THE NETWORK AS KNOWLEDGE: GENERATIVE RULES AND THE

Strategic Management JournalStrat. Mgmt. J.,21: 405–425 (2000)

THE NETWORK AS KNOWLEDGE: GENERATIVERULES AND THE EMERGENCE OF STRUCTURE

BRUCE KOGUT*The Wharton School, University of Pennsylvania, Philadelphia, Pennsylvania, U.S.A.

The imputation problem is how to account for the sources of the value of the firm. I proposethat part of the value of the firm derives from its participation in a network that emerges fromthe operation of generative rules that instruct the decision to cooperate. Whereas the value offirm-level capabilities is coincidental with the firm as the unit of accrual, ownership claims tothe value of coordination in a network pit firms potentially in opposition with one another. Weanalyze the work on network structure to suggest two types of mechanisms by which rents aredistributed. This approach is applied to an analysis of the Toyota Production System to showhow a network emerged, the rents were divided to support network capabilities, and capabilitieswere transferred to the United States.Copyright 2000 John Wiley & Sons, Ltd.

The thesis of this article is that a structure of anetwork is an emergent outcome generated byrules that guide the cooperative decisions of firmsin specific competitive markets. The observeddifferences in the patterns of cooperation acrossindustries are not happenstance. They reflectrather the implicit operation of these cooperativerules and the competing visions that come toshape a network. The emergence of the structuralpattern of cooperation is not the result of anabstract and static choice between market or firm,or market versus hybrid cooperative forms ofgovernance. Structure is emergent in the initialconditions of a specific industry.

The structure of an industry is interesting,because it represents capabilities of coordinationamong firms, as well as claims on the propertyrights to profits to cooperation. The conventionalemphasis on the opposition of market, contract

Key words: networks; knowledge; value of the firm;rent distributors*Correspondence to: Professor Bruce Kogut, The WhartonSchool, University of Pennsylvania, Steinberg Hall-DietrichHall, Philadelphia, PA 19104-6370, U.S.A.

CCC 0143–2095/2000/030405–21 $17.50Copyright 2000 John Wiley & Sons, Ltd.

and firm represents largely a static view of theboundary choice facing a firm. The commonproposition that a viable firm is worth at leastthe sum of its parts rests on the supposition thatinternal organized coordination is a source ofeconomic value. Winter (1987) has insightfullycoined this problem of valuation as a question ofimputation. In his formulation, a firm is the unitof account to which rents accrue. Imputation is,then, the assignment of a firm’s value to itsconstituent parts, e.g. patents, people, andmachines. The determination of a firm’s value isthe result of the uniqueness of these factors, theirtransferability, and their resistance to imitation bycompetitors. If these factors, and the knowledgeof how to coordinate them, were imitable, thenthere would be no rent to impute. These con-ditions are equivalent to concluding that there areno rents to public knowledge.

This reasoning quickly points to the boundary ofthe firm as representing more than a legal definitionof the firm as theunit of accrual, but also assignifying a qualitative barrier between the knowl-edge held privately and the knowledge that ispublic. Yet, this barrier has often been exaggerated,

406 B. Kogut

for knowledge even when shared among two firmsmay not be a public good in the conventional senseof this term, that is, information that is accessible toother parties at zero marginal cost. For coordinationamong firms itself entails principles of organizationthat can be idiosyncratic to their relationship andthat code for particular kinds of capabilities, suchas speed of product development or minimizinginventories.

The accounting for knowledge stumbles, conse-quently, on an important problem: how do weimpute the knowledge external to the firm thatnevertheless contributes to its profitability? Thispuzzle is related to the interpretation of “totalfactor productivity” in macroeconomic growthaccounting studies. After accounting for the con-tribution of inputs to economy-wide growth, theresidual is attributed to exogenous technicalchange, institutional factors, or externalities.Alternatively, it is possible to specify explicitlythe ‘imputed’ source of productivity gains:

yi = at + gam + Oi aiXi + R

where yi is output of firm i, at is a shift parameter,aiXi are weighted inputs (such as the value ofcapital and labor), R is a residual, andgam isthe weighted value imputed to the fixed effect ofmembership in a network (all terms are in logs).This formulation proposes that “network capa-bility” is a source of imputed value to the produc-tivity of a firm.

The most tangible expression of the directvalue of external knowledge to the firm is thecompelling evidence that rapid product develop-ment depends on the reliance on outside suppliers.Both Clark, Chew and Fujimoto (1987) andMansfield (1988) found that time to market wasspeeded through a policy of outsourcing to sup-pliers. The capability to commercialize productscan in this case be seen to rest on the successfulexploitation of the knowledge of other firms. Inthis sense, the competitive capabilities of a firmrest not only on its own knowledge or on itsknowledge of the network. The capabilities ofthe firm, rather, are dependent upon the principlesby which cooperation among firms is coordinatedand supported in the network.1

1Most, if not all studies of networks treat knowledge as thequestion of knowing who has knowledge and the access of

Copyright 2000 John Wiley & Sons, Ltd. Strat. Mgmt. J.,21: 405–425 (2000)

This view of the network as knowledge con-fronts four analytical challenges:

1. What is meant by network capabilities?2. How do we understand the generative rules

that drive the emergent structure of the net-work?

3. How does structure influence the competingclaims to rents among members to a network?

4. If structure encodes emergent knowledge, howis the network transformed into intentionalreplication of this knowledge through time?

The contribution of this paper is to understandnetworks as arising out of generative rules thatguide the formation of relationships and code fororganizing principles of coordination. These rulesare “sorting” provisions that indicate the matchbetween firms and the nature of their cooperation.The rules code for principles (e.g., “shareresearch” or “supply just on time”) and in turnlead to network capabilities that are not specificto a firm, but represent joint gains to coordinationand learning. We explain that the structural pat-terns that emerge in a network define two kindsof rents, one that accrues to a broker, the othermore broadly to the members of a closed group.Because these capabilities are quasi-public goodsto members and yet firms are the units ofaccrual—not the network, a central issue is howthe rents to this coordination are made bothexclusionary and sustainable in the face of poten-tial defection.

Our reasoning rests upon three central ideas ofwhat define a firm: unit of accrual, governancestructure to resolve agency problems throughresidual claims, and a repository of coordinatingcapabilities and social identity. While these threeideas are operative in the analysis below, westress, in particular, the latter property of the firmin order to analyze how capabilities are generatedby network coordination. To shift the understand-ing of networks as simply a resolution to agencyconflicts, or as access to information, to theircapabilities in promoting variety and yet coordi-nation in specific industry settings is a primaryambition of this paper.

this knowledge through cooperative relationships. (See Powell,1990, for example.)

The Network as Knowledge 407

Specialization and variety in market andnetworks

A definition of an economic network is the pat-tern of relationships among firms and institutions.In this definition, an idealized market is a polarcase of a network in which firms transact atspot prices and are fully connected in potentialtransactional relations but are disconnectedthrough their absence of cooperative agreements.Few markets exist of this type. Rather, mostmarkets consist of sub-sets of firms and insti-tutions (e.g., universities) that interact moreintensely with each on a long-term basis. Thesepatterns of interactions encode the structuralrelationships that represent the network.

This definition fails to convey the observationthat the structure of a network implies principlesof coordination that not only enhance the individualcapabilities of member firms, but themselves leadto capabilities that are not isolated to any onefirm. It is important to the following argument todistinguish between information and coordination(or what is also called know-how). At a minimum,the ability of a firm to access information in anetwork constitutes an advantage, e.g., the effect ofaccessing the technology of a research center onits subsequent innovations. Of course, this accessis most likely the outcome of a bargain, in whichthe two parties arrive at an understanding of contri-bution and compensation. This sort of access,stressed by Powell (1990) among others, exem-plifies the informational benefits of enhancing afirm’s capabilties through relationships.2

Cooperation, however, can also engender capa-bilities in the relationship itself, such that theparties develop principles of coordination thatimprove their joint performance.3 Such principlesmight be rules by how supplies are delivered,such as by just in time or mass production. Or

2See Gulati (1998) for a discussion of information and net-work formation and Palmer, Jennings and Zhou, 1993, forcohesion studies on innovative diffusions. Early studies onthe idea of competition among alliance networks are Nohriaand Garcia-Pont, 1991, and Gomes-Casseres, 1994. Theintriguing idea of “communities of practice” is also akin tothis perspective, in which “connectivity”—rather than positionin a network—is a source of advantage. See Brown andDuguid (1991).3Studies on the capabilities of networks are explored in manystudies on Italian and German networks, e.g., Lorenzoni andBaden Fuller (1995), Lipparini and Lomi (1996), Herrigel(1993) and Piore and Sabel (1994).


they might involve more complex rules governingthe process by which innovations are collectivelyproduced and shared. In this sense, the networkis itself knowledge, not in the sense of providingaccess to distributed information and capabilities,but in representing a form of coordination guidedby enduring principles of organization.

The proposition that part of the value of a firmcan be imputed to the capability of its embeddednetwork is implicit in the treatment of the divisionof labor as handled by Smith through Stigler.The now classic question in the analysis of theboundaries between markets and firms can berephrased as the following: if networks are struc-tured by organizing principles of coordinationthrough a division of labor among firms, thenwhy are these firms not organized within a singlecommon governance structure?

Networks offer the benefit of both specializa-tion and variety generation. The superior abilitiesof markets to generate variety is a commonplacebelief that is, nevertheless, problematic. The con-verse of this statement is that firms are superiorvehicles for the accumulation of specialized learn-ing. To understand variety, we must also under-stand why specialization and variety are antitheti-cal within the firm, but define complementswithin a network.

Smith’s famous essay recognized the power ofthe market to achieve variety through specializa-tion in the division of labor (Smith, 1965 edition).Smithian efficiencies in specialization were dueto the inherent learning by doing by completingrepetitive tasks, as well as the reduction in losstime due to changing tools and tasks. At theheart of Smithian efficiencies is the implicationthat learning by doing, despite the initial endow-ments of equal competence among individuals,accumulates to lower the costs of subsequentproduction. Smith saw the division of labor asderived from the dynamic learning through spe-cialization. He posited that people were largelysimilar in their a priori talents; differentiation intospecialized competence was the outcome, not theprecursor, to the division of labor. In other words,specialization through a division of labor is thedriver of the acquisition of competence and,consequently, of knowledge.

The perspective of the firm as a repositoryof knowledge embraces Smith’s observation onexperience-derived learning through a division oflabor as posing both a static coordination problem

408 B. Kogut

as well as determining dynamic paths of knowl-edge acquisition. Firms are social communitiesthat permit the specialization in the creation andreplication of partly tacit, partly explicit organiz-ing principles of work.

But why are firms required to provide thiscoordination? The behavioral sciences provide animportant insight. (See Kogut and Zander, 1996,for a review.) Boundaries to a firm representmore than the legal unit of accrual; they providethe cognitive representation of what constitutesthe object of membership, that is, of identity.Through identity, individuals anchor their percep-tions of self and other and attach meaning tomembership in a firm, as well as in the categoriesof skill that define a division of labor (e.g.,“worker” or “accountant”). By this anchoring,they employ focal rules by which action is coordi-nated and intention communicated through com-mon categorization of self and other. Moreimportantly, through identification to occupationand firm, individuals are guided and motivatedalong coordinated paths of joint learning. Bound-aries matter, because within the cusp of thesesocial membranes, identities are circumscribed.The behavioral foundations to why the knowledgeof the firm is bounded are to be found in the basichuman motivation of belonging and membership.

If benefits of identity are to lower the costs ofcommunication and coordination, they come at acost. For identities represent a norm which indi-cates avenues of exploration; by implication, theyalso prohibit certain paths. Organization by firmis variety reducing. The great power of the marketis not only its information properties, but also itsfunction as a generator of variety in innovationsand capabilities that are subject to selection. The“market”, as an assemblage of firms pursuingdifferent visions and organized by distinct iden-tities, generates a variety that individual firmscannot manufacture internally without decrementto division of labor and the salience of focalrules, i.e., to the organizing principles (inclusiveof compensation and incentive schemes), bywhich work is coordinated.

In effect, a rent arises out of the scarcityvalue not only of land or technology, but also ofbehavioral coordination within the firm. Yet, atthe same time, networks also provide capabilitiesto coordinate behavior among firms. This dynamicbetween the capability of the firm and market liesat the heart of Stigler’s argument that a firm moves


from vertical integration to disintegration of itsactivities according to the process by which amarket “learns” to supply inputs at a lower costthan it can itself (Stigler, 1951).

These observations are important because theyforce a recasting of the received wisdom onthe relative merits of markets—which we haveindicated is a network—and firms. Curiously, theinitial statement of Coase is compatible with theview that the structure of external relationshipsinfluences boundary between firm and market.Coase (1937) posited that this boundary is deter-mined by the internal costs of production andmanagement relative to the costs of market searchand procurement.

Coase, however, left unexamined the issue howstructure reduces the costs of search and coordi-nation. Just as Stigler pointed to learning in themarket, Chandler’s (1962) early contribution wasto explore how higher-order organizing principlesof divisionalization could reduce the costs ofinternal complexity. His history of the innovationin internal hierarchy pointed to the role thatstructure plays in reducing the costs of coordi-nation and authority. More importantly,divisionalization increased the internal variety ofthe firm by separating potentially competingvisions into relatively contained units.

Figure 1 presents a diagram of the Coasianfirm as the base case. Since Coase acknowledgedmarket search and management costs, it is reason-able to think of these costs as increasing as thevariety of products are produced internally orsourced externally. We index these costs asincreasing in relation to a set of products thatreflect the identity of the firm. At low variety, afirm produces at lower cost than purchasing fromother firms; this condition is simply to guaranteethe existence of the firm. At some point, theinternal management of increasing varietybecomes more expensive than sourcing varietyfrom the external network known as the market.

It is then straightforward to diagram the effectof a Chandlerian innovation; it increases internalvariety at given levels of cost. Similarly, we canthink of Stigler’s life cycle notion of integrationand de-integration as implying an improvementin the capability of the network. Just as a firmlearns, so does the network insofar that supplierscome to substitute for internal production. Ineffect, the variety of the firm decreases as theknowledge diffuses to the market. Figure 1


Figure 1. Static costs of sourcing variety with organizational and institutional learning

illustrates this idea by showing how improvementin the efficiency of the market reduces the internalvariety of the firm. As the market learns, theneed to integrate vertically dampens. It is, in fact,exactly the increased knowledge in the suppliernetwork, as realized through the innovations inthe Toyota system, that has forced a radical disin-tegration of American auto assemblers. Gulati andLawrence’s (1999) observations on the extensionof coordination to the external value chain aredrawn from the historical diffusion of these inno-vations in the auto industry.

Simple rules and emergent networks

A network is then a collection of firms, eachensconced in an identity that supports specializa-tion and a dynamic of learning and exploration.But the network, unlike the firm, does not consistof an authority relationship that can enforce anorganizational structure on its members. InChandler’s history, an entrepreneur imposes an


innovation on the firm with the support of topmanagement. How does a network learn to coor-dinate in the absence of authority?

The seeds of the answer to this question lie inHayek’s contention that the market is the engineof variety (Hayek, 1988). Hayek’s contributionis often credited for his observation that marketsare superior to hierarchies for embedding infor-mation in prices. But Hayek, even in his heralded1945 article on information, meant somethingmore. Knowledge is held tacitly, raising the prob-lem of how central planners could ever know asmuch as decentralized firms. Specialization isself-preserving, even if markets generate infor-mation as to their valuation and accessibility,because prices can be communicated, but notcompetencies. The dynamic advantage of the mar-ket is the generation of variety through an“extended order” that supports coordinationamong specialized firms.

Hayek’s notion of the extended order begs thequestion what generates the structure of this

410 B. Kogut

order, or what we would label the network. Obvi-ously, markets differ from each other. Theextended order supporting variety is hardly homo-geneous across industries. Is the emergence ofthese network structures random or do they reflectthe operation of operating principles that act asgenetic rules?

Our claim is that the structure of a networkarises from inherent characteristics of technol-ogies that populate an industry, as well as socialnorms and institutional factors that favor the oper-ation of particular rules. Technologies of massproduction, which characterize some industries insome countries, influence the choices that firmsmake to cooperate or not. Industries characterizedby science-based technologies tend toward rulesthat promote cooperation between research centersand firms. As these rules generate the structureof a network, the structure itself influences sub-sequent behavior.

For clarity, consider the simple example of thetit-for-tat rule as analyzed by Axelrod and Hamil-ton (1981). By analyzing the convergence proper-ties in a population in which agents use differentrules for cooperating and defecting, they foundthat particular rules, especially one that rewardsfor cooperation and sanctions for defections onthe next round, tend to dominate. Convergence,however, is frequency dependent and thus vulner-able to tipping in either direction. The implicationof this analysis is that structure that isolates“cooperators” tends toward self-organizing com-munities of cooperation.4 There is no authority,and yet the network self-organizes—that is,converges—toward the dominance of rule (e.g.,tit-for-tat) over the other. Once achieved, theresultant structure supports a general capability ofcooperation without even enacting the rule itself.

There are many studies whose results implythe operation of rules generating a dynamic ofself-organization. These rules need not be techno-logical in origin, but can also reflect institutionaland cultural norms. They are also deeply embed-

4For an insightful analysis of how structure influences dif-fusion (e.g., cooperation) by agents, see Boorman 1974. Theparallel between the proposal to understand structure as emer-gent and rule-based shares obvious affinities to the literatureon complex adaptive systems; see Axelrod and Cohen (2000)for a treatment. The genetic algorithms in the vein of Holland(1992) are an appropriate research strategy by which tothink about rules appropriate for stylized industry conditions.Empirically the identification of rules may be fairly simple,as I try to illustrate in the subsequent examples in this article.


ded in the social identity of the actors. They mayseem “irrational” from a perspective of economicoptimization. But what is critical to understand,as we will explain later, is that rules generatestructure that dissuades rule-breaking behavior.

That identities underline the preference forparticular rules is central to White’s argumentthat networks are manifestations of the physicsof identity and control (White, 1992). This claimis implicit in White’s early work on the structuralimplications of kinship rules (White, 1963). (Thisanalogy is not far-fetched in light of the penchantfor using familial descriptions in the allianceliterature, such as “parent” company or joint ven-ture as a “child”.) Societies differ in their rulesby which kinship encourages and prohibits mar-riage. As a consequence, kinship rules generatedistinctive trees. Rules that permit marriagebetween first cousins generate radically differentsocietal patterns of kinship than those that forbidmarriage only between paternal cousins. Theidentity of what constitutes family is the foun-dation to the origins of the rules that governfamilial replication.

In traditional societies, identity and familydetermined the economic and social networks.The study by Padgett and Ansell (1993) infersthe rule, based on the analysis of the marriageand economic networks of Florentine familiesin the fifteenth century, that aristocratic familiesdid not interact socially or economically withthe new families. These families were situatedin a fairly simple economic and social networkdetermined by economic and marital rules. Bymild violations of the rule forbidding economicties, the Medici family influenced the structuresuch that they were two times more central inthe network than the next family (Wassermanand Faust, 1994).

The rules of kinship and social prestige haveclear analogues in the implication of identity withinthe social community of industrial and financialfirms. For example, the results observed in the studyby Podolny and Stuart (1995) on cooperation arebased on the rule that high status firms do not allywith low status firms. As a consequence, there is asorting behavior by which prestigious firms aregrouped by strong ties (cohesive ties with nointermediaries) and engage in weaker ties to lessprestigious firms, sometimes as a form of endorse-ment (Stuart, Hoang, and Hybels, 1999).

Despite our emphasis on the neglected social


Figure 2. Biotechnology

influences on cooperative rules, technological fac-tors are obviously critical to understanding net-works in modern economies. The study by Axel-rod et al. (1995) on standard setting for theUnix operating system assumes that because oftechnological complementarities, firms areencouraged to cooperate. Competitive pressuressuggest a mating rule in which firms prefer toally with distant rivals.5 In this case, a sortingrule is derived from competitive rivalry: avoidcooperation with near rivals. From this simplerule, they show that individual cooperativedecisions among “agents” (i.e., firms) generate adistinctive structure over time, with two groupsformed around competing standards. The resultsof their simulations, using empirical data for vari-able values, underscore how identities correspondstrategically to competing visions of the future(e.g., distributed versus central computation, cableversus satellite transmission).

5Note the study of Baum, Calabrese and Silverman (2000)has the contrary assumption that distance provides the benefitsof new information, a weak tie argument. As we note through-out this paper, both assumptions (or rules) can be right,depending upon the industry context (e.g., standard settingversus competence seeking).


The tendency of some industries to convergetoward what Gomes-Casseres (1994) calls com-peting constellations varies widely across indus-tries. The emergence of structure in a network issensitive to specific industry settings. Competingaround standards is not, for example, a featurein the pharmaceutical industry. The rule in theAmerican biotechnology industry was the follow-ing: start-up firms should form alliances withestablished companies. The origin of the ruleslies principally in the lack of financial resourcesand marketing and distribution capabilities of thestart-up companies. Venture capitalists, concernedby the “burn rate” of the initial capital provisionsto these companies, required relationships toavoid costly expenditures and to signal the qualityof the drug portfolio.

The outcome of these rules is a pattern ofalliances that as early as 1983, as shown inFigure 2, are marked by the creation of severalfragmented star and hub sub-graphs that revealan emergent structure. (The structure is generatedon the basis of the block-model data in Table 3of Walker, Kogut and Shan, 1997.6) Sometimes

6I thank Jon Brookfield for suggesting this graph.

412 B. Kogut

established companies are relatively central; ina few instances, a new biotechnology companyemerges as central. Overall, the network structureis very sparse, and yet there is an identifiablestructure.

The rules in the biotechnology industry thatgenerate the relational structure are themselvesproducts of the non-random distribution of capa-bilities that distinguish start-ups and pharmaceu-tical companies. Start-ups, consisting of molecularbiologists, lacked certain capabilities. But byimplication, pharmaceutical companies wereunable to integrate the new science, built uponparticular professional identities, with their tra-ditional research endeavors. Identification limited,at least initially, the internal variety of pharma-ceutical companies (see Zucker and Darby, 1995).Specialized by differentiated capabilities, theirmutual need suggested a rule of relationship for-mation that generated distinctive patterns in thestructuration of a cooperative network.

For this reason, the emergence of structure inbiotechnology industries outside the United Statesfollowed a different trajectory. Here we see theimportance of understanding the conjunction oftechnological and social influences. Becausescientists in France identify professionally withnational scientific laboratories, small firms wereimpaired in attracting the critical scientific talent(Gittelman, 1999). In this network, laboratorieshave remained critical nodes in the network, withdense ties formed with national laboratories.Thus, different ideas of professional prestige inconjunction with the technological properties ofgenetic engineering research resulted in a dramati-cally different network structure in France, onebuilt around laboratories and large firms.

This dynamic between internal capabilities,ensconced in specific identities and organizationalstructures, and the external knowledge in themarket drives a co-evolution between the emer-gent properties in the firm and network. Eventhough markets and firms are organized by differ-ent principles, there is nevertheless a correspon-dence in their structure and properties. We returnhere to Smith’s and Stigler’s arguments that dif-ferentiation in the knowledge of the firm andmarket influence boundary decisions.

The findings of Gittelman suggest the groundedspeculation that the dialectic between specificmarkets and individual firm competence drives aco-evolution that enjoys a reflection in the struc-


ture of the network. An example serves to illus-trate this correspondence. The excellent study byAnnalee Saxenian (1994) compares the structureof semiconductor and computer industry networksin the Silicon Valley and Route 128. She foundthat the two regions had very different networkstructures, even though the technologies andindustries were the same. In Figure 3, we graphher ethnography. The top panel shows a hier-archical structure in Route 128; a few large firmsdominate smaller companies. The Silicon Valleyshows a decentralized network. Is it a coincidencethat the internal structure of the Silicon Valleyis described as flat and that of the Route 128firm as hierarchical? Why should the internalstructure correspond to the external structure?

In the case of the Silicon Valley, there is aninstitutional foundation that supports the flow ofinformation and matches engineer to project andfirm. Obviously, it is rarely in the interest of thecurrent employer to see proven research talentsexit their firm, and it is in their interest todiscourage involuntary exits. The evolution of alabor market for talent counters potential negativesanctions by the current employer. That is, thereare a sufficient number of job opportunities sothat in the event the engineer exits in the future(because the hiring firm dies or the new projectopportunity ends), subsequent sanctions by formeremployers are unlikely to be effective. A marketconsisting of many small networked firms cannotgenerate effective sanctions on the mobility ofengineers. There is, as a consequence of high

Figure 3. Saxenian’s ethnographic description


mobility, less motivation to build a vertical hier-archy by which to promise future rewards.

A labor market that is dominated by a fewlarge firms permits sanctions through refusal ofthese firms to rehire the engineer or through theirsignals to other client firms in the area. Thesesanctions need be no more than the loss ofrelative ranking in the internal hierarchies of thesedominant firms. (Note that the internal labor mar-kets of Silicon Valley are characterized as flat;tall hierarchical ranks are not viable if labormarket mobility is high and work is organizedby projects.)7 If a regional market does not sup-port labor mobility, then individual engineers arelikely to seek internal advancement, or—and itis important to stress this implication—to migratefrom the region. (See Almeida and Kogut, 1999,for evidence.)

The theoretical link between internal and exter-nal structure begins from the recognition thatfirms and markets are jointly emergent phenom-ena embedded in spatially-defined labor networks.Their structure reflects the emergent propertiesthat influence information and incentives, as wellas the know-how and coordination, that informfirm and individual strategies. The structuralopportunities through labor market has a powerfuleffect on differentiating the orientation of pro-fessional identities. In the hierarchical network ofRoute 128, engineers identify themselves withinternal labor markets; the Silicon Valley encour-ages identification along professional competencein projects.

The comparison between the Silicon Valleyand Route 128 raises the important distinctionbetween emergence and intentionality in networkstructure. Networks are rarely formed by design,but rather they emerge initially in response to theinstitutional and technological opportunities of anindustry or field. During this process of formation,relationships develop outinformational propertiesthat drive a matching process among firms. How-ever, over time, knowledge that is initially infor-mation gradually becomes encoded in persistingstructures that influence subsequent behavior in

7This line of argumentation comes closest to Coleman (1990:439ff.), who assumes that external labor markets are competi-tive markets. He thus does not consider the relationshipbetween the type of external and internal labor markets andits effects on the location of innovation. For an analysis ofa related problem, see Anton and Yao (1994).


two distinct ways: as a conduit of informationand as the basis ofcoordinated action.

Structure in a network is thus not determinedjust by exogenous factors, but is an expressionof competing and evolving rules that guide thebehaviors of interacting entities. Sometimes,inherent technological characteristics favor theemergence of particular rules. A technology thatenjoys scale economies tends to generate largefirms; another technology, such as microproc-essors (see below), tends toward network exter-nalities. These characteristics influence size distri-butions and the structure of a network. Similarly,institutional contexts (e.g., socialist or capitalist,German or American legal environments, etc.)influence the origins and formation of networks.Institutions, such as governments, sometimes dic-tate rules. A rule that establishes monopoliescompared to another that regulates prices hasdramatic implications for the emergence of indus-try structure and organizing principles of coordi-nation and competition. The effects of govern-ment discretion generated widely differentindustrial and relational structures among coun-tries (Hughes, 1983). Because markets and firmsare not simply given constructs, but arise fromvaried institutional origins and technologicalinfluences, there are no generic rules that areexogenous and known a priori. Rather, the sys-temic interaction of technological, social, andinstitutional factors influence the evolution of net-work structure.

Capabilities and rents

In an economic network, a firm is legally consti-tuted as the unit of accrual. Hence, cooperationand coordination in a network pose the questionswhether there are rents in networks and to whomdo they accrue. An answer to these questionsrequires an understanding of the location advan-tages to only information access in a networkcompared to membership in a coordinated net-work. In other words, we are interested in under-standing the conditions by which certain networkstructures generate value that is captured differen-tially by participating firms through their coordi-nation.

Call the first type of advantage a Burt rent.Burt describes the generation of network structureas the outcome of the competitive struggle amongegos motivated by envy and self-interest. The key

414 B. Kogut

construct for Burt is the notion of “non-redundant” ties. A tie is non-redundant if it rep-resents the only path between two nodes as con-stituted by individuals, firms, or even industries.Entities that have multiple unique (i.e., non-redundant) ties with other nodes who are notconnected occupy powerful brokerage positionscalled “structural holes.” Burt has argued thatfirms that are positioned in structural holes aremore powerful because they arbitrate the infor-mation flows between groupings of firms whohave loose ties with each other. Such networkshave a hierarchical structure, even if there aremany hierarchies. In this network, the rent accruesto the firm bridging the structural hole.

The second type of advantage can be calleda Coleman rent. Coleman (1990) stressed thatredundant ties among firms (or actors) result ina resolution to collective action problems. Coordi-nation is improved through repeated exchangeamong stable members to the group. This networktends to be flatter. In this case, depending uponthe quality of the interaction, the rent accrues tomembership in the group, with the actual allo-cation to individual members determined by rulesof adjudication and relative bargaining power.

For example, Uzzi (1996) describes networksin the textile industry as characterized by problemsolving, communication, and trust; Uzzi’s descrip-tion of a network is, in many ways, identical tothe advantages of a firm, as listed above, regard-ing coordination, learning, and communication.The advantage of a textile network is, however,the flexibility to explore new relationships andopportunities, but within a relatively closed cliquethat supports long-term trust among members.

Though Burt and Coleman networks are dis-tinct, they both generate potentially a rent tocoordination, though with very different impli-cations. The Burt rent accrues to the entrepre-neurial broker located in a structural hole. Thisis an efficient network insofar thatinformationflows throughout the network at the cost of main-taining a minimal number of relationships. Bro-kers, thus, increase the efficiency of the overallnetwork and capture, as a result, a rent for thisservice. Just as in studies of monopoly, the calcu-lation of whether the net welfare gain to thesystem is positive or negative depends upon theconjunction of the incentives for the broker toact in alignment with collective interests.

Studies by Burt (1997) indicate significant


rents to brokers, but do not assess the globalwelfare benefits. (The term “social capital” sug-gests a resource that improves collective welfareby fostering cooperation.) For example, his studyon promotion and remuneration of Americanmanagers indicates that occupying a structuralhole increases the size of bonuses (holding therest constant) for senior offices increased by$291,000. It could be that individuals who fillstructural holes are scarce entrepreneurs whoimprove internal coordination through controlover scarce resources and hence the performanceof the system. (Burt’s analysis of their attributesis ambiguous to these effects.) In this case, therole of a broker can be welfare improving foreveryone since a service for the network isrendered to everyone’s advantage.

However, in some kinds of networks, one par-ty’s gain is another’s loss. A zero-sum implicationis inherent, in fact, exactly in rank tournamentsfor promotion that serve as Burt’s example. Ifwe view promotion as a problem of mobilitychains, then one person’s promotion is indeedanother’s disappointment. In this case, local rentis at best a distributional transfer, ignoring thepossibility for dysfunctional consequences andinviting strategic behavior on the part of domi-nated individuals or firms. A zero-sum (if notnegative sum) game does not comply with mostdefinitions of social capital. Clearly, redistributionis implied in the studies on the structure of input-output activities that indicate some industriesoccupy structural holes (Burt, 1992).

A Coleman rent is associated with the benefitsof trust supporting coordination in long-termrelationships. This rent is not due to informationalefficiency, for as Burt forcefully notes, a redun-dant network is not a minimal communicationtree. However, dense relationships have the attri-bute of supporting monitoring and coordinationby matching incentives to contribution. One couldalso argue that dense networks also foster a senseof collective identity that supports coordinatedexchange.

Coleman (1990) distinguishes between inde-pendent and global viability in associations. Theformer consists of contributions of individualsto an organization (or closed network) with aproportional reward. Global viability, which Cole-man believes cannot sustain an organization overtime, rewards people at their reservation price ofpersistence in the club, while allowing for intra-


organizational payments to members in an amountthat violates rules of proportionality.8 WhereasBurt implies that this group rent flows to thebroker, a Coleman network claims that the gainsto superior coordination must be distributed inways to assure participation. Thus, differentnotions of viability are a critical distinctionbetween Coleman and Burt networks.

The importance of understanding viability isimplicit in the examples discussed earlier. Forexample, the study by Axelrod et al. (1995) reliedupon Nash equilibrium. A coalition is only viableto the extent there is no improvement for anyfirm to defect to the other coalition. For a Saxen-ian network to function, individuals in a hierarchymust view this as more rewarding than defectingto a start-up. A hierarchical firm is not viable inSilicon Valley, because individuals will chooseto switch jobs in that institutional setting.

The study by Walker, Kogut, and Shan (1997)explored both Burt and Coleman types of net-works. The early history of this industry revealeda network that was relatively unstructured andmore like a market. Certainly, while entrepreneurshad important affiliations to sources of ideas (e.g.,universities) or finance, horizontal ties amongfirms were weak.9 In this type of network, mar-ket-like relationships emerge through firms com-municating information regarding e.g., prices andspecifications. Coordination in this instance hap-pens through transactions governed by price sig-nals. Learning takes place through the revelationof cooperative or dishonest reputations.

Over time, a more complex network emerges.Figure 2, shown earlier, represents a relationalstructure that reveals both structural holes andColeman-type networks. The pharmaceutical com-pany marked II is an isolate that has non-redundant ties with 6 start-up firms; it clearlyoccupies a structural hole. However, some firms,such as those in group IV, engaged in moredense transactions that are suggestive of formativetype of Coleman networks. The analysis of sub-sequent relationships revealed that Coleman net-work based on coordination, inclusive of mutualknow-how exchange, emerged. Because of this

8This description is similar to the theory of clubs (and toOlson-type of selective incentives among small groups). Thereare many rules by which individuals can be rewarded thatsatisfy their reservation price (i.e., minimum for staying amember). See Cornes and Sandler (1996) for a summary.9See Powell, Koput and Smith-Doerr (1996) on ties.


social capital, firms belonging to the same groupstended to cooperate with each other subsequentto their initial cooperation. Network structurebegan to replicate itself in stable patterns ofenduring cooperation. It is not simply that bio-technology relationships are enduring across yearsthat explains this persistence. It is rather thatthese formative groups formed progressively moreclosed cliques; the flow of new relationships wasinfluenced by Coleman-type incentives for coop-erating firms to deepen their cooperation. Thispattern is also implied in the findings of Gulati(1995) who shows that partners tend to ally withthose close to them in the network and withwhom they have previously allied.

The emergent properties of networks ride onself-organizing processes that tend to freeze thestructure among firms over time into stable pat-terns of interactions. The Walker et al. (1997)study noted that a danger to a Coleman networkis that it limits search and can reduce variety.Uzzi proposes that the optimal network structurein the textile industry has a high density ofrelationships among firms, yet while allowingnew entrants and the possibility of further explo-ration. Since the advantage of the market is thegeneration of variety, too much structure reducesinnovation. Of course, to the extent firms whodefect from cooperation are eliminated, thisreduction is desirable. On the other hand, theconstraints on individual experimentation increasedue to requirements to orchestrate coordinationwith other actors. The more networks take onthe properties of firm organization, coordinationdeprives individual firms of potential avenues ofexploration. Thus, neither Burt, nor Colemanstructures can be ranked a priori for their welfaremerits; additional structure to the analysis isfirst required.10

It may help to analyze this line of debate by

10See the discussion in Walker et al. (1997) on the potentialfor dense relationships to drown out experimentation andlearning. Rowley, Behrens, and Krackhardt (2000) provide anexcellent review of the relationship of strong and weak tiesto exploration, with evidence indicating the importance ofindustry context in evaluating the relationship of structure toindividual firm performance. See also the discussion byWalker (1998) on the search models indicating that weak tiesproliferate from the objective of maximizing informationaccess. This notion of the proliferation of weak ties is similarto Khanna, Gulati, and Nohria’s (1994) notion of alliancescope. See also Baum et al. (2000), and the comparisonby Zaheer and Zaheer (1997) of structural holes and weaktie arguments.

416 B. Kogut

anchoring the discussion in a few empirical facts.The primary observation is that alliance networksare exceedingly sparse. If there are 100 firms ina network and if we only count ties as non-directional (we ignore who sends and whoreceives), then we expect there to be 4950 poten-tial ties. (We code a tie as 0 or 1, regardless ofhow many agreements two firms may have witheach other). If we think about these ties as form-ing different coalitions of players, then we havea 2 to the n problem defining the solution spacefor potential membership in 2 coalitions—2because a firm can choose to join or not join.).For a 100 firm network, there are then 2100

possible combinations in membership. Despite thecombinatorial richness in potential alliances,empirically, we see networks that are sparse (i.e.,the actual ties are far below the maximum) andthat engage in rather limited experimentation overtime regarding the changing identity of coalitionpartners.11

It would seem that sparse networks favor aBurt-like description of many structural holes.But another way to think about the strategicimplications of stability in what appears to beself-organizing patterns of cooperation is to askhow sensitive are these groupings to defection ofpartners. (In the theory of graphs, this exerciseis akin to asking how robust is structure when xof k edges are reassigned randomly among nvertices.) After all, where there are rents, thereare bound to be strategies to alter the structureby new alliances. An important test for the likeli-hood of success of a strategy is to compareColeman-type structures (i.e., dense ties amongfew actors) and Burt-type structures (broker firmswith few redundant ties among its satellites) forrobustness.

The recognition that the results of this compari-son are easy to predict reflects the stability ofstructuredespitevery sparse networks. Social net-works are fairly stable to random perturbations(at moderate probability) because structures tendto be localized.12 This statistical result is strength-ened if we admit ties are not randomly assigned,but tend to follow generative rules that encourage

11See Kogut and Walker, 1999, for a discussion of someexamples.12For simulations of the robustness of local structure (that is,small worlds) in sparse networks, see Watts and Strogatz(1998).


cooperation among firms already cooperating.After all, the advantage of membership in a localclub promotes further cooperation among mem-bers. There is, in other words, considerable orderdespite rather low density of cooperation and thepotential for a multitude of alliance structures. Itis not surprising, then, that industries (or anycommunity) vary in their structures, and yet sharethe property of tending toward self-replicatingpatterns of cooperation.

Yet, despite the tendency of a Coleman net-work to generate incentives for replication evenin sparse networks, we are unlikely to arrive ata robust finding regarding global or group wel-fare. In one industry, the advantages of standardsetting suggest that one rule for determining thedecision to cooperate is to join a big coalition(Axelrod et al., 1995). Following this rule, forexample, in Uzzi’s textile industry butts upagainst a more appropriate rule to sort by prestigeamong low and high quality designers. It is notsurprising that empirical results seem contradic-tory, when they reflect differences in the proper-ties of given networks.

Similarly, caution is required to assume thatBurt networks are not stable and converge overtime to a market network. (Burt, 1999, viewsstructural holes as dynamically unstable.) It isimportant to avoid the logical fallacy of attribu-ting persistence to genesis. Once structures arecreated, we need to ask what sustains them.Because action is constrained in emergent net-works, the late recognition of how defection canbenefit an individual firm may be inconsequential;that is, the firm cannot act upon it. Partners maynot be available and may be unwilling to defect;switching may be costly and promise only uncer-tain advantages. Again, the notion of viability iscritical. Defection may be prohibited because ofsanctions imposed by the group. Or, for Padgett’sand Ansell’s medieval Florence or for Poldony’sand Stuart’s semiconductor firms, defection isdeterred because the identities of membership ina group dissuade alliances with less prestigiousfamilies or firms. For either motive—economicor social sanctioning, the preferred strategy maybe one of local replication, such as imitating thesupplier strategy of a competitor, than creatingshort-cuts in the hope of destabilizing a fairlyimmobile structure. In theory, there is no clearreason to believe that structural holes are notsustainable.


Property rights and network centrality

Of course, another way in which the emergenceof structure is not only influenced, but also sus-tained, is through property rights. A basic prop-erty right is the ownership by an individual ofthe use of natural endowments, such as skill. Ofcourse, if human capital were alienable from theperson, it too would flow like a resource througha network, diminishing its value. It is, however,the stickiness of human capital that influences aperson’s eligibility to play a brokering role. Thetacitness of firm knowledge similarly makes afirm less susceptible to the competitive imitationof its claim to broker.

This confounding of position in a network andattributes that makes one firm more central thananother poses an econometric problem of se-lection bias. Rents may accrue to “quality” peopleor firms who therefore occupy structural holes.

In this regard, ownership of an innovation thatis property right protected is an attribute that influ-ences the generation and appropriation of rentsindependent of network effects. Yet, even here, thecausality behind the generation of rents is complex,and the structure of the network becomes anendogenous feature in competition among inno-vators. It is important to emphasize that cooperativeagreements are frequently concessions that permitthe utilization of one firm’s knowledge by another.Examples are the licensing of technology or thedecision to cooperate in a joint venture in whichfirms contribute know-how. It is a common featurethat such agreements prohibit the selling of thetechnological rights to other firms, thus preventingundesired strategic short-cuts.

Property rights can particularly have a powerfuleffect on networks if the resource is scarce insofarthat it constitutes a “bottleneck” technology, suchas an operating system standard for software, oran electrical grid, or the telecommunication“pipe” to a residence. Possession of rights to abottleneck resource can lead clearly to a mo-nopoly position. However, it may also lead toisolation, rather than centrality, if a firm decidesto exploit its position without cooperation. Hencecooperation is critical.

Cooperation is likely to result when a firmowns a scarce resource, and yet is competingagainst other firms who offer alternatives. Theseproperty rights to a bottleneck resource areespecially valuable when coupled with “network


externalities.” These externalities arise when theconsumption or use of a good by one person orfirm makes it attractive to another to do the same—the classic example being a computer operatingsystem. In these joint conditions of externalitiesin competitive environments, cooperation isencouraged.

Such externalities, for example, exist inmicroprocessors. Since software is written formicroprocessor standards and people want to usethe same software, there exist externalities thatfavor the dominance of one standard over another.For a microprocessor firm, the logical strategy isto grab the largest size of the market. Somewhatcounter-intuitively, it would want to induce entryinto its market as long as these entrants agreedto license its technology and standards. In fact,cooperation exploded in the microprocessorindustry until Motorola and Intel achieved domi-nance; National Semiconductor did not achievethe same penetration and, interestingly, main-tained a higher level of alliance activity. Becauseall entrants were required basically to cooperateon the standards, these three firms were eachcentered in the middle of a star of relationships.Centrality thus was the outcome of network exter-nalities coupled with a strong property rightregime.13 Thus, in this case, the strategy to appro-priate rents through technology licensing gener-ated the structure, rather than structure simplydetermining the rents to a broker or Colemangroup.

It is proverbially axiomatic that in a hub andspoke structure, such as associated with the domi-nance of Intel in microprocessor licensing, thecentral firm is in the better position to reap therents in the network. (Think of this prediction asidentical to the measure of market power by salesconcentration.) Certainly, any bottleneck positionshould be associated with differential marketpower. The bottleneck could be property rightsto the limited number of gates at an airport(which clearly is reflected in a “hub and spoke”transportation pattern), the communications pipe-line to the home, water reservoirs in a desert,etc. All of these bottlenecks produce structuralholes with the gains to the broker.

In many industries, property rights to bottle-necks are not a characteristic of the competitive

13This analysis is given in Kogut, Walter, and Kim (1996).

418 B. Kogut

landscape. Firms would certainly benefit fromtrying to replicate the rent capture imposedthrough hierarchical dominance. And yet, theclaim of Uzzi and others is that rent generationcan be superior in a dense clique with thick tiesamong the players because of improved coordi-nation and problem-solving. In such cases, thepreservation of cooperation is maintained becauseexclusion to the club deprives the defectingmember from sharing the group rents. (Thissanctioning possibility is the basis of the Nashcomparisons behind the Axelrod, et al., 1995,simulation discussed earlier.) In other words,rents to coordination again provide self-organizingincentives to members to maintain the networkstructure.

Table 1 summarizes the discussion of therelationship of network structure and propertyrights to bottleneck resources. The consequencesfor rent generation and distribution depend on theassessment of the viability of competing rulesfor cooperation. The empirical studies of variousindustries indicate that certain rules came todominate, but in the context of particular histori-cal and institutional settings. Thus, the rule forcooperation that appears to have proliferated inmicroprocessors is to share technology, while notgiving up control to the bottleneck resource itself.In some industries where property rights arestrong, but there is no bottleneck technology (e.g.,in pharmaceuticals), the emergent structure does

Table 1. Competing rules and structural predictions (Ignoring social rules)

Regulatory Technological Feasible Structure Industry Example Competing Rules

1. Strong property Bottleneck Central players with no Microprocessors, Induce entry byrights resource isolates software operating licensing vs. dominate

systems by superior technology

2. Strong property No bottleneck Weak hierarchies with Pharmaceuticals Cooperate for financerights resource many isolates vs. dominate by

superior technology

3. Weak property Bottleneck Many closed hierarchies Autos Source widely whilerights resource with no isolates switching often vs.

build competence infew single-sourcesuppliers

4. Weak property No bottleneck Decentralized relational Information and Seek new informationrights resource networks financial markets vs. rely on existing

relationships


not consist of competition among a few hubs,but reveals a complex structure with many centralfirms and also many isolates. One surmises thatin this industry, centrality might reflect the “qual-ity” of a firm rather than its control over abrokerage position. This may be the reverse inthe financial industry, where in fact the positionin an information flow results in the capture ofrents. Quality is an attribute of the position.While trading relationships are still embedded instructure (Baker, 1984), the incentives to cooper-ate are attentive to positions of prestige and rank.Here indeed we have Burt rents accruing to struc-tural holes.

An interesting case is where property rights toa given scarce resource are not strong, but therestill exists discernible structure among competinghubs. Automobile assembly is a good example,whereby assemblers have some power over accessto distribution channels and customer loyalty butthey do not have property right control overunique assembly skills. Indeed, entry by newcompanies has been an important element in thehistory of this industry.

The dependence of structure on historical andinstitutional conditions is also revealed in thevariety of competing rules and their implicationsfor the emergence of distinct network configur-ations. The evolution of the Toyota System illus-trates precisely the migration from a hub-and-spoke structure to a more cooperative self-


organizing supplier system because the generativerules of cooperation changed over time. Becausethe work on supplier chains in the automobileindustry consists of a rich set of studies thathighlight the creation of capabilities through net-works, we use this industry in the followingsection to provide a holistic exploration of theideas of organizing principles as generative rulesand the relationship of emergent structure andrents.

An illustration: Toyota production system

The Toyota Production System constitutes one ofthe most important organizational innovations ofrecent decades, yet it did not emerge out of aconscious design but out of an emergent process(Fruin and Nishiguchi, 1993). The Japanese mar-ket consisted after the war of a demand for highvariety, a plethora of auto companies none ofwhich operated plants built to the scale of themass production facilities of American com-panies. Furthermore, Japanese suppliers wereinitially inferior in their capabilities compared toassemblers (Nishiguchi, 1994). Over time, this‘dual labor market’ evolved into a new divisionof labor based upon the continuous upgrading ofsupplier competence and their participation inproject design. These new tight supplier relation-ships created capabilities that increased speed tomarket, quality, and new model cycle times, withsupplier networks as the organizing principle todeliver this capability.

The inter-organizational model emerged inJapan during a period from 1965 to the early1980s. Over this time period, the productionstructure shifted toward reciprocal, multilateralrelations and a concern with specific rights oftransaction rather than residual rights of owner-ship. From using subcontractors mainly as buffersin the 1950s, assemblers were after 1960 commit-ted to upgrade their subcontractors’ technicalcapabilities. The composite know-how ofassemblers was transferred to suppliers throughteaching, along with assembly lines. The emer-gence of contract assembly and subsystems manu-facture noticeably changed the logic of supplierrelations toward “collaborative manufacturing.”There were obviously gains to both buyers andsuppliers in the form of increased returns ofhigher order organizing principles.

Though historically emergent, the transfer of


the Toyota Production System represents theefforts to transfer networks by design to othercountries (Florida and Kenney, 1991; Dyer,1996). However, the rules by which suppliernetworks emerged in Japan are different than therules that guide its design and transfer. (Hereagain, we must separate out genesis frompersistence.) In effect, the emergence of inter-supplier networks were guided by these two rules:

1. The dual labor market is not violated by inte-grating suppliers into Toyota.

2. Supplier capabilities are improved through thetransfer of competence to them.

The first rule grew out of the recognized limitsof a Japanese company toward extending thereciprocal agreements among employees to sup-pliers. Membership in a large company entailedexpectations of long-term employment in returnfor strong identification of the employee withthe firm. Contrary to Stigler’s hypothesis of theintegration of weak suppliers, the social expec-tations underlying the employment relationshipprecluded the integration of suppliers into thecore firm.14 The second rule developed partly asa consequence of the first, as well as throughgovernment policy to protect weaker suppliersagainst the dominance of large assemblers(Nishiguchi, 1994).

Many subcontractors welcomed the new sub-contracting system, since it brought with it morestable contractual relations through increased assetspecificity, more opportunities for technologicallearning, and improved growth prospects(Nishiguchi, 1994). From using subcontractorsmainly as buffers in the 1950s, assemblers wereafter 1960 committed to upgrade their subcontrac-tors’ technical capabilities. The composite know-how of assemblers (including the know-how tooperate assembly lines) was transferred to sup-pliers through teaching. The emergence of con-tract assembly and subsystems manufacturenoticeably changed the logic of supplier relationstoward “collaborative manufacturing.” There wereobviously gains to both buyers and suppliers inthe form of increased returns of higher orderorganizing principles.

Still, these two rules generated initially a net-

14This relationship is described in many places, including Aoki(1990), Ablegglen and Stalk (1985), and Nischiguchi (1994).

420 B. Kogut

work structure that was strikingly hierarchicalin structure, creating a hub-and-spoke centralityaround Toyota. In this regard, the Toyota supplierstructure was organized along hierarchical linesthat were not substantially different than thosefound among General Motors and its suppliers.However, the rule to respect the dual labor marketprohibited the extensive vertical integration seenamong American firms. Toyota’s value addedcontribution to its autos was, and remains, rad-ically less than General Motors’ share of valuein its assembled cars. (See also Dyer, 1996, whofound the internal value-added of an Americanassembler to be twice that of a Japanese.)

The dynamic that transformed this hierarchyinto a more closed relationship stemmed from thelogic of a series of organizational and processinnovations that began within Toyota and eventu-ally diffused out to suppliers. The initial inno-vations of Toyota centered around the introduc-tion of customer-driven production (kanban)manufacturing. The danger of this kind of pro-duction system is that working capital growthserves as the buffer to respond flexibly to changesin customer demand. Almost by contradiction,Toyota innovated by minimizing inventory levelsthrough the innovation of just-in-time (JIT) deliv-eries. These systems were coupled with powerfulanalytical tools, such as value analysis and qualitycontrol, which generated information used tominimize costs.15

If kanban and JIT were powerful innovations,rapid introduction would appear to be fosteredthrough vertical integration. In fact, through verti-cal integration, a firm should have more powerto enact JIT because it has the authority to doso. Supplier resistance to these innovations clearlyfrustrated the chief architect of Toyota’s inno-vations, Ono, who complained bitterly over theresistance to these new methods (Cusumano,1985). Lieberman and Demeester (1999) docu-ment the slow diffusion of these methods bytracking inventory levels among suppliers, findingthat they continue to decrease for Toyota sup-pliers through the 1970s, with Nissan and otherJapanese firms lagging considerably.

The gradual extension of these new organizingprinciples of customer-driven productiontransformed supplier relations from the exploi-

15Ohmae’s (1982) early study is a brilliant explanation of thestrategic implications of value analysis of Japanese firms.


tation of subcontractors to collaborative manufac-turing based on multilateral problem solving.Over time, asset-specific contractual relationsincreased, as exemplified by contract assemblersand system-component producers. Structurally,Japanese subcontractors were reorganized overtime into tiers through a concentration of orders,intensified specialization, and increased depen-dence on particular customers (Fruin and Nishigu-chi, 1993). In the new tightly tiered structure,approximately 180 first-tier suppliers contract toseveral thousand lower-tier subcontractors that, inturn, contract to tens of thousands third tiersuppliers. This structure maintained variety inidentity and competence among suppliers, whilegenerating long-term coordination among thekey participants in the Toyota System.

This tiered structure implies that at a giventime, subcontracting philosophies at plants withinthe same firm vary widely. Nishiguchi comparesphilosophies within the same buyer firm, rangingfrom “bargaining” subcontractor managers whoused subcontractors to protect regular workerswhen demand fluctuated to pro-subcontractingmanagers who engaged in intense contacts, train-ing and problem solving. Nishiguchi’s results areconsistent with asset specificity not being thecause but the consequence of a particular strategy.(In fact, he finds asset specificity to be a conse-quence of the supply strategy.) Yet, at the sametime, a strategy that points to asset-specific con-tractual relations needs inter-firm relationalmechanisms that enable them to function.

The increased reliance by Toyota on first-tiersuppliers generated important organizational inno-vations designed for the new network. Throughrepeated interaction between firms in the network,a series of innovations emerged that supportedthe acquisition of skills specific to the relation-ships, or what Asanuma (1989) has called‘relation-specific’ capital. (See also Dyer, 1996,and Dyer and Singh, 1998, for an exposition ofthis idea.) These innovations included joint pricedetermination based on objective value analysis,joint design based on value engineering, the targetcost method of product development, profit-sharing rules, subcontractor proposals, black boxdesign, resident engineers, subcontractor grading,quality assurance through self-certified subcon-tractors, and just-in-time delivery circumscribedby bonus-penalty programs. These innovationsrepresented the shift from main purchasing func-


tion of the customer shifted from downstreamprice negotiation to the assessment of subcontrac-tor performance and the coordination of inter-firm functions.

Asanuma’s description of the heterogeneity ofthe Toyota supplier system indicates a networkin which a set of tiered suppliers acts to differen-tiate the status of suppliers. An important distinc-tion is between suppliers that work according todrawings supplied by the core firm, and thosethat submit their own drawings to the core firmfor approval. Over time, the latter groups ofmore innovative suppliers evolve toward greaterindependence because of an increase in volumeproduced by a supplier, or an increase in thescope of activities performed by a supplier.

Through monitoring and supplier qualificationrequirements, the core plant selectively developsrelationships with suppliers. Suppliers are evalu-ated according to how well they have performedon earlier contracts. Often partial ownership issought in the suppliers that rank the highest interms of performance and potential capabilities.Moreover, suppliers earn points in the supplierrating system for codifying methods so that theycan be used by other suppliers, leading to lowercosts for the core firm. By this dynamic, continu-ous learning lead to improvement in productivityof the whole supplier networks. All types ofsuppliers had to develop some skills to maintainthe relation to the core firm, other than purelytechnological capabilities.

In effect, Toyota evolved a system that reliedupon self-organization to resolve the contradictionbetween its two rules: the inability to integratewhile requiring improvement in supplier skills.The tiered system forced firms to prove theircompetence and yet also created incentives forthem to codify and share their knowledge. Thus,the first two rules were augmented by a third:

3. To participate in the first tier, suppliers arerequired to prove, codify, and share their com-petence with each other.

Unlike the operation of the competitive hierarchyto American assemblers, this hierarchy thenevolved to move the single locus of innovationfrom the core assembler (Toyota) to the suppliersas well. Nishiguchi (1994) calls this “clusteredcontrol”; in our terminology, it represents a Cole-man network of dense ties between the members.


This structure achieved “independent viability”in Coleman’s terms by devising rent-sharing rulesthat supported decentralized innovations. Initialprices are set in the light of planned productioncosts, based on the supplier’s and the core firm’sexperience with similar parts. It clearly recognizesthe firm as the “unit of accrual” and structuresdecentralized incentives to promote coordination. Ifthe supplier should improve the process and outpacethe planned experience-derived cost reductions, itretains the savings. The pricing mechanism reflectsnot only an anticipation of learning on the part ofthe suppliers; it provides an incentive to beat thetarget. However, at the same time, rule threedictates that these improvements flow to othersuppliers. As a consequence, improvements atone supplier flow dynamically to others.

Independent viability of membership is a neces-sary factor in the self-organizing character to theToyota Supply system.16 Consider a firm thatwithheld technology and sought to free-ride onthe efforts of others. The intense informationflows permit easy monitoring; sanctions need beno more than exclusion from the first-tier clubthat shares rents. No wonder in times of crisis,cooperation can be organized without hierarchicalcontrol, a key attribute of a self-organizing sys-tem. (See Nishiguchi and Beaudet, 1998, for adiscussion of self-organization following a fire ata top-tier supplier.) It is important to observethat this system, from the point of view of Toy-ota, represents a less costly expenditure of timeeven if it involves a dense set of ties. Becausemonitoring is coupled with cooperation in tech-nology transfer, it is also occasion to learn fromother supplier’s experiences (see Sabel, 1996;Helper, MacDuffie, and Sabel, 1998).

Dense networks provide an important capabilityof knowledge acquisition, in conjunction withalso generating information required for monitor-ing and enforcement. Thus, monitoring occurs notas a function of an overt sanction mechanism,but rather through the operation of professionalidentities that support the transfer of technology

16This viability is similar to Axelrod et al.’s (1995) impositionof Nash equilibrium: firms do not switch coalitions if thechange does not improve their utility. Of course, firms mightwant to switch into a Toyota coalition and free-ride; hencethe importance of the third screening rule.

422 B. Kogut

among suppliers and Toyota.17 Whereas a broker-age role is efficient from the perspective of moni-toring and sanctioning, the self-organizing proper-ties of dense relationships benefits from puttingthe stress on the rapid flow of competenciesthrough frequent and on-going relationships.

The organizing principles of the Toyota Systemsupport the capabilities of providing variety andspeed to the market. Capable suppliers in a net-work provide competing variety based uponspecialized competence. Black box modularitypermits specialization, and yet demands a highdegree of coordination. The rapid diffusion ofproduction know how serves to reduce the costs,while the tight coordination of suppliers andassembly in design and production reduces overalltime to the market.

These capabilities did not reside in any givenfirm, but were created by the knowledge of howto coordinate among firms with a history ofcooperation. The unit of accrual is the individualfirm; there is no holding company structure tothe network. Yet, to remove a firm from thisnetwork would be to deprive it of importantcapabilities that it could not immediately recreate,even if it could access equally capable suppliers.These capabilities are not simply the static onesof reducing inventory, but of encouraging inno-vations by technology transfer and incentives.

The Toyota System thus created a structure inwhich variety could be maintained among strongsuppliers without disrupting the centrality of theassembler in the system. Toyota thus did not fallprey to Stigler’s solution of vertical integrationto overcome inefficiencies in the network. How-ever, each firm remained the unit of accrual.These adaptive innovations, thus, posed the Willi-amson problem that property claims to thisknowledge were weak, hence creating the incen-tive for suppliers to maintain secrecy. Thesethreats were resolved, though, by three mecha-nisms. Coordination generated rents that inducecooperation by acting as an bond or efficiencywage to deter defection; defection would lead todeprivation from future rents. Second, the processof technology transfer itself created informationthat enabled monitoring within the network asopposed as through hierarchical control. But there

17It is not surprising that suppliers are often conflicted intheir loyalties, especially at times when these networks aretransferred to new locations.


is also a third mechanism, the value of enhancecoordination through the identity of members inthe high status Toyota Production system.

The Toyota system embodies the irony thatthese principles of coordination arose due torestrictions on vertical integration. But if theemergence of the Toyota network could not havebeen foreseen, its transfer by replication in theUnited States and elsewhere entails intentionalcoordination among network members. Yet, theseintentions do necessarily demand authority toinsure replication. Toyota does not need to orderits American suppliers: create the Toyota Pro-duction System by this blueprint. Replication neednot follow a description of who cooperates withwhom and how. For the incentives to replicateimplies the possession of capabilities that are rentgenerating, hence rewarding members to re-createthe structure. While the origins of the ToyotaSystem suggest the operation of unintendedconsequences in the evolution of the network, theintentional adherence to these rules suggests afunctional understanding of their causal conse-quences.

CONCLUSIONS

We began with the observation that value is nota mystical entity. The source of its imputation isnot always clear, as witnessed in the lack ofconsensus over the interpretation of a residual,called total factor productivity. In recent years,we have come to understand better that animportant source of value for a firm lies in thecapabilities supported by organizing principles ofwork. These principles constitute what is meantas the knowledge of the firm.

The study of networks as knowledge under-stands capabilities achieved through coordinatedaction at multiple levels of analysis. At one level,knowledge is the principles defining coordinationin a division of labor that anchor identities ofindividuals and groups within firms. At anotherlevel, the boundary of firm and network are mal-leable definitions determined by shifting identitiesand their co-evolving capabilities. Operating uponthese levels is the domain of generative rules ofcooperation and competition.

The network generated by rules of cooperationdifferentiates firms by their structural positions.Since firms but not networks are units of accrual


and selection, there exists, therefore, a potentialdivergence between the distribution of these rentsand the contribution of individual firms. Some-times, this divergence is mitigated through thecoincidence of structural position and propertyright claims. However, in situations in whichknowledge is diffuse among a group of firms,coordination can become prey to concerns overcooperation. Embedding a monitoring and sanc-tion mechanism into a cycle of positive returnsattached to technology transfer drove the parti-cular success of the supplier system of the ToyotaProduction System. And by devising crediblerules that guaranteed independent viability, Toy-ota could, by intention, replicate the network(even if particular members changed) in newlocations.

Networks are more than just relationships thatgovern the diffusion of innovations and norms,or explain the variability of access to informationacross competing firms. Because they are theoutcome of generative rules of coordination, net-works constitute capabilities that augment thevalue of firms. These capabilities, e.g., speed tomarket, generate rents that are subject to privateappropriation. It is through an understanding ofnetworks as knowledge encoding coordinationwithin and between specialized firms in specificcooperative and competitive structures that the“missing” sources of value can be found.

ACKNOWLEDGEMENTS

This paper is supported through the Reginald H.Jones Center of the Wharton School. I wouldlike to thank Gordon Walker for many helpfuldiscussions, Udo Zander who contributed substan-tially to parts of this paper, and the anonymousreviewers.

REFERENCES

Ablegglen, J. and G. Stalk, Jr. (1985).Kaisha: TheJapanese Corporation. Basic Books, New York.

Almeida, P. and B. Kogut (1999). “Localization ofknowledge and the mobility of engineers in regionalnetworks,” Management Science, 47, pp. 905–917.

Anton, J. and D. Yao (1995). “Start-ups, spinoffs, andinternal projects.”Journal of Law and Economicsand Organization, 11, pp. 362–378.

Aoki, M. (1990). “Toward an Economic Model of the


Japanese firm,”Journal of Economic Literature, 28,pp. 1–27.

Asanuma, B. (1989). “Manufacturer-supplier relation-ships in Japan and the concept of relation-specificskill,” Journal of the Japanese and InternationalEconomies, 3, pp. 1–30.

Axelrod, R. and M. Cohen (2000).Harnessing Com-plexity. Free Press, New York.

Axelrod, R. and W. Hamilton (1981). “The evolutionof cooperation in biological systems.” In R. Axelrod(ed.), The Evolution of Cooperation. Basic Books,New York, pp. 88–105.

Axelrod, R., W. Mitchell, R. Thomas, D. S. Bennettand E. Bruderer (1995). “Coalition formation instandard-setting alliances,”Management Science, 41,pp. 1493–1508.

Baker, W. (1984). “The social structure of a nationalsecurities market,”American Journal of Sociology,89, pp. 775–811.

Baum, J. A. C., T. Calabrese and B.S. Silverman.(2000). “Don’t go it alone: Alliance networks andstartups’ performance in canadian biotechnology,”Strategic Management Journal, Special Issue,21,pp. 267–294.

Boorman, S. (1974). “Island models for takeover by asocial trait facing a frequency-dependent barrier in aMandelian population,”Proceedings of the NationalAcademy of Sciences, 71, pp. 2103–2107.

Brown, J. S. and P. Duguid (1991). “Organizationallearning and communities of practice,”OrganizationScience, 2, pp. 40–57.

Burt, R. (1992).Structural Holes: The Social Structureof Competition. Harvard University Press, Cam-bridge, MA.

Burt, R. (1997). “The contingent value of social capi-tal,” Administrative Science Quarterly, 42,pp. 339–365.

Burt, R. (1999). “The network structure of social capi-tal,” mimeo, University of Chicago.

Chandler, A. (1962).Strategy and Structure: Chaptersin the History of the Industrial Enterprise. MITPress, Cambridge, MA.

Clark, K., B. Chew and T. Fujimoto (1987). “Productdevelopment in the world auto industry,”BrookingsPapers on Economic Activity, 3, pp. 729–771.

Coase, R. (1937 orig. pub. date; 1995). “The natureof the firm.” In: S. Estrin and A. Marin (eds),Essential Readings in Economics. St. Martin’s Press,New York.

Coleman, J. (1990).Foundations of Social Theory.Harvard University Press, Cambridge, MA.

Cornes, R. and T. Sandler (1996).The Theory ofExternalities, Public Goods and Club Goods, 2ndedn. Cambridge University Press, New York.

Cusumano, M. (1985).The Japanese Automobile Indus-try. Technology and Management at Nissan andToyota. Harvard University Press, Cambridge, MA.

Dyer, J. (1996). “Does governance matter?Keiretsualliances and asset specificity as sources of Japanesecompetitive advantage,”Organization Science, 7,pp. 649–666.

Dyer, J. and H. Singh (1998). “The relational view:

424 B. Kogut

Cooperative strategy and sources of interorgani-zational competitive advantage,”Academy of Man-agement Review, 23, pp. 660–679.

Florida, R. and M. Kenney (1991). “Transplantedorganizations: The transfer of Japanese industrialorganization to the U.S,”American SociologicalReview, 56, pp. 381–398.

Fruin, M. and T. Nishiguchi (1993). “Supplying theToyota production system: Intercorporate organi-zational evolution and supplier subsystems.” In B.Kogut (ed), Country Competitiveness: Technologyand the Organizing of Work. Oxford UniversityPress, New York, pp. 225–246.

Gittelman, M. (1999). “Scientists and networks: Acomparative study of cooperation in the French andAmerican biotechnology industry,” Ph.D. thesis,Wharton School, University of Pennsylvania.

Gomes-Casseres, B. (1994). “Group versus group: Howalliance networks compete.”Harvard BusinessReview, 72, pp. 62–74.

Gulati, R. (1995). “Social structure and alliance forma-tion pattern: A longitudinal analysis,”AdministrativeScience Quarterly, 40, pp. 610–652.

Gulati, R. (1998). “Alliances and networks,”StrategicManagement Journal, 19(4), pp. 293–318.

Gulati, R. and P. Lawrence (1999). “Organizing verticalnetworks: A design perspective,” paper presented at theSMJ Special Issue Conference on Strategic Networks.

Hayek, F. (1988).The Fatal Conceit. The Errors ofSocialism. University of Chicago Press, Chicago, IL.

Helper, S., J. P. MacDuffie and C. Sabel (1998).“Boundaries of the firm as a design problem,” Regi-nald H. Jones Center Working Paper Series, WP98-11, The Wharton School.

Herrigel, G. (1993). “Large firms, small firms, and thegovernance of flexible specialization: The case ofBaden Wurttemberg and socialized risk.” In BruceKogut (ed.), Country Competitiveness. MacMillan,London. pp. 15–35.

Holland, J. (1992).Adaptation in Natural and ArtificialSystems: An Introductory Analysis with Applicationsto Biology, Control, and Artificial Intelligence. MITPress, Cambridge, MA.

Hughes, T. (1983).Networks of Power Electrificationin Western Society, 1880–1930. Johns Hopkins Uni-versity Press, Baltimore, MD.

Khanna, T., R. Gulati, and N. Nohria (1998). “Thedynamics of learning allainces: Competition,cooperation, and relative scope,”Strategic Manage-ment Journal. 19(3), pp. 193–210.

Kogut, B. and G. Walker (1999). “The small worldof firm ownership in Germany: Social capital andstructural holes in large firm acquisitions – 1993–1997,” working paper, Reginald H. Jones Center,The Wharton School.

Kogut, B., G. Walker, and D. J. Kim (1996).“Cooperation and entry induction as a function oftechnological rivalry,” Research Policy, 24,pp. 77–95.

Kogut, B. and U. Zander (1996). “What firms do?Coordination, identity, and learning,”OrganizationScience, 7, pp. 502–514.


Lieberman, M. and L. Demeester (1999). “Inventoryreduction and productivity growth: Linkages in theJapanese automotive industry.” ManagementScience. 45, pp. 466–485.

Lipparini, A. and A. Lomi (1996). “Relational struc-tures and strategies in industrial districts: An empiri-cal study of interorganizational relations in the Mod-ena biomedical industry,” mimeo, University ofBologna.

Lorenzoni, G. and C. Baden-Fuller (1993). “Creatinga strategic center to manage a web of partners.”California Management Review. 3, pp. 146–163.

Mansfield, E. (1988 orig. pub. date 1995). “The speedand cost of industrial innovation in Japan and theUnited States: External vs. internal technology,”Innovation, Technology and the Economy: SelectedEssays of Edwin Mansfield, Vol. 1. Elgar: Brook-field, VT.

Nishiguchi, T. (1994).Strategic Industrial Sourcing.Oxford University Press, New York.

Nohria, N. and C. Garcia-Pont (1991). “Global strategiclinkages and industry structure,”Strategic Manage-ment Journal, Summer Special Issue,12, pp. 105–124.

Ohmae, K. (1982).The Mind of the Strategist: TheArt of Japanese Business. McGraw-Hill, New York.

Nishiguchi, T. and A. Beaudet (1998). “The Toyotagroup and the Aisin Fire,” Sloan ManagementReview. 40, pp. 49–59.

Padgett, J. and C. Ansell (1993). “Robust action andthe rise of the Medici, 1400–1434,”American Jour-nal of Sociology, 98, pp. 1259–1319.

Palmer, D., P. D. Jennings, and X. Zhou (1993). “Lateadoption of the multidivisional form by large U.S.corporations: Institutional, political, and economicaccounts,” Administrative Science Quarterly, 38,pp. 100–131.

Piore, M. and C. Sabel (1984).The Second IndustrialDivide: Possibilities for Prosperity. Basic Books,New York.

Podolny, J. and T. Stuart (1995). “A role-based ecologyof technological change,”American Journal of Soci-ology, 100, pp. 1224–1260.

Powell, W. (1990). “Neither market nor hierarchy:Network forms of organization,” Research inOrganizational Behavior, 12, pp. 295–336.

Powell, W., K Koput and L. Smith-Doerr (1996).“Interorganizational collaboration and the locus ofinnovation: Networks of learning in biotechnol-ogy,” Adminstrative Science Quarterly, 41,pp. 116–145.

Rowley, T., D. Behrens and D. Krackhardt (2000).“Redundant governance structures: An analysis ofstructural and relational embeddedness in the steeland semiconductor industries,”Strategic Manage-ment Journal, Special Issue,21, pp. 369–386.

Sabel, C. (1996). “Learning by monitoring: The insti-tutions of economic development.” In N. Smelserand R. Swedberg (eds.),The Handbook of EconomicSociology. Princeton University Press, Princeton, NJ,pp. 137–165.

Saxenian, A. (1994).Regional Advantage: Culture and


Competition in Silicon Valley and Route 128. Har-vard University Press, Cambridge, MA.

Smith, A. (1965). The Wealth of Nations. ModernLibrary, New York.

Stigler, G. (1951). “The division of labor is limited bythe extent of the market,”Journal of Political Econ-omy, June, pp. 185–193.

Stuart, T., H. Hoang, and R. C. Hybels (1999).“Internorganizational endorsements and the perform-ance of entrepreneurial ventures.”AdministrativeScience Quarterly, 44, pp. 315–349.

Uzzi, B. (1996). “The sources and consequences ofembeddedness for the economic performance oforganizations: The network effect,”American Socio-logical Review, 61, pp. 674–698.

Walker, G. (1998). “Strategy and network formation.” InJ. Baum and H. Rao (eds.),Advances in StrategicManagement. JAI Press, Greenwich, CT, pp. 149–165.

Walker, G., B. Kogut, and W. Shan (1997). “Socialcapital, structural holes and the formation of anindustry network.” Organization Science, 8,pp. 109–125.

Wasserman, S. and K. Faust (1994).Social Network


Analysis: Methods and Applications. CambridgeUniversity Press, New York.

Watts, D. and S. Strogatz (1998). “Collective dynamicsof ‘small-world’ networks,” Science, 393, pp. 440–442.

White, H. (1963).An Anatomy of Kinship. Prentice-Hall, Englewood Cliffs, NJ.

White, H. (1992). Identity and Control. A StructuralTheory of Social Action. Princeton University Press,Princeton, NJ.

Winter, S. (1987). “Knowledge and competence asstrategic assets.” In D. Teece (ed.),The CompetitiveChallenge – Strategies for Industrial Innovation andRenewal. Ballinger, Cambridge, MA.

Zaheer, A. and S. Zaheer (1997). “Catching the wave:Alertness, responsiveness, and market influence inglobal electronic networks,”Management Science,43, pp. 1493–1509.

Zucker, L. and M. Darby (1995). “Present at therevolution: Transformation of technical identity fora large incumbent pharmaceutical firm after thebiotechnology breakthrough,” NBER WorkingPaper 5243.

Documents

THE NETWORK AS KNOWLEDGE: GENERATIVE RULES AND THE