynamicsandqueuingtheory

    Learn

    ingand

    adaptation

    LinandShaw(1998),Swaminathanetal.(

    1998);

    Tan(1999),C

    hatfield(2001),Iwanagaan

    d

    Namatam

    e(2002),R

    ivkinandSiggelkow

    (2002),W

    olfram(2002),Andersonetal.(2003),

    Skvoretz

    (2003),Stiller(2003),C

    hatfield

    etal.

    (2004),C

    hiles,Meyer,andHench(2004)

    ,

    Mizraji(2004),Thadakamallaetal.(

    2004),

    Hordijka

    ndKauffman(2005),Suranaet

    al.

    (2005),C

    arlisleandMcMillan(2006),

    Lichtenstein,Dooley,andLumpkin(2006),

    McCarthy,Tsinopoulos,A

    llen,and

    Rose-Anderssen(2006),Sawaya(2006),

    Varga

    andAllen(2006);Goldberg,Sastry,andLlora

    (2007),L

    ichtenstein,Carter,Dooley,and

    Gartner(2007),Pathak,Dilts,andBiswas

    (2007),S

    toica-KluVerandKluVer(2007)

    Cellularautomata

    Self-organization

    Agent-basedmodelingoforganizationsand

    supplynetworks

    AgentsandSchemas

    GeneticalgorithmsonCASDesign

    Fitnesslandscapes

    Fitnessmodeling,NKmod

    els

    Case-studyapproachforin

    vestigating

    organizationalstrategy,innovation,evolution,

    fluctuation,positivefeedback,stabilization,and

    recombinationandnewpr

    oductdevelopment

    EmpiricalstudyofCASan

    daction-based

    research

    Neuralnetworkmodelingofagentschemas

    Agentlearningmechanism

    s

    Heterogeneousagentdecis

    ionmodels

    Dynamicnetworkmodelin

    g

    Logisticalequationmodelingofinnovation

    dynamism

    C

    ontinued

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    10/34

    556 Complexity and Adaptivity in Supply Networks

    Table1:(Continued)

    Research

    Contribution

    SignificantDevelopment

    ContributionRelatedto

    RepresentativePublications

    Technical

    Nonlineartimeseriesanalysis

    Emergenceofpatterns

    Shalizi(2001),Kumara,Ranjan,Surana

    ,and

    Narayanan

    (2003),

    Suranaetal.

    (2005),

    Bhan

    andMjolsness(2006),

    BrahaandYaneer(2007),

    SchillingandPhelps(2007)

    Computationalmechanicsan

    d-machines

    Attractorreconstruction

    Bifurcationdiagramsandchaosanalysis

    Chaosidentification

    Applicationofstatisticalmechanicsformodeling

    andanalyzingCASnetworks

    Allianc

    eformation

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    11/34

    Pathak et al. 557

    research developments and advancements over the last 6 years across multiple

    different areas.

    On careful examination, we note an interesting trend. Almost all of the re-

    search contributions and advancements listed in Table 1 have occurred predomi-nantly outside the OM and SCM discipline. This observation is further supported

    by the observation that the special issue of Management Science on Complexity

    Theory (Amaral & Uzzi, 2007) does not carry a single article that deals purely

    with supply chain issues. Thus, it is clear that, while other areas such as industrial

    engineering, computer science, physics, organizational science, research and devel-

    opment, and strategic management, to name a few, are strongly pursuing research

    based on CAS perspectives, OM and SCM research is not keeping pace.

    One of the greatest contributions of the CAS perspective may be its abil-

    ity to incorporate increasing realism and empirical data into research models that

    can be understood in a practical business setting (Anderson, 1999). This has beendemonstrated with CAS research both in diverse applications (ecology, social re-

    tirement models, and zoology) with high realism (Van Winkle, Rose, & Chambers,

    1993; Grimm, 1999; Axtell, 2003) and in uses of empirical data from business

    organizations (Nilsson & Darley, 2006; Sawaya, 2006).

    Consider the parallels that exist between work by Albert, Jeong, and Barabasi

    (2000) on error and attack tolerance of complex networks and research by Hen-

    dricks and Singhal (2003) regarding supply network resilience under disruption.

    Findings indicate that the heterogeneous dyads in scale-free networks, such as

    those found in the Internet, biological-cell, and social-network connectivity, ex-hibit higher tolerance to random errors but lower tolerance to targeted attack than

    the more homogenous, exponential-style networks. These findings can be leveraged

    to hypothesize how different supply-network topologies give rise to different levels

    of supply-network resiliency under disruptions related to either random failure or

    targeted attack, potentially leading to important implications for industry manage-

    ment decisions. In fact, Thadakamalla et al. (2004) have shown how knowledge

    can be generated about survivability and resiliency of supply networks using con-

    cepts shown in the work of Albert et al. (2000). The work of Braha and Bar-Yam

    (2007) utilizes statistical properties of a complex network to show how the struc-

    tural information flows in distributed product development networks have similarproperties to other social, biological, and technological networks. It would be inter-

    esting to follow Braha and Bar-Yams suggestion regarding applying theirfindings

    about statistical properties of intraorganizational product development network to

    a supply network context, as this may result in new insights on how interfirm and

    intrafirm properties connect and evolve.

    Recent advancements made by Rivkin and Siggelkow (2007) toward extend-

    ing CAS research of organizations (Levinthal, 1997; McKelvey, 1999) using the

    NK model offitness from theoretical biology (Kauffman & Levin, 1987; Kauffman

    & Weinberger, 1989) to questions of adaptability in individual organizations could

    have important lessons for the study of supply chains. Rivkin and Siggelkow (2007)leverage empirical research demonstrating patterns of interactions within decision

    processes to show that the number of local optima is highly correlated with the

    decision-interaction patterns. Therefore, if there are many local optima, the relative

    value of exploration decreases. The implication is that the value of exploration of

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    12/34

    558 Complexity and Adaptivity in Supply Networks

    opportunities versus the exploitation of existing opportunities varies depending on

    how rugged and dynamic the landscape is.

    From a supply chain management perspective, the results and findings on

    adaptability and use of NK models have been demonstrated for supply base man-agement (Choi & Krause, 2006). Also important are the number of suppliers (N)

    and the level of interrelationships among the suppliers (K) and the degree of dif-

    ferentiation of these suppliers. In particular, the significance of interrelationships

    could have further implications for buyerbuyer or suppliersupplier coopetition

    (simultaneous competition and cooperation) in supply networks (Bengtsson &

    Kock, 2000; Choi, Zhaohui, Ellram, & Koka, 2002). For instance, supplier firms

    are typically under the control of the buying company through established work

    routines and contractual terms, yet they are able to make decisions on their own

    behalf. In this regard, the tension between control and emergence might be applica-

    ble to suppliersupplier relationships and thus may provide an interesting contextfor CASN studies.

    Another use of NK models can be found in the manufacturing-strategy litera-

    ture. Levinthal and Warglien (1999) show how Japanese automotive manufacturers

    use robust design to achieve single-peaked landscapes (landscapes with very low

    interaction levels among agents as compared to the total number of agents). They

    state that in die change operations, using pear-shaped clamps that can be smoothly

    brought to fit in only one way thereby driving even approximate movements into

    the right direction, reduces errors on the production line. The landscape in this

    case is designed by the physical shape of the task environment (p. 346). Thisexample illustrates how NK models can be conceptualized to reduce variability in

    a production network. If we apply this concept to SCM, one can argue that quality

    management practices can use similar concepts from NK models for managing

    buyersupplier relationships in order to reduce variability of the quality of the

    products that the suppliers send to their buyers, thus leading to a single-peaked

    landscape as suggested by Levinthal and Warglien (1999). For instance, when

    Honda uses a consistent supplier-management approach not only with their first-

    tier suppliers but also with their second- and third-tier suppliers (Choi & Hong,

    2002), one might view this as an attempt to create a single-peaked landscape in the

    supply network.Discussions and examples so far suggest that the CAS perspective holds

    promise for enriching and extending the current body of knowledge in the OM and

    SCM disciplines. We provide a detailed discussion of potential research directions

    later in the article, but we first discuss some underlying issues and challenges.

    CRITICAL ISSUES AND CHALLENGES IN CASN RESEARCH

    For more than 50 years, research studies have enriched our understanding of various

    OM and SCM issues (Beamon, 1998). The use of analytical models, simulation

    methods, and empirical approaches have greatly enhanced knowledge and im-proved decision-making processes. Analytical modeling-based studies have ma-

    tured from their initial years into explicit considerations of various operational

    decisions, the stochastic nature of demand, and the combinatorial possibilities of

    available scenarios and options. Empirical research has grown to provide insights

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    13/34

    Pathak et al. 559

    regarding strategic issues, managerial perceptions, and measurements of key op-

    erational issues. Undoubtedly, the scope of problems being investigated in extant

    literature is becoming richer and scholars are attacking complicated issues that were

    previously outside the scope of investigation for tractability reasons (Vonderembse,Uppal, Huang, & Dismukes, 2006). Addressing complicated issues, however, does

    not equate to addressing complexities.

    Complexity vs. Complicatedness

    The distinction between complicated research and complexity-oriented research

    is important for ensuring a broad-based research agenda. Cilliers (2000) suggests

    that something that is complicated can be intricate, but the relationship between

    the components is fixed and well defined. For instance, a jumbo jet is a complicated

    system that is amenable to taking individual components apart and putting them

    back together. In contrast, a complexsystem is characterized in terms of the nonlin-

    ear dynamic interactions of the individual parts. Furthermore, while a complicated

    system can be viewed as the sum of its parts, a complex system cannot be viewed

    that way; one cannot predict the behavior of a complex system by examining the

    behavior of its individual parts. These emergent properties of complex systems are

    due to the nonlinear dynamic relationship between the individual components.

    In a recent special issue on complex systems inManagement Science, Amaral

    and Uzzi (2007) provide the following commentary that further illuminates the

    differences between complicatedness and complexity (p. 1033):

    In contrast to simple systems, such as the pendulum, which has a small numberof well-understood components, or complicated systems, such as Boeing jet,which have many components that interact through predefined coordinationrules (Perrow, 1999), complex systems typically have many components thatcan autonomously interact through emergent rules. In management contexts,complex systems arise whenever there are populations of interacting agentsthat can act on their limited and local information. The agents and the largersystem in which they are embedded operate by trading their resources withoutthe aid of a central control mechanism or event a clear understanding of howactions of (possibly distant) agents can affect them.

    Amaral and Uzzi (2007) comment on the complexity in the supply chain arenaand emphasize the increasingly decentralized decision making, networkwide dis-

    semination of innovations, and the need to find approaches to make lean supply

    chains robust against random failures and targeted breakdowns. The authors pro-

    pose a complexity-based perspective for future investigations of various business

    issues.

    Parallel to the investigation of complicated issues that continue to be exam-

    ined, research initiatives are needed that examine complexity in OM and SCM.

    This endeavor can potentially illuminate several critical issues, such as intercon-

    nected supply networks and learning and adaptivity within supply networks that

    are currently rare in SCM literature.

    Challenges of Theory Development with a CAS Perspective

    In general, theory building requires careful application of structural methods

    to identify phenomena. Once identified, the phenomena must be validated by

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    14/34

    560 Complexity and Adaptivity in Supply Networks

    designing and conducting research studies (Meredith, 1998). Throughout this pro-

    cess, careful attention must be given to the level of rigor such that the research ad-

    heres to appropriate methodological guidelines. The results obtained, as well as any

    relevant insights, must have clear application to the phenomena within the boundaryconditions and be generalizable for the theory to be integrated into a wider body of

    knowledge. Here, we examine some of the unique theory-development challenges

    that must be overcome if a coherent body of knowledge is to be developed around

    CAS principles.

    First, the complexity of supply networks will press limits on researchers

    ability to understand the internal interactions between constructs and mechanisms

    of larger-scope phenomena. For example, operations research has successfully

    leveraged game theory to understand competitive and cooperative phenomena both

    within and between organizations (Cachon & Lariviere, 1999). Although these

    investigations provide insight into optimal monopolistic or duopolistic decisions,there are limits to modeling the nonlinear dynamics and adaptations inherent in

    the oligopoly or free-market structures that dominate our economy. As discussed

    previously, when several locally optimal policies interact in a complex supply

    network, the resulting nonlinear dynamics of global behavior can be unpredictable.

    Therefore, game-theoretic studies can be enriched by adopting the CAS perspective

    to help examine the applicability, impact, and robustness of theirfindings within the

    larger, more realistic supply network contexts in which game theory is intractable.

    One reason for the growing popularity of CAS across several disciplines is its

    ability to incorporate more realism in building theories, providing opportunityfor greater relevance, and supplying an understanding of the way phenomena act

    in otherwise intractable environments. CAS provides an approach to rigorously

    examine situations that closely map reality, yet simultaneously requires continuous

    extension and refinement to unravel unexpected behaviors that supply chains and

    networks are capable of producing.

    A second challenge is that OM and SCM as disciplines currently lack metrics

    for evolution and dynamism in supply networks. For example, many phenomena

    in supply networks occur over time, and it will be crucial to examine the evolution

    of the supply network over an extended time horizon. Such a behavior could be

    measured and depicted using attractors and the corresponding lags at which attrac-tors are reconstructed (Williams, 1997). Furthermore, because phenomena in an

    evolving supply chain occur at different levels, they must be captured at the firm,

    topology, and systems levels. For example, investigation of supply chain disruptions

    would require simultaneous consideration of agent-level metrics such as capacity

    and fitness, topology-level metrics such as degree distribution and path length, and

    system-level metrics such as robustness and efficiency. Given that empirical data

    collection can be problematic whenever real organizations are involved, empirical

    studies aimed at examining dynamic and evolutionary behavior inherent in sup-

    ply networks will require resourceful approaches to operationalize and integrate

    underlying constructs based on data collected from multiple system levels.Third, developing robust theories in the presence of adaptation presents a

    formidable task. In a system of entities with changing policies, careful analysis of

    the impact of interactions among these policies will be required. For example, Texas

    and California are preparing to restructure their power markets from zonal to nodal

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    15/34

    Pathak et al. 561

    models next year (Alaywan, Wu, & Papalexopoulos, 2004; Ercot, 2007). Power

    generators and wholesalers are planning to adapt their policies (e.g., trade strategies,

    scheduling, risk management) to take advantage of almost continuous shifts in

    pricing and transmission congestion across 3,0004,000 locations. Attempting toascertain common overarching principles in such CASNs may require approaches

    uncommon to operations and supply chain research like longitudinal data collection

    and data analysis without resorting to linearity assumptions. Research design and

    validation techniques will require resourcefulness when exploring both new and

    previously identified phenomena in the presence of dynamically changing and

    interacting entity behaviors.

    It may be possible to glean supply network information from publicly avail-

    able data or company archival data sources in order to understand factors affecting

    the dynamic behavior of the network. Such information, assuming it can be found,

    can be used to inform model development and validate models of supply networks.Because of the dynamic nature of CASN, rich longitudinal data of both quanti-

    tative and qualitative nature are important to accurately assess entity adaptation

    and its impact on system-level behavior. This likely requires close collaboration

    between academic researchers and practitioners who are dedicated to understand-

    ing the complexities that affect organizations in a supply network in order to make

    the commitment to this type of research effort. For example, structure, schema,

    and performance of various constituent organizations of a supply network might

    be sampled at regular intervals over time in order to understand the dynamic and

    emergent behavior of the system.Finally, while borrowing concepts and ideas developed in other disciplines

    can be innovative and useful, one must remember to take great care when relating

    a phenomenon found in a few studies to a wider range of situations. As seen in

    physics, abstraction of phenomena to larger- or smaller-scale systems does not

    always hold true, and any attempt to do so must be done thoughtfully and with

    great care (Feynman & Weinberg, 1986). Likewise, the impact of complexity and

    adaptation observed in one system may not hold true when applied in other systems.

    Such CASN characteristics make research in this area difficult, but, fortunately, OM

    and SCM disciplines could learn from other disciplines, such as organizational

    science, economics, computer science, and evolutionary biology, to name but afew. These disciplines have been extremely careful in generalizing their results and

    have intelligently combined a diverse range of methods and tools (as summarized

    in Table 1) to effect a slow paradigm shift.

    FUTURE DIRECTIONS OF CASN RESEARCH

    One key way in which CASN ideas and theories might be leveraged is in bridging

    the researchreality gap. For instance, tapping existing CAS research and apply-

    ing it to supply network contexts will move the field beyond a static, isolated

    dyadic buyersupplier framework. As indicated previously in this article, Braha

    and Bar-Yam (2007) studied the statistical properties of organizational networks

    that focus on product development. They show that structure of information-flow

    networks have properties that are similar to those displayed by other social, biolog-

    ical, and technological networks. They conclude their study by suggesting that the

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    16/34

    562 Complexity and Adaptivity in Supply Networks

    intraorganizational properties they studied might be applied to an interorganiza-

    tional level at which business organizations form the networks (i.e., supply net-

    works). Thus, by shifting the unit of analysis to the firm level, existing knowledge

    from an external discipline can be used for researching supply network problems.In this section, we attempt to highlight some of the issues that must be ad-

    dressed in order to develop a useful CASN research framework. We start by sug-

    gesting a CASN definition. We then elaborate on how supply network theory may be

    developed, building on CAS phenomenon. We finish by discussing some unique

    CASN research design, measurement, and methodological issues for validation

    purposes and list some potential CASN research questions.

    Defining CASN

    Aformaldefinition of CASN is one step toward furthering the use of CAS principlesin examining supply networks. Formulating such a definition is not a trivial task

    and will require an iterative process with inputs, from a variety of experienced

    researchers. What we propose here should be taken as a starting point for a formal

    discussion from which an acceptable definition might emerge.

    A CASN is a system of interconnected autonomous entities that make choices

    to survive and, as a collective, the system evolves and self-organizes over time.

    CASN consists of four key elements: (i) organizational entities exhibiting adap-

    tivity, (ii) a topology with interconnectivity between multiple supply chains, (iii)

    self-organizing and emergent system performance, and (iv) an external environ-ment that coevolves with the system. Each of these fundamental elements within

    a CASN can maintain several properties, such as capacity and service level (en-

    tity); path length, redundancy, and clustering (topology); efficiency and flexibility

    (system); and demand, dynamism, and risk (environment). The properties of these

    elements can be used to describe the state of a CASN at a moment in time or

    over a finite span of time. It is the interactions across these entities over time and

    the evolution of their properties that the SCM discipline seeks to understand more

    fully. Some of these properties may already have well-accepted measurements or

    metrics, such as a firms inventory holding costs, while others, such as supply chain

    agility, may require additional refinement.

    Building SCM Theory by IdentifyingCAS Phenomena

    A theory states how interrelated constructs are impacted by mechanisms creating a

    phenomenon (Schmenner & Swink, 1998). Future development of CASN theory-

    building efforts likewise should begin by viewing the properties associated with

    entities, topology, system, and environment as interrelated constructs. Mechanisms

    that alter these constructs are initiated by entities residing both inside and outside

    the CASN. For example, participating entity decisions such as supplier selection,shifting priorities (allocation of resources), or procedural modifications may im-

    pact not only internal constructs such as capacity, service level, or inventory but

    also system constructs like supply network efficiency, flexibility, and redundancy.

    Similarly, entities that exist in the external environment of the CASN can initiate

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    17/34

    Pathak et al. 563

    mechanisms such as modification of infrastructure or changes in regulatory policy

    that may impact CASN constructs.

    The constructs associated with each of the fundamental CASN elements are

    clearly interrelated. Changes in any one entity construct may lead to alterationof topology that impacts overall system properties, which, in turn, may lead to

    changes in the surrounding environment. Ultimately, the states of the entity, topol-

    ogy, system, and environmental constructs impact decision making within each

    participating entity. Individual-entity decision making may spawn changes that cy-

    cle through the CASN and eventually lead to an altered system and environment

    that impacts future decisions. Therefore, theory development about how various

    CASN elements interact can improve understanding of the impact of decisions

    made within each entity as well as their impact on other elements in the supply net-

    work. For example, the vertical integration decision taken by an original equipment

    manufacturer (OEM) determines the components or subcomponents that it wouldoutsource. Furthermore, a firm could decide to sole-source or engage several sup-

    pliers. These decisions would directly affect the network topology. The sourcing

    strategy and the associated network topology impact the OEMs flexibility to cater

    to potential demand fluctuations. In the event that the OEM is unable to satisfy a

    portion of demand due to supply shortages (e.g., due to capacity constraints at the

    sole supplier), the service level of the OEM gets adversely affected. This illustrates

    how entity decisions, network topology, system characteristics, and environmental

    characteristics are closely intertwined with each other.

    Unique CASN Research Design Issues

    While physical and temporal scales are often quite naturally defined and addressed

    in fixed and well-delineated relationships in complicated research, the nonlinear

    dynamic relationships in a CAS often span multiple scales. Defining the appropri-

    ate system scale is essential if the CASN behavior under study is to be observed

    consistently. Also, any constructs external to both the entities and the topological

    relationships constituting the system that impact the behavior must be integrated

    into the theoretical model, while superfluous variables must be eliminated. In ad-

    dition to the system scale, defining the environmental scope of the system is also

    paramount. Properly specifying these various types of scales enhances the valueof the research and also helps to focus the emphasis of study on key factors.

    System scale and unit of analysis

    Because of the recursive nature of systems both within and outside the CASN, it is

    important to select the appropriate physical scale or unit of analysis within which

    the theory is valid. Just as physics has discovered (Feynman & Weinberg, 1986)

    where, at the nano-scale level, normal laws of Newtonian physics break down,

    attempting to analyze a CASN phenomenon in too small or too large a context

    may yield comparatively perplexing results. Descriptions of the physical scalemust specify the range of entities that constitute the system as well as the types of

    relationships that are considered to form the interrelations within the topology.

    In addition to defining the physical scale of the system, the proper scaling

    of time is important as well. Different types of phenomena may occur over longer

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    18/34

    564 Complexity and Adaptivity in Supply Networks

    or shorter periods of time; therefore, certain research designs may require either a

    lengthier period of study or more frequent measurements than others. For example,

    examining how changes in fuel-efficiency regulations impact supplier selection

    policies in the automobile industry might require a longer time period of study thaninvestigating interfirm behavior in online reverse auctions. Clearly, there must be

    multiple scales and potential units of analysis for systems as complicated as supply

    networks. An illustration of this is the problem with multiple levels of validation

    that are common to interorganizational and agent-based models in general (Carley,

    2003). Even here, one key feature is the systems-level behavior that emerges over

    time. Therefore, while there may be many factors that are important at an entity

    level, systems-level behavior must include observation of the systems behavior

    that is creatively derived from the state and behavior of the constituent entities.

    Environmental scope

    As discussed previously, the system and its surrounding environment coevolve over

    time (Lewin, Long, & Carroll, 1999). Changes in either of these elements impact

    how decisions are made by CASN entities. Therefore, it is important to consider

    both the properties of the system and the environmental constructs that are related to

    the phenomenon of interest in any theory set forth. For example, using agent-based

    simulation, Siggelkow and Rivkin (2005) studied how environmental turbulence

    and complexity affect the formal design of the organizations. From an empirical

    perspective, Anderson and Tushman (2001) studied the effect of environmental

    constructs such as uncertainty, munificence, and structural complexity on firmsurvival. They found that uncertainty was the main reason that firms go out of

    business. These are examples of how inclusion of environmental constructs is

    important for research in CASN.

    Just as it is important to determine the proper physical and temporal scales,

    finding the appropriate number and type of environmental constructs to include

    in a theory is important when balancing the needs for validity and tractability.

    Examples of potentially important constructs are demand, dynamism, uncertainty

    (both aleatory and epistemic), risk, munificence, and ecological factors. As in

    any research, however, caution must be exercised when selecting environmental

    constructs, as inclusion of too many may lead to models that are unwieldy whileinclusion of too few may yield insufficient explanatory power of the phenomena.

    Leveraging models, measurements, and methodologies for validation

    A model of CASN behavior should precisely state how to measure the relevant

    constructs, how the constructs are related, and how certain mechanisms affect

    those constructs. Only when these issues are clearly stated can the theory be val-

    idated and examined for consistency with the phenomena under study across a

    wide range of situations. However, in addition to precise and internally consis-

    tent theoretical statement, a model should also allow for integration of other con-structs and mechanisms so that further theory refinement can make a significant

    improvement. Different validation methodologies have various strengths and weak-

    nesses and some are more easily accepted within a discipline than others. In a field

    such as SCM, in which so many constructs are interrelated, this observation holds

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    19/34

    Pathak et al. 565

    particularly true. For example, in the 1980s just-in-time inventory movement high-

    lighted the inefficiencies of classic inventory models that were developed using

    mathematical optimization techniques. The interrelationship of inventory levels

    with other important operational aspects such as push/pull strategy, setup times,capital costs, multiskilled employees, and strong supplier relationships were not

    explicitly considered in the classic inventory models, partly due to the constraints

    placed by the methodological orientation. Yet, in hindsight it is clear that an ex-

    plicit consideration of these interrelationships in research investigations pertaining

    to inventory models would have been a worthy undertaking much earlier. While

    theories with a small number of constructs may lend themselves well to analyti-

    cal validation, integrating components across multiple theories or exploring single

    theories with a large number of constructs may require empirical investigation.

    Regardless of how a new or reformulated theory is created, it is important

    to ensure the possibility of validation and refinement of the resultant theory. In-deed, when building CASN theories, such validation can be accomplished via

    many different methodologies such as analytical, simulation-based, empirical, or

    archival. For example, analytical models of interorganizational industrial systems

    have existed for many years and have been the focus of many researchers ef-

    forts. Within small physical-scale models, closed-form mathematical equations

    have been leveraged to expose detailed relationships between multiple variables

    within and across organizational boundaries. Mathematical programming opti-

    mization models have also been leveraged to provide insight for improved decision

    making. However, analytical tractability for the most realistic situations (e.g., in aCASN) is often limited in its ability to obtain solutions for problems of reasonable

    size.

    Thus, analytical efforts of a CASN may require a different orientation from

    the optimization approach that is currently commonplace in studies investigating

    supply chain issues. The impact of uncertainties within a many-entity environment

    may overwhelm the limited robustness of small-scale globally optimal solutions.

    Furthermore, the adaptive nature of CASN entities must allow for reactive decision

    making within, and in response to, their changing surroundings. New investigations

    of analytical models that seek to mitigate risk and improve decisions through

    maintaining multiple alternative policies that can be implemented contingent uponspecific changes in larger-scale theoretical models should lead to improved supply

    chain performance.

    Methodologically, computer-based simulations have been leveraged for

    interorganizational supply network research as well (Lin & Shaw, 1998;

    Swaminathan, Smith, & Sadeh, 1998; Tan, 1999; Chatfield, 2001; Chatfield et al.,

    2004; Sawaya, 2006; Pathak et al., 2007). Some of the earliest work in the area was

    performed by Forrester (1961), who used simulation to examine system dynamics

    within a supply chain. Simulations of CASNs can allow for entities to adjust their

    decisions in response to their environments as well as the actions of other entities.

    Such a methodology is powerful in that it can generate results about larger-scalesystemic behavior in ways that are analytically intractable. Simulations also provide

    a method for examining the dynamic behavior of systems in addition to potential

    steady-state behavior. Unfortunately, when compared to the specific results often

    obtained from analytical models via proofs or bounds, the ability of simulations

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    20/34

    566 Complexity and Adaptivity in Supply Networks

    may be limited when definitively extrapolating the inner workings of large-scale

    systems to the overall system behavior.

    Consider the example of the beer game (Sterman, 1989) in which local firms

    are making reordering decisions (small-scale decision change) that lead to thebullwhip effect due to excessive ordering at each tier in the supply network (large-

    scale performance change). Such an effect has been investigated using agent-based

    computer simulation. One of the interesting effects that has been observed in these

    simulations has been an overall unstable behavior (in the form of wild orderfluctua-

    tions) under certain simulation conditions in which the local agents have unlimited

    memory about the order fulfillment history of their suppliers and the order history

    of their customers (Sawaya, 2006). This is due to the agents overreaction to late

    orders, whereby the agents keep placing larger and larger orders as they adjust their

    reorder point to compensate, leading to fluctuations and system instability.

    The example highlights the possibility of generating extraneous system ef-fects due to a particular implementation of the simulation model with specific be-

    haviors. In this beer-game context, when the memory of the agents is limited, the

    system instability is reduced. It is challenging to use simulation to prove anything,

    but it allows researchers to understand something important about the likelihood

    of different outcomes. Naturally, simulation is subject to many of the same limita-

    tions as analytical and other models, for example, lack of robust empirical data to

    drive or motivate the simulation, the inherent assumptions, or artifacts introduced

    because of the way the simulation has been implemented. Therefore, caution must

    be exercised and simulation studies probably need to be augmented with rigorousadditional research efforts via empirical and analytical methodologies that thor-

    oughly examine the connections between small-scale decisions and large-scale

    performance in a CASN.

    Empirical methodologies are likely to be an important contributor to CASN

    theory-development efforts as they establish a link to industry reality, providing

    validation and ensuring the practicability of model prescriptions. Because one of

    the advantages of the CASN view of supply networks is its ability to incorporate

    increasing realism into models and theories of supply networks, empirical data

    are essential for the development of CASN theory. Empirical methods will always

    carry significant motivational weight in the OM and SCM disciplines. However,researchers often face challenges with data collection and with the complexities that

    empirical data introduce into supply-network conceptualizations and models. One

    example of empirical research comes from Choi and Hong (2002), in which they

    use an inductive case-study approach to build propositions about supply networks.

    In any case, as researchers become more familiar with the power of CASN, they

    will perhaps be less hesitant to incorporate complicated real-world data into theory

    and models of supply networks. It is also possible that, as various organizations

    recognize the benefits of more complex supply-network representations, they will

    be more willing to allocate the necessary resources for detailed empirical data

    collection and analysis.Finally, archival data methodologies can aid in the collection of data to inves-

    tigate the evolution of supply networks. For example, Utterback (1994) determined

    the dynamics of industrial growth by using census data and Christensen (1997) used

    archival data on disk drives and their makers over time to develop the theory of

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    21/34

    Pathak et al. 567

    disruptive technology. Such data can be mined to examine how a particular industry

    evolved and to investigate what other evolutionary paths might have been followed.

    Within a CASN context, Pathak (2005) used archival demand data from the U.S.

    automobile industry to investigate factors affecting the evolution and growth in asupply network.

    The complexity and multidimensionality of a CASN paradigm, as well as the

    diversity of research questions, rule out the use of a single approach. A combination

    of approaches is necessary to adequately explore difficult issues such as multidi-

    rectional causalities, simultaneous and time-lagged effects among variables, non-

    linearities, cyclical feedback mechanisms, and path dependencies. Furthermore,

    the normal means of applying methodologies may require modification for appli-

    cation within the CASN context. Creatively combining the strengths of analytical,

    simulation, empirical, and archival methodologies will be essential when gener-

    ating, establishing, and refining theories within an integrated body of knowledge.As an example, consider leveraging multiple methodologies in developing new

    strategies for bullwhip mitigation within a CASN context (Murray, 2007). Analyt-

    ical methodologies are capable of determining how order variance can be reduced

    by strategically leveraging negatively correlated demand streams or demand in-

    formation from multiple downstream supply network participants. Simulation can

    provide verification of analytical results while extending them to examine the in-

    direct cost reductions that result at firms further upstream. Empirical studies could

    be used to investigate the applicability of these mitigation strategies in real-world

    supply networks or perhaps even identify where they are already in use. Further,archival data can be used to demonstrate the prevalence of the problem in an in-

    dustry.

    Based on the discussions thus far, it is clear that future CASN research offers

    an exciting perspective to extend known problems and also a new set of problems

    to address. In Table 2, we summarize sample research questions that could be

    addressed by embracing the complexity and adaptivity perspective.

    CONCLUSIONS AND IMPLICATIONS

    SCM research examines the systems that span organizational boundaries. To date,the field has amassed a large and insightful collection of research that focuses on

    dyadic relations and phenomena that arise in tightly coupled, integrated systems

    (Beamon, 1998; Vonderembse et al., 2006). Largely absent from this body of work

    has been research that examines the broader, network-level effects that exist in real-

    life supply networks. In such networks, cause and effect are not simple, behavior is

    dynamic, and the actions of any firm in the network can potentially affect any other

    firms in the network. Complexity science provides a conceptual and methodological

    framework that enables consideration of these network-level issues.

    In this position paper we present a CASN perspective as a means to sup-

    plement and augment existing SCM theories and practices. For example, whilethe issue of visibility is central to research that examines collaborative plan-

    ning and inventory management among members of a supply chain, a CASN

    perspective would require researchers to extend the concept of visibility to an

    entire network of firms that may only be indirectly connected to the buying

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    22/34

    568 Complexity and Adaptivity in Supply Networks

    Table2:Pote

    ntialresearchissuesandquestionsforbuildingcomplexadaptivesupplynetwork(CAS

    N)theory.

    PotentialArea

    ofContribution

    PotentialCASNRese

    archIssues

    AssociatedCASNResearchQuestions

    Theoretical

    Interfirminteractio

    nsaffectingCASNtopology

    HowdodifferentC

    ASNtopologiesgiveriseto

    StatisticalpropertiesofCASNtopologies

    supplynetworkres

    iliencyunderdisruptions?

    Fitnessofindividu

    alentities;fitnessofCASN

    Howdointerfirma

    ndintrafirmpropertiesconnect

    Effectofenvironm

    entonCASNevolutionbothat

    andevolveinaCA

    SN?

    individualandsystemlevels

    Howcanconcepts

    offitness,exploration,and

    Multiplefeedback

    loopsandtheireffectson

    exploitationbeuse

    dforstudyingcollaborative

    evolutionandperformance

    buyer-supplierrela

    tionships?

    Complexityandre

    dundancyofinformationflow

    Howcanpolicymakerssetgloballyoptimal

    andtheireffectsonCASNevolutionand

    policiesbyinfluencinglocalfirmbehaviorina

    performance

    CASN?

    Decisionmakingcriteriaatthefirm,system,and

    Whatarethekeyd

    ecisioncriteriathatadecision

    environmentlevelsthataffectCASNevolution

    makerneedstoknowfromafirmsperspective,

    PolicydesignforCASN

    fromasystempers

    pective,andfromaregulatory

    Notionoflooseco

    uplingamongfirmsinasupply

    bodysperspective?

    network

    Whatistheroleofinformationsystemsinfostering

    Coevolutionofsupplychainstrategyandsupply

    loosecouplingamongfirmsinasupplynetwork?

    networkstructure

    Howdoescollabor

    ativedecisionmakingsustain

    Examinationofstrongandweaktiesamongfirms

    andprosperinasu

    pplynetwork?

    inasupplynetwork

    Whatisthesystem

    -wideimpactofopportunistic

    behaviorbyasinglefirminasupplynetwork?

    Howdodivergentsupplynetworksindivergent

    industriesimpacte

    achother?

    Whataretheimplicationsforlong-termstrategy

    processinlightofthecomplexandadaptive

    natureofsupplyne

    tworks?

    C

    ontinued

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    23/34

    Pathak et al. 569

    Table2:(Continued)

    PotentialArea

    ofContribution

    PotentialCASNRese

    archIssues

    AssociatedCASNResearchQuestions

    Methodologica

    l

    Analyticalmethod

    s

    Howcandynamic

    systemsmodelinganddynamic

    Dynamicsystem

    smodeling

    networkanalysisb

    eusedforstudying

    Dynamicnetworkanalysisandmodeling

    time-dependentevolutionofCASN?

    Hamiltonian-basedoptimization,genetical-

    Canoptimization-basedapproachesbeusedfor

    gorithms,andreliability-baseddesign-optimization

    studyingtime-dependentbehaviorand

    methods

    representingevolutiontrajectories?Whatarethe

    Statisticalphysics

    limitations?Canth

    isbeovercomebycombining

    Evolutionaryga

    metheory

    multiplemethodologies?

    Empiricalmethod

    s

    Howcanonemodelsophisticatedagentsthatcan

    Survey

    learn,adapt,andre

    spondtouncertaintiesinherent

    Casestudy

    inaCASN?

    Econometrics

    Couldepistemicuncertaintiespresentwithina

    Archivaldataan

    alysis

    firmbemodeled,q

    uantified,andanalyzedsoas

    Longitudinalstu

    dy

    toimprovetheeffe

    ctivenessofdecisions?

    Ethnography

    Howcouldcellularautomatonsbeusedfor

    Actionresearch

    investigatingcoopetitivedynamicsinaCASN?

    Behavioralexpe

    riment

    Couldasystems-dynamicsapproachbecombined

    withreliability-bas

    eddesignoptimization

    methodsforinvestigatingoptimalpolicy-design

    issuesinCASN?

    WhataretherelevantscalesforcoreCASN

    constructs,suchas

    complexity,adaptivity,and

    dynamism,t

    hatcanbeusedinsurveyresearch? C

    ontinued

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    24/34

    570 Complexity and Adaptivity in Supply Networks

    Table2:(Continued)

    PotentialArea

    ofContribution

    PotentialCASNRese

    archIssues

    AssociatedCASNResearchQuestions

    Simulationmethods

    Whatarepotential

    approachestocollectdatathat

    Agent-basedsim

    ulation

    areamenableforanalyzingcomplexadaptive

    Cellularautomaton

    behaviorofsupply

    networks?

    Systemsdynamics

    Whatarepotential

    testsforvalidityofCASN

    Evolutionarygametheory

    results?

    Neuralnetworks

    Evolutionaryalg

    orithms

    Technical

    Stabilityanalysis

    HowcouldLyapun

    ovanalysisbeusedfor

    Causalityanalysis

    addressingstabilityissuesinCASN?

    Couldbifurcationdiagramsbeusedforanalyzing

    thepresenceofchaosinanevolvingCASN

    (numberoffirmsa

    swellasthelinkageschange)? C

    ontinued

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    25/34

    Pathak et al. 571

    Table2:(Continued)

    PotentialArea

    ofContribution

    PotentialCASNRese

    archIssues

    AssociatedCASNResearchQuestions

    Attractorreconstruction

    Howcancomputationalmechanicsand

    causal-stateidentificationalgorithmsbeusedfor

    identifyingcausalcomponentsinanevolving

    CASN?

    Couldeconometric

    toolssuchasGranger

    causalityanalysisandvectorautoregression

    modelsbeusedforanalyzinglongitudinaldata

    generatedbyCASNresearch?

    Howcouldattracto

    rsbeusedfordesigning

    optimaldecisionsforafirm?Couldagent-based

    modelingandoptimizationmethodologiesbe

    combinedfordesigningoptimalpoliciesaround

    attractorsthatarepresentinaCASN?

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    26/34

    572 Complexity and Adaptivity in Supply Networks

    firm. Thus, as the practices of supply chain managers change over the future

    from a dyadic-only perspective to more of a network perspective, new research

    concerning supplier selection and supplier relations should be conducted in or-

    der to identify new best practices emerging from such new types of decisionmaking.

    To perform CASN research, we believe that supply chain researchers will

    need to draw from a rich variety of research methodologies. Whereas most existing

    supply chain research has focused on variance studies using surveys, discrete-event

    simulation, case studies of dyads, or analytical models, CASN research requires

    agent-based and computational models, process models that are dynamic and gen-

    erative, and case studies of larger ensembles of firms. Both computational and

    qualitative methods provide means to capture complex cause and effect, nonlin-

    earity, ambiguity, and dynamism; however, these are difficult methodologies to

    implement in a rigorous way, and so CASN researchers will possibly have to de-fine and uphold extremely high methodological standards in order for their work

    to be valid and have impact.

    A CASN perspective has the potential to be particularly important to decision-

    making activities in a supply network. For a supply network manager, a CASN

    perspective offers a new language and a new mental model from which to view

    the business world, draw interesting insights, and make decisions. A CASN per-

    spective may aid a supply network manager in making decisions while keeping

    the adaptivity of other firms, the complexity of the overall system, and the sur-

    rounding environment in mind. Furthermore, a CASN perspective will help enableresearchers to study the effects of decision making at the network level, as a supply

    network is ultimately a complex web of decision making.

    Supply networks today are being forced to take a growing amount of in-

    formation into account as more data continue to become available both from the

    surrounding environmental context and from increased numbers of evolving sup-

    ply network partners. Organizations that are unable to interpret and leverage vast

    amounts of information from changing and interconnected sources may face legal

    liabilities and will likely fail to maintain adequate performance in the competitive

    environment. Thus, information and decision-science researchers are likely to play

    an important role in helping to determine the future of decision making withinthese CASN contexts.

    A paradigm shift toward embracing and integrating principles from complex-

    ity science has already occurred in many other disciplines. Recent SCM research

    that draws analogy between supply networks and CAS suggests this discipline may

    be embarking on a similar change (Swaminathan et al., 1998; Choi et al., 2001;

    Surana et al., 2005). We urge the SCM research community to leverage the CAS

    perspective for integrating existing knowledge and further investigating the com-

    plexity and adaptivity that inherently exist within supply networks. These efforts

    would benefit from a generally accepted foundation within which theories can be

    combined and on which future efforts can build. Creation of such a foundationis well beyond the scope of any single article such as this. What is required is

    both authoritative identification of, and agreement on, the conceptually appropri-

    ate and empirically valid constructs that can be applied to supply network systems

    framed as CAS. With such a foundation, the SCM field will be poised for both

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    27/34

    Pathak et al. 573

    integrating existing knowledge into a structured body of knowledge, thus extend-

    ing its relevance and applicability to real-world industry. [Invited.]

    REFERENCES

    Abell, B., Serra, R., & Wood, R. (1999). Strategic thinking and the new science

    (Book review). Emergence, 1(2), 7179.

    Alaywan, Z., Wu, T., & Papalexopoulos, A. D. (2004). Transitioning the Califor-

    nia market from a zonal to a nodal framework: An operational perspective.

    Presentation made at IEEE Power Engineering Society, Power Systems Con-

    ference and Exposition, New York.

    Albert, R., Jeong, H., & Barabasi, A. L. (2000). Error and attack tolerance ofcomplex networks. Nature, 406, 378382.

    Aldunate, R. G., Pena-Mora, F., & Robinson,G. E. (2005). Collaborative distributed

    decision making for large scale disaster relief operations: Drawing analogies

    from robust natural systems. Complexity, 11(2), 2838.

    Allen, P. M., & Strathern, M. (2003). Evolution, emergence, and learning in com-

    plex systems. Emergence, 5(4), 833.

    Amaral, L. A. N., & Uzzi, B. (2007). Complex systems-A new paradigm for the

    integrative study of management, physical, and technological systems. Man-

    agement Science, 53, 10331035.

    Anderson, P. (1999). Complexity theory and organization science. Organization

    Science, 10, 216232.

    Anderson, P., & Tushman, M. L. (2001). Organizational environments and industry

    exit: The effects of uncertainty, munificence and complexity. Industrial and

    Corporate Change, 10, 675711.

    Anderson, P. E., Jensen, H. J., Oliveira, L. P., & Sibani, P. (2004). Evolution in

    complex systems. Complexity, 10(1), 4956.

    Anderson, R., Issel, L., & McDaniel, R., Jr. (2003). Nursing homes as complex

    adaptive systems: Relationship between management practice and resident

    outcomes. Nursing Research Policy, 52(1), 1221.

    Arthur, W. B., Durlauf, N. B., & Lane, D. (1997).Economy as an evolving complex

    system II process and emergence in the economy. Santa Fe, NM: Santa Fe

    Institute.

    Axtell, R. A. (2003). Toward behavioral realism in retirement models: From micro

    simulation to agent-based modeling. Presentation made at the Conference

    on Improving Social Insurance Programs, University of Maryland, College

    Park, MD.

    Barabaasi, A.-L. (2002). Linked: The new science of networks. Cambridge, MA:

    Perseus Books.

    Beamon, B. M. (1998). Supply chain design and analysis: Models and methods.

    International Journal of Production Economics, 55, 281294.

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    28/34

    574 Complexity and Adaptivity in Supply Networks

    Bengtsson, M., & Kock, S. (2000). Coopetition in business networksto co-

    operate and compete simultaneously. Industrial Marketing Management, 29,

    411426.

    Bhan, A., & Mjolsness, E. (2006). Static and dynamic models of biological net-works. Complexity, 11(6), 5763.

    Braha, D., & Yaneer, B.-Y. (2007). The statistical mechanics of complex prod-

    uct development: Empirical and analytical results. Management Science, 53,

    11271145.

    Brown, S. L., & Eisenhardt, K. M. (1998). Competing on the edge: Strategy as

    structured chaos. Boston: Harvard Business School Press.

    Burke, M. A., Fournier, G. M., & Prasad, K. (2006). The emergence of local norms

    in networks. Complexity, 11(5), 6583.

    Cachon, G., & Lariviere, M. (1999). Capacity choice and allocation: Strategic

    behavior and supply chain performance. Management Science, 45, 1091

    1108.

    Carley, K. (2003). Validating computational models. CASOS working paper,

    Carnegie Mellon University, Pittsburgh, PA.

    Carley, K. M. (forthcoming). Dynamic network analysis in the summary of the

    NRC workshop on social network modeling and analysis. In R. Breiger & K.

    M. Carley (Eds.), National Research Council.

    Carlisle, Y., & McMillan, E. (2006). Innovation in organizations from a complexadaptive systems perspective. E:CO, 8(1), 29.

    Chatfield, D. C. (2001). SISCO and SCMLSoftware tools for supply chain sim-

    ulation modeling and information sharing. Doctoral dissertation, The Penn-

    sylvania State University, State College, PA.

    Chatfield, D. C., Kim, J. G., Harrison, T. P., & Hayya, J. C. (2004). The bullwhip

    effectimpact of stochastic lead time, information quality, and information

    sharing: A simulation study. Production and Operations Management, 13,

    340353.

    Chiles, T., Meyer, A., & Hench, T. (2004). Organizational emergence: The originand transformation of Branson, Missouris musical theaters. Organization

    Science, 15, 499520.

    Choi, T. Y., Dooley, K. J., & Rungtusanatham, M. (2001). Supply networks and

    complex adaptive systems: Control versus emergence.Journal of Operations

    Management, 19, 351366.

    Choi, T. Y., & Hong, Y. (2002). Unveiling the structure of supply networks: Case

    studies in Honda, Acura, and Daimler Chrysler. Journal of Operations Man-

    agement, 20, 469493.

    Choi, T. Y., & Krause, D. R. (2006). The supply base and its complexity: Implica-tions for transaction costs, risks, responsiveness, and innovation. Journal of

    Operations Management, 24, 637652.

  • 7/28/2019 Complexity and Adaptivity in Supply Networks

    29/34

    Pathak et al. 575

    Choi, T. Y., Zhaohui, W., Ellram, L., & Koka, B. R. (2002). Supplier-supplier

    relationships and their implications for buyer-supplier relationships. IEEE

    Transactions on Engineering Management, 49, 119130.

    Christensen, C. M. (1997).Innovators dilemma. Boston: Harvard Business SchoolPress.

    Cilliers, P. (2000). Rules and complex systems. Emergence, 2(3), 4050.

    Dagnino, G. B. (2004). Complex systems as key drivers for the emergence of a

    resource- and capability-based interorganizational network. E:CO Special

    Double Issue, 6(12), 6169.

    Dooley, K., Corman, S., McPhee, R., & Kuhn, T. (2003). Modeling high-resolution

    broadband discourse in complex adaptive systems.Nonlinear Dynamics, Psy-

    chology, & Life Sciences, 7(1), 6185.

    Downs, A., Durant, R., & Carr, A. N. (2003). Emergent strategy development for

    organizations. Emergence, 5(2), 528.

    Ercot. (2007). Ercot nodal transition plan, accessed September 12, 2007, available

    at http://nodal.ercot.com/docs/po/index.html.

    Feynman, R., & Weinberg, S. (1986).Elementary particles and the laws of physics:

    The 1986 Dirac Memorial Lectures. New York: Cambridge University Press.

    Fonseca, M. G. D., & Zeidan, R. M. (2004). Epistemological considerations on

    agent-based models in evolutionary consumer choice theory. E:CO, 6(3),

    48.Forrester, J. W. (1961). Industrial dynamics. Cambridge, MA: MIT Press.

    Goldberg, D. E., Sastry, K., & Llora, X. (2007). Toward routine billion-variable

    optimization using genetic algorithms. Complexity, 12(3), 2729.

    Global Logistics and Supply Chain Strategies. (2007). Supply chain com-

    plexity masters: Boeing. For Boeing, a new aircraft means a revamped

    supply chain. 11(3), 3841, accessed September 12, 2007, available at

    http://glscs.texterity.com/glscs/200703/?pg = 38.

    Grimm, V. (1999). Ten years of individual-based modelling in ecology: What have

    we learned and what could we learn in the future.Ecological Modelling, 115,129148.

    Haslett, T., & Osborne, C. (2003). Local rules: Emergence on organizational

    landscapes. Nonlinear Dynamics, Psychology, and Life Sciences, 7(1), 87

    98.

    Hendricks, K., & Singhal, V. (2003). The effect of supply chain glitches on share-

    holder wealth. Journal of Operations Management, 21, 501522.

    Hordijk, W., & Kauffman, S. A. (2005). Correlation analysis of coupled fitness

    landscapes. Complexity, 10(6), 4149.

    Iwanaga, S., & Namatame, A. (2002). The complexity of collective decision. Non-

    linear Dynamics, Psychology,