BALANCING EXPLORATION AND EXPLOITATION IN ALLIANCE …€¦ · whereas exploitation involves “the use and devel-opment of things already known” (1993: 105). More ... Exposure

BALANCING EXPLORATION AND EXPLOITATION INALLIANCE FORMATION

DOVEV LAVIEUniversity of Texas at Austin

LORI ROSENKOPFUniversity of Pennsylvania

Do firms balance exploration and exploitation in their alliance formation decisionsand, if so, why and how? We argue that absorptive capacity and organizational inertiaimpose conflicting pressures for exploration and exploitation with respect to the valuechain function of alliances, the attributes of partners, and partners’ network positions.Although path dependencies reinforce either exploration or exploitation within eachof these domains, we find that firms balance their tendencies to explore and exploitover time and across domains.

Scholars studying exploration and exploitationin organizational learning have assumed a strategicposture by recognizing the essential trade-offs firmsmake in undertaking these activities, yet little isknown about the organizational mechanisms thatdrive firms’ tendencies to engage in either activityor about whether and how firms balance the twoactivities. The fundamental conceptualizationshighlighting the merits of balancing conflictingneeds for exploration and exploitation (Levinthal &March, 1993; March, 1991) have been incorporatedonly in simulation studies (e.g., Levinthal, 1997;Rivkin & Siggelkow, 2003) or elucidated by anec-dotal evidence (e.g., Tushman & O’Reilly, 1997).Studies such as those cited have noted that alter-

native organizational forms, such as decentralizedversus centralized structures and organic versusmechanistic ones, are better suited for engaging ineither exploration or exploitation within firms’ or-ganizational boundaries (Brown & Eisenhardt,1997; Nickerson & Zenger, 2002; Siggelkow &Levinthal, 2003). However, they do not address thequestion of balance in interfirm relationships. Sim-ilarly, a host of empirical studies in the allianceliterature have struggled to identify industry con-ditions or clusters of firms that demonstrate ten-dencies to either explore or exploit, paying lessregard to whether balance can be achieved. Weattempt to fill this gap in organizational learningresearch by offering theory and evidence that dem-onstrate why and how firms balance these tenden-cies over time and across domains.

We focus on the challenges of balancing explora-tion and exploitation in alliance formation deci-sions (Koza & Lewin, 1998) following the growinginterest in interorganizational learning. In this con-text, most studies have focused on external indus-try forces, suggesting that turbulence and marketuncertainty may generate either exploitation (Beck-man, Haunschild, & Phillips, 2004; Rothaermel,2001b), exploration (Rowley, Behrens, & Krack-hardt, 2000), or both (Koza & Lewin, 1998). Park,Chen, and Gallagher (2002) noted that only re-source-poor firms form exploitation alliances inturbulent industries. Yet even studies that examinefirm characteristics have produced mixed evidenceon the antecedents of exploration and exploitation.For instance, Rothaermel and Deeds (2004) notedthat exploitation increases with firm size, whereasBeckman and her coauthors (2004) showed thatfirm size also contributes to exploration. These

We thank special issue coeditor Ken Smith and threeanonymous AMJ reviewers for their helpful comments.We benefited from the feedback received from RanjayGulati, Pam Haunschild, George Huber, Dan Levinthal,Toby Stuart, and Jim Westphal. Additional feedback wasreceived from participants in the 2005 Harvard BusinessSchool Corporate Entrepreneurship Conference and par-ticipants of management colloquia held at the WhartonSchool, University of Pennsylvania, and the McCombsSchool of Business, University of Texas at Austin. Wealso thank Christopher Ray and Mike Hendron for theirresearch assistance. The data used in this study werederived primarily from Dovev Lavie’s doctoral disserta-tion, “The Interconnected Firm: Evolution, Strategy, andPerformance” (University of Pennsylvania, 2004). We aregrateful for the financial support received from the MackCenter for Technological Innovation at the WhartonSchool. An abbreviated prior version of this paper waspublished in the Best Paper Proceedings of the 2005annual meeting of the Academy of Management, inHonolulu.

� Academy of Management Journal2006, Vol. 49, No. 4, 797–818.

797

studies offer interesting insights but focus on dif-ferent conceptualizations of the exploration-exploi-tation construct.

We posit that prior research on the antecedents ofexploration and exploitation has produced incon-sistent evidence because each study examined ex-ploration and exploitation within a single domain,disregarding the conflicting organizational pres-sures that influence learning in various domains. Inthis study, we explicitly distinguish three domainsin which exploration and exploitation can be pur-sued and balanced. The alliance literature has tra-ditionally associated exploration and exploitationwith the value-adding activities of alliances—thatis, the value chain function that they serve—thusconceptualizing a “function domain.” Taking thiscourse, researchers have identified knowledge-gen-erating R&D alliances as exploration alliances andknowledge-leveraging marketing alliances as ex-ploitation alliances (Koza & Lewin, 1998; Rothaer-mel, 2001b). Setting a somewhat different course,we also encompass the network positions of part-ners (conceptualizing a structure domain) and theirprofiles (conceptualizing an attribute domain).Structure exploration refers to a firm’s decision toform alliances with partners with whom it has noprior ties (Beckman et al., 2004), whereas attributeexploration refers to a firm’s forming alliances withpartners whose organizational attributes consider-ably differ from those of its prior partners. In ourframework, we further acknowledge the interde-pendence between exploration and exploitation(Levinthal & March, 1993) by conceptualizing theseactivities as resting on a single continuum ratherthan prevailing as two independent organizationalchoices.

Our main contribution to theory involves delin-eating distinct domains of exploration-exploitationand advancing the notion that firms balance explo-ration and exploitation over time within domainsas well as across these domains. For example, afirm may engage in an R&D alliance (function ex-ploration) with a prior partner (structure exploita-tion) who differs in size and industry focus fromthe firm’s other partners (attribute exploration).Moreover, a firm may shift from exploitation toexploration or vice versa within domains over time(e.g., transitioning from prior partners to new part-ners). To further advance this theory, we supple-ment the traditional focus on external industryforces by arguing that firms face internal organiza-tional pressures for exploration-exploitation.Whereas organizational inertia (Hannan & Free-man, 1984) fosters exploitation, absorptive capacity(Cohen & Levinthal, 1990) encourages exploration.We posit that although firms encounter challenges

in balancing these conflicting pressures within do-mains, they can reconcile these pressures by dy-namically balancing exploration and exploitationacross domains. Our analysis of a comprehensivesample of alliances formed by United States–basedsoftware firms between 1990 and 2001 providesempirical support for these ideas.

THEORY AND HYPOTHESES

The exploration-exploitation framework distin-guishes two broad patterns of learning behaviors.March defined them as follows: “Exploration in-cludes things captured by terms such as search,variation, risk taking, experimentation, play, flexi-bility, discovery, innovation. Exploitation includessuch things as refinement, choice, production, effi-ciency, selection, implementation, execution”(1991: 71). Levinthal and March added that explo-ration involves “a pursuit of new knowledge,”whereas exploitation involves “the use and devel-opment of things already known” (1993: 105). Morerecently researchers have elaborated these ideas byconsidering their implications not only for intra-but also for interorganizational learning (e.g., Child,2001; Grant & Baden-Fuller, 2004; Holmqvist, 2003;Ingram, 2002; Lane & Lubatkin, 1998; Larsson,Bengtsson, Henriksson, & Sparks, 1998). They haverecognized that collaboration with partners facili-tates learning by accessing new knowledge residingoutside a firm’s boundaries and by collaborativelyleveraging existing knowledge with partners. Thus,alliances, which are voluntary arrangements amongindependent firms involving exchange, sharing, orjoint development or provision of technologies,products, or services (Gulati, 1998), have become anoteworthy vehicle for exploration and exploitation.

The Three Domains of Exploration-Exploitationin Alliance Formation

We identify three separate domains of explora-tion and exploitation that together describe an alli-ance. We begin by asking three questions: What isthe value chain function of the alliance (function)?Whom does the focal firm partner with (structure)?To what extent does the firm’s partner differ fromprior partners (attribute)? Table 1 summarizes thesedistinctions.

Function exploration-exploitation. The explora-tion-exploitation framework was introduced in theinterorganizational context by Koza and Lewin(1998), who noted that firms may form alliances toexploit existing knowledge or to explore new op-portunities. Subsequent research has thus focusedon the value chain function that alliances serve

798 AugustAcademy of Management Journal

(Koza & Lewin, 2000; Park et al., 2002; Rothaermel,2001b; Rothaermel & Deeds, 2004). Firms that en-gage partners in R&D that may lead to innovativetechnologies and applications can be said to partic-ipate in exploration, whereas firms that rely onalliances for commercializing and using existingtechnologies or employing complementary partnercapabilities undertake exploitation. In this sense,exploration alliances engage in upstream activitiesof the value chain, enabling partners to share tacitknowledge and develop new knowledge. In con-trast, exploitation alliances engage in downstreamactivities such as commercialization and marketingthat leverage and combine partners’ existing capa-bilities through exchanges of explicit knowledge(Rothaermel, 2001b). The distinction between ac-quiring and generating new knowledge through ex-ploration and accessing, integrating, and imple-menting existing knowledge through exploitation(Grant & Baden-Fuller, 2004) has been thus linkedto firms’ polar tendencies to engage in R&D alli-ances versus marketing alliances (Park et al., 2002;Rothaermel, 2001b; Rothaermel & Deeds, 2004).

Structure exploration-exploitation. The struc-ture domain of exploration-exploitation takes intoaccount the network positions of a firm’s partners.Recurrent alliances between firms are considered a

form of exploitation, and alliances formed withnew partners are considered exploration. When afirm forms recurrent alliances with a select group ofpartners, it can rely on existing arrangements andchannels to facilitate access and transfer of knowl-edge already prevailing in its immediate alliancenetwork. In this regard, Beckman and colleagues(2004) argued that forming additional allianceswith existing partners is a form of exploitation inwhich a firm reinforces its existing relationships inorder to use its current knowledge base. Hence, theproximate network positions of partners facilitatesthe flow of knowledge and information and en-hances the efficiency of collaboration (Verspagen &Duysters, 2004). By forming alliances with familiarpartners, firms can also rely on prior experienceand interfirm trust to enhance the predictabilityand reliability of collaboration (Chung, Singh, &Lee, 2000; Gulati, 1995a; Gulati & Gargiulo, 1999;Li & Rowley, 2002); such a pattern of alliance for-mation corresponds to March’s (1991) notion ofexploitation. In contrast, when partners have noprior ties to a firm, the firm cannot rely on directexperience with these partners, but it can broadenits reach and seek knowledge that cannot be chan-neled through its immediate network. In keepingwith March (1991), because a search for partners

TABLE 1Domains of Exploration-Exploitation

Domain Function Structure Attribute

Answers the question What value chain function doesthe alliance serve?

Whom does the firm partnerwith?

To what extent does the partnerdiffer from prior partners?

Focus Alliance type Network structure Partner profileExploration (March, 1991)

(search, variation, risktaking, experimentation,play, flexibility,discovery, innovation)

Forming a knowledge-generatingR&D alliance

Forming an alliance with anew partner that has noprior ties to the firm

Forming an alliance with apartner whose organizationalattributes differ from those ofprior partners

Exploitation (March, 1991)(refinement, choice,production, efficiency,selection,implementation,execution)

Forming a knowledge-leveragingmarketing/production alliance

Forming recurrent allianceswith a partner that hasprior ties to the firm

Forming an alliance with apartner whose organizationalattributes are similar to thoseof prior partners

Content of learned knowledge Value chain knowledge such asnew technologies or marketinformation and expertise inexisting technologies

Remote knowledge andinformation on partners’identities andaccessibility or immediateknowledge and in-depthfamiliarity with specificpartners

Exposure to organizationaldiversity or specialization ina specific set of partnerattribute configurations

Relevant references Koza & Lewin (1998); Rothaermel(2001); Rothaermel & Deeds(2004)

Baum, Rowley, Shipilov, &Chuang (2005); Beckman,Haunschild & Phillips(2004); Verspagen &Duysters (2004)

Gulati, Lavie, & Singh (2003);McGrath (2001); Darr &Kurtzberg (2000)

2006 799Lavie and Rosenkopf

beyond a firm’s local network offers new opportu-nities yet greater uncertainty and risk, we concep-tualized it as a form of exploration.

Attribute exploration-exploitation. Unlike thefunction domain, which defines the nature of alli-ance relationships, and the structure domain,which relates to the prior network positions of part-ners, the attribute domain refers to intertemporalvariance in the organizational attributes of a firm’spartners. Following March (1991), we associateexploration with experimentation and variation inroutines, processes, technologies, and applications.Exploration enhances adaptation to environmentalchanges by increasing variance in these organiza-tional attributes (McGrath, 2001) and by supporting“long jumps” or reorientations (Levinthal, 1997)that enable a firm to adopt new attributes and attainnew knowledge outside its domain (Rosenkopf &Nerkar, 2001). A deviation from a systematic pat-tern of alliance formation with partners that sharecertain organizational attributes is thus consideredan exploratory behavior. In contrast, when a firmpersistently forms new alliances with partners thatare similar to its prior partners with respect toattributes such as size and industry focus, it canapply established heuristics and effective gover-nance mechanisms for assimilating external knowl-edge (Darr & Kurtzberg, 2000) and can also effi-ciently accumulate and apply its partneringexperience in the learning process (Gulati, Lavie, &Singh, 2003; Hoang & Rothaermel, 2005). Such per-sistence in alliance formation leads to repetition-based improvement, experiential learning, and spe-cialization, which are associated with exploitation(Levinthal & March, 1993). Within the attribute do-main, firms’ alliance networks range from exhibit-ing consistency in partners’ attributes (exploita-tion) to showing frequent deviation from such apattern (exploration).

The three domains are conceptually distinct, inpart, because the content of learned knowledge var-ies across domains (see Table 1). Nevertheless, theymay be empirically related.

Normative versus Behavioral Perspectives onBalancing Exploration and Exploitation

The notion of balance refers to equilibrium be-tween conflicting tendencies. Existing research re-veals a striking contrast between normative as-sumptions and behavioral tendencies with respectto the balance between exploration and exploita-tion. Prescriptions about whether firms shouldstrive to manage the trade-off between explorationand exploitation are inconsistent with observationsabout firms’ tendencies to balance these activities

in actual practice. On the one hand, researchershave normatively assumed that firms should seekto balance exploration and exploitation becauseboth short-term productivity and long-term innova-tion are essential for organizational success andsurvival (March, 1991: 87). They have urged firmsto pursue both effectiveness and efficiency and tointegrate organizational renewal and control,which can be correspondingly enhanced via explo-ration and exploitation. Rivkin and Siggelkowhighlighted “the need for an organization to strikethe balance between search and stability” (2000:308), and Siggelkow and Levinthal explicitly re-ferred to the “premise that adaptive entities arecharged to maintain a balance of exploration andexploitation” (2003: 651). Similarly, Tushman andO’Reilly argued that “organizations can sustaintheir competitive advantage by operating in multi-ple modes simultaneously—managing for short-term efficiency by emphasizing stability and con-trol, as well as for long-term innovation by takingrisks and learning by doing” (1997: 167).

On the other hand, despite the undesirable out-comes and self-destructive nature of adaptive pro-cesses (March, 1991), failure and success traps maylead to excessive exploration or exploitation, re-sulting in imbalance (Levinthal & March, 1993).Thus, in practice, researchers have long recognizedthe obstacles that firms face when simultaneouslypursuing exploration and exploitation, highlight-ing the contradictory natures of activities designedto achieve efficiency and those aimed at flexibilityand adaptation (Abernathy, 1978).1 Firms may seekto overcome these internal organizational trade-offsby engaging in mergers and acquisitions; these ac-tivities yield loosely coupled subunits that are notbound by the same routines and culture as theparent firms and thus maintain buffers betweenexploratory and exploitative activities. Alliancescan also serve as vehicles allowing a firm to exploreexternal opportunities while maintaining an inte-grated internal organization. Nonetheless, empiri-

1 Organizational research suggests that firms cope withthese trade-offs by frequently alternating between incon-sistent organizational designs (Brown & Eisenhardt,1997), establishing buffers between specialized subunits(Christensen, 1998; Levinthal, 1997), or maintaining am-bidextrous organizations that integrate these culturallyand organizationally differentiated subunits at the corpo-rate level (Benner & Tushman, 2003; Tushman &O’Reilly, 1997). These ideas are mostly normative be-cause they entail substantial implementation challengesand require empirical validation to determine whetherthey indeed lead to a balance between exploration andexploitation.


cal research indicates that exploitation crowds outexploration (Benner & Tushman, 2002; Sorenson &Stuart, 2001).

In particular, alliance research offers evidence onfirms’ tendencies to focus on one of these types ofactivity, explaining what may lead to imbalance,rather than to balance, between exploration andexploitation. The bulk of studies have highlightedexogenous industry forces, such as industry turbu-lence and market uncertainty, that exacerbatefirms’ tendencies to explore or exploit in their alli-ances (Beckman et al., 2004; Rothaermel, 2001b),Even those studies that identify firm characteristicsthat generate idiosyncratic tendencies to explore orexploit under certain industry conditions (Park etal., 2002; Rothaermel, 2001b; Rothaermel & Deeds,2004) shed almost no light on the organizationalmechanisms that guide these tendencies, the trade-offs that they entail, and firms’ attempts to balancethese tendencies in alliance formation decisions.The balancing of exploration and exploitation inparticular domains of alliance formation has thusremained a normative assumption rather than aproven behavioral pattern. In this study, we beginto reconcile the discrepancy between this assump-tion and firms’ behavioral tendencies.

Conflicting Organizational Pressures forExploration and Exploitation

In studying whether and how firms balance ex-ploration and exploitation in alliance formation de-cisions, one should consider not only externalstimuli in the form of industry conditions but alsothe internal organizational pressures that guidefirms’ responses to these stimuli. These fundamen-tal pressures can influence firms’ tendencies evenin stable industry conditions (Nickerson & Zenger,2002). In the organizational learning perspective(Levitt & March, 1988; March, 1991), inertia impelsfirms toward exploitation, whereas search activi-ties, backed by absorptive capacity, drive explora-tion. The balancing of exploration and exploitationin alliances is challenging given the simultaneousco-existence of these conflicting organizationalpressures.

Pressures for exploitation. Pressures for exploi-tation often derive from organizational inertia,which is evident “when the speed of reorganizationis much lower than the rate at which environmen-tal conditions change” (Hannan & Freeman, 1984;151). Inertia results from internal forces, such asirreversible managerial commitments and historicdecisions, as well as from external forces, such asinstitutional legitimation (Hannan & Freeman,1984). Inertia intensifies as established routines

and skills become embedded in decision-makingprocesses and are applied almost automatically inresponse to external stimuli (Nelson & Winter,1982). When a new problem arises, the firm withinertia engages in local search for relevant experi-ences (Cyert & March, 1963; Gavetti & Levinthal,2000), which yields consistent responses. Hannanand Freeman (1984) noted that inertia elicits ac-countable, reproducible, and reliable organization-al outcomes and thus reduces uncertainty and vari-ability in accordance with March’s (1991) notion ofexploitation. In the context of alliances, inertialpressures encourage firms to rely on organizationalroutines for selecting partners, establishing alliancegovernance mechanisms, allocating resources, andcoordinating and monitoring alliances (Kale, Dyer,& Singh, 2002). Hence, inertia may independentlyinhibit exploration in one or more domains of alli-ance formation.

First, in the function domain, firms that committo existing technologies (Burgelman, 1994; Kelly &Amburgey, 1991) are less likely to explore newtechnologies through their alliances. Accordingly,Rothaermel (2001b) found that incumbents in thepharmaceutical industry benefited by exploitingcomplementary assets rather than by exploring newtechnologies with biotechnology partners. Hence,firms may tend to apply their existing knowledgerather than incur the extensive learning costs ofknowledge-generating R&D alliances. Inertial pres-sures to reduce technical uncertainty and organiza-tional risk further limit firms’ engagements inknowledge-generating alliances because these alli-ances entail substantially more interaction, collab-oration, and exchange of tacit knowledge than domarketing alliances (Rowley et al., 2000). Hence,organizational inertia may facilitate exploitation inthat domain. Second, inertia may reduce structureexploration by promoting partner-selection rou-tines that impel firms to enhance the predictability,stability, and reliability of their alliances. Firmscan pursue these objectives by forming recurrentalliances with prior partners that are instituted onfamiliarity, trust, and established collaborationpractices (Gulati, 1995a). Thus, inertia favors exist-ing partners despite the potential merits of newpartners, resulting in structure exploitation. Fi-nally, even when partner selection routines fail toyield relevant partners, firms may still leverage es-tablished routines to identify partners that match acertain profile. In so doing, they specialize andbecome efficient in managing alliances, thus rein-forcing attribute exploitation.

Pressures for exploration. Whereas inertiadrives firms’ tendencies to exploit, absorptive ca-pacity facilitates counterpressures by furnishing


the mechanism via which firms can identify theneed for and direction of exploratory activities. Ex-ploration is guided not only by inventing but alsoby learning from others (Huber, 1991; Levitt &March, 1988) and by employing external knowl-edge (March & Simon, 1958). Absorptive capacity,defined as the ability to value, assimilate, and ap-ply external knowledge (Cohen & Levinthal, 1990),helps firms identify emerging opportunities andevaluate their prospects, thus enhancing explora-tion. It adjusts firms’ aspiration levels, so that theybecome attuned to learning opportunities and moreproactive in exploring them. Indeed, prior researchhas demonstrated how absorptive capacity en-hances organizational responsiveness and directsscientific and entrepreneurial discovery (Deeds,2001; Rosenkopf & Nerkar, 2001). It also increasesthe likelihood of identifying external opportunitiesand can therefore lead to exploration in one ormore domains of alliance formation.

First, absorptive capacity motivates the searchfor new technologies and the assimilation of exter-nal knowledge, thus facilitating the formation ofknowledge-generating R&D alliances. Although ex-ternal knowledge (Huber, 1991) can be graftedthrough corporate acquisitions or employee recruit-ment, R&D alliances offer a cost-efficient and time-sensitive mode of learning (Kumar & Nti, 1998).Hence, absorptive capacity leads to function explo-ration. Second, absorptive capacity encourages thepursuit and assimilation of external knowledge andthus motivates firms to identify new partners thatcan furnish such knowledge. Firms can broadentheir knowledge bases by forming alliances withpartners with whom they have no prior ties. Fur-ther, absorptive capacity reinforces structure explo-ration since it enhances receptivity to externalknowledge and enables firms to apply and internal-ize the knowledge learned from new partners(Mowery, Oxley, & Silverman, 1996). For similarreasons, absorptive capacity encourages firms toenrich their knowledge bases by seeking partnersthat differ from their prior partners with respect toattributes such as size and industry focus under theassumption that these partners offer access tounique knowledge bases and experiences. Thus,absorptive capacity extends the range of partneringopportunities and enables firms to communicate,understand, and collaborate more effectively with adiverse group of partners (Lane, Salk, & Lyles,2001). Consequently, firms can experiment withnew and characteristically different partners.

The challenge of balance. The tension betweeninertia and absorptive capacity sheds light on theinherent trade-offs between exploration and exploi-tation that prevail within specific domains. These

trade-offs emerge not only because of the need toallocate resources for refinement versus develop-ment of technologies or because of the viability ofimmediate short-term returns versus uncertainlong-term returns (March, 1991). The challenge ofbalance in the function, structure, and attributedomains also derives from the fundamentally con-flicting domain-specific pressures imposed by in-ertia and absorptive capacity. Inertia reinforceslearning from firms’ own experience, but absorp-tive capacity enhances receptivity to externalknowledge and thus promotes learning from out-siders (Levitt & March, 1988). Whereas absorptivecapacity leads to variation, inertia leads to empha-sis on the selection and retention stages of thelearning cycle (Zollo & Winter, 2002). Unlike ab-sorptive capacity, which generates alternatives andmay result in long jumps, inertia compels localsearch and choice (Gavetti & Levinthal, 2000). Byinvesting in absorptive capacity firms attenuate in-ertial forces and constrain their ability to refine andenhance the efficiency of existing routines (March,1991). As firms develop organizational routinesand submit to inertial forces, they subdue theirabsorptive capacity and adaptability to unfoldingenvironmental events (Hannan & Freeman, 1984).Contradicting the normative assumption of bal-ance, we thus expect either exploration or exploi-tation to dominate alliance formation decisions atany given time in specific domains. Two questionsremain open: (1) Which activity is likely to domi-nate in each domain? and (2) How can firms bal-ance their exploration and exploitation activitieswhen forming alliances?

Path Dependencies in Exploration andExploitation within Domains

The tendency to underscore either exploration orexploitation within domains can be ascribed topath dependencies, whereby “a firm’s previous in-vestments and its repertoire of routines (its ‘his-tory’) constrain its future behavior” (Teece, Rumelt,Dosi, & Winter, 1994: 17). Path dependence in ex-ploitation emerges because inertia facilitates rou-tine-based experiential learning. Specifically, afirm’s routines represent persistent patterns of be-havior based on past experience (Nelson & Winter,1982) that are “the outcome of trial and error learn-ing and the selection and retention of prior behav-iors” (Gavetti & Levinthal, 2000: 113). Routines thatbecome parts of firms’ repertoires are likely to bethose that have been previously shown to producefavorable outcomes. In turn, these outcomes maylead to path dependence because the frequency ofemploying a routine increases its efficiency and the


likelihood of desirable outcomes, which in turnfurther reinforce its application (Levinthal &March, 1993; Levitt & March, 1988). Hence, firms’accumulated exploitation experience reinforces es-tablished routines within each domain. Inertia-driven exploitation is thus likely to intensify withfirms’ prior exploitation experience.

Absorptive capacity is path dependent as wellsince the ability to explore new opportunities andevaluate, understand, and acquire new knowledge,depends on firms’ past experience in relevantknowledge domains (Zahra & George, 2002). Themore extensive the scope of firms’ prior searchactivities has been, the more familiar they becomewith their external environments, and the moreeffectual their channels and mechanisms for ex-ploring external opportunities become. The broadknowledge base and attention to changing industryconditions and emerging technologies that evolvewith absorptive capacity motivate the search fornew technologies, experimentation, and learningfrom external sources (Levitt & March, 1988).Therefore, exploration tendencies, guided by ab-sorptive capacity, intensify with firms’ prior explo-ration experience.

In the function domain, the leveraging of existingtechnologies entails nurturing practices for coordi-nating joint marketing engagements, managing in-direct sales, and developing supply chain allianceprograms. Such practices typically rely on firms’accumulated experience in function exploitation,whereas experience in R&D alliances may hinderthe evolution of these practices because of the dis-crepancy between the natures of downstream andupstream alliances (Rowley et al., 2000). Firms thathave concentrated their efforts on forming down-stream alliances are likely to favor these alliancesover upstream alliances since this focus allowsthem to accumulate and apply their experience in arelevant context without encountering significantadjustment costs (Argote & Ophir, 2002). Similarly,firms that have previously leveraged their absorp-tive capacity to accumulate experience in manag-ing R&D alliances become more receptive towardexternal technologies and can establish the com-munication channels, knowledge acquisition, andassimilation procedures needed to further pursueknowledge-generating R&D alliances. Experiencein function exploration is therefore self-reinforcing.

Similarly, knowledge-sharing routines and rela-tional mechanisms that enhance collaboration andmitigate appropriation hazards in alliances are pri-marily partner-specific (Gulati et al., 2003). There-fore, the benefits arising from firms’ past invest-ments in relation-specific assets, trust building,and informal arrangements (Dyer & Singh, 1998),

while facilitating learning in subsequent allianceswith the same partner, cannot be applied as effi-ciently in alliances with other partners. Firms thataccumulate experience in recurrent alliances withthe same partners thus tend to leverage this expe-rience in structure exploitation. Structure explora-tion, in turn, is path dependent because firms thatfrequently work with new partners can develop theflexibility, receptivity, and diversity needed for in-teracting with unfamiliar partners and capitalizingon their potentially distinct knowledge bases.Structure exploration experience enhances the ca-pacity of firms to understand new partners andlearn from them, whereas firms that concentrate onforming recurrent alliances with the same partnersmay lack the versatility needed for forming alli-ances with partners who are not already membersof their immediate alliance networks.

Finally, the application of partner-selection rou-tines that favor partners who match a certain organ-izational profile becomes more prevalent amongfirms that have accumulated sufficient experiencewith prior partners that match that profile. By con-tinuously allying with a homogenous group of part-ners, firms can determine the merits of this prac-tice, and through experiential learning contributeto the evolution of these partner-selection routines(Levinthal & March, 1993). In turn, firms that haveaccumulated experience with a heterogeneousgroup of partners can develop broad absorptive ca-pacity for interacting and exchanging knowledgewith characteristically distinct partners. By over-coming potential epistemological impediments toeffective knowledge transfer, these firms are moti-vated to engage in attribute exploration. In sum,exploration and exploitation in alliance formationare self-reinforcing within each domain.

Hypothesis 1. Firms encounter path depen-dence in exploration and exploitation withinthe function, structure, and attribute domains,so that prior experience in exploration (exploi-tation) will reinforce the tendency to explore(exploit) within each domain.

Balancing Exploration and Exploitationacross Domains

Given the conflicting pressures imposed by iner-tia and absorptive capacity, balancing within do-mains is organizationally challenging and entailssubduing natural behavioral tendencies and cogni-tive constraints (Levinthal & March, 1993). Ratherthan fully dismissing the normative assumptionthat firms seek to balance exploration and exploi-tation, we argue that given organizational impedi-


ments, firms may avoid the inefficiencies thatemerge from seeking to reconcile exploration andexploitation within domains by pursuing alterna-tive forms of balance.

In particular, we suggest that firms may balanceexploration and exploitation across domains in al-liance formation decisions. Prior research has dem-onstrated that firms can coordinate exploration ef-forts in different areas, such as across technologicaland organizational boundaries (Rosenkopf &Nerkar, 2001) or across technological and geo-graphical domains (Rosenkopf & Almeida, 2003).Similarly, in accordance with our distinctionamong the function, structure, and attribute do-mains in alliance formation, the quest for balancemay motivate firms to explore in some of thesedomains while exploiting in others. For instance,firms may form recurrent R&D alliances (engagingin function exploration) with existing partners (en-gaging in structure exploitation) to leverage famil-iarity and established alliance management rou-tines. Such duality in firms’ tendencies is feasiblebecause the pressures of inertia and absorptive ca-pacity within domains may not necessarily conflictacross domains.

The organizational trade-offs between explora-tion and exploitation prevail primarily within do-mains, yet firms can simultaneously nurture organ-izational routines that regulate exploitation in onedomain while investing in absorptive capacity tosupport exploration in other domains. For exam-ple, the ability to conduct market research foremerging technologies, develop best practices forlearning from partners, and assimilate knowledgein the course of joint R&D alliances does notcounter firms’ efforts to develop long-term relation-ships, nurture interfirm trust, make relation-specific investments, and use informal governancemechanisms, which are essential in recurrent alli-ances with the same partners (Gulati, 1995a). Con-sidering the conflicting pressures imposed by iner-tia and absorptive capacity and firms’ inherenttendencies to specialize in either exploration orexploitation within each domain, firms maycounter their tendencies to explore in one domainby exploiting in another.

Moreover, firms that simultaneously exploreacross the three domains may face undesirable andperhaps unnecessary levels of uncertainty and risk.They face technical risk in new technology devel-opment as well as managerial challenges associatedwith the need to collaborate with unfamiliar ordiverse partners. In turn, firms that simultaneouslyexploit across all domains limit their search activ-ities and constrain potential technical and marketopportunities by focusing on the refinement of ex-

isting knowledge, fostering only established inter-firm ties, and restraining network heterogeneity,thus limiting long-term prospects. Thus, in view ofthe need for balance between exploration and ex-ploitation and the behavioral impediments to suchbalance within domains, we expect firms to exploitin some domains while exploring in others.

Hypothesis 2. Firms tend to balance explora-tion and exploitation across the function,structure, and attribute domains, so that thetendency to explore (exploit) in one domainwill be compensated by the tendency to exploit(explore) in some other domains.

Intertemporal Balancing of Exploration andExploitation within and across Domains

The conflicting pressures of inertia and absorp-tive capacity constrain firms’ abilities to simulta-neously balance exploration and exploitationwithin domains. The self-reinforcing trends as-cribed to prior experience in exploration or exploi-tation further limit firms’ abilities to evade thesetendencies. We thus expect exploration and exploi-tation trends within domains to be moderate, ratherthan punctuated or frequently changing (Romanelli& Tushman, 1994). Nevertheless, firms may still beable to balance exploration and exploitation bygradually shifting from one learning activity to theother within certain domains.

In so arguing, we follow field research that ex-plains how firms engage in time-paced transitionsthat enable them to operate in the present whileplanning and executing future organizationalchange through sequenced steps (Brown & Eisen-hardt, 1997). Our argument is also akin to the no-tion that subtle changes in firms’ perceptions oftheir environments improve sequential attentionand adaptation over time (Gavetti & Levinthal,2000) and that by modulating between discrete or-ganizational choices over time, firms may enjoytemporal efficiency unachievable through eitherchoice alone (Nickerson & Zenger, 2002). Hence, inthe organizational literature, simulation studieshave already demonstrated the merits of intertem-poral sequencing of different organizational struc-tures, suggesting that firms may engage in explora-tion followed by gradual refinement that dislodgesthem from their current developmental trajectories(Siggelkow & Levinthal, 2003). Such practices,therefore, enable firms to avoid competency traps(Levitt & March, 1988) and gradually balance ex-ploration and exploitation over time. Stated differ-ently, firms may strive for balance by exploring at acertain point in time and then diligently shifting


toward exploitation or vice versa. The transitionbetween exploration and exploitation requiresfirms that have traditionally followed establishedroutines to enhance their absorptive capacity,while firms that have developed expertise in iden-tifying external opportunities and capitalizing onnew knowledge must regulate some organizationalprocedures that improve efficiency.

In the context of alliance formation, firms canbalance exploration and exploitation over time bygradually adjusting their tendencies to explore orexploit within each domain. One possible path mayinvolve a gradual shift toward exploitation as firmsconclude early R&D efforts and proceed to commer-cialization and production (Rothaermel & Deeds,2004). Thus, firms initially engage in knowledge-generating R&D alliances and progressively shift tomarketing alliances. An alternative path from ex-ploitation to exploration may emerge if firms ex-haust current initiatives and re-engage in techno-logical exploration. Similarly, firms that haveengaged in recurrent alliances with a select groupof partners may realize that they have fully lever-aged their existing relationships and begin tosearch for partnering opportunities with new andpossibly diverse partners. They can experimentwith a small number of new partners without nec-essarily jeopardizing their existing relationships. Inturn, firms that have engaged in ad hoc allianceswith occasional partners can gradually rationalizetheir alliance networks and enter recurrent alli-ances with selected partners. For example, Unisys,an information technology firm, quadrupled itsnumber of alliances between 1990 and 1995 byforming ad hoc reseller relationships with newpartners. Since 1995, its number of alliances hasremained stable, but Unisys began to engage moreextensively in recurrent initiatives, joint activities,and systems integration projects with existing part-ners (Lavie, 2004). This pattern illustrates a gradualshift to structure exploitation.

Finally, firms may seek organizationally distinctpartners to balance homogeneous networks, or in-stead, begin concentrating on partners that match acertain profile. The transition to attribute explora-tion cannot be consummated instantaneously,however, because an isolated alliance formationevent cannot drastically alter the characteristic pro-file of a firm’s prior partners. Additionally, theadjustment of absorptive capacity needed for capi-talizing on such diversity is time-consuming. Thus,the balancing of exploration and exploitationwithin domains is likely to occur gradually. Tran-sitions within domains may be feasible only overprolonged periods of time because self reinforcing

pressures of inertia and absorptive capacity decel-erate them.

Hypothesis 3. Firms tend to gradually balanceexploration and exploitation within the func-tion, structure, and attribute domains, so thatthey shift from exploration to exploitation orvice versa over time.

So far we have argued that firms may overcomeinherent path dependencies in exploration and ex-ploitation over time within domains. The remain-ing question is which direction such transitions arelikely to take. We next posit that the temporal ex-ploration and exploitation trajectories within do-mains are interdependent across domains, so thatfirms can achieve a balance even when at any giventime they face conflicting tendencies within andacross domains.

Under the normative assumption that firms seekto balance exploration and exploitation (March,1991), a shift between exploration and exploitationwithin a certain domain, even if promoting localequilibrium within that domain, may steer firmsaway from global equilibrium. Under these circum-stances, firms can compensate for such deviationby countering exploration tendencies in one do-main with exploitation tendencies in another. Forexample, a firm that shifts its focus from R&D alli-ances to marketing alliances over time may inten-sify its search for new partners and thus balanceincreasing tendencies to exploit in the functiondomain with tendencies to explore in the structuredomain. The coordination of these trends acrossdomains would enable the firm to conserve its in-vestments in the development of organizationalroutines and in the nurturing of absorptive capac-ity. Instead of countering inertial forces in one do-main and weakening absorptive capacity in an-other, firms may divert their inertial forces andabsorptive capacity from one domain to another tothe extent that the corresponding investments androutines are somewhat fungible across domains.For example, firms that have developed informa-tion channels for identifying prospective partnerscan rely, to an extent, on the same channels forgathering information on the organizational at-tributes of new and existing partners. By simulta-neously balancing exploration and exploitationacross domains and over time, firms strive towardan overall balance between exploration andexploitation.

Hypothesis 4. Firms tend to simultaneouslybalance exploration and exploitation over timeand across domains, so that over time, in-creases in the level of exploration (exploita-


tion) in one domain will be compensated bydecreases in the level of exploration (exploita-tion) in some other domains.

METHODS

Research Setting and Sample

We designed our study as a pooled time seriesanalysis of alliances formed by U.S. software firms.The U.S. software industry (SIC codes 7371-7374)offered a suitable setting because intensive allianceformation in this industry enhanced the variance,reliability, and meaningfulness of variables. Also,the high proportion of publicly traded firms madefinancial information readily available and attenu-ated potential size- and age-related biases. Finally,our sample was representative, since the world-wide software industry has been dominated byUnited States–based firms.

The study’s time frame spanned the years 1990 to2001, although we tracked alliances back to 1985when computing variables such as partnering ex-perience and exploration experience that requiredinformation on historic alliances. The five-yearwindow was used under conventional assumptionsin alliance research (Stuart, 2000). The initial sam-ple of focal firms included all 547 publicly tradedUnited States–based software firms that were activein 2001. Because the analysis was longitudinal, 170firms with less than five years of COMPUSTATrecords were discarded. We also eliminated fivesubsidiaries of other sampled firms and five firmsthat had no alliances. Because of missing data, thelagging of independent variables, and the require-ment for a minimum number of observations perfirm in the computation of some variables, the ef-fective sample size ranged between 252 and 337firms. Selection bias was ruled out in view of lackof differences between the 337 sampled firms andthe remaining 314 public firms in the industry.2

Data Collection

Following Anand and Khanna (2000b), we firstrelied on the Securities Data Corporation (SDC)database in compiling records of alliances formedby each focal firm between 1985 and 2001. We thencorrected these records by searching alliance an-nouncements and status reports in press releasesusing the Factiva database, corporate Web sites,and Securities and Exchange Commission (SEC)filings accessed through the Edgar database. Mostalliance announcements were cross-validated by atleast two sources. By relying on multiple sourcesand tracking follow-up announcements and statusreports, we minimized the occurrence of alliancesthat were announced but not realized. To furthervalidate our data, we reviewed some of our alliancelistings with a select group of corporate executivesin charge of alliances. Following these procedures,alliance records were corroborated, corrected,added, or eliminated. For instance, we droppedseveral resale, licensing, and supply relationshipsthat resembled arm’s-length transactions ratherthan collaborative alliances.

Overall, we identified 19,928 alliances involving8,469 unique partners. On average, a focal firmformed 58.96 alliances between 1990 and 2001.Only 24.7 percent of the identified alliances werereported in the SDC database. Unlike Anand andKhanna (2000b), we retained the additional recordssince we employed the firm-year rather than thealliance as the unit of analysis; in our case, elimi-nation of records could have biased measures thatentailed complete “ego-network” data.3, 4

2 The remaining firms included inactive firms, firmswith fewer than five years of COMPUSTAT records, andfirms headquartered in foreign countries. Insignificantdifferences were found in total assets (t � 1.19, p � .23),revenues (t � 0.49, p � .63), number of employees (t �0.01, p � .99), R&D expenses (t � 1.02, p � .31), selling,general, and administration expenses (t � 1.14, p � .25),net income (t � 1.11, p � .27), operating income (t �1.14, p � .25), cash (t � 1.57, p � .12), long-term debt (t ��0.08, p � .94), earnings per share (t � 1.45, p � .15),and other measures. These results suggested that oursample was representative of public firms in the softwareindustry.

3 Only 72 alliances were identified during the years1985–89, an annual average of 0.21 alliances per firm;during 1990–2001, the average annual number of alli-ances was 4.91. This difference in averages reflects thesurge in alliance formation during the 1990s and the factthat many of the firms in our sample did not commenceoperations before 1990. In fact, only 22 firms engaged inalliances between 1985 and 1989, thus mitigating con-cerns about potential “left-censoring” in the calculationof partnering experience and structure exploration as aresult of exclusion of alliances formed prior to 1985.

4 When comparing the proportions of different types ofalliance agreements in our final sample to those origi-nally reported in SDC, we found that our data offeredmore extensive coverage of nonequity alliances (t �25.85, p � .001) and alliances with foreign partners (t �25.73, p � .001). The proportions of marketing (t � 34.36,p � .001), original equipment manufacturing (OEM) orvalue-added reseller (VAR) (t � 22.89, p � .001), andR&D (t � 36.17, p � .001) agreements were also higher inour data than in the SDC data, but the proportions ofsupply (t � �4.16, p � .001), licensing (t � �26.87, p �


For each alliance, we coded the announcementdate, partners’ identities, public status, and agree-ment types. An alliance could involve more thanone type of agreement. The reliability of our codingprocedure was enhanced by having one of the au-thors complete all the coding drawing on detailedguidelines that included alliance definitions,search techniques, and coding schemes. Interraterreliability reached 97.8 percent when a researchassistant repeated the procedure for a subsample ofsix randomly selected firms. Firm- and partner-specific data, such as annual historical SIC code,total assets, revenue, long-term debt, R&D ex-penses, and net income, were extracted from COM-PUSTAT, which served as a single source for archi-val data, thus enhancing the reliability of ourmeasures. The 2,777 publicly traded partners in thesample accounted for 63.6 percent of the alliances,thus limiting potential biases that may have arisenfrom the lack of financial information for privatepartners. This missing information did not affectour measures, with the exception of the attributeexploration variable, which relied on financial in-formation and could be calculated more accuratelywhen the proportion of publicly traded partnerswas higher.5

We considered the firm-year the unit of analysisas our dependent variables captured firm-level ten-dencies. Thus, we transformed the data for the19,928 alliances to 2,451 firm-year observations bypooling the data across all alliances formed by eachfocal firm in a given year. The effective sample sizein multivariate analysis ranged between 972 and1,946 observations because of the operation-alization of our measures and missing values.6

Dependent Variables

We operationalized exploration-exploitationwith a combined continuous measure rather thanwith two separate indicators under the assumptionthat exploration inhibits exploitation and viceversa, so that these two activities conflict (Abern-

athy, 1978; March, 1991). This assumption wasconsistent with the significant, negative correlationthat we observed between, for example, upstreamand downstream alliance formation in the functiondomain (r � �.71, p � .001).

Function exploration. We followed Koza andLewin’s (2000) distinction between exploration, ex-ploitation, and hybrid alliances that integratedownstream and upstream activities. From allianceannouncements, we coded a categorical indicatorof whether each alliance involved a knowledge-generating R&D agreement (coded 1); an agreementbased on existing knowledge involving joint mar-keting and service, OEM/VAR, licensing, produc-tion, or supply (0); or a combination of R&D andother agreements (0.5). Unlike internal R&D thatdraws directly from a firm’s existing knowledge,R&D agreements in the software industry entailmoving outside of the firm’s technical knowledgebase or at least integrating internal knowledge withthe external knowledge of partners, thus represent-ing exploration. The following is an example of anannouncement of an alliance we classified as anR&D agreement:

Business Wire. 12 March 2001—Cadence DesignSystems, Inc. and Agere Systems today announcedthe formation of a strategic alliance to develop chipinput/output (I/O) planning capability. This tech-nology alliance will lead to the development of aunique methodology that will promote the co-de-sign of integrated circuits (ICs) and IC packaging tospeed time-to-market. Cadence and Agere haveteamed to develop this new technology, to helpclose the gap in the design flow between IC designand IC packaging environments. With its experiencein high-speed design, Agere is an excellent ally inthe co-development of this new technology. Theagreement includes a contractual commitment byAgere to provide Cadence with engineering re-sources for a two-year period. Cadence will deliverto Agere functionality that is based on a jointly-developed product requirement specification.

It is worthwhile clarifying that we completed thecoding taking the perspective of the focal firm. Forexample, when a focal firm marketed a solutiondeveloped by its partner without engaging in jointR&D efforts, the alliance was coded as a marketingagreement rather than an R&D agreement. Our func-tion exploration measure was calculated as the av-erage value of the alliance agreement indicatoracross all alliances formed by firm i in year t. Val-ues ranged from 0 to 1; high values indicated func-tion exploration, whereas low values indicatedfunction exploitation.

Structure exploration. For each alliance formedby firm i, an indicator received a value of 1 if the

.001), and royalties (t � �2.03, p � .05) agreements werelower. These results rule out the possibility that the SDCdatabase covers more substantial types of alliances.

5 Compared to private partners, public partners en-gaged in more strategic long-term alliances (t � 25.61,p � .001) and favored joint ventures (t � 6.22, p � .001)and R&D alliances (t � 29.69, p � .001) over marketingalliances (t � �22,43, p � .001).

6 Missing values occurred, for instance, because SECregulations did not require firms to report R&D invest-ments and because financial information was unavail-able for privately owned partners.


firm had no joint prior alliances with partner j and0 if such alliances existed. For each firm, structureexploration was calculated as the average value ofthis indicator across all alliances formed by firm iin year t. In order not to classify a firm’s first alli-ance as structure exploration by default, we ex-cluded 181 firm-year observations correspondingto years in which firms formed their first and onlyalliances. Auxiliary analysis revealed, however,that our findings remained unchanged when theseobservations were retained. Our findings were alsorobust to alternative operationalizations, such asaveraging counts of prior ties as opposed to dummyindicators. Values again ranged from 0 to 1, withhigh values indicating structure exploration andlow values, structure exploitation.

Attribute exploration. To enhance construct va-lidity, we incorporated multiple partner attributesin our attribute exploration measure. We calculatedfour indicators representing distinctive partner at-tributes in the year of alliance formation, includingpartners’ size (asset value), propensity to invest inmarketing (advertising intensity), financial strength(logarithm of the ratio of cash to long-term debt),and industry focus (four-digit SIC code). For eachalliance, with respect to the first three attributes(k � 1,2,3), we calculated the absolute differencebetween the attribute of the partner and the averageattributes of the ten prior partners of the firm using

the formula PADjk � � PAjk �1

10�p � j � 10

j � 1

PApk�,

where PAjk is the value of attribute k of partner j.For the partner industry measure (k � 4), we useda dummy indicator receiving a value of 1 when theprimary four-digit SIC code of the new partner wasnot included in the list of primary SIC codes of theten prior partners of the firm, and 0 otherwise.7 For

each attribute k, we then calculated the partnerattribute difference measure (PADikt) as the averagePADjk across all alliances formed by firm i in year t.Since the four PADikt measures were not signifi-cantly correlated (average interitem correlation �.02), we used Euclidian distance rather than factorscores to compute attribute exploration. Beforecomputing this variable, we standardized eachPADikt measure by subtracting its mean value anddividing the result by its standard deviation. Thisprocedure produces PADikt measures that werecomparable across attributes. Finally, we applied

the Euclidean distance formula ��k � 1

4

sPAD2ikt to

compute the aggregated attribute exploration vari-able for firm i in year t. Using linear transformation,we normalized this variable to range between 0 and1, with high values indicating attribute explorationand low values indicating exploitation.8

Independent Variables

To test Hypothesis 1, we calculated firms’ accu-mulated exploration experience within each do-main for each firm-year. We relied on the sameformulas used for constructing our explorationmeasures, but instead of incorporating the alliancesformed by firm i in year t, we counted all thealliances formed by that firm between 1985 and thepreceding year (t � 1). We preferred this measure tosuch alternatives as one-year lagged explorationvariables because we were interested in the overalltendencies of firms to engage in exploration andexploitation over time rather than in their tempo-rary tendencies relative to a preceding year. Tostudy balance across domains (Hypothesis 2), weincorporated simultaneous measures of explorationin alternative domains when testing for explorationtendencies in a given domain. Finally, we mea-sured time on the basis of the year in which alli-ances were formed. For each firm-year, our time

7 The restriction to ten prior alliances overcame a po-tential bias, as the accuracy of measures depended on thenumber of prior alliances. The more alliances formed inthe past, the more stable the measure of deviation frompast behavior. Thus, records relating to firms with part-nering histories comprising fewer than 10 alliances wereexcluded, and moving averages based on a consistentpartnering history window were calculated for the re-maining observations. We experimented with differentwindows ranging from 5 to 15 prior alliances, and theresults were robust within this range. The history win-dow of 10 alliances was selected because measures be-came less stable with shorter history windows, but thenumber of omitted observations became substantial withlonger history windows. The choice of this particularhistory window was also derived from assumptionsabout organizational memory and the adaptation processbased on a firm’s partnering history. Finally, in a supple-

mental analysis, we incorporated additional indicators ofattribute exploration based on the governance mode andstrategic significance of alliances as well as on partners’countries of origin and R&D investments. The findingswere consistent but less significant with these alternativemeasures.

8 We also considered an alternative measure of at-tribute exploration based on differences between partnerattributes and focal firm attributes instead of focusing ondifferences across partners. This measure produced nosignificant findings, possibly because firms can engagenot only in exploration but also in exploitation by spe-cializing in forming alliances with partners that are or-ganizationally distinct from themselves.


clock ranged between 1 and 12, with 1 correspond-ing to 1990 and 12 corresponding to 2001.

Control Variables

Using COMPUSTAT data, we controlled for var-ious time-variant, firm-specific factors that mightinfluence tendencies to engage in exploration andexploitation. Firm size, which has produced conflict-ing impacts on exploration-exploitation in prior stud-ies (Beckroan et al., 2004; Rothaermel & Deeds, 2004),was measured as the value of a firm’s assets in apreceding year. Additionally, we controlled for firms’R&D intensity in the preceding year, a variable repre-senting their innovation capacity and internal explo-ration efforts that may affect external exploration ac-tivities through alliances. We controlled for firm agein a preceding year because, as firms mature andbecome more dependent on their established routinesand skills (Hannan & Freeman, 1984), they grow lesslikely to change their strategic orientations (Kelly &Amburgey, 1991) and engage in exploration. Simi-larly, we controlled for prior partnering experience,which has been associated with organizational inertia(Li & Rowley, 2002) and might account for path de-pendence in alliance formation decisions in the struc-ture domain (Chung et at., 2000; Gulati & Gargiulo,1999). Prior alliances might also contribute to firms’innovativeness (Ahuja, 2000; Tsai, 2001) and inter-nalization of partners’ capabilities (Mowery et al.,1996), thus enhancing function exploration. Follow-ing prior research (e.g., Anand & Khanna, 2000a; Ho-ang & Rothaermel, 2005), for each firm-year we com-puted partnering experience as a count of all prioralliances formed by a focal firm with any partnerbetween 1985 and the preceding year.

In addition, we controlled for firms’ past finan-cial performance, which might reinforce estab-lished routines and drive out exploration(Levinthal & March, 1993). Firm profitability, anaccounting measure of performance, was based onthe ratio of net income to total assets in the preced-ing year. We also controlled for firm solvency, mea-sured with the log-transformed ratio of cash tolong-term debt in the preceding year, since theavailability of financial funds may facilitate exper-imentation and slack-induced search (Bourgeois,1981; Levinthal & March, 1981). Additionally, wecontrolled for firms’ merger and acquisition activ-ity, following prior research that has identified ac-quisitions as an alternative to alliance formation(Hennart & Reddy, 1997; Koza & Balakrishnan,1993; Villalonga & McGahan, 2005). We countedthe number of targets that each firm acquired ormerged with in a given year, using SDC data. Fi-

nally, intertemporal trends were controlled for byincluding a series of year dummy variables.

Analysis

Table 2 reports descriptive statistics and correla-tions, and Table 3 reports the results of our analysisof the panel data using cross-sectional time seriesregressions with random-effects models and gener-alized least square (GLS) estimators.9 We ruled outconcerns of potential autocorrelation in the data onthe basis of the Wooldridge (2002) test, ensuringthat our findings were insensitive to the incorpora-tion of first-order autoregressive errors generatedfrom an AR(1) process. Our findings were also ro-bust to the use of maximum-likelihood estimators(MLE) instead of GLS estimators. Variance inflationfactors were considerably lower than the criticalvalue, thus ruling out potential multicolinearity.We treated missing values with listwise deletion,which accounts for the variation in model samplesizes. Wald chi-square fit statistics are reportedwith the results of our GLS models in Table 3.

For each dependent variable, we report hierar-chical models in Table 3. We tested our hypotheseswith the full models (4a, 4b, and 4c), in which theyear dummies were replaced with a continuousvariable that allowed us to test Hypothesis 3. Wetested this hypothesis by verifying that the timeeffect was significant, monotonic, and negative insign if the predicted level of the dependent variableindicated initial exploration or positive in sign if itindicated initial exploitation. We tested Hypothe-sis 4 by comparing the valence of the time effectsacross domains.

RESULTS

Table 2 reveals low correlations among the depen-dent variables that are consistent with the decompo-sition of the exploration-exploitation construct

9 We followed prior research (Beckman et al., 2004;Gulati, 1995b; Stuart, 1998) in using random-effects mod-els because, unlike these efficient models, fixed-effectsmodels severely reduce degrees of freedom and may gen-erate unstable results for panels over short time periods(our panels included between 3.9 to 5.8 observations perfirm on average). In addition, fixed-effects models pre-dict annual changes in dependent variables, whereas wewere primarily interested in explaining overall explora-tion-exploitation. Moreover, fixed-effects models wouldhave excluded our time variable, as each observation isuniquely identified by a firm-year combination. Hence,the use of fixed-effects models in this study would haveseverely constrained our analysis.


TA

BL

E2

Des

crip

tive

Sta

tist

ics

and

Cor

rela

tion

s

Var

iabl

esn

Mea

ns.

d.

Min

i-m

um

Max

i-m

um

12

34

56

78

910

1112

13

1.F

un

ctio

nex

plo

rati

on2,

587

0.46

0.35

01

2.S

tru

ctu

reex

plo

rati

on2,

406

0.89

0.19

01

�.1

0***

3.A

ttri

bute

exp

lora

tion

1,34

80.

200.

100

1�

.04

�.0

14.

Fir

msi

zet

�1

2,21

60.

381.

480

52.1

5.0

2�

.09*

**.0

7*5.

Fir

mR

&D

inte

nsi

tyt

�1

1,96

50.

351.

100

21.9

8�

.00

.01

.07*

�.0

5*6.

Fir

mag

e t�

12,

578

13.0

58.

290

60�

.04†

�.0

3�

.05

.08*

**�

.12*

**7.

Par

tner

ing

exp

erie

nce

t�

12,

587

22.2

455

.37

01,

301

.07*

**�

.22*

**.0

8**

.53*

**�

.03

.11*

**8.

Fir

mp

rofi

tabi

lity

t�

12,

210

�0.

210.

77�

10.0

76.

68.1

2***

�.0

3�

.09*

**.0

8***

�.2

5***

.17*

**.0

7***

9.F

irm

solv

ency

t�

12,

204

4.59

4.51

�12

.07

15.4

2.0

7***

�.1

1***

�.0

4�

.08*

**.0

0�

.02

.19*

**.1

3***

10.

Acq

uis

itio

ns t

2,58

70.

822.

020

44.0

2�

.09*

**.0

4.4

7***

�.0

6**

.14*

**.4

7***

.13*

**.0

211

.F

un

ctio

nex

plo

ra-

tion

exp

erie

nce

t�

1

2,24

40.

470.

280

1.2

8***

�.1

2***

�.0

5†.0

1.0

5*�

.05*

.09*

**.0

5*.0

7**

.02

12.

Str

uct

ure

exp

lora

-ti

onex

per

ien

cet

�1

2,07

00.

930.

090.

431

�.1

4***

.25*

**�

.00

�.1

8***

�.0

2�

.03

�.4

2***

�.0

6**

�.1

7***

�.1

6***

�.2

2***

13.

Att

ribu

teex

plo

ra-

tion

exp

erie

nce

t�

1

1,14

00.

270.

080

1�

.11*

**.0

2.2

4***

.06†

.12*

**�

.02

�.0

0�

.07*

�.0

4.0

5†�

.24*

**.1

1***

14.

Tim

e2,

587

7.99

3.05

112

�.0

7***

�.0

6**

.19*

**.0

4†.0

8***

.20*

**.2

4***

�.1

1***

.13*

**.0

7***

�.1

1***

�.1

1***

.15*

**

†p

�.1

0*

p�

.05

**p

�.0

1**

*p

�.0

01T

wo-

tail

edte

sts.

TA

BL

E3

Res

ult

sof

Ran

dom

-Eff

ects

GL

SM

odel

sfo

rE

xplo

rati

on-E

xplo

itat

ion

inA

llia

nce

For

mat

ion

a

Ind

epen

den

tV

aria

bles

Fu

nct

ion

Exp

lora

tion

Str

uct

ure

Exp

lora

tion

Att

ribu

teE

xplo

rati

on

Mod

el1a

Mod

el2a

Mod

el3a

Mod

el4a

Mod

el1b

Mod

el2b

Mod

el3b

Mod

el4b

Mod

el1c

Mod

el2c

Mod

el3c

Mod

el4c

1990

0.01

0.06

0.04

0.03

0.03

0.08

�0.

13†

�0.

06�

0.06

1991

0.04

0.06

†�

0.00

�0.

01�

0.00

�0.

04�

0.07

�0.

10�

0.10

1992

0.06

*0.

06*

�0.

00�

0.04

�0.

04�

0.11

*�

0.10

*�

0.10

†�

0.11

†

1993

0.01

0.03

0.05

�0.

08**

�0.

10**

�0.

06�

0.07

†�

0.09

†�

0.10

*19

940.

040.

05†

0.07

*�

0.02

�0.

03�

0.10

*�

0.14

***

�0.

14**

�0.

15**

1995

0.07

**0.

07*

0.13

***

�0.

09**

�0.

10**

*�

0.05

�0.

11**

�0.

08†

�0.

09*

1996

0.10

***

0.09

***

0.08

**�

0.03

�0.

03�

0.02

�0.

12**

*�

0.11

**�

0.12

**19

970.

09**

*0.

08**

0.04

�0.

01�

0.01

0.01

�0.

14**

*�

0.14

***

�0.

14**

*19

980.

030.

03�

0.02

�0.

06*

�0.

07*

�0.

07*

�0.

09**

�0.

07*

�0.

08*

1999

0.04

0.04

�0.

02�

0.06

*�

0.06

*�

0.04

�0.

04�

0.04

�0.

0420

000.

04†

0.03

0.00

�0.

03�

0.02

�0.

020.

010.

000.

00F

irm

size

t�

1�

0.01

�0.

01�

0.02

�0.

020.

07*

0.04

0.04

0.05

�0.

02�

0.03

�0.

03�

0.02

Fir

mR

&D

inte

nsi

tyt

�1

0.01

0.00

�0.

01�

0.01

0.01

0.01

0.03

0.03

0.04

0.01

0.01

0.00

Fir

mag

e t�

1�

0.02

�0.

02�

0.01

�0.

010.

000.

01�

0.02

�0.

02�

0.08

*�

0.07

�0.

07�

0.07

Par

tner

ing

exp

erie

nce

t�

10.

10**

0.10

**0.

06*

0.08

*�

0.32

***

�0.

18**

*�

0.13

**�

0.13

***

0.09

†0.

070.

060.

05F

irm

pro

fita

bili

tyt

�1

0.10

***

0.10

***

0.07

*0.

07*

0.01

0.02

0.03

0.03

�0.

09*

�0.

08�

0.09

�0.

08F

irm

solv

ency

t�

10.

000.

00�

0.01

�0.

01�

0.06

*�

0.04

†�

0.04

�0.

04�

0.04

�0.

05�

0.05

�0.

06A

cqu

isit

ion

s t�

0.01

�0.

01�

0.00

�0.

00�

0.01

�0.

000.

020.

010.

020.

030.

030.

02

Fu

nct

ion

exp

lora

tion

exp

erie

nce

t�

1

0.14

***

0.30

***

0.29

***

Str

uct

ure

exp

lora

tion

exp

erie

nce

t�

1

0.20

***

0.20

***

0.20

***

Att

ribu

teex

plo

rati

onex

per

ien

cet

�1

0.08

*0.

09*

0.09

*

Fu

nct

ion

exp

lora

tion

t�

0.09

*�

0.08

*0.

060.

06S

tru

ctu

reex

plo

rati

ont

�0.

05*

�0.

04*

�0.

02�

0.02

Att

ribu

teex

plo

rati

ont

0.01

0.00

�0.

01�

0.01

Tim

e�

0.12

***

0.08

*0.

24**

*

n,

firm

-yea

rs1,

946

1,81

11,

188

1,18

81,

885

1,71

41,

187

1,18

71,

192

972

972

972

n,

firm

s33

732

927

127

133

032

127

127

127

125

225

225

2O

bser

vati

ons/

firm

5.8

5.5

4.4

4.4

5.7

5.3

4.4

4.4

4.4

3.9

3.9

3.9

Wal

d�

253

.29*

**88

.43*

**17

6.88

***

141.

83**

*14

5.24

***

175.

95**

*12

2.75

***

109.

35**

*67

.42*

**48

.37*

**51

.13*

**45

.29*

**R

2-w

ith

in0.

020.

000.

010.

000.

020.

000.

010.

000.

070.

040.

040.

03R

2-b

etw

een

0.05

0.27

0.39

0.38

0.22

0.48

0.31

0.31

0.02

0.07

0.08

0.08

R2-o

vera

ll0.

040.

090.

190.

170.

070.

090.

100.

090.

060.

080.

080.

07

aS

tan

dar

diz

edco

effi

cien

tsar

ere

por

ted

.†

p�

.10

*p

�.0

5**

p�

.01

***

p�

.001

Tw

o-ta

iled

test

s.

to distinctive domains. Specifically, our data dis-prove the premise that the identity of partners (ex-isting versus new) dictates the extent to which theycharacteristically differ from prior partners. Themean values of the dependent variables indicatetendencies toward structure exploration (y � 0.89)and attribute exploitation (y � 0.20).

However, exploration and exploitation are rela-tively evenly represented in the function domain(y � 0.46). In addition, the high correlations amongfirm size, acquisitions, and partnering experiencesuggest that large firms with substantial partneringexperience engage more extensively in mergers andacquisitions. Our models reveal overall fit statistics(R2s) ranging between 0.04 and 0.17. However, theyare more powerful in explaining variation in explo-ration-exploitation across firms, as reflected in val-ues for R2-between ranging from 0.05 to 0.48.

With respect to the control variables, our resultssuggest that prior partnering experience is associatedwith exploration in the function domain (� � 0.08,p � .05) and exploitation in the structure domain(� � �0.13, p � .001). Supporting Beckman et al.(2004), we found that firms with extensive partneringexperience were more likely to seek prior partners fortheir new alliances. However, experienced firms alsotended to engage more extensively in R&D alliances,which entail greater risk, resource commitment, andinteraction than marketing alliances (Rowley et al.,2000). Perhaps for similar reasons, function explora-tion was also positively related to profitability (� �0.07, p � .05). This finding is consistent with theassertion that resource-poor firms tend to function-ally exploit rather than explore in dynamic markets(Park et al., 2002).

In support of Hypothesis 1, model 4 reveals ten-dencies for imbalance within each domain. Specif-ically, experience in function exploration rein-forces exploration in that domain (� � 0.29, p �.001); structure exploration experience leads tostronger exploration in the structure domain (� �0.20, p � .001); and experience with diverse part-ners—that is, attribute exploration experience—in-creases the tendency to explore in the attributedomain (� � 0.09, p � .05). These effects persistedeven when we controlled for exploration in alter-native domains and were insensitive to the incor-poration of first-order autocorrelation regressors.

In keeping with Hypothesis 2, we found a nega-tive association between function exploration andsimultaneous structure exploration (� � �0.04, p �.05) and vice versa (� � �0.08, p � .05). Thesefindings suggest that firms that concentrate onforming R&D alliances also engage in a greaternumber of alliances with prior partners, whereasthose that frequently experiment with new partners

favor downstream alliances.10 Still, Table 3 showsno significant association between function orstructure exploration and tendencies to explore inthe attribute domain. Nevertheless, as Figure 1 il-lustrates, firms in our sample tended to compensatefor exploration in the structure domain with ex-ploitation in the attribute domain. Systematic dif-ferences were also observed between function andattribute exploration, suggesting balance acrossdomains.11

Additionally, in support of Hypothesis 3, wefound that firms tended to modify their explorationtendencies over time irrespective of the reinforcinginfluences of exploration experience and the avail-ability of alternative exploration modes. As the sig-nificant time effects indicate, function exploitationincreases over time (� � �0.12, p � .001), whileexploration intensifies in the structure (� � 0.08,p � .05) and attribute (� � 0.24, p � .001) domains.The time trends in the function and attribute do-mains are fully consistent with our hypothesis be-cause they demonstrate a decrease in an initiallyhigh level of exploration in the function domain(y1990 � 0.61, � � 0) and an increase in an initiallylow level of exploration in the attribute domain(y1990 � 0.13, � � 0). In the structure domain,however, the trend suggests intensifying explora-tion (See Figure 1).12 We verified the monotonicity

10 We ruled out the alternative explanation that thesefindings could be fully ascribed to the prevalence of R&Dconsortia in the software industry that frequently engagethe same partners in recurrent R&D alliances. The nega-tive association between activities in the function andstructure domains remained significant when we con-trolled for the average number of participants in alliancesor excluded multipartner alliances (9.05 percent of theannounced alliances).

11 We verified that the differences in levels of explo-ration-exploitation across domains were statistically sig-nificant using the Friedman’s distribution-free test formultiple pairwise comparisons. This nonparametric testwas appropriate because a Shapiro-Wilk test revealedthat our dependent variables violated the normality as-sumption (W � 0.99, 0.85, 0.76, p � .001). We ran thistest for subsamples segregated by year in order not toviolate the independence assumption. Its results indi-cated differences across all three domains (p � .001).Since this test entailed listwise deletion in unbalancedsamples, we also ran the Tukey-Kramer test, which al-lows for unequal sample sizes but assumes a Gaussiandistribution. Finally, we verified that our results wererobust under the Dunnett T3 pairwise comparisons test,for which unequal variances of variables is assumed.These results are available from the authors.

12 We ruled out the possibility that the trend in thestructure domain was driven by the increasing popular-


of the trends in the three domains by introducingquadratic and cubic terms of the time variable,which turned out insignificant.13

Finally, the fact that the time effects in model 4

were positive in the structure and attribute do-mains while negative in the function domain sug-gests that, supporting Hypothesis 4, over time firmstend to balance tendencies to explore in one do-main by exploiting in other domains.14 In particu-lar, the exacerbated tendency to explore in thestructure domain could be understood by consid-ering its balancing effect on intensifying exploita-tion in the function domain (see Figure 1). Hence,the divergence in the direction of intertemporaltrends across domains suggested that firms simul-taneously balance exploration and exploitationacross domains and over time.

DISCUSSION

The fertile existing research on exploration-ex-ploitation adopts March’s (1991) balance hypothe-sis but falls short of furnishing empirical evidencethat firms indeed conform to this expected behav-ior. Our findings call into question the idea thatfirms can balance exploration and exploitationwithin given domains, thus accounting for their

ity of alliance formation in the software industry, whichmight have accounted for increasing accessibility of newpartners over time, by introducing a control for the an-nual average number of alliances formed in this industry.This auxiliary analysis revealed no significant effect ofthis popularity measure on structure exploration. Wealso verified that the time effects could be ascribed tofirms’ tendencies rather than to entry of new firms intothe industry by incorporating an interaction between thetime variable and a dummy moderator indicatingwhether a firm was an incumbent that operated before1990. This interaction was insignificant, but the directtime effect remained significant.

13 We also tested monotonicity by replacing the con-tinuous time variable with a dummy variable represent-ing a transition in a specific year. The analysis was re-peated for transitions in 1990–2000. The transitioncoefficients produced for each year had the same valancein the function domain after 1993 (indicating exploita-tion tendencies); had the same valence in the structuredomain after 1992 (exploration tendencies); and had thesame valence throughout the study’s time frame in theattribute domain (exploration tendencies). Inconsistentcoefficients were insignificant. These results are avail-able from the authors.

14 Consistent results of Friedman and Tukey-Kramertests for differences across domains in consecutive yearsoffered additional support. These results are availablefrom the authors.

FIGURE 1Intertemporal Trends in Domains of Exploration-Exploitation


conflicting tendencies to explore or exploit in spe-cific domains of alliance formation (e.g., Beckmanet al., 2004; Rothaermel & Deeds, 2004). We ascribethese conflicting tendencies to internal pressures ofinertia and absorptive capacity, thus complement-ing research on exogenous industry conditions thatmay drive exploration and exploitation (Koza &Lewin, 1998; Park et al., 2002; Rothaermel, 2001;Rowley et al., 2000). By disentangling distinctivedomains of exploration-exploitation, we revealhow firms can still balance their exploration andexploitation tendencies. In fact, the distinctionsamong alliance formation domains serve a similarrole to that of the buffers between internal organi-zational units that presumably support firms’ con-current exploration and exploitation efforts (Ben-ner & Tushman, 2003; Christensen, 1998;Levinthal, 1997; Tushman & O’Reilly, 1997). Werefute the assumption that firms simultaneouslybalance exploration and exploitation within eachdomain, yet we show how balance is achievedacross domains and over time.

Our findings are illustrated in Figure 1, whichdepicts predicted exploration levels over timebased on model 4 while all other variables are heldat their mean levels. The figure suggests that firmsstrive to balance exploration and exploitationwithin the function domain, resulting in almostequal proportions of R&D and marketing agree-ments. They also balance exploration and exploita-tion across domains, as indicated by the high levelof structure exploration versus the low level ofattribute exploration. Finally, firms balance explo-ration and exploitation over time, as revealed in theopposed time trends across domains. Specifically,as firms proceed from exploration to exploitation inthe function domain, they tend to intensify theirexploration efforts in the structure and attributedomains.15 This process leads to a consistentmidrange level of the compound exploration-ex-ploitation measure that averages the three domain-specific measures.

A New Perspective on Balancing Exploration andExploitation in Alliances

Our framework suggests that internal pressuresfor exploration and exploitation constrain firms’expected learning behaviors within domains. Nev-ertheless, firms appear to balance their tendencies

to explore and exploit with respect to the nature oftheir alliances or choice of partners over time andacross domains. For this reason, studies that focusonly on one domain are sensitive to the choice ofdomain and depict only a partial picture of firms’balancing efforts. By spanning a fuller range ofdomains and longer time frames, scholars can morefully uncover the balance of exploration and ex-ploitation. Although we focused here on alliances,we believe that similar patterns can be observedwithin organizational boundaries, and we thus ex-tend prior research on the oscillation between or-ganizational forms and the dynamics of explora-tion-exploitation (Brown & Eisenhardt, 1997;Nickerson & Zenger, 2002; Siggelkow & Levinthal,2003).

We acknowledge the challenges associated withthe balancing process (Abernathy, 1978), notingthat the path dependencies that derive from priorexploration-exploitation experience limit firms’abilities to offset the pressures of inertia and ab-sorptive capacity within domains. We identifywhat may be termed second-order exploitation,whereby firms leverage their experience to enhancethe efficiency of either exploration or exploitationactivities. Consequently, balancing within domainsrequires prolonged adjustment to overcome thesepath dependencies.

Accordingly, we advance a dynamic perspectiveon balance wherein, over time, firms adjust theirtendencies to engage in exploration or exploitationwithin domains. For instance, firms that engage inrelatively high proportions of knowledge-generat-ing R&D alliances turn over time to knowledge-leveraging marketing and production alliances.This sequence is consistent with the product devel-opment cycle (Rothaermel & Deeds, 2004) in whichfirms leverage partners’ technologies before capital-izing on their market access. In addition, we dem-onstrate that as firms gradually shift from joint R&Dto collaborative production and marketing, theytend to experiment with new and diverse partnersthat can potentially offer a wider range of resourcesand capabilities. These trends occur independentof the maturation of firms and the entry of newfirms into the software industry. Although firmsexperience path dependencies in exploration-ex-ploitation that hinder their reaction in the shortterm, gradual adjustment of tendencies within do-mains occurs over time. This adjustment was ap-parent in the function and attribute domains, but inthe structure domain path dependencies were toostrong to overturn, resulting in intensifying explo-ration. Moreover, we demonstrate that temporal ad-justments in learning activities tend to be traded offacross domains. A shift toward exploitation in one

15 Perhaps the stronger time trend in the function do-main relative to the other domains can be attributed tothe initial intermediate level of exploration-exploitation,which inhibits path dependencies in that domain.


domain is accompanied by shifts toward explora-tion in others.

Hence, our study demonstrates how firms simul-taneously balance exploration and exploitationacross domains. Indeed, software firms reveal con-flicting tendencies to seek new partners while en-suring that these partners’ organizational profilesare quite similar to those of prior partners. Thisbalance across the structure and attribute domainsenables firms to access potentially new knowledgebases while reducing the risk of partner unfamiliar-ity (Gulati, 1995a) and leveraging prior experiencein managing similar alliances. In the same vein, thecontrasting tendencies in the function and struc-ture domains further demonstrate how firms avoidthe inherent tension between inertia and absorptivecapacity by teaming up with “old buddies” whenengaging more extensively in highly demandingR&D alliances. The established communicationchannels, governance mechanisms, and collabora-tion routines available with prior partners supportthe interactivity, coordination, and resource-sharing needs of these upstream alliances. A bal-ance is maintained because firms that underscoretechnology development with their R&D alliancestend to be conservative with respect to whom theypartner with. No dominant approach for pursuingeither exploration or exploitation within domainsnecessarily exists as long as balance is maintainedacross domains. Balancing across domains enablesfirms to become both innovative and efficient inmanaging alliances while reducing complexity,risk, and uncertainty.

In keeping with March (1991), firms act as adap-tive systems in a state of equilibrium between ex-ploration and exploitation. At any time within agiven domain, a firm may emphasize either explo-ration or exploitation, yet across domains and overtime, balance is maintained. By recognizing theevolutionary dynamics and multiple facets of ex-ploration and exploitation, our study bridges thegap between the normative assumption that firmsshould strive to balance exploration and exploita-tion and the observation that in practice firms dem-onstrate polar temporal tendencies to explore orexploit in certain domains.

Limitations and Directions for Future Research

Future research could address some of the limi-tations of our study. First, our interorganizationallearning perspective could be extended by consid-ering its interplay with intraorganizational learn-ing. The fundamental pressures imposed by inertiaand absorptive capacity guide exploration and ex-ploitation not only outside but also inside firm

boundaries. Thus, the notion of balancing acrossdomains and over time may apply to intraorganiza-tional learning through, for example, internal de-velopment or mergers and acquisitions. Hence, fu-ture research could uncover the internal domainsof exploration and exploitation and study whetherand how firms balance exploration and exploita-tion across organizational boundaries. Followingthe balance hypothesis, we would expect firms thatengage in internal exploitation to pursue explora-tion in their alliances. By juxtaposing intra- andinterorganizational exploration-exploitation, firmsmay be able to overcome trade-offs in resource al-location (Cheng & Kesner, 1997) and knowledgecreation (Rosenkopf & Nerkar, 2001) within andacross organizational boundaries.

Second, future research might examine varioustime intervals in assessing temporal adjustments inexploration and exploitation, which could revealnot only path dependencies but also cyclical pat-terns. Longer time intervals should also allow forthe incorporation of firm fixed effects, unlike therandom-effects models that we used. The more crit-ical question is what drives the time trends that weobserved within domains. The balance hypothesispredicts conflicting trends across domains, butwithin each domain, firms’ tendencies may bedriven by triggers such as exogenous industryevents, corporate leadership changes, or resourceallocation constraints (Nickerson & Zenger, 2002;Park et al., 2002). A related question is whether thebalance we observed results from conscious andproactive strategy (Christensen, 1998; Tushman &O’Reilly, 1997) or is, rather, a by-product of firms’ad hoc isolated engagements within each domain.Perhaps the relatively modest model fit statisticspoint to this random component. Field studiescould shed more light on the rationale behind thisbalancing behavior and better isolate managerialdiscretion from natural evolutionary paths in expli-cating tendencies and transitions.

Third, future research might test our frameworkin other industries. We focused on the softwareindustry, since its intensity of alliance formationenabled us to effectively track patterns of explora-tion-exploitation. Organizational pressures in otherindustries may vary and produce different patternsacross domains. Specifically, the software industryis turbulent and thus experiences stronger pres-sures for exploration than stable industries (Rowleyet al., 2000). Nonetheless, although firms in otherindustries may demonstrate different patternswithin certain domains, we expect that they wouldstill strive toward balance across domains and overtime.

Fourth, future research could overcome some of


our empirical limitations by directly measuring in-ertia and absorptive capacity instead of incorporat-ing them as latent mediating constructs. Such aresearch design might enhance internal validity by,for example, better relating exploration to absorp-tive capacity, as opposed to other intangible assets,which might instead account for embeddedness,heterogeneity, and innovation in alliance networks.Yet direct measurement might require surveys andthus limit researchers’ ability to measure and iden-tify time trends. With respect to construct validity,future research might develop alternative operation-alizations of the three domains. For instance, ourattribute exploration measure incorporates certainorganizational attributes that could be comple-mented with other relevant attributes. In differentcontexts, certain attributes may be more dominantthan others, but scholars should be attuned to dataavailability constraints, especially with respect toprivately held partners. Another avenue for futureresearch would involve the simultaneous analysisof exploration and exploitation at the dyadic level,as a given alliance can be exploratory for one part-ner and exploitative for another.

Finally, future research might elaborate on theperformance implications of balancing explorationand exploitation in the context of alliance forma-tion (He & Wong, 2004; Rothaermel, 2001). Ourstudy confirms the conventional wisdom that firmsseek to balance exploration and exploitation(March, 1991) but leaves open the question ofwhether such balance eventually enhances firmperformance. Despite its limitations, our study of-fers essential contributions to research on explora-tion and exploitation by demonstrating how firmsbalance these tendencies over time and acrossdomains.

REFERENCES

Abernathy, W. J. 1978. The productivity dilemma. Bal-timore: Johns Hopkins University Press.

Ahuja, G. 2000. Collaboration networks, structural holes,and innovation: A longitudinal study. Administra-tive Science Quarterly, 45: 425–455.

Anand, B. N., & Khanna, T. 2000a. Do firms learn tocreate value? The case of alliances. Strategic Man-agement Journal, 21: 295–317.

Anand, B. N., & Khanna, T. 2000b. The structure oflicensing contracts. Journal of Industrial Econom-ics, 48(1): 103–135.

Argote, L., & Ophir, R. 2002. Intraorganizational learning.In J. A. C. Baum (Ed.), The Blackwell companion toorganizations: 181–207. Oxford, U.K.: Blackwell.

Baum, J. A. C., Rowley, T., Shipilov, A. V., & Chuang,

Y-T. 2005. Dancing with strangers: Aspiration per-formance and the search for underwriting syndicatepartners. Administrative Science Quarterly, 50:536–575.

Beckman, C. M., Haunschild, P. R., & Phillips, D. J. 2004.Friends or strangers? Firm-specific uncertainty, mar-ket uncertainty, and network partner selection. Or-ganization Science, 15: 259–275.

Benner, M. J., & Tushman, M. 2002. Process managementand technological innovation: A longitudinal studyof the photography and paint industries. Adminis-trative Science Quarterly, 47: 676–706.

Benner, M. J., & Tushman, M. L. 2003. Exploitation,exploration, and process management: The produc-tivity dilemma revisited. Academy of ManagementReview, 28: 238–256.

Bourgeois, L. J. 1981. On the measurement of organiza-tional slack. Academy of Management Review, 6:29–39.

Brown, S. L., & Eisenhardt, K. M. 1997. The art of con-tinuous change: Linking complexity theory andtime-paced evolution in relentlessly shifting organi-zations. Administrative Science Quarterly, 42:1–34.

Burgelman, R. A. 1994. Fading memories: A process the-ory of strategic business exit in dynamic environ-ments. Administrative Science Quarterly, 39: 24–56.

Cheng, J., & Kesner, I. 1997. Organizational slack andresponse to environmental shifts: The impact of re-source allocation patterns. Journal of Management,23(1): 1–18.

Child, J. 2001. Learning through strategic alliances. In A.Dierkes, B. Antal, J. Child, & I. Nonaka (Eds.), Hand-book of organizational learning and knowledge:657–680. Oxford, U.K.: Oxford University Press.

Christensen, C. M. 1998. The innovator’s dilemma:When new technologies cause great firms to fail.Boston: Harvard Business School Press.

Chung, S., Singh, H., & Lee, K. 2000. Complementarity,status similarity and social capital as drivers of alli-ance formation. Strategic Management Journal, 21:1–22.

Cohen, W. M., & Levinthal, D. A. 1990. Absorptive ca-pacity: A new perspective on learning and innova-tion. Administrative Science Quarterly, 35: 128–152.

Cyert, R. M., & March, J. G. 1963. A behavioral theory ofthe firm (2nd ed.). Englewood Cliffs, NJ: Prentice-Hall.

Darr, E. D., & Kurtzberg, T. R. 2000. An investigation ofpartner similarity dimensions on knowledge trans-fer. Organizational Behavior and Human DecisionProcesses, 82: 28–44.

Deeds, D. L. 2001. The role of R&D intensity, technicaldevelopment and absorptive capacity in creating en-


trepreneurial wealth in high technology start-ups.Journal of Engineering and Technology Manage-ment, 18: 29–47.

Dyer, J. H., & Singh, H. 1998. The relational view: Coop-erative strategies and sources of interorganizationalcompetitive advantage. Academy of ManagementReview, 23: 660–679.

Gavetti, G., & Levinthal, D. A. 2000. Looking forward andlooking backward: Cognitive and experientialsearch. Administrative Science Quarterly, 45: 113–137.

Grant, R. M., & Baden-Fuller, C. 2004. A knowledge ac-cessing theory of strategic alliances. Journal of Man-agement Studies, 41: 61–84.

Gulati, R. 1995a. Does familiarity breed trust? The impli-cations of repeated ties for contractual choices.Academy of Management Journal, 35: 85–112.

Gulati, R. 1995b. Social structure and alliance formationpatterns: A longitudinal analysis. AdministrativeScience Quarterly, 40: 619–652.

Gulati, R. 1998. Alliances and networks. Strategic Man-agement Journal, 19: 293–317.

Gulati, R., & Gargiulo, M. 1999. Where do interorganiza-tional networks come from? American Journal ofSociology, 104: 1439–1493.

Gulati, R., Lavie, D., & Singh, H. 2003. The nature ofpartnering experience and the gains from alli-ances. Paper presented at the 62nd Annual Meetingof the Academy of Management, Seattle.

Hannan, M. T., & Freeman, J. H. 1984. Structural inertiaand organizational change. American SociologicalReview, 49: 149–164.

He, Z.-L., & Wong, P.-K. 2004. Exploration vs. exploita-tion: An empirical test of the ambidexterity hypoth-esis. Organization Science, 15: 481–494.

Hennart, J. F., & Reddy, S. 1997. The choice betweenmergers/acquisitions and joint ventures: The case ofJapanese investors in the United States. StrategicManagement Journal, 18: 1–12.

Hoang, H., & Rothaermel, F. 2005. The effect of generaland partner-specific alliance experience on jointR&D project performance. Academy of Manage-ment Journal, 48: 332–345.

Holmqvist, M. 2003. A dynamic model of intra- andinterorganizational learning. Organization Studies,24: 95–103.

Huber, G. P. 1991. Organizational learning: The contrib-uting processes and the literatures. OrganizationScience, 2: 88–115.

Ingram, P. 2002. Interorganizational learning. In J. A. C.Baum (Ed.), The Blackwell companion to organiza-tions: 642–633. Oxford, U.K.: Oxford BlackwellBusiness.

Kale, P., Dyer, J. H., & Singh, H. 2002. Alliance capability,stock market response, and long-term alliance suc-

cess: The role of the alliance function. StrategicManagement Journal, 23: 747–767.

Kelly, D., & Amburgey, T. L. 1991. Organizational inertiaand momentum: A dynamic model of strategicchange. Academy of Management Journal, 34: 591–612.

Koza, M. P., & Balakrishnan, S. 1993. Information asym-metry, adverse selection and joint-ventures. Journalof Economic Behavior and Organization, 20: 99–117.

Koza, M. P., & Lewin, A. Y. 1998. The co-evolution ofstrategic alliances. Organization Science, 9: 255–264.

Koza, M. P., & Lewin, A. Y. 2000. Managing partnershipsand strategic alliances: Raising the odds of success.European Management Journal, 18: 146–151.

Kumar, R., & Nti, K. O. 1998. Differential learning andinteraction in alliance dynamics: A process and out-come discrepancy model. Organization Science, 9:356–367.

Lane, P. J., & Lubatkin, M. 1998. Relative absorptivecapacity and interorganizational learning. StrategicManagement Journal, 19: 461–477.

Lane, P. J., Salk, J. E., & Lyles, M. A. 2001. Absorptivecapacity, learning, and performance in internationaljoint ventures. Strategic Management Journal, 22:1139–1161.

Larsson, R., Bengtsson, L., Henriksson, K., & Sparks, J.1998. The interorganizational learning dilemma:Collective knowledge development in strategic alli-ances. Organization Science, 9: 285–305.

Lavie, D. 2004. The evolution and strategy of intercon-nected firms: A study of the Unisys alliance network.Academy of Management Best Paper Proceedings:E1–E6.

Levinthal, D. A. 1997. Adaptation on rugged landscapes.Management Science, 43: 934–950.

Levinthal, D. A., & March, J. G. 1981. A model of adaptiveorganizational search. Journal of Economic Behav-ior and Organization, 2: 307–333.

Levinthal, D. A., & March, J. G. 1993. The myopia oflearning. Strategic Management Journal, 14(spe-cial issue): 95–112.

Levitt, B., & March, J. G. 1988. Organizational learning. InW. R. Scott & J. F. Short (Eds.), Annual review ofsociology, vol. 14: 319–340. Palo Alto, CA: AnnualReviews.

Li, S. X., & Rowley, T. J. 2002. Inertia and evaluationmechanisms in interorganizational partner selection:Syndicate formation among U.S. investment banks.Academy of Management Journal, 45: 1104–1119.

March, J. G. 1991. Exploration and exploitation in organ-izational learning. Organization Science, 2: 71–87.

March, J. G., & Simon, H. A. 1958. Organizations. NewYork: Wiley.


McGrath, R. G. 2001. Exploratory learning, innovativecapacity, and managerial oversight. Academy ofManagement Journal, 44: 118–131.

Mowery, D. C., Oxley, J. E., & Silverman, B. S. 1996.Strategic alliances and interfirm knowledge transfer.Strategic Management Journal, 17(winter specialissue): 77–91.

Nelson, R., & Winter, S. 1982. An evolutionary theory ofeconomic change. Cambridge, MA: Belknap Press.

Nickerson, J. A., & Zenger, T. R. 2002. Being efficientlyfickle: A dynamic theory of organizational choice.Organization Science, 13: 547–566.

Park, S. H., Chen, R., & Gallagher, S. 2002. Firm resourcesas moderators of the relationship between marketgrowth and strategic alliances in semiconductorstart-ups. Academy of Management Journal, 45:527–545.

Rivkin, J. W., & Siggelkow, N. 2003. Balancing search andstability: Interdependencies among elements of or-ganizational design. Management Science, 49: 290–311.

Romanelli, E., & Tushman, M. L. 1994. Convergence andupheaval: Managing the unsteady pace of organiza-tional evolution. Academy of Management Journal,37: 1141–1166.

Rosenkopf, L., & Almeida, P. 2003. Overcoming localsearch through alliances and mobility. ManagementScience, 49: 751–766.

Rosenkopf, L., & Nerkar, A. 2001. Beyond local search:Boundary spanning, exploration, and impact in theoptical disk industry. Strategic Management Jour-nal, 22: 287–306.

Rothaermel, F. T. 2001. Incumbent’s advantage throughexploiting complementary assets via interfirm coop-eration. Strategic Management Journal, 22: 687–699.

Rothaermel, F. T., & Deeds, D. L. 2004. Exploration andexploitation alliances in biothechnology: A systemof new product development. Strategic Manage-ment Journal, 25: 201–222.

Rowley, T., Behrens, D., & Krackhardt, D. 2000. Redun-dant governance structures: An analysis of structuraland relational embeddedness in the steel and semi-conductor industries. Strategic Management Jour-nal, 21: 369–386.

Siggelkow, N., & Levinthal, D. A. 2003. Temporarily di-vide to concur: Centralized, decentralized, and rein-tegrated organizational approaches to explorationand adaptation. Organization Science, 14: 650–669.

Sørenson, O., & Stuart, T. E. 2001. Syndication networksand the spatial distribution of venture capital invest-ments. American Journal of Sociology, 106: 1546–1586.

Stuart, T. 2000. Interorganizational alliances and the per-formance of firms: A study of growth and innovationrates in a high-technology industry. Strategic Man-agement Journal, 21: 719–811.

Teece, D. J., Rumelt, R. P., Dosi, G., & Winter, S. 1994.Understanding corporate coherence: Theory and ev-idence. Journal of Economic Behavior and Organi-zation, 23: 1–30.

Tsai, W. 2001. Knowledge transfer in interorganizationalnetworks: Effects of network position and absorptivecapacity on business unit innovation and perfor-mance. Academy of Management Journal, 44: 996–1004.

Tushman, M. L., & O’Reilly, C. A. 1997. Winning throughinnovation: A practical guide to leading organiza-tional change and renewal. Boston: Harvard Busi-ness School Publishing.

Verspagen, B., & Duysters, G. 2004. The small world ofstrategic technology alliances. Technovation, 24:563–571.

Villalonga, B., & McGahan, A. M. 2005. The choiceamong acquisitions, alliances, and divestitures.Strategic Management Journal, 26: 1183–1208.

Wooldridge, J. M. 2002. Econometric analysis of crosssection and panel data. Cambridge, MA: MIT Press.

Zahra, S. A., & George, G. 2002. Absorptive capacity: Areview, reconceptualization, and extension. Acad-emy of Management Review, 27: 185–203.

Zollo, M., & Winter, S. G. 2002. Deliberate learning andthe evolution of dynamic capabilities. OrganizationScience, 13: 339–351.

Dovev Lavie ([email protected]) is an as-sistant professor of management at the University ofTexas at Austin. He received his Ph.D. from the WhartonSchool of the University of Pennsylvania. His currentresearch interests include value creation and appropria-tion in alliance networks and applications of the re-source-based view and the dynamic capabilities ap-proach in technology-intensive industries.

Lori Rosenkopf ([email protected]) is anassociate professor at the Wharton School of the Univer-sity of Pennsylvania. She received her Ph.D. in manage-ment of organizations from Columbia University. Loristudies interorganizational networks and knowledgeflows in various high-tech industries by analyzing par-ticipation in technical committees; alliances; the mobil-ity of technical professionals; and patents within techno-logical communities.


Documents

BALANCING EXPLORATION AND EXPLOITATION IN ALLIANCE …€¦ · whereas exploitation involves “the use and devel-opment of things already known” (1993: 105). More ... Exposure