GOVERNING COLLABORATIVE ACTIVITY: INTERDEPENDENCE …faculty.insead.edu/vikas-aggarwal/documents/[2]...

Strategic Management JournalStrat. Mgmt. J., 32: 705–730 (2011)

Published online EarlyView in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/smj.900

Received 30 October 2008; Final revision received 20 October 2010

GOVERNING COLLABORATIVE ACTIVITY:INTERDEPENDENCE AND THE IMPACTOF COORDINATION AND EXPLORATION

VIKAS A. AGGARWAL,1* NICOLAJ SIGGELKOW,2 and HARBIR SINGH2

1 INSEAD, Fontainebleau, France2 Wharton School, University of Pennsylvania, Philadelphia, Pennsylvania, U.S.A.

We examine the performance implications of selecting alternate modes of governance in interor-ganizational alliance relationships. While managers can choose from a range of modes to governalliances, prior empirical evidence offers limited guidance on the performance impact of thischoice. We use an agent-based simulation of interfirm decision making to complement empiricalstudies in this area. Our results point to a complex interplay between interdependencies, gov-ernance structures, and firms’ search capabilities. Different patterns of interdependence createvarying needs with respect to coordination and exploration, while at the same time different gov-ernance modes, coupled with organizational search capabilities, supply varying degrees of thesefactors. Firm performance in an alliance relationship improves when the needs and supplies ofcoordination and exploration are matched. We find situations in which stronger organizationalsearch capabilities can backfire, leading to lower exploration within the alliance relationship,and hence to lower firm performance. Moreover, we show that for higher levels of interdepen-dence, coordination can become more critical for firm performance than exploration: unless it istied to coordination, exploration can be ineffective in alliance settings. Copyright 2010 JohnWiley & Sons, Ltd.

INTRODUCTION

Interfirm collaborative relationships are a growingphenomenon with significant organizational andperformance consequences (Reuer, 2004). Engag-ing in such relationships requires determining howto govern the shared activities—a particularlyimportant consideration when the activities associ-ated with the collaboration interact with the otheractivities of the participating firms. Broader firm-level consequences can often result from gov-ernance mode selection. In resource-constrainedenvironments, for example, firms may be unable

Keywords: alliances; governance modes; coordination;exploration; interdependence; simulation∗ Correspondence to: Vikas A. Aggarwal, INSEAD, Boulevardde Constance, 77305 Fontainebleau Cedex, France.E-mail: [email protected]

to invest in the capabilities necessary to pursuedifferent modes of governance across multiplealliance relationships. As a result, developing abusiness model around a single, particular modecan become a necessary strategic choice.

While governance mode choice can have strate-gic implications, there is limited direct evidencefor the impact of different modes on firm per-formance. Prior work in this area has generallyaddressed two issues. First, does cooperative activ-ity matter? And second, what factors compel firmsto select alternate modes of governance in cooper-ative settings? Empirical evidence for the impli-cations of cooperative activity has been mixed:while some work documents the positive effectsof such activity on firm performance using metricssuch as stock market returns (Das, Sen, and Sen-gupta, 1998) and patenting output (Shan, Walker,

Copyright 2010 John Wiley & Sons, Ltd.

706 V. A. Aggarwal, N. Siggelkow, and H. Singh

and Kogut, 1994), other studies suggest that coop-erative activity plays a more limited role in influ-encing firm-level outcomes relative to factors suchas internal capabilities (Lee, Lee, and Pennings,2001). More robust evidence exists with respectto the antecedents of such activity. Two streamsof prior work have addressed the determinants ofalliance governance: a transaction cost economics(TCE) view, which generally concludes that morehierarchical forms of governance are associatedwith transactions characterized by increased appro-priation hazards (e.g., Pisano, 1989; Oxley, 1997);and an organizational approach, which examinesthe choice between alternate modes of gover-nance taking into account firm-level considerations(e.g., Gulati and Singh, 1998; Kale and Puranam,2004; Dyer, Kale, and Singh, 2004; Villalonga andMcGahan, 2005; Aggarwal and Hsu, 2009).

Although ample work has assessed alliance gov-ernance mode determinants, less is known aboutthe performance implications of different modes.Sampson (2004) makes important headway towardaddressing this by focusing on the implicationsof ‘governance misalignment.’ Using TCE as atheoretical prior, this study focuses on the inno-vation implications of governance choices that runcounter to TCE predictions. The empirical findingthat misaligned choices can have adverse inno-vation effects is compelling, as it motivates thenotion that mode choice has performance effectsdetermined in part by the nature of interfirm char-acteristics. Moreover, it suggests the need for adeeper understanding of the mechanisms influ-encing firm performance as governance modesvary.

Our central research question addresses thislink between governance mode and performance:how does the governance mode used to managedecisions in the context of an interorganizationalrelationship between two firms impact firm perfor-mance? To examine this question, we employ anagent-based simulation that enables us to developa more nuanced view of the implications of gov-ernance choice. Part of the motivation for thisapproach is to better understand some of the factorsunderlying the implications of cooperative activ-ity that may be more difficult to address usingempirical methods. Prior studies exploring firmperformance in alliance settings, for example, haveexplored moderating effects such as alliance activ-ity type (e.g., marketing vs. technical) and relative

firm size, while abstracting away from the particu-lar implications of firm characteristics and gover-nance mode. Omitting such characteristics may bethe reason for the mixed results around the impli-cations of cooperative activity noted above. By uti-lizing a simulation methodology, we are affordeda degree of flexibility in experimental design notpossible with alternative approaches.

We use a simulation model with the aim of gen-erating a novel set of insights (following from aset of assumptions) that can guide future theoreti-cal and empirical work. To develop our model, wedraw from a rich body of literature that has usedagent-based simulations to address issues of orga-nizational strategy (e.g., Levinthal, 1997) and fromprior work examining interorganizational relation-ships. We model a range of governance structures,different patterns of interdependencies, and vary-ing levels of organizational capabilities. To anchorour analytical explorations we first outline a con-ceptual framework of firm performance that buildson the information processing and contingency the-ories of the firm. With this framework in mindwe then use the simulation to generate furtherinsight into the mechanisms that underly the per-formance effects of alternate governance modes.In the final section, we discuss theoretical impli-cations and outline several specific insights andhypotheses that emerge from this study.1

THEORY AND LITERATURE

Our aim in this section is to develop a conceptualframework that describes a number of mechanismsunderlying firm performance in alliance settings.We begin by discussing governance mode choiceand interfirm interdependence; we then turn tothe factors of coordination and exploration. Wemotivate our framework with a discussion of priorliterature that includes work on organization designand information processing. The framework, inturn, motivates the simulation model discussed inthe subsequent section.

1 Many of the theoretical mechanisms we discuss might also berelevant in intrafirm settings, and our theoretical developmentand discussion moreover draw on a broad range of the organiza-tional design literature examining within-firm issues. Since ourfocus is on alliance governance, however, we construct a simu-lation model that more closely mirrors situations involving twodistinct and interacting firms. As a result, we make assumptionsaround the nature of organizational interdependence and decisionmaking that may be less appropriate for an intrafirm setting.

Copyright 2010 John Wiley & Sons, Ltd. Strat. Mgmt. J., 32: 705–730 (2011)DOI: 10.1002/smj

Governing Collaborative Activity 707

Governance modes and interdependence

An important question in any interorganizationalrelationship is how to govern the shared activities.An early stream of work that examines the gov-ernance choice issue drew from transaction costtheory (Williamson, 1975; Klein, Crawford, andAlchian, 1978). In this context, appropriation con-cerns arising from contractual hazards become acentral consideration. Gulati and Singh (1998) pickup on this thread and discuss appropriation con-cerns as an important determinant of governancestructure, while at the same time suggesting thatcoordination costs play an equally important role.They draw from the organization design literature(e.g., Thompson, 1967; Galbraith, 1977) to suggestthat interdependence is a particularly importantfactor in determining how firms structure alliancerelationships. Interdependence in an alliance set-ting can encompass factors such as sharing com-plementary technologies and production facilities,as well as joint product development (these areamong the activities Gulati and Singh [1998] dis-cuss in measuring interdependence).

The importance of interdependence, both amongactivities in which firms are jointly engaged andbetween the firms’ alliance and nonalliance activ-ities, stems from the information processing needsof the interacting firms. In his discussion of intraor-ganizational design, Galbraith (1977: 40) notesthat ‘in order to coordinate interdependent roles,organizations have invented mechanisms for col-lecting information, deciding, and disseminatinginformation to resolve conflicts and guide inter-dependent actions.’ Information plays a similarrole in alliance settings: just as greater levelsof internal interdependence lead to higher taskuncertainty and increased information process-ing needs within an organization (Tushman andNadler, 1978), greater levels of interdependencewithin and between the activities of firms inan alliance lead to higher alliance-related infor-mation processing needs. As a result, alliancedesign choices are driven by the need to captureand effectively manage interdependence-relatedinformation.

Bounded rationality (Simon, 1945; Cyert andMarch, 1963) in many ways underpins the infor-mation processing perspective. If actors were un-boundedly rational, a single decision maker couldsimply optimize across all combinatorial possi-bilities; yet with limits on individual information

processing ability, a more decentralized approachwill likely prevail (Mintzberg 1979). In the con-text of alliance relationships, therefore, firms arelikely to (at least partially) decentralize decisionmaking in order to deal with information flowsarising from interfirm interdependence. As priorwork examining within-firm organizational struc-tures has demonstrated, information flows are acore consideration in optimally configuring organi-zations, particularly with varying levels of decom-posability (Burton and Obel, 1980; Simon, 1996).More generally, just as environmental contingen-cies drive different choices of within-firm orga-nization (e.g., Lawrence and Lorsch, 1967) wewould expect that no single alliance governancechoice would apply equally effectively acrossall circumstances. Varying information processingrequirements associated with different patterns ofinterdependence between firms will likely neces-sitate employing alternative structures to governthe associated decisions. This discussion thereforesuggests:

Proposition 1. The pattern of activity interdepen-dence between firms in an alliance creates per-formance differences among different alliancegovernance modes.

Coordination, exploration and firmperformance

To understand how interdependence can influencethe link between governance choice and perfor-mance, we seek to better understand the leversthat firms employ to influence performance inalliance settings. Coordination is one such lever:the nature and functioning of coordination mecha-nisms and the associated failures to appropriatelycoordinate activities have been a central set ofconcerns for both the classic and the more recentorganization design literature (e.g., Simon, 1945;Galbraith, 1977, Burton and Obel, 1984; Rivkinand Siggelkow, 2003). While many of these stud-ies have focused on intrafirm interactions, workgoing back to (at least) Schelling (1960) has dis-cussed issues of coordination across multiple orga-nizations. In the particular context of alliancerelationships, Gulati and Singh (1998) suggestthat firms take coordination concerns into accountwhen making governance mode decisions. More-over, a recent stream of alliance research lookingat firm adaptation and the development of routines



over time suggests that the ability to effectivelycoordinate activities among firms can be an impor-tant driver of firm performance (e.g., Zollo, Reuer,and Singh 2002, Gulati, Lawrence, and Puranam2005).

In addition to the ability to coordinate amongactivities, a firm’s ability to explore and find newactivities can be an important driver of perfor-mance in an alliance. Exploration is a core elementof the behavioral theory of the firm (Cyert andMarch, 1963), with search processes critical toa firm’s ability to adapt and evolve (Nelson andWinter, 1982; March, 1991; Levinthal, 1997). AsMarch (1991: 71) suggests, exploration, a conceptcaptured by ideas such as ‘search, variation, risktaking, experimentation, play, flexibility, discov-ery, innovation,’ is critical for organizations. More-over, search can be exploratory to varying degrees,from narrow, local search to broader search. Thedrivers of exploration have been addressed in agrowing body of work, with factors such as organi-zational structure shown to play an important role(e.g., Siggelkow and Rivkin, 2005, 2006 and refer-ences therein). We build on this work in the presentstudy, while at the same time shifting the focus toexploration in an interorganizational setting.2

The prior literature suggests, therefore, thatcoordination and exploration have important ef-fects on firm performance. Moreover, the degree ofcoordination and exploration that will be requiredin an alliance setting is, in turn, likely to be influ-enced by the pattern of interfirm interdependence:greater interdependence increases the challenges ofcoordination (e.g., Galbraith, 1977) and requiresa higher degree of exploration because interde-pendencies create more rugged performance land-scapes that can cause firms to get stuck with verysuboptimal choices (e.g., Levinthal, 1997). Thisdiscussion therefore suggests:

Proposition 2a. Different patterns of activityinterdependence between firms create different

2 Our definition of ‘exploration’ refers to organizational actionsrelated to finding new activity configurations. Our use of theterm exploration is thus consistent with what is more gener-ally termed ‘search’ in prior literature (see, e.g., Siggelkow andRivkin’s (2006) discussion of exploration). March’s (1991) fur-ther distinction between exploitation and exploration is implicitlyembedded in our model (see below), as we vary the breadth ofsearch undertaken by managers from narrow (akin to ‘exploita-tion’) to broad (akin to ‘exploration’ in March’s more particularsense).

needs for coordination and exploration in analliance relationship.

While interdependence patterns create varyingneeds for coordination and exploration, the supplyof these factors is influenced both by the alliancegovernance structure and by organizational capa-bilities. More specifically, in alliance relationships,a primary consideration is the degree of autonomyheld by the individual firms. Two core consider-ations in such a setting are (1) the informationflows enabling individual actors to make deci-sions, and (2) the overarching processes governingthese decisions. We refer to the collective set offactors that characterize the underlying informa-tion flow and decision processes as the alliancegovernance structure, and suggest that the partic-ular structure employed influences the supply ofcoordination and exploration in interorganizationalrelationships. We discuss these structures in moredetail in the following section.

Beyond the particular governance structure em-ployed, internal firm capabilities are also likelyto play an important role. A broad stream of theorganizational strategy literature has focused onthe role of capabilities in the formation, gover-nance, and performance of collaborative relation-ships (e.g., Kale, Dyer, and Singh, 2002; Zolloet al., 2002; Colombo, 2003; Leiblein and Miller,2003; Mayer and Salomon, 2006). In this paper,we conceptualize capabilities as a set of factorsthat enable an organization to engage, to varyingdegrees, in the search for better configurations ofperformance-relevant activities. We use the term‘organizational search capabilities’ to denote thesefactors in our conceptual framework and simula-tion model.3 Such capabilities represent the endproduct of investments by firms in their ability togenerate new ideas and evaluate alternative coursesof action. As a result, these capabilities may bestatic in the context of an individual alliance rela-tionship, but adjustable over time as firms altertheir investments in decision-evaluation resources.This discussion thus suggests that as firms seek torespond to the coordination and exploration needsarising from interfirm interdependence, they willdraw upon the joint effects of governance modes

3 We defer a more detailed discussion of these capabilities tothe Simulation Model section, where we describe the particulartypes of such capabilities that we model in this paper.



and organizational search capabilities. As a result,we have:

Proposition 2b. Different governance structuresand organizational search capabilities supplydifferent degrees of coordination and explo-ration in an alliance relationship.

Finally, as much of the organizations literature(e.g., Lawrence and Lorsch, 1967; Galbraith, 1977;Burton and Obel, 1995; Volberda, 1996) collec-tively indicates, higher performance stems fromsituations in which firms are adequately equippedto deal with the demands arising from their par-ticular form of interdependence. As our discussionhas suggested, in the case of alliance relationships,coordination and exploration are core considera-tions. Moreover, as Tushman and Nadler (1978:619) note, ‘too much capacity will be redundantand costly; too little capacity will not get the jobdone.’ Thus, firms in an alliance will be morelikely to achieve higher performance when theyare able to strike a balance between the needs andsupply of coordination and exploration. This dis-cussion thus suggests:

Proposition 3. Firm performance in an alliancerelationship improves when coordination andexploration needs are matched with the degreeof coordination and exploration supplied.

The conceptual framework illustrated inFigure 1 summarizes our discussion thus far: thedemands for coordination and exploration areinherently influenced by the pattern of interde-pendence among the activities in which the firms

involved in the interorganizational relationship areengaged. As interdependence increases, so too willthe need to coordinate actions among firms; sim-ilarly, a higher degree of interdependence willcreate greater landscape complexity, necessitatinghigher levels of exploration (e.g., Levinthal, 1997).While interdependence has implications for coordi-nation and exploration demands, governance modeand organizational search capabilities have sup-ply implications for these factors: mode choiceaffects the nature and extent of coordination amongfirms, as well as the ability of firms to exploretheir environment. Likewise, organizational searchcapabilities affect the supply of exploration and,in the interplay with the chosen governance mode,coordination. In the next section, we describe asimulation model that enables us to explore thisframework in greater detail.

SIMULATION MODEL

To develop our simulation model, we build on andextend prior work that has used the NK framework(Kauffman, 1993) to model firm decision making(e.g., Levinthal, 1997; Lenox, Rockart, and Lewin,2006). This literature generally envisions firms assets of interdependent choices, an approach charac-teristic of an activity systems view of organizations(Porter, 1996; Siggelkow, 2001, 2002). In partic-ular, a firm is seen as having to resolve a setof N ‘policy choices.’ For instance, a firm mayhave to decide whether to increase its product vari-ety, whether to engage in various marketing cam-paigns, or whether to increase its budget for salesforce training. The NK model further assumes that

Explorationsupplied

Explorationneeds

Coordinationsupplied

Coordinationneeds

Interdependencepattern

Governance mode andorganizational

search capabilities

Firmperformance

Figure 1. Conceptual model: governance mode, capabilities and interdependence



the benefit that is derived from each policy choiceis affected by K other policy choices. For instance,the value that an increase in product variety maygenerate could depend on whether the sales forcehas been recently trained, or not. In sum, eachunique configuration of the N choices generates aparticular performance value. Thus, we can imag-ine a performance landscape on N + 1 dimensions:N ‘horizontal’ dimensions representing the spaceof all possible alternatives for each of the N policychoices, and one ‘vertical’ dimension representingthe performance level resulting from each overallchoice configuration (of the N decisions). Land-scape complexity depends on interactions amongthe policy choices, and firms search the landscapeattempting to move to higher performance levelsby altering individual choices.

The two principal components of NK modelsare thus (1) a mechanism to create performancelandscapes (i.e., the mapping from choices to per-formance), and (2) a set of decision rules thatdescribe how firms search the landscape (i.e., howfirms generate and assess alternative choice config-urations). In describing our model, we focus firston extending the concepts of interdependence andperformance landscapes to a two-firm setting. Wethen discuss our model of search and governancein this context.

Patterns of interdependence

We model two firms interacting within the con-text of an alliance relationship. Firms 1 and 2 areeach composed of a set of binary policy choices,which we denote as F1 and F2. We decompose thepolicy choice sets of each firm into two groups:(1) non-alliances choices under the control of therespective firm and (2) choices that are part ofthe alliance relationship. For Firm 1, we thushave F1 = {N1, A1}, where N1 represents the setof non-alliance choices, and A1 represents the setof alliance-dedicated activities in which Firm 1is engaged. For example, a research and devel-opment (R&D) alliance between two technologyfirms might consist of joint work toward a partic-ular product. Choices with respect to this R&Dalliance that Firm 1 has to resolve, for exam-ple, whether or not to dedicate a particular engi-neer to this alliance, would be included in A1.Other choices that do not fall into the provinceof the alliance such as how much to invest inbranding an existing product would be included

in N1. We denote the number of policy choicesin sets N1 and A1 with n1 and a1, and defineanalogous values for Firm 2. The total number ofpolicy choices in the system of two firms is thusN ≡ n1 + a1 + a2 + n2.

Having defined the policy choices for each firm,we can define the pattern of interactions amongthem. Figure 2a illustrates an interaction matrixfor two firms with n1 = 4, a1 = 2, a2 = 2, andn2 = 4. The matrix specifies which policy choicesare affected by which other choices. An X inrow i, column j , signifies that the resolution ofthe j th policy choice affects the value of the i th

policy choice. Using this matrix, we can specifyexactly which policy choices affect which otherpolicy choices. For instance, in the example givenin Figure 2a, d1 is affected by d1 through d8.Likewise, d9 is affected by d5 through d12. For easeof notation, we denote individual policy decisionsby di , where i is indexed from 1 to N , andsequentially number the decisions in the sets N1,A1, A2, and N2. Thus for Figure 2a, we haveN1 = (d1, d2, d3, d4), A1 = (d5, d6), A2 = (d7, d8),and N2 = (d9, d10, d11, d12). We further define thefull vector of decisions in the entire system of thetwo firms as d ≡ (d1, d2, . . . , dN). Since decisionsare binary, d is thus a string of N 0’s and 1’s.

To analyze the role played by different pat-terns of interdependence on optimal governancemode choice, we construct a set of specific inter-dependence patterns. These patterns follow a log-ical sequence, increasing in overall complexity,with the differences between any two consecutivepatterns arising from the addition of a particulartype of interdependence. Figure 2b shows the pat-terns we use, with shaded areas representing thepresence of interactions. (Each shaded ‘box’ inFigure 2b is completely filled with X’s.) We modela system of 12 policy choices with n1 = 4, a1 = 2,a2 = 2, and n2 = 4. Pattern 1 is ‘fully decompos-able,’ that is, it has interactions occurring solelywithin each group of policy choices. For instance,choices in N1 only affect other choices in N1;similarly for choices in A1, A2, and N2. Pattern2, ‘pure alliance interaction,’ introduces interac-tions among the alliance choices of the two firms,that is, between A1 and A2. Thus, in this patternall the activities within the alliance affect eachother, but none of the alliance activities interactwith any of the non-alliance activities. Pattern 3,‘firm-alliance interaction,’ introduces interactionsamong all the alliance and non-alliance choices



A. Interaction matrix example

N1 A1 A2 N2

d1 d2 d3 d4 d5 d6 d7 d8 d9 d10 d11 d12

N1

A1

A2

N2

d1

d2

d3

d5

d6

d7

d8

d9

d10

d11

d12

d4Firm

1Fi

rm 2

Alli

ance

Firm 1 Firm 2

Alliance

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

B. Patterns of interdependence

Pattern 1: Pattern 2:

‘Fully decomposable’ ‘Pure alliance interaction’

N1 A1 A2 N2 N1 A1 A2 N2

N1 N1

A1 A1

A2 A2

N2 N2

Pattern 3: Pattern 4:

‘Firm-alliance interaction’ ‘Full interdependence’

N1 A1 A2 N2 N1 A1 A2 N2

N1 N1

A1 A1

A2 A2

N2 N2

Figure 2. Interfirm interdependence

within each firm. Thus, in this case, choices withinA1 and A2 interact with choices within N1 andN2. All activities within the alliance continue tointeract with each other. (The pattern depicted inFigure 2a corresponds to Pattern 3.) Lastly, in Pat-tern 4, ‘full interdependence,’ all activities, eventhe non-alliance activities of the two firms, interactwith each other.4

4 While this set of patterns is not exhaustive of the full set ofpossible interdependencies, they are sufficient for our purpose in

Performance landscapes

We turn next to the mechanism for assessingperformance in our model. Prior literature (e.g.,Levinthal, 1997; Rivkin and Siggelkow, 2003)has used the ‘contributions’ of individual policychoices to describe performance in an NK setting.We follow a similar approach. For each policy

that they enable us to create varying types of coordination needsthat we can then systematically examine.



choice di in our system, we construct a contri-bution value function, Ci . The contribution valuefunction takes as arguments each of the ki + 1choices affecting di (this includes di itself, as wellas the other ki choices affecting di as defined by theinteraction matrix). Each contribution value func-tion Ci is constructed as follows: for each uniqueconfiguration of policy choice di and the ki otherchoices affecting it, we draw a random ‘contri-bution value’ from the uniform distribution [0,1].5

Once all possible contribution values for each indi-vidual choice are defined, the performance for anyparticular set of policy choices is computed asthe sum of the contribution values associated withthese choices.

To reduce statistical artifacts that could arisefrom the stochastic nature of the landscape gen-eration process, we normalize performance by thehighest value possible on any given landscape.Let d ∗ be the policy choice configuration thatleads to the highest performance �∗, that is, �∗ =∑12

i=1 Ci(d ∗). Then, for instance, if the performanceof Firm 1 is determined only by the contributionsof the first six policy choices, the performance ofFirm 1 for any choice configuration d is given by∑6

i=1 Ci(d)/

�∗. In the case of an alliance betweenFirm 1 and Firm 2 involving policies 4 through 8,Firm 1’s performance would be determined by thecontributions of both Firm 1’s non-alliance policychoices (d1 through d4) and by the policy choicesof the alliance (d5 through d8). For instance, if Firm1 obtains portion α of the alliance performance,Firm 1’s overall performance would be computedas

(∑4i=1 Ci(d) + α

∑8i=5 Ci(d)

) /�∗.

Agents and search capabilities

We turn now to the set of decision-making rulesthat govern agents’ behavior in our simulation.We define an agent as a decision maker havingauthority over some subset of policy choices in thetwo-firm system. For instance, a particular agentmight have authority (and care about the perfor-mance of) all of the non-alliance activities of Firm1. One should note that we use the term ‘agent’ for

5 As an example, if the interaction matrix has eight X’s inrow i (including the X in column i), then the contributionvalue function Ci can take on 28 = 256 possible values for Ci

depending on how the vector d is configured (i.e., Ci(d) willdepend only on the values of the eight choices noted by X’s inthe interaction matrix).

expositional simplicity. An ‘agent’ in our modelneed not be (and in practice rarely would be) asingle person. An agent in our model is the rele-vant ‘decision-making body’ that is responsible formaking decisions concerning a set of activities, andthus could be, for instance, a steering committee,or a set of managers.

For a given simulation run, we begin by plac-ing the agents at a random point on the landscape(i.e., assigning random starting values to their pol-icy choices). In each subsequent period, the agentsdecide whether to alter the policy choices undertheir control (i.e., to change a policy from a 0to a 1, or vice versa). This is done by evaluat-ing a set of possibilities, which is influenced bythe organizational search capabilities, and select-ing the best option from this set. The determina-tion of what constitutes a ‘best option’ is, as wedescribe below, dependent on the particular gover-nance structure. Before discussing the governancestructures in more detail, we describe our con-ceptualization of organizational search capabilities,which we model as the ability of agents to generateand evaluate alternative configurations for the setof policy choices they control. This set of param-eters can be thought of as the degree to whichorganizational decision-making units are endowedwith resources necessary to make decisions in analliance context.

We model two dimensions of such capabilities:(1) the ability to make simultaneous decisions overa larger vs. smaller number of the policy choicescontrolled, which we term the ‘search radius’ asper prior literature (e.g., Siggelkow and Rivkin,2005); and (2) the ability to evaluate a larger vs.smaller number of alternatives in a given period.

For the search radius (parameter SR), we modela base case in which agents can change only asingle policy choice in each period (SR = 1), aswell as a more complex case in which agents canchange up to two policy choices simultaneouslyin each period (SR = 2). For the number ofalternatives (parameter ALT), we model a casewhere agents only evaluate a single, randomlychosen alternative in each period (ALT = 1), aswell as a more complex case where all alternativesthat lie within the search radius are evaluated (ALT= max). For example, assume an agent controlsthree policies that are currently configured as 000.If the agent’s search radius is 1, and she evaluatesall alternatives, she considers 100, 010, and 001.If she evaluates only a single alternative, she will



randomly pick one of these, for example, 010. Ifher search radius is 2, she can also consider optionsthat are different from the current configuration intwo policies such as 110. In this case, were she toevaluate all alternatives in her search radius, shewould evaluate 3

2C + 31C = 6 alternatives. In sum,

we model four different levels of organizationalcapabilities: Level A: SR = 1 & ALT = 1; LevelB: SR = 1 & ALT = max; Level C: SR = 2 &ALT = 1; Level D: SR = 2 & ALT = max.

Governance modes

We turn next to the different governance modes,which determine the specifics of the decision-making process. We construct these modes alongfour dimensions: (1) number of decision makers(agents), (2) order of decision making, (3) metricsused to evaluate the implications of choices, and(4) nature of oversight and hierarchy around thedecision-making process. These dimensions rep-resent key elements of organization and alliancedesign (Simon, 1945; Galbraith, 1977; Reuer,2004), enabling us to create a range of governancemodes that differ in the underlying informationflow within and across organizations. We discussthe motivation underlying each dimension and theresulting governance modes in the remainder ofthis section.

The number of decision makers in the system isan important dimension of variance in structuringalliance governance. Since we are modeling a two-firm system, having two separate decision makers,one for each firm, is a natural baseline. Alterna-tively, a more integrated level of governance mightentail having a single decision maker, while a sit-uation in which there are two decision makers forthe firms, along with a third decision maker for thealliance function, would be a natural characteristicof hybrid governance forms. A three-agent modelmight thus have a decision-making body specificto alliance policy choices (e.g., a joint committeefrom both firms managing the alliance activities).As a second dimension of variance, we considerthe question of how decision making is ordered,that is, who gets priority in the decision-makingprocess. The order of decision making has impli-cations for information flow within organizationsand, as discussed in depth in the previous section,such issues are likely to be central to the gover-nance of interorganizational relationships.

The third dimension of variance relates to themetrics used by individual agents to make deci-sions. Such decision metrics can be thought of asthe incentives for individual organizational units,which derive from the particular contractual fea-tures of the alliance (e.g., Robinson and Stuart,2007). We model decision metrics (which aredefined solely by the governance mode) as the setof contribution values an agent takes into accountin making choices. As an example, in one particu-lar governance mode an agent will consider just thecontribution values of the decisions over which shehas control. In an alternate governance mode, theagent will consider not only the contribution val-ues of her own decisions, but also some portion ofthe contribution values of other agents’ decisions.6

Finally, the fourth dimension of variance weconsider is the nature of hierarchy among agents inthe model. Alliances can incorporate elements ofhierarchical governance (e.g., Hennart 1993) suchthat decisions taken by one agent need to be ratifiedby another before they can be implemented. Suchhierarchy can, like the decision metrics, derivefrom the particular contractual arrangements ofthe alliance. It is different, however, in that itdescribes the overall organization of decision mak-ing within the organization, rather than the morenarrowly prescribed underlying incentives for indi-vidual agents.

Having described the four dimensions alongwhich we model the governance of a particu-lar relationship, we now describe the governancestructures we use in our analysis (Figure 3). Wemodel four structures, which we term ‘modular,’‘self-governing alliance,’ ‘ratified alliance,’ and‘integrated.’ The modular and integrated forms lieat opposite ends of the governance spectrum. Withthe modular form, there are two agents, each con-trolling the policy choices of either Firm 1 orFirm 2 (thus, in our model, six policy choicesfor each agent). These agents make choices simul-taneously, and only consider the profits for their

6 Defining agents’ decision metrics in this way implies thatwe abstract away from a direct analysis of factors such asshirking that may result from, for example, under-enforcementof contracts. Although agents do act opportunistically in thattheir profit calculations consider only their own predefined set ofcontribution values to the possible detriment of the other agentsand the system as a whole, we do not directly model situationswhere issues of trust are at play, or where agents directly seekto mislead. This is consistent with the notion of self-interest in,for example Klein et al. (1978), but may not include the ‘guile’component of Williamson’s (1975) definition of opportunism,where there are hazards associated with the contract itself.



Governancestructure

Agents andpoliciescontrolled

Decision-makingsequence

Metrics for evaluatingdecisions

Oversight andhierarchy

Modular

(MOD)

Two agents

Control F1 andF2, respectively

Firms 1 and 2(simultaneously)

Own-firm profit,F1 and F2

None

Self-governingalliance

(SGA)

Three agents

Control N1,A1-A2, and N2,respectively

1. Alliance

2. Firms 1 and 2(simultaneouslywith respect toN1 and N2)

Alliance profit for A

Own-firm profit plusportion (α) of allianceprofit for F1and F2

None

Ratifiedalliance

(RAT)

Three agents

Control N1,A1-A2, and N2,respectively

1. Firms 1 and 2(simultaneouslywith respect toN1 and N2)

2. Alliance

3. Firms 1 and 2ratify allianceproposals

Alliance profit for A

Own-firm profit plusportion (α) of allianceprofit for F1 and F2

F1 and F2 mustratify allianceproposals

Integrated

(INT)

One agentControls all policychoices

Single decisionmaker

Own-firm profit,combining F1+F2

Implicit fulloversight

Figure 3. Summary of governance structures

own firm in making decisions,∑6

i=1 Ci(d)/

�∗

and∑12

i=7 Ci(d)/

�∗, respectively. This is a simplecase of an arms-length interorganizational relation-ship with no formal decision-making structure togovern joint activities. At the other end of the spec-trum, we have the integrated mode, where a singleagent manages choices for both Firms 1 and 2(thus, 12 total policy choices in our model). In thiscase, the agent takes into account the total com-bined profit of Firms 1 and 2,

∑12i=1 Ci(d)

/�∗,

and the individual firms thus operate as a quasi-integrated entity.

In addition, we model two hybrid alliance forms,both of which have an alliance manager who man-ages the alliance choices (the four choices in thesets A1 and A2 in our model), while agents forFirms 1 and 2 manage the non-alliance choicesof the two firms (N1 and N2 respectively). Inboth cases, the alliance function takes into accountonly its own profits

∑8i=5 Ci(d)

/�∗ when mak-

ing decisions. Each individual firm, in turn, takesinto account the profits associated with its ownchoices, along with a portion (α) of the allianceprofit (we set α to 1

/2 in our analysis), that

is,(∑4

i=1 Ci(d) + 12

∑8i=5 Ci(d)

) /�∗ for Firm 1

and(∑12

i=9 Ci(d) + 12

∑8i=5 Ci(d)

) /�∗ for Firm

2. The main difference between the two alliancemodes lies in the degree of importance placedon the agendas of the individual firms relative tothe agenda of the alliance. In the self-governingalliance mode, the alliance agent is allowed tomove first, and the agents for Firms 1 and 2 movesecond, knowing the alliance’s move in the currentperiod. Thus, in this governance mode, the agendaof the alliance is given a high priority.

By contrast, in the ratified alliance mode, theagents for Firms 1 and 2 move first, followed bythe alliance agent who knows the choices made bythe two firms in the current period. The allianceagent, however, cannot directly implement changesin policy choices. Rather, the agent ranks theavailable alternatives and sends this ranked set tothe two firm agents, who must ‘ratify’ the allianceagent’s proposals. Ratification requires that bothfirm agents accept an alliance proposal; a firmagent accepts a proposal if the proposal does notreduce profit for the agent (conditional on the movethe agent already made at the start of the period). Inthis case, the agendas of Firms 1 and 2 come first,and are accounted for as the alliance proposals areconsidered.



RESULTS AND ANALYSIS

We begin this section by explaining our ana-lytic approach and presenting a number of initialresults. We then focus on interdependence, gov-ernance structure and capabilities in more detail,outlining how these factors impact the demandand supply of coordination and exploration. Inthis context, we develop a set of metrics forcoordination and exploration that allow us tomore precisely show the underlying mechanismsat work. With these metrics, we then discuss themechanisms underlying the relationship betweenappropriate governance choice and patterns ofinterdependence.

Analytic approach and motivating results

We conducted a broad range of simulations aimedat understanding the implications of joint varia-tion in interdependence, exploration ability, andgovernance mode. As noted previously, we modela 12-policy choice system of two firms in whichn1 = 4, a1 = 2, a2 = 2, and n2 = 4. Since the sys-tem itself, along with the factors we model (likethe patterns of interdependence), is symmetric forboth firms, the performance results for the twofirms in the system are equal when run over alarge number of landscapes. To facilitate compari-son with the integrated structure (which comprisesall 12 policy choices), we report values for the fullsystem. Performances of individual firms are sim-ply one-half of the values reported for the entiresystem.

Each ‘period’ in the simulation consists of theagents making a set of decisions with respect totheir activities as per the governance mode underwhich they are operating (see Figure 3).7 We runthe simulation for 200 periods on a given landscapein order to observe the long-run performance of thefirms in the system. To ensure that our results arenot driven by statistical artifacts, we repeat eachsimulation 10,000 times. Thus any reported per-formance value for a particular time period (e.g.,

7 For example, with the ratified alliance mode, the followinghappens within a single period: the agents for Firms 1 and 2simultaneously make a decision with respect to their activitysets N1 and N2; the alliance then evaluates alternatives for theiractivity set A1-A2 (taking into account the new choices withinN1 and N2) and sends a ranked set of the alternatives to theagents for Firms 1 and 2. Finally, the Firm 1 and 2 agents decidewhether to ratify the choice made by the alliance.

period 200) on a particular pattern of interdepen-dence is an average over 10,000 simulation runs(i.e., landscapes). In general, except where explic-itly noted otherwise, reported performance differ-ences are statistically significant at the one percent(or lower) confidence level.

We begin the analysis by examining the perfor-mance outcomes of the various governance formsunder Pattern 1, the fully decomposable pattern.Figure 4 illustrates the performance of the fourgovernance modes on this interdependence patternfor the four different levels of organizational capa-bility. For capability level A (narrow search radiusand one alternative evaluated in each period), inthe short run, the integrated mode has the lowestlevel of performance, while the alliance modes per-form best. (For this pattern, the two alliance modesgenerate the same performance; hence their linesare indistinguishable in the figure.) In the long run,however, each of the governance modes convergesto the same level of performance. A broadening ofthe search radius (moving from capability level Ato level C) preserves the relative short-run differ-ences between governance modes, while enablingeach of the modes to arrive at a higher long-runlevel of performance. An increase in the numberof alternatives (moving from level A to level B, orfrom C to D) increases the speed with which thegovernance forms converge to the long-run maxi-mum, yet does not affect the level of the long-runperformance.

The Pattern 1 results are directly linked to itsdecomposable nature, that is, to the fact that thereare no interdependencies either between the indi-vidual firms and the alliance, or between the firmsthemselves. As a result, performance speed isdriven by the number of agents in the system,since each agent is able to act in parallel with andindependent of the others. The integrated struc-ture, which relies on one central agent to makeall decisions, thus has the lowest short-run per-formance, while the two alliance forms, each ofwhich has three agents who can work in paral-lel, have the highest short-run performance. Themodular form, with two agents, falls in the middle.In the long run, however, given that the system canbe fully decomposed, each mode eventually arrivesat the same performance level. Changes in organi-zational capability operate as expected: an increasein the number of alternatives increases the speedwith which each mode increases its performance,



1.00

0.95

0.90

0.85

0.80

Per

form

ance

0.75

0.70

0.650 20 40 60

MOD SGA RAT INT

80 100

Period120 140 160 180 200

1.00

0.95

0.90

0.85

0.80

Per

form

ance

0.75

0.70

0.650 20 40 60

MOD SGA RAT INT

80 100

Period

Capability Level C Capability Level D

Capability Level BCapability Level A

120 140 160 180 200

1.00

N2A2

A1

N1

N1 N2A1 A2

0.95

0.90

0.85

0.80

Per

form

ance

0.75

0.70

0.650 20 40 60

MOD SGA RAT INT

80 100

Period120 140 160 180 200

1.00

0.95

0.90

0.85

0.80

Per

form

ance

0.75

0.70

0.650 20 40 60

MOD SGA RAT INT

80 100

Period120 140 160 180 200

+ ALT+ SR

These figures graph performance (vertical axis) for the two-firm system over time periods 0 to 200 (horizontal axis) for the four governance modes on interdependence pattern 1 (fully decomposable), with varying levels of organizational capabilities. Each point on each line represents the average of 10,000 simulation runs. MOD, SGA, RAT, and INT refer to the modular, self-governing alliance, ratified alliance, and integrated forms, respectively.

Figure 4. Performance levels of four governance modes: interdependence pattern 1

while an increased search radius enables all firmsto reach higher performance levels over time.8

8 Since interdependencies exist within the set of activities eachmanager controls (e.g., within N1), each manager still may getstuck on a local peak within her ‘sub-landscape.’ An increase inthe search radius reduces the probability that managers will getstranded on such low, local peaks, and hence increases long-runperformance.

Turning next to Figure 5, we examine the per-formance of the different governance modes onPattern 4, the fully interdependent pattern, apply-ing the same varying levels of organizational capa-bilities as before. Here, the results are much morecomplex. Turning first to the case of capabilitylevel A, we see that in contrast to the Pattern1 case, performance across the four governance



1.00

0.95

0.90

0.85

0.80

Per

form

ance

0.75

0.70

0.650 20 40 60

MOD SGA RAT INT

80 100

Period120 140 160 180 200

+ ALT+ SR

1.00

0.95

0.90

0.85

0.80

Per

form

ance

0.75

0.70

0.650 20 40 60

MOD SGA RAT INT

80 100

Period120 140 160 180 200

1.00

N2

N2

A2

A2

A1

A1N1

N1

0.95

0.90

0.85

0.80

Per

form

ance

0.75

0.70

0.650 20 40 60

MOD SGA RAT INT

80 100

Period120 140 160 180 200

1.00

0.95

0.90

0.85

0.80

Per

form

ance

0.75

0.70

0.650 20 40 60

MOD SGA RAT INT

80 100

Period120 140 160 180 200

Capability Level C Capability Level D

Capability Level BCapability Level A

These figures graph performance (vertical axis) for the two-firm system over time periods 0 to 200 (horizontal axis) for the four governance modes on interdependence pattern 4 (full interdependence), with varying levels of organizational capabilities. Each point on each line represents the average of 10,000 simulation runs. MOD, SGA, RAT, and INT refer to the modular, self-governing alliance, ratified alliance, and integrated forms, respectively.

Figure 5. Performance levels of four governance modes: interdependence pattern 4

modes differs in the long run. In addition, modesvary in their relative performance over time. Theintegrated mode initially outperforms the otherthree modes; in the long run, however, it is thelowest performing mode. The modular form, bycontrast, initially underperforms, but is the best inthe long run. Likewise, as we vary organizationalcapabilities, we can observe complex changes: anincrease in either search radius or the number of

alternatives that are considered in each period isbeneficial to the integrated form in the long run.By contrast, the modular form suffers as organi-zational capabilities are increased along either ofthese dimensions. Lastly, while the self-governingalliance outperforms the ratified alliance when thenumber of considered alternatives is high (capabil-ity levels B and D), the ratified alliance has higherperformance, at least in the short run, when the



search radius is high and the number of consideredalternatives is low (capability level C).

The results of Patterns 1 and 4 illustrate severalpoints. First, we find initial support for Propo-sition 1: patterns of interdependence can have asignificant impact on the relative value of individ-ual governance modes. Second, once the pattern ofinterdependence is no longer simple, the relation-ship between governance mode and performance isnot trivial, especially when one takes into accountdifferent levels of organizational capabilities andmeasures performance at different time points. Togain a better understanding of the results, andto identify the underlying mechanisms that cre-ate this complex set of findings, it is helpful toreturn to our conceptual model of coordinationand exploration. In particular, we will describe indetail the demands that different patterns of inter-dependence create with respect to coordination andexploration and how well the different governancemodes match these demands. As we will show,the degree to which coordination and explorationdemands are matched can help explain observedperformance heterogeneity.

Coordination and exploration as underlyingmechanisms

To study the effect of interdependence patternson the performance created by different gover-nance modes, we focus on how particular pat-terns differ systematically where interdependenciesoccur: between alliance and non-alliance choices;within alliance policy choices; and between thenon-alliance choices of the two firms. These threeclasses of interdependencies create three corre-sponding dimensions of coordination dem ands:(1) coordination between the individual firms andthe alliance; (2) coordination within the allianceitself; and (3) coordination between the non-alliance activities of Firms 1 and 2. As summa-rized in Panel A of Table 1, the only coordina-tion required by Pattern 1 is among a few choiceswithin the alliance, that is, within A1 and withinA2. Pattern 2, which also contains interactionsbetween the alliance activities of the individualfirms (i.e., between A1 and A2), requires moreintra-alliance coordination. Pattern 3 requires addi-tional coordination between the activities of theindividual firms and the alliance. Finally, Pattern4 demands coordination between the non-allianceactivities for Firms 1 and 2 as well.

Patterns 1–4 differ not only in their locations ofinterdependencies but also in the number of inter-dependencies. Generally, the larger the numberof interdependencies, the more ‘rugged’ a perfor-mance landscape becomes, that is, the more localpeaks it contains (e.g., Levinthal, 1997). (A pol-icy configuration d is a local peak if all policyconfigurations that differ from d in only one pol-icy choice have lower performance.) The higherthe number of local peaks, the higher the risk ofgetting stranded on a low-performing local peak;and consequently, the higher the level of explo-ration needed to reach high performance. As aresult, the number of local peaks that are con-tained in a landscape is a good measure for explo-ration demands. The right-most column of PanelA of Table 1 contains the number of local peaksfor each interdependence pattern. As expected, thenumber of local peaks increases from 18 for Pat-tern 1 to 314 for Pattern 4. (For comparison, alandscape with 12 policy choices has 212 = 4, 096possible policy configurations.) In sum, the pat-terns differ in the types and degree of coordina-tion, as well as the degree of exploration theydemand.

We now turn our attention to the different gover-nance modes. Variation across these modes alongthe four key design dimensions (number of agents;order of decision making; decision evaluation met-rics; and oversight and hierarchy) generates sub-stantial differences in the types and degree ofcoordination. To facilitate comparison with the dif-ferent types of coordination that are demanded bythe different interdependence patterns, Panel B ofTable 1 summarizes the coordination that is sup-plied by the different governance structures alongthe three coordination dimensions (individual firm-alliance; within alliance; and between Firms 1and 2).

The modular mode supplies very little coordi-nation along any of the coordination dimensions.Non-alliance activities are completely uncoordi-nated between the two firms as decisions aremade simultaneously and independently of oneanother. Likewise, the modular mode has very lit-tle within-alliance coordination. Since own-firmalliance activities are controlled by each firm (forinstance Firm 1 controls d5 and d6), a limitedamount of coordination is created between at leasta subset of alliance activities (i.e., Firm 1 coor-dinates between d5 and d6, and Firm 2 betweend7 and d8). Similarly, only limited coordination



Table 1. Demand and supply of coordination and exploration

A. Level of coordination and exploration demanded by the interdependence pattern

Coordination demanded

Between Firms 1, 2,and alliance

Within alliance Between Firms 1, 2 Exploration demanded(number of local peaks)

Pattern 1 None Little None 18Pattern 2 None High None 35Pattern 3 High High None 198Pattern 4 High High High 314

B. Level of coordination supplied by governance modes

Between Firms 1, 2, andalliance

Withinalliance

BetweenFirms 1, 2

Sources of coordination supply

MOD Low Low None • No formal mechanism, but individualfirms control subset of allianceactivities

SGA Low High None • Sequential moves within period• Firms consider alliance performance

RAT Medium High None • Sequential moves within period• Firms consider alliance performance• Ratification stage

INT High High High • Single decision maker

C. Level of exploration supplied by governance modes

Exploration level Exploration influences

MOD Very high • Parochial decision making by firm managersSGA High • Parochial policy evaluations and implementation by allianceRAT Medium • Parochial policy evaluations by alliance

• Ratification stepINT Low • Systemic viewpoint

MOD, SGA, RAT, and INT refer to the modular, self-governing alliance, ratified alliance, and integrated governance modes,respectively.

occurs between alliance activities and the non-alliance activities of each firm. For instance, Firm1 coordinates between d1-d4 and d5 and d6, butignores the interactions between d1-d4 and d7

and d8.Within the self-governing alliance, strong coor-

dination is achieved among the alliance activ-ities since they are all controlled by a singlealliance agent. A small amount of coordinationalso occurs between the non-alliance activitiesand the alliance activities. This coordination arisesfrom the sequential moves within a given periodand from the individual firms taking into accounta portion (α) of the alliance performance in mak-ing their decisions. Thus, when firm managers

make decisions concerning the non-alliance activ-ities, they would not implement policy changesthat (drastically) undermine alliance performance.No coordination, however, is achieved between thenon-alliance activities of both firms.

The ratification alliance mode adds more coordi-nation; here, in addition to the coordination sourcesmentioned for the self-governing alliance mode,the additional ratification step adds greater coor-dination between the non-alliance policies and thealliance, since the alliance is not allowed to imple-ment choices that might be very detrimental to oneof the firms. Finally, the integrated mode, with asingle decision maker, achieves full coordinationacross all decisions.



In our simulations, to measure the amount ofcoordination that is provided by the various gov-ernance modes given a particular interdependencepattern and organizational capability level, weobserve the number of times that a coordinationfailure occurs. We measure coordination failuresas incidents in which total profits decline from theprior period. Thus, in such a period, the variousagents implemented policy changes that in theirentirety reduced overall profits, a clear form ofcoordination failure.

We now turn our attention to exploration. Explo-ration is affected by both the governance structureand the organizational capability level. Differentstructures create different degrees of explorationby imposing higher or lower levels of decision-making constraints on agents. In the modular case,for example, the two agents act purely indepen-dently, without regard to the effects that theiractions may have on the other firm’s performance.Such independence can result in very high explo-ration. In contrast, the integrated firm, always hav-ing to take all ramifications into account, tendsto be much less explorative. Exploration sup-ply for the alliance modes lies in between. Theself-governing mode creates exploration throughthe unfettered alliance agent, who is allowed tomove first and implement choices from a purelyparochial point of view. The ratified alliance struc-ture, one would expect, creates less exploration. Inthis case, while the alliance is still allowed to makeparochial evaluations of alternatives, the ratifica-tion step dampens the exploration the firms engagein. Table 1, Panel C summarizes this discussion.

The effects of organizational capabilities onexploration are straightforward at the level of indi-vidual agents. An increase in the search radiusincreases the range of possible alternatives that anagent might consider, and thus increases explo-ration. Likewise, an increase in the number ofalternatives that an agent considers increases, atleast in the short run, the area of the landscape anagent might explore. How an increase in explo-ration at the level of individual agents translates,however, to exploration at the level of the organi-zation can be less straightforward. For instance,as Siggelkow and Rivkin (2006) have shown,increases in exploration at a low level within ahierarchical organization can backfire and createless exploration for the organization as a whole.

In our simulations, to measure the degree towhich exploration is supplied by different com-binations of governance structures and organiza-tional capabilities for different interdependencepatterns, we create two measures: percent ofchoices evaluated and number of sticking points.We construct the percent of choices evaluated bycalculating the total number of different contri-bution values (Ci’s) considered by agents in thesystem over time, and dividing this by the totalnumber of contribution values that exist for a land-scape. (For instance, Pattern 1 involves 144 con-tribution values while Pattern 4 involves 49,152contribution values.) The second exploration met-ric, sticking points, measures the number of pointson the landscape from which the firm will not makea move (Rivkin and Siggelkow, 2003). Higher val-ues of this metric suggest a governance structure-capability combination that is less prone to broaderexploration.

Optimal governance with varyinginterdependence

Armed with a more detailed understanding ofexploration and coordination supplies and de-mands, and with our battery of coordination andexploration measures, we now return to a morein-depth discussion of the Pattern 1 and 4 results.We begin with Pattern 1; Table 2 reports the per-formance results for Pattern 1 at periods 10 and200, along with the metrics discussed above forexploration and coordination. We begin by not-ing that this pattern, being fully decomposable,requires little coordination and exploration. On thispattern, for any given level of organizational capa-bility, all governance modes perform equally wellin the long run, reaching the same level of per-formance over time. There are instances, however,in which performance differs across modes in theshort run.

To understand the sources of the performancedifferences, we turn to the metrics for coordina-tion and exploration. As the right-most column ofTable 2 indicates, none of the governance modesexperiences any coordination failure on this pat-tern. As a result, performance variation is solelydriven by differences in exploration generated bythe different modes. As a matter of fact, theperformance ordering of the four governancemodes in period 10, for all levels of organiza-tional capability, is exactly the ordering along the



Table 2. Performance, exploration, and coordination, interdependence pattern 1

Degree of exploration

Performance Percent of choices evaluated∗

Capabilitylevel

Gov.mode

p.10 p.200 p.10 p.200 Stickingpoints

Coordinationfailures

A MOD 0.898 0.931 34.7 45.2 18.0 0SGA 0.916 0.930 40.1 45.2 20.3 0RAT 0.917 0.931 40.0 45.2 19.0 0INT 0.849 0.931 25.3 45.2 18.5 0

B MOD 0.951 0.951 55.3 55.3 17.9 0SGA 0.951 0.951 55.3 55.3 19.8 0RAT 0.951 0.951 55.3 55.3 18.5 0INT 0.951 0.951 55.3 55.3 18.5 0

C MOD 0.910 0.987 47.0 82.6 2.1 0SGA 0.944 0.986 53.6 81.4 2.5 0RAT 0.944 0.986 53.6 81.4 2.3 0INT 0.855 0.983 34.3 80.3 2.2 0

D MOD 0.991 0.991 90.3 90.3 2.2 0SGA 0.993 0.993 89.6 89.6 2.3 0RAT 0.993 0.993 89.6 89.6 2.0 0INT 0.990 0.990 90.7 90.7 2.2 0

MOD, SGA, RAT and INT refer to the modular, self-governing alliance, ratified alliance, and integrated forms, respectively.Performance refers to the sum of Firm 1 and Firm 2 performance at periods 10 and 200. Exploration is measured using twodifferent metrics: (1) percent of choices evaluated: the running total of all contribution values accessed by the two firms at periods10 and 200 as a result of the firms’ decision-making processes divided by the total number of such values on the landscape; and(2) sticking points: the average number of points on the landscape from which the two-firm system will not move. Coordinationfailures are measured as the number of periods between the beginning of the simulation and period 200 that total profit (Firm 1and Firm 2) declines from one period to the next. All measures (except sticking points) are evaluated and averaged across 10,000simulations. Sticking points are averaged over 100 simulation runs.∗ Pattern 1 has a total of 144 possible contribution values that can be evaluated.

exploration metric percent of choices evaluatedby period 10. For instance, for capability levelA, this exploration metric is lowest for the inte-grated mode at 25 percent, and highest for theself-governing and ratified alliance modes at 40percent, with the modular mode in between at 35percent. This is consistent with the ordering of per-formance values at period 10.

The performance benefits in early periods forthe alliance modes illustrate the benefit of split-ting activities among different agents in the sys-tem; because the various activities can be easilydecomposed in this pattern, a larger number ofagents who can work in parallel results in a higherdegree of exploration without resulting in coor-dination failures across activities. In the long run,however, at Period 200, the four governance modesreach the same level of exploration, and indeed thesame level of performance.

With this simple pattern of interdependence,increases in organizational capabilities also create

expected effects. An increase in the search radius(moving from capability level A to level C) leadsprimarily to an increase in long-run explorationand, consequently, to higher long-run performancelevels. The percent of choices evaluated metricincreases from about 45 percent in level A to about81 percent for level C; likewise, the number ofsticking points decline for all governance modesfrom about 18 to 2. At the same time, an increasein the number of alternatives considered (capabil-ity levels A to B, and C to D) results primarilyin increased speed with which the various gover-nance modes converge to the long-run performancelevel. With capability levels B and D, for all gov-ernance modes, the measures of exploration andperformance are essentially identical between peri-ods 10 and 200.

As we have seen previously, results for Pattern4, the fully interdependent pattern, are much morecomplex. Pattern 4 requires a high degree of bothcoordination and exploration. Moreover, as argued



Table 3. Performance, exploration, and coordination, interdependence pattern 4

Degree of exploration

Performance Percent of choices evaluated∗

Capabilitylevel

Gov.mode

p.10 p.200 p.10 p.200 Stickingpoints

Coordinationfailures

A MOD 0.721 0.816 0.3 1.2 83.1 10.8SGA 0.740 0.807 0.4 1.0 99.3 6.4RAT 0.757 0.800 0.4 0.7 192.4 3.4INT 0.764 0.800 0.2 0.5 313.9 0

B MOD 0.721 0.750 1.1 1.7 83.1 46.1SGA 0.787 0.804 0.9 1.1 96.0 20.5RAT 0.764 0.770 0.9 0.9 189.6 29.7INT 0.837 0.837 0.7 0.7 313.9 0

C MOD 0.727 0.808 0.3 4.3 8.0 35.1SGA 0.750 0.860 0.5 3.5 22.0 20.5RAT 0.766 0.853 0.5 2.2 49.6 10.0INT 0.784 0.882 0.3 2.5 48.9 0

D MOD 0.674 0.733 4.6 11.6 8.0 71.7SGA 0.769 0.822 3.1 4.5 20.4 30.8RAT 0.766 0.780 2.7 3.1 46.9 42.5INT 0.888 0.888 4.2 4.2 48.9 0

MOD, SGA, RAT and INT refer to the modular, self-governing alliance, ratified alliance, and integrated forms, respectively.Performance refers to the sum of Firm 1 and Firm 2 performance at periods 10 and 200. Exploration is measured using twodifferent metrics: (1) percent of choices evaluated: the running total of all contribution values accessed by the two firms at periods10 and 200 as a result of the firms’ decision-making processes divided by the total number of such values on the landscape; and(2) sticking points: the average number of points on the landscape from which the two-firm system will not move. Coordinationfailures are measured as the number of periods between the beginning of the simulation and period 200 that total profit (Firm 1and Firm 2) declines from one period to the next. All measures (except sticking points) are evaluated and averaged across 10,000simulations. Sticking points are averaged over 100 simulation runs.∗ Pattern 4 has a total of 49,152 possible contribution values that can be evaluated.

in Table 1, the different governance modes differwidely in both the coordination and the explo-ration that they provide. We begin our discussionof Pattern 4 with capability level A. In this case,in the short run at period 10, the integrated formhas the highest performance, followed in order bythe ratification, self-governing alliance, and mod-ular forms. The coordination failures metric mir-rors well our previous discussion of coordinationsupplied by the various governance modes (recallTable 1, Panel B): the integrated mode providesthe highest degree of coordination (0 coordinationfailures), followed by the ratified mode (on aver-age 3.4 coordination failures), the self-governingmode (6.4 failures), and the modular mode (10.8failures). It is interesting to note that in thiscase, coordination failures predict short-run per-formance remarkably accurately. Moreover, thisfinding holds for all capability levels. Even if afirm has higher short-term exploration, coordina-tion appears to trump in the short run. For instance,

as in Pattern 1, both the ratified and the self-governing modes have higher short-term explo-ration than the integrated mode due to the parallelsearch processes that they are able to engage in.However, more search, if uncoordinated, is likelyto lead to some changes that are performance detri-mental. These changes are particularly harmful inthe short run, since firms do not have enough timeto potentially undo these maladaptive adjustments.As a consequence, the integrated firm, despite itssmaller degree of short-term exploration, outper-forms the ratified and self-governing (and modular)modes in the short run.

The explanation of long-run results is more com-plicated as both coordination and exploration playcritical roles. In the long run, firms have to engagein both to have high performance, yet trade-offsarise. As the exploration and coordination met-rics in Table 3 indicate, there is a general trade-offbetween exploration and coordination: the modes



that create much exploration, such as the modu-lar mode, also create many coordination failures,while the modes that are good at coordination, suchas the integrated mode, are less able to create broadexploration.

Coming back to the results for capability levelA, we can observe that in the long run the short-run performance results are exactly reversed. In thelong run, the modular mode has the highest levelof performance, followed by the self-governingalliance, and the ratified and integrated forms.The two long-run exploration measures (percentof choices evaluated at period 200 and the num-ber of sticking points) both confirm our previousdiscussion of exploration supplied by the variousgovernance modes (recall Table 1, Panel C). Forinstance, while the integrated mode has 314 stick-ing points, the modular form gets stuck on onlyabout 83 points. Likewise, the modular form evalu-ates in 200 periods about 590 different contributionvalues (1.2% of all possible) while the integratedform looks at only 246 (0.5%). The differencesin exploration are reflected in performance, as inthis case the long-run exploration metrics predictwell the performance ordering of the governancemodes.

This direct match between long-run explorationand long-run performance does not hold, however,for all levels of organizational capability. As orga-nizational capabilities are increased, their effectson coordination failures and exploration differacross governance modes; consequently, the rel-ative balance of exploration and coordination thatis supplied by each mode changes. For instance,an increase in the search radius or the num-ber of alternatives is an unambiguous positivefor the integrated mode. Coordination failures areunaffected (they stay at zero) while explorationincreases. As a result, performance increases ascapabilities increase. For all the other governancemodes, however, increases in organizational capa-bilities are not unambiguously beneficial. Gener-ally, an increase in organizational search capa-bilities not only increases exploration but alsoincreases coordination failures. The more broadlymanagers search and the more alternatives theyevaluate, the more likely it is that at least onemanager (within the firms or within the alliance)will implement a change that is performance detri-mental for the other agents in the system. Forinstance, for the self-governing alliance, the coor-dination failures metric increases from 6.4 (with

capability level A) to 20.5 (for capability levelsB and C) to 30.8 (for capability level D). Sim-ilar patterns can be observed for the ratificationand the modular modes. The impact of increasingsearch capabilities at the level of the agents onlong-run firm performance is, hence, ambiguous.For instance, while the modular mode increases itsexploration dramatically between capability levelsA and D (the number of sticking points falls bya factor of 10 and the percent of choices evalu-ated increases by a factor of 10), its coordinationfailures also increase by a factor of 7. As we hadseen in the short-run results, if coordination fail-ures are too prevalent, exploration does not yieldperformance benefits: while it is helpful to searchmore broadly, if the outcome of the search is unco-ordinated, firms’ performances will suffer. As aresult, the long-run performance of the modularmode actually decreases from 0.816 to 0.733 asorganizational capabilities are increased from levelA to level D.

In sum, the results highlight several key points.First, as suggested by Proposition 3, the matchbetween exploration and coordination demandsand supply helps explain the performance out-comes of the various governance modes. In Pattern1, in which little coordination was needed, perfor-mance differences among modes is well explained,both in the short and the long run, by the degreeof exploration that is supplied. In Pattern 4, inwhich coordination demands are very high, coor-dination supply plays a pivotal role in explain-ing performance. For short-run performance, ittracks performance perfectly. For long-run perfor-mance, if exploration is very low (as with capa-bility level A), exploration can trump coordinationin explaining performance differences; otherwise,coordination again appears to be more importantin explaining performance. Second, in the absenceof many interdependencies, increases in organi-zational search capabilities are beneficial for allgovernance modes. In the presence of rich interac-tions, however, increases in organizational searchcapabilities can backfire. Even though increasesin organizational capabilities may have counter-intuitive effects on performance, their effects canbe explained by their consequences on explorationand coordination. While exploration increases withgreater organizational search capabilities, coordi-nation failures increase as well. Consequently, per-formance can suffer.



Table 4. Relative effects of coordination and exploration

Dependent variable: performance (4-1) period 10 (4-2) period 200

Coordination failures −0.156 [0.002] −0.161 [0.001]Exploration 0.189 [0.001] 0.135 [0.000]P2 dummy −0.012 [0.001] −0.016 [0.001]P3 dummy −0.005 [0.001] −0.010 [0.001]P4 dummy −0.046 [0.001] −0.028 [0.001]P3 × coord. failure −0.085 [0.002] −0.105 [0.002]P4 × coord. failure −0.191 [0.002] −0.269 [0.006]P2 × exploration 0.013 [0.001] 0.013 [0.006]P3 × exploration 0.948 [0.008] 0.528 [0.000]P4 × exploration 0.915 [0.013] 0.329 [0.000]Constant 0.830 [0.001] 0.873 [0.001]Model R2 0.547 0.595

This table shows OLS regressions of firm performance (sum of Firm 1 and Firm 2) at period 10 or period 200. Exploration ismeasured by percent of choices evaluated and coordination is measured by coordination failures. All coefficients are significant atthe p < 0.001 level. Each specification contains 640,000 observations: 4 patterns, 4 governance modes, and 4 capability levels across10,000 simulations. Standard errors are indicated in brackets.

Results for Patterns 2 and 3 can be similarlyexplained. However, rather than discussing eachsimulation separately, we summarize our resultscompactly by reporting (in Table 4) the resultsof ordinary least squares (OLS) regressions ofperformance (sum of Firm 1 and Firm 2) onour measures of exploration (percent of choicesevaluated ) and coordination (coordination fail-ures). The underlying dataset pools observationsat either period 10 (short run) or 200 (long run)for the four patterns, four governance modes,and four capability levels. We thus have 640,000observations for the short-run and the long-runregressions (4 × 4 × 4 × 10,000 simulations). Tocontrol for pattern-level influences on performance,we include pattern dummies for Patterns 2–4 (thusPattern 1 is the baseline); to examine the varyingrole of coordination and exploration as the degreeof interdependence changes, we include interactioneffects for these dummies with the coordinationand exploration measures. Since there are no coor-dination failures on Pattern 1, coordination is onlyinteracted with Pattern 3 and Pattern 4, so thatPattern 2 is the baseline in this case. For a givenpattern, the effect of either coordination or explo-ration is thus captured by the sum of the maineffect and the pattern interaction.

The results in Table 4 confirm more systemat-ically a number of observations we have madein our prior discussion. First, as interdependen-cies increase, coordination becomes increasinglyimportant: the interaction of coordination failures

with the pattern dummies, for example, is signif-icantly greater in magnitude for Pattern 4 thanfor Pattern 3. With increasingly dense interactions,coordination becomes paramount in explainingperformance. Second, exploration is always impor-tant, but becomes less so with higher degrees ofinteraction. These results are true for both the shortand long run, with coordination even more impor-tant for higher interaction patterns in the long run.9

Figure 6 illustrates the economic significance ofthese regression results, graphing the magnitudeof the performance change associated with a onestandard deviation change in either coordination orexploration.10 As this figure illustrates, for denserpatterns coordination can become even more crit-ical than exploration.

This last result is noteworthy. Most research onthe value of exploration has stressed the positiverelationship between the value of exploration andthe degree of interdependency (e.g., Rivkin andSiggelkow, 2007). While we confirm this rela-tionship, we find that in an interorganizationalsetting coordination may actually become even

9 All coefficients are significant with p < 0.001; in addition,for each specification, all coefficients are significantly differentfrom one another with p < 0.001, with the exception of theinteraction terms P3 × exploration and P4 × exploration forperiod 10, where the significance of the difference between thesetwo variables is with p < 0.05.10 For ease of comparison, the negative of the coordinationfailure effects are shown; thus the figure shows the effect of a onestandard deviation increase in exploration, and a one standarddeviation decrease in coordination failures.



0.7

0.6

0.5

0.4

0.3

0.2

0.1

0

0.7

0.8

0.6

0.5

0.4

0.3

0.2

0.1

0

Pattern 1 Pattern 2 Pattern 3

ExplorationCoordination

Pattern 4

Pattern 1 Pattern 2 Pattern 3

ExplorationCoordination

Pattern 4

Performance impact, period 10:Effect of a one standard deviation change in coordination or exploration

Performance impact, period 200:Effect of a one standard deviation change in coordination or exploration

These figures show the change in performance associated with a one standard deviation change in coordination or exploration. Values are calculated using the estimated coefficients from Table 5 evaluated at the mean value of coordination (coordination failures) and exploration (percent of choices evaluated) for each pattern. Coordination values shown represent the negative of the coordination failures effect to facilitate exposition.

Figure 6. Magnitude of coordination and exploration effects

more important than exploration. Exploration perse, when not coordinated, can backfire.

DISCUSSION AND CONCLUSION

In this paper, we have sought to better understandhow modes of governance affect the performanceof firms in alliance relationships. While collabo-ration is an important component of firm strategyand a growing body of work has focused on deter-minants of governance choice in this setting, thereis limited empirical evidence concerning the linkbetween mode choice and firm performance. Wedrew from a body of work that has used simulationmethodologies to address issues central to strategyand organization design. In the remainder of thissection we review our core results and theoreticalpredictions, outline the implications of some of the

assumptions made in our model, and discuss howour study relates to and contributes to literature inthis area.

Results and theoretical predictions

We begin by reviewing our core results, notinghow they relate to the propositions that werederived from the prior literature, and offering aset of specific predictions for future work. In manycases, our analysis has enabled us to uncover addi-tional nuances relative to the propositions thatcan help guide future theoretical and empiricalresearch. Beginning with Proposition 1, we findthat the pattern of interdependence among theactivities of firms engaged in an alliance does havea crucial impact on the optimal governance struc-ture. In addition to confirming this prediction, wealso find that the time horizon over which firms



seek to maximize their performance has an impor-tant effect on optimal governance as well: as wesaw with the fully interdependent pattern and capa-bility level A (Figure 5), governance modes thatcreate high short-term performance may not yieldhigh long-term performance, and vice versa. Thus,knowing the pattern of interdependence alone maynot suffice to make a decision on the optimal gov-ernance mode.

Consistent with Propositions 2a and 2b, wefind that the need for and supply of coordina-tion and exploration is affected by the pattern ofinterdependence, governance modes, and organi-zational search capabilities. We discover consid-erable subtlety, however, on the supply side, asgovernance choice and search capabilities interactwith one another. As we saw in Table 3, increasesin organizational search capabilities can increasenot only exploration but also coordination failures.Moreover, the degree to which search capabilitiesincrease coordination failures is strongly mediatedby the governance mode. For instance, while theintegrated mode was unaffected, coordination fail-ures in the modular mode soared with increases inorganizational search capabilities. This interactioncan make simple predictions as to which gover-nance mode provides more coordination difficult.For instance, while the degree of coordination fail-ures generally followed the trend we anticipated inPanel B of Table 1 (most coordination is providedby the integrated mode, followed by the ratifiedalliance, the self-governing alliance, and the modu-lar mode), we found situations in which the ratifiedalliance actually creates more coordination failuresthan the self-governing alliance. This happens inparticular when interdependencies are numerousand agents evaluate many alternatives (see Table 3,capability levels B and D).11

11 The reason why the ratified alliance has more coordination fail-ures in this case than the self-governing alliance is quite subtle.Recall, in the ratified alliance, the managers of the non-allianceactivities move first. Given high interdependency and many alter-natives considered, it is likely that at least one manager will findand implement a change that is performance enhancing for herown firm, but potentially performance destroying for the otherfirm. In the self-governing alliance, the alliance moves first. Inthis case, the alliance achieves higher performance (since it canoptimize parochially and does not need ratification). This higherperformance increases the bar for the non-alliance managers forfinding a better alternative. (Recall, non-alliance managers takeone-half of the alliance performance into account in making eval-uations). As a result, in this case, non-alliance managers are lesslikely to find a (parochially) performance enhancing move that,given the high interdependency, could undermine the perfor-mance of the other firm.

Finally, consistent with the arguments leadingto Proposition 3, we find that firm performancein an alliance relationship improves when coor-dination and exploration needs are matched withcoordination and exploration supplied. Our results,however, also reveal that the match on the coordi-nation and the exploration dimension is not alwaysequally important. As the degree of interdepen-dence increases, the value of having a good matchon the coordination dimension increases. For veryhigh levels of interdependence, coordination canactually trump exploration. Even though high lev-els of interdependence require a high degree ofexploration, unless this exploration is tied to coor-dination, it can be ineffective.

As discussed at the beginning of this arti-cle, empirical results concerning the performanceimpact of alliances are mixed; moreover, thereexists only limited empirical evidence examiningthe performance implications of alternate modes ofgovernance. The results of our model suggest thatclear-cut empirical results may require a greaterfocus on the interaction of factors such as interde-pendence, capabilities, and governance mode. Inparticular, our results offer three specific predic-tions that could guide future empirical work:

(1) When activity interdependence is pervasive,in the absence of strong organizational searchcapabilities, the performance benefit of a fullyintegrated alliance governance mode relativeto less integrated modes will decline over time.

(2) When activity interdependence is pervasive, inalliances with weak coordinating mechanisms,firm performance may decline as firms’ searchcapabilities become stronger.

(3) When activity interdependence is pervasive,the relative importance of selecting high coor-dination governance modes as compared tohigh exploration modes will increase.

Simulation model assumptions

Next, we discuss several assumptions embeddedin our model and elaborate on the implications forour results (and for future extensions) of relaxingsome of these assumptions. First, we modeled analliance situation with two firms. It is not uncom-mon, however, for alliances to involve more thantwo firms. In such cases, we would expect that sim-ilar issues of coordination and exploration wouldcome into play. The main question, then, is how



firms can design a set of governance structures thattakes into account the complexities associated withhaving a greater number of involved parties. Forinstance, an appropriate methodology for ratify-ing joint decisions in such a setting would likelyinvolve a more complex set of rules given thelarger number of players. At the same time, wewould expect that the core issue is likely to be thesame as in a two-firm setting: how can the need forcoordination and exploration arising from the pat-tern of interfirm interdependence be matched withthe supply of coordination and exploration arisingfrom interfirm governance of decision making andfirm capabilities?

A second modeling assumption involves the por-tion of the alliance profit each firm takes intoaccount in making decisions (the parameter α inour model). In our analysis, we have assumed thatboth firms put equal weight on alliance perfor-mance. One might imagine, however, situations inwhich the alliance is considered of greater impor-tance by one firm than by the other. How are ourresults affected in such a situation? We take a firstcut at exploring this by examining our results inthe case where Firm 1 has α = 0.75 and Firm 2has α = 0.25 (for the alliance modes). Rerunningour model on Patterns 1 and 4, and focusing onthe two alliance modes (self-governing and rati-fied), we find that, as expected, on Pattern 1 thereare no significant differences between the origi-nal, ‘balanced’ alliance results and the alternative,‘asymmetric’ alliance results (due to the lack ofinterdependencies between the alliance and non-alliance activities). With Pattern 4, however, wefind an interesting set of results: alliance asymme-try reduces the performance of Firm 1 (with the0.75 weight), while increasing the performance ofFirm 2, for both the self-governing and ratifiedmodes. The firm with a higher emphasis on thealliance suffers largely because the other firm nowhas a stronger incentive to implement choices thatimprove performance of its non-alliance activities(N2) but that are detrimental to the alliance andhence to the other firm. (Since α is low, Firm 2bears less of the negative externality it imposesonto the alliance.) This parochial behavior can leadto situations in which even joint profits are lowerthan in the symmetric case. These results raise thequestion for future research of how firms can cre-ate modes with effective coordination when firmsdiffer in the degree to which they consider the jointalliance to be important.

A third modeling assumption relates to theimplications of predefining a task allocation struc-ture. Because we prespecify the tasks that comprisethe alliance, we are effectively making the choiceof alliance scope exogenous. This choice, how-ever, is likely intertwined with a host of otherissues related to interdependence, firm capabili-ties, and governance mode. We conducted severalrobustness analyses to better understand the impli-cations of thinking about alliance boundaries inthis way. The task decomposition we have usedin this study has been a 4-4-4 split between Firm1, the alliance, and Firm 2. We evaluated twoalternative task decomposition structures: a 5-2-5and a 3-6-3 structure, on which we examined theresults of the self-governing and ratified alliancemodes. The results suggest that the alliance bound-ary issue can be explained in connection withthe governance structures: the results of the 5-2-5 structure closely resemble the modular results,while the 3-6-3 structure creates results that tendto resemble those of the integrated mode. Thus,while the choice of boundaries may in reality bean endogenous consideration for firms, we canthink about these as more continuous versions ofthe governance structures. The modes we evalu-ate, therefore, represent discrete points among thecontinuum of choices available to firms, with taskallocation being subsumed as an issue within thegovernance structure decision.

Lastly, we should note that our findings haveparallels with results from prior efforts using quitedifferent simulation methods to address organiza-tional issues. Burton and Obel (1980, 1984), forexample, utilize a decomposed linear program-ming methodology (Dantzig, 1963; Baumol andFabian, 1964) to address intrafirm organizationissues. They find that firm performance under alter-nate divisional forms (e.g., M-form vs. U-form)is influenced by the degree of decomposabilityof the firm’s underlying technology. In particular,their results suggest that the benefits of the moredecentralized M-form structure are more apparentfor more nearly decomposable technologies. Thisrelates to our own finding where on Pattern 1,which is fully decomposable, the more decentral-ized ‘ratification’ mode performs better than themore centralized ‘integrated’ mode (with the oppo-site generally true on the less decomposable Pat-tern 4). It is reassuring to note that different model-ing approaches can result in parallels such as these.On the other hand, different approaches can also



lead to new insights, as our results linking varia-tion in interdependencies of various types (withinand across firms and their alliance activities) withalternate alliance governance structures suggest.

Contributions and implications

Our study has a number of implications for the-ory and future research in this area. The obser-vation that coordination challenges can trumpexploration in interorganizational settings under-scores the importance of evaluating the contingentimplications of interdependence when structuringcooperative relationships. This is consistent withGulati and Singh’s (1998) discussion of coordina-tion as a relevant concern when structuring alliancegovernance. Moreover, because we abstract awayfrom the technology- and industry-related factorsthat underlie transaction cost-related considera-tions of appropriability (e.g., Pisano, 1989; Oxley,1997) we can focus more specifically on iden-tifying the sources of coordination and explo-ration supply, allowing a more targeted approach tounderstanding the trade-offs associated with gov-ernance mode design. For example, while someprior work has drawn on arguments related tocapabilities and interdependence in discussing thedeterminants of governance mode, much of thiswork has utilized fairly broad proxies for hier-archical controls such as the use of equity (e.g.,Colombo, 2003). Having a more contingent viewof the potential implications of specific governancemode design considerations (as we attempt to doin this article) can enable more targeted researchapproaches.

A core set of results in this study relates tothe ways in which governance structure inter-acts with organizational search capabilities. Theseresults are particularly intriguing given the atten-tion paid to the role of firm-level factors such asalliance capabilities in recent studies. For exam-ple, Kale et al. (2002: 748), building on the ideathat greater alliance experience can have positiveperformance effects (Anand and Khanna, 2000),discuss the role of a ‘dedicated alliance function’in capturing and managing alliance-related knowl-edge, suggesting that such a function can enablethe development of alliance capabilities throughlearning and evolutionary processes (March, 1991;Nelson and Winter, 1982). In much the same spirit,Zollo et al. (2002) develop the idea of interorga-nizational routines, which also follow a similar

evolutionary path. Our concept of organizationalsearch capabilities shares a number of featureswith these various concepts: they are assets tiedto the firm and developed over time that endowfirm agents with superior evaluation capabilitiesin the context of interorganizational relationships.Whereas prior studies have generally stressed thepositive relationship between alliance capabilitiesand firm performance, our analysis suggests thatthere are contingencies under which investments inthe development of these capabilities may actuallybackfire.

This study also contributes to the broader con-versation around the performance consequences ofinterorganizational relationships. Various streamsof prior work have examined the effects of alli-ances on different metrics of firm performance,looking at outcomes such as firm survival (e.g.,Baum and Oliver, 1991), innovation output (e.g.,Shan et al., 1994) and stock market valuation (Daset al., 1998). While these studies generally findpositive effects, albeit with a range of contin-gencies, other studies paint a more mixed view,for example when comparing interorganizationalrelationships with firm capabilities with respectto their impact on performance (e.g., Lee et al.,2001). Much of the variance in alliance-relatedfirm performance outcomes is likely due to partnerand interfirm characteristics (e.g., Stuart, 2000); byillustrating how factors such as search capabilitiesand interdependence can influence the governance-performance link, our study can shed new light onthis area, with future work focused more specifi-cally on empirically testing the predictions of ourmodel.

A key aim of this study has been to develop a setof insights that can be used to guide future empir-ical work. In this context, the use of factors suchas interorganizational patterns of interdependencedoes increase the burden of collecting empiricaldata that is more fine-grained and complex. Atthe same time, however, the simulation results canhelp generate more refined hypotheses that havethe opposite effect of simplifying the data collec-tion task. With more specific priors (e.g., in moreinterdependent settings, modes that enable greatercoordination are preferred to modes that enablegreater exploration), future work can explore thegovernance implications of interorganizational col-laboration in a more targeted way using empiricaland case-based studies.



ACKNOWLEDGEMENTS

We thank Howard Brenner for programming assis-tance, and Richard Bettis, Gino Cattani, SendilEthiraj, David Hsu, Dan Levinthal, two anony-mous reviewers, and conference participants inthe Strategic Management Society Conference,the Atlanta Competitive Advantage Conference,and the Academy of Management meetings forthoughtful comments. We gratefully acknowledgethe Mack Center for Technological Innovation atthe Wharton School and the INSEAD-WhartonCenter for Global Research and Education for gen-erous funding.

REFERENCES

Aggarwal VA, Hsu DH. 2009. Modes of cooperativeR&D commercialization by start-ups. StrategicManagement Journal 30(8): 835–864.

Anand BN, Khanna T. 2000. Do firms learn to createvalue? The case of alliances. Strategic ManagementJournal , March Special Issue 21: 295–315.

Baum JAC, Oliver C. 1991. Institutional linkages andorganizational mortality. Administrative Science Quar-terly 36(2): 187–218.

Baumol WJ, Fabian T. 1964. Decomposition, pricing fordecentralization and external economics. ManagementScience 11(1): 1–32.

Burton RM, Obel B. 1980. A computer simulation testof the M-form hypothesis. Administrative ScienceQuarterly 25(3): 457–466.

Burton RM, Obel B. 1984. Designing Efficient Organi-zations: Modeling and Experimentation . Elsevier Sci-ence: Amsterdam, the Netherlands.

Burton RM, Obel B. 1995. Strategic OrganizationalDiagnosis and Design: Developing Theory forApplication . Kluwer Academic: Norwell, MA.

Colombo MG. 2003. Alliance form: a test of thecontractual and competence perspectives. StrategicManagement Journal 24(12): 1209–1229.

Cyert RM, March JG. 1963. A Behavioral Theory of theFirm . Prentice Hall: Englewood Cliffs, NJ.

Dantzig GB. 1963. Linear Programming and Extensions .Princeton University Press: Princeton, NJ.

Das S, Sen PK, Sengupta S. 1998. Impact of strategicalliances on firm valuation. Academy of ManagementJournal 41(1): 27–41.

Dyer JH, Kale P, Singh H. 2004. When to ally andwhen to acquire. Harvard Business Review 82(7–8):108–115.

Galbraith JR. 1977. Organization Design . Addison-Wesley: Reading, MA.

Gulati R, Lawrence PR, Puranam P. 2005. Adaptationin vertical relationships: beyond incentive conflict.Strategic Management Journal 26(5): 415–440.

Gulati R. Singh H. 1998. The architecture of cooperation:managing coordination costs and appropriation

concerns in strategic alliances. Administrative ScienceQuarterly 43(4): 781–814.

Hennart J. 1993. Explaining the swollen middle: whymost transactions are a mix of ‘market’ and hierarchy.Organization Science 4(4): 529–547.

Kale P, Dyer JH, Singh H. 2002. Alliance capability,stock market response, and long-term alliance success:the role of the alliance function. Strategic ManagementJournal 23(8): 747–767.

Kale P, Puranam P. 2004. Choosing equity stakesin technology-sourcing relationships: an integrativeframework. California Management Review 46(3):77–99.

Kauffman S. 1993. The Origins of Order: Self-organiza-tion and Selection in Evolution . Oxford UniversityPress: New York.

Klein B, Crawford RG, Alchian AA. 1978. Verticalintegration, appropriable rents, and the competitivecontracting process. Journal of Law and Economics21(2): 297–326.

Lawrence PR, Lorsch JW. 1967. Organization andEnvironment . Harvard Business School Press: Boston,MA.

Lee C, Lee K, Pennings JM. 2001. Internal capabilities,external networks, and performance: a study ontechnology-based ventures. Strategic ManagementJournal , June–July Special Issue 22: 615–640.

Leiblein MJ, Miller DJ. 2003. An empirical examinationof transaction- and firm-level influences on the verticalboundaries of the firm. Strategic Management Journal24(9): 839–859.

Lenox MJ, Rockart SF, Lewin AY. 2006. Interdepen-dency, competition, and the distribution of firm andindustry profits. Management Science 52(5): 757–772.

Levinthal D. 1997. Adaptation on rugged landscapes.Management Science 43(7): 934–950.

March JG. 1991. Exploration and exploitation inorganizational learning. Organization Science 2(1):71–87.

Mayer KJ, Salomon RM. 2006. Capabilities, contractualhazards, and governance: integrating resource-basedand transaction cost perspectives. Academy ofManagement Journal 49(5): 942–959.

Mintzberg H. 1979. The Structuring of Organizations.Prentice Hall: Englewood Cliffs, NJ.

Nelson RR, Winter SG. 1982. An Evolutionary Theory ofEconomic Change. Belknap Press: Cambridge, MA.

Oxley JE. 1997. Appropriability hazards and governancein strategic alliances: a transaction cost approach.Journal of Law, Economics, & Organization 13(2):387–409.

Pisano GP. 1989. Using equity participation to supportexchange: evidence from the biotechnology industry.Journal of Law, Economics, & Organization 5(1):109–126.

Porter M. 1996. What is strategy? Harvard BusinessReview 74(6): 61–78.

Reuer JJ. 2004. Strategic Alliances: Theory and Evidence.Oxford University Press: New York.

Rivkin J, Siggelkow N. 2003. Balancing search andstability: interdependencies among elements of



organizational design. Management Science 49(3):290–311.

Rivkin J, Siggelkow N. 2007. Patterned interactionsin complex systems: implications for exploration.Management Science 53(7): 1068–1085.

Robinson D, Stuart T. 2007. Financial contracting inbiotech strategic alliances. Journal of Law andEconomics 50(3): 559–595.

Sampson R. 2004. The cost of misaligned governancein R&D alliances. Journal of Law, Economics, &Organization 20(2): 484–526.

Schelling TC. 1960. The Strategy of Conflict . HarvardUniversity Press: Cambridge, MA.

Shan W, Walker G, Kogut B. 1994. Interfirm cooperationand startup innovation in the biotechnology industry.Strategic Management Journal 15(5): 387–394.

Siggelkow N. 2001. Change in the presence of fit: therise, the fall, and the renaissance of Liz Claiborne.Academy of Management Journal 44(4): 838–857.

Siggelkow N. 2002. Evolution toward fit. AdministrativeScience Quarterly 47(1): 125–159.

Siggelkow N, Rivkin J. 2005. Speed and search: design-ing organizations for turbulence and complexity.Organization Science 16(2): 101–122.

Siggelkow N, Rivkin J. 2006. When exploration back-fires: unintended consequences of multi-level orga-nizational search. Academy of Management Journal49(4): 779–795.

Simon HA. 1945. Administrative Behavior. Free Press:New York.

Simon HA. 1996. The Sciences of the Artificial . MITPress: Cambridge, MA.

Stuart TE. 2000. Interorganizational alliances and theperformance of firms: a study of growth andinnovation rates in a high-technology industry.Strategic Management Journal 21(8): 791–811.

Thompson JD. 1967. Organizations in Action. McGraw-Hill: New York.

Tushman ML, Nadler DA. 1978. Information processingas an integrating concept in organizational design.Academy of Management Review 3(3): 613–624.

Villalonga B, McGahan AM. 2005. The choice amongacquisitions, alliances and divestitures. StrategicManagement Journal 26(13): 1183–1208.

Volberda HW. 1996. Toward the flexible form: howto remain vital in hypercompetitive environments.Organization Science 7(4): 359–374.

Williamson OE. 1975. Markets and hierarchies: analysisand antitrust implications. Free Press: New York.

Zollo M, Reuer JJ, Singh H. 2002. Interorganizationalroutines and performance in strategic alliances.Organization Science 13(6): 701–713.


GOVERNING COLLABORATIVE ACTIVITY: INTERDEPENDENCE …faculty.insead.edu/vikas-aggarwal/documents/[2]...

Documents

Transcript of GOVERNING COLLABORATIVE ACTIVITY: INTERDEPENDENCE …faculty.insead.edu/vikas-aggarwal/documents/[2]...