OLD TECHNOLOGY MEETS NEW TECHNOLOGY: COMPLEMENTARITIES

Strategic Management JournalStrat. Mgmt. J., 29: 47–77 (2008)

Published online 25 June 2007 in Wiley InterScience (www.interscience.wiley.com) DOI: 10.1002/smj.634

Received 16 December 2003; Final revision received 8 May 2007

OLD TECHNOLOGY MEETS NEW TECHNOLOGY:COMPLEMENTARITIES, SIMILARITIES, ANDALLIANCE FORMATION

FRANK T. ROTHAERMEL1* and WARREN BOEKER2

1 College of Management, Georgia Institute of Technology, Atlanta, Georgia, U.S.A.2 Department of Management and Organization, University of Washington BusinessSchool, Seattle, Washington, U.S.A.

Alliance formation is commonplace in many high-technology industries experiencing radicaltechnological change, where established firms use alliances with new entrants to adapt totechnological change, while new entrants benefit from the ability of established players tocommercialize the new technology. Despite the prevalence of these alliances, we know littleabout how these firms choose to ally with specific firms given the range of possible partnersthey may choose from. This study explores factors that lead to alliance formation betweenpharmaceutical and biotechnology companies. We focus on the alliance tie as the unit of analysisand argue that dyadic complementarities and similarities directly influence alliance formation.We then introduce a contingency model in which the positive effect of complementarities andsimilarities on alliance formation is moderated by the age of the new technology firm. We drawtheoretical attention to the intersection between levels of analysis, in particular, the intersectionbetween dyadic and firm-level constructs. We find that a pharmaceutical and a biotechnologyfirm are more likely to enter an alliance based on complementarities when the biotechnologyfirm is younger. Another noteworthy finding is that proxies for broad capabilities appear to beat least as effective, if not more so, in predicting alliance formation compared to fine-grainedscience and technology-related indicators, like patent cross-citations or patent common citations.We conclude by suggesting that future studies on alliance formation need to take into accountinteractions across levels; for example, how dyadic capabilities interact with firm-level factors,and the advantages and disadvantages of more or less fine-grained measures of organizationalcapabilities. Copyright 2007 John Wiley & Sons, Ltd.

INTRODUCTION

The formation of partnerships, alliances, and jointventures between firms has increased at a dra-matic rate over the last few decades (Hagedoorn,2002). This phenomenon has inspired significantresearch into the question of why firms enter

Keywords: alliance formation; technological change; mea-surement of capabilities; multilevel research; pharmaceu-tical biotechnology industry∗ Correspondence to: Frank T. Rothaermel, College of Manage-ment, Georgia Institute of Technology, 800 West Peachtree St,NW, Atlanta, GA 30308-1149, U.S.A.E-mail: [email protected]

alliances. Extant research has suggested that firmsare motivated to enter alliances to overcome mar-ket failures (Williamson, 1985), accrue marketpower (Porter and Fuller, 1986), learn from oneanother (Hamel, Doz, and Prahalad, 1989), sharerisks (Ohmae, 1989), access complementary assets(Arora and Gambardella, 1990; Rothaermel, 2001),enhance legitimacy (Baum and Oliver, 1991),build new competences (Hennart, 1991), enternew markets and technologies (Kogut, 1991),enhance innovativeness (Shan, Walker, and Kogut,1994) and new product development (Rothaer-mel and Deeds, 2004), and improve early perfor-mance (Baum, Calabrese, and Silverman, 2000).

Copyright 2007 John Wiley & Sons, Ltd.

48 F. T. Rothaermel and W. Boeker

Others have suggested that a firm’s propensityto form alliances depends on the firm’s strategicand social position (Eisenhardt and Schoonhoven,1996); technical, commercial, and social capital(Ahuja, 2000); and its resources and external envi-ronment (Park, Chen, and Gallagher, 2002). Whilewe seem to have a fairly good understanding ofwhy firms enter alliances, the question of partnerchoice has received less attention.

Yet, the importance of alliance partner choicehas long been recognized (Harrigan, 1985). Pio-neering research into the question of who allieswith whom has drawn on resource dependencetheory and suggested that organizations enter intorelationships motivated by strategic interdependen-cies (Oliver, 1990). Scholars have also argued thatstatus similarities (Podolny, 1994) and social net-works (Gulati, 1995b) play an important role inpredicting partner choice. More recent work hassuggested that complementarities arising from dif-ferent geographic locations combined with statussimilarity and social capital predict alliance forma-tion (Chung, Singh, and Lee, 2000). Firms’ par-ticipation in technical committees has also beenfound to explain subsequent alliance formation(Rosenkopf, Metiu, and George, 2001).

Extant empirical research that explains who part-ners with whom has focused predominantly on hor-izontal alliances between established firms (e.g.,Gulati, 1995a, 1995b; Ahuja, 2000; Chung, et al.,2000; Garcia-Pont and Nohria, 2002). Rather thanfocusing on alliance formation between establishedfirms, we focus on the evolution of technologyfields in an industry and on the antecedents ofalliances between firms imprinted under an oldtechnology and firms imprinted under a new tech-nology (Stinchcombe, 1965). Such a focus seemsparticularly salient given the theoretical emphasisresearchers have placed on strategic alliances as atool for established firms to adapt to technolog-ical change (Teece, 1992; Hill and Rothaermel,2003). Prior work, with its focus on horizontalpartnerships between established firms in existingtechnologies, has offered insights into the impor-tance of alliance behavior for market power andmarket expansion, but has focused little atten-tion on the manner in which firms ally to meetthe needs of a dynamic and technologically com-plex external environment. The setting for thisstudy is the intersection between the pharmaceu-tical and the biotechnology industry. Our focus is

on the choice of alliance partner between incum-bent technology firms and new technology firms.The incumbent technology firms in this study arethe pharmaceutical companies founded under thetechnology paradigm of chemical screening, andthe new technology firms are biotechnology firmsfounded under a new technology paradigm basedon molecular biology.

Extant literature indicates that strategic interde-pendence and status similarities increase the proba-bility of alliance formation (Oliver, 1990; Podolny,1994; Gulati, 1995b; Chung et al., 2000). Wheninvestigating the effect of dyadic complementari-ties on alliance formation, we follow prior researchand focus on complementarities arising from non-overlapping niches (Gulati, 1995b; Chung et al.,2000). In addition to non-overlapping niches, wedevelop a second measure to assess the poten-tial for complementarities that combine differentcompetences along the value chain. We combinea biotechnology firm’s competence in drug devel-opment—an upstream value chain activity—witha pharmaceutical firm’s competence in regula-tory management, marketing, and distribution—alldownstream value chain activities. Prior researchhas demonstrated that pharmaceutical firms pos-sess distinctive competences along the value chain(De Carolis, 2003).

When analyzing the effect of dyadic similaritieson alliance formation, we build on the pioneeringwork of Mowery, Oxley, and Silverman (1996,1998), and use patent cross-citation and patentcommon citation rates to proxy technological sim-ilarities. In addition to these more fine-grainedmeasures for science and technology relatedness,we develop a third measure that captures dyadicsimilarity, based on each firm’s competence ininnovation. In particular, we measure dyadic sim-ilarity through overall patenting propensity, whichcaptures the firm’s broad-based capabilities in gen-erating innovation (Stuart, 2000).

Prior work has not considered that the proba-bility of alliance formation might be contingentupon how dyadic constructs may interact withfirm-specific factors. Here we focus on the ageof the new technology firm, as a proxy for thefirm’s legitimacy, power, and credibility. We positthat alliance formation driven by complementari-ties between established and new technology firmsis more likely when the new technology firm isyounger. As the new technology firm ages, weargue that alliance formation between established

Copyright 2007 John Wiley & Sons, Ltd. Strat. Mgmt. J., 29: 47–77 (2008)DOI: 10.1002/smj

Old Technology Meets New Technology 49

and new technology firms is more likely to bebased on similarities. Thus, we contend that theage of the new technology partner moderates theeffect of complementarities in a negative fash-ion, while it moderates the impact of similari-ties in a positive fashion. We conclude that thedyadic effects of complementarities and similar-ities on alliance formation are contingent uponfirm-level characteristics of the new technologyventure.

THEORY AND HYPOTHESESDEVELOPMENT

Complementarities and alliance formation

Past research cites the pooling of complemen-tary skills and resources to create added value asone of the primary motives for creating strategicalliances (Arora and Gambardella, 1990; Teece,1992; Rothaermel, 2001). With the increased em-phasis on core competences and the possession of‘best in class’ capabilities over the last decade,strategic alliances have been promoted as an oppor-tunity for two firms to combine and create a syner-gistic entity (Nohria and Garcia-Pont, 1991; Dyerand Singh, 1998). A strategic interdependenceperspective on alliance formation suggests thatdependencies create conditions that favor alliances,and that firms create alliances with those partnerswho can best provide the complementary assetsand skills they need. Extant research examiningresource interdependencies between specific firmshas tied alliance formation to the availability ofspecific competences, typically functional capabil-ities within and across firms, such as research anddevelopment (R&D), production, marketing, anddistribution.

The central perspective of these arguments isthat each partner has certain areas of strength thatmay compensate for the weaknesses of their poten-tial alliance partner—a perspective that has beensupported by a number of empirical studies (Har-rigan, 1985; Lorange and Roos, 1992; Burgers,Hill, and Kim, 1993). Prior research has providedconsistent evidence across different industries thatfirms that occupied complementary niches weremore likely to form alliances (Gulati, 1995b;Chung et al., 2000). One common example is thebenefit of alliance formation between large, estab-lished firms and small firms (often recent entrants

into the market) where the established manufac-turing, sales, marketing, and distribution channelsof the large firms can be utilized by the smallerpartners, which may excel in product innovationor new product development. In these cases, firmsattempt to leverage the critical knowledge avail-able from another partner to build on their ownset of capabilities. Teece (1986, 1992) goes so faras to argue that established competitors in emerg-ing industries risk obsolescence if they are unableto form partnerships with innovative startups thatcan provide them with a steady stream of futureideas. Rothaermel (2001) found evidence of theimportance of complementary assets in allianceformation by providing support for the notion thatincumbents may benefit from radical technolog-ical change through allying with new entrants, ifthe incumbents possess specialized complementaryassets necessary to commercialize the new technol-ogy. When applied in the context of the dyad, thesearguments suggest that firms seek out partners thatcan help them manage strategic interdependenciesby offering superior capabilities in areas where thefocal firm may be weaker.

Within the empirical setting of our research,biotechnology represents a scientific base (molec-ular biology) that is significantly different fromthe knowledge base of pharmaceuticals (organicchemistry). Past work has indicated that a scientistwho is trained in the framework of drug discov-ery and development based on chemical synthesisloses on the average around 80–100 percent ofhis or her skills when attempting to transition tothe emerging framework of drug discovery anddevelopment based on molecular biology (Rothaer-mel, 2001). The new biotechnology is thus consid-ered to be competence destroying for pharmaceu-tical companies (Powell, Koput, and Smith-Doerr,1996; Stuart, Hoang, and Hybels, 1999). Biotech-nology firms possess R&D competences and capa-bilities that traditional pharmaceutical companiescan profitably draw upon to maintain their innova-tive presence. The fact that pharmaceutical com-panies marketed and distributed seven of the top-ten selling biotechnology drugs in the late 1990s,even though none of the drugs were developed bythe pharmaceutical companies, demonstrates theimportance of the distribution of biotechnologyproducts by pharmaceutical companies (Ernst &Young Biotechnology Reports). Emerging indus-tries such as biotechnology may offer significantopportunities for cooperation between small new



entrants and large established firms that are seek-ing to exploit technological spillovers in an attemptto realize the potential for commercialization. Sit-uations in which both parties can benefit usuallyoccur when complementarities between partnersallow them to compensate for each other’s weak-nesses while leveraging each other’s comparativestrengths.

Hypothesis 1: Complementarities between anincumbent technology firm and a new technol-ogy firm increase the probability of alliance for-mation.

Similarities and alliance formation

Although a focus on complementary assets, skills,and knowledge may provide a relatively straight-forward explanation of the formation of alliancesbetween some firms, such a perspective potentiallyignores how firms overcome the uncertainties asso-ciated with such partnerships (Kale, Singh, andPerlmutter, 2000). Market risk and a high levelof uncertainty typically characterize the initiationof any alliance or partnership (Hamel et al., 1989).The creation of an alliance involves a very care-ful assessment on the part of each partner as towhat the partner and the alliance might offer andwhether the benefits of the alliance exceed thepotential downside risks.

To ameliorate some of the fears that the alliancerepresents, firms may focus on other character-istics and signals from their potential partnersthat they feel may increase the likelihood thatthe alliance will work. A preference to partnerwith firms of similar status, for example, hasbeen found to consistently predict alliance forma-tion in the investment banking industry (Podolny,1994; Chung et al., 2000). In more technology-and science-intensive industries like biotechnol-ogy, patenting is often considered a signal of thequality for a firm, in particular in the absence ofmore tangible signals like the successful commer-cialization of new products.1 In this industry, firmsthat patent more than their peers are consideredto be on the technological frontier of their field(Sorensen and Stuart, 2000). Thus, highly inno-vative firms may search out each other as alliance

1 Very few biotechnology ventures have successfully commer-cialized new biotechnology drugs at this stage of industryevolution.

partners based on similarities in overall technologystrategy. This seems particularly pertinent in thepharmaceutical industry because it is characterizedby distinct strategic groups, in which one groupcompetes on drug discovery and development ofpatent-protected proprietary drugs, while a secondstrategic group competes on generic, me-too prod-ucts, where the patent protection has expired (Cooland Schendel, 1987, 1988). In addition, work byMowery et al. (1998) showed that firms with sim-ilarities in technological capabilities, proxied bypatent cross-citation and patent common citations,were more likely to form an alliance. Similarly,work by Lane and Lubatkin (1998) demonstratedthat pharmaceutical and biotechnology firms thatdraw from the same knowledge base and domi-nant logics were more likely to partner and createbetter-performing alliances. Their research adoptedan organizational learning perspective to exam-ine the role that absorptive capacity played inthe dyad’s ability to value, assimilate, and utilizeexternal knowledge, showing that similarities inbasic knowledge and organizational practices (e.g.,compensation practices and levels of formaliza-tion and centralization) helped the partners oper-ate together more effectively. Assuming that thereare good strategic reasons to create an alliance,firms may be most interested in partnering withfirms that they feel are similar in other dimensions,like innovation orientation or technological over-lap, both of which seem to be particularly salientin high-technology industries.

Hypothesis 2: Similarities between an incum-bent technology firm and a new technology firmincrease the probability of alliance formation.

Complementarities, similarities, and newtechnology firm age

Biotechnology firms, as more recent market en-trants, are frequently at the forefront of the intro-duction of radical new technologies (Tushman andAnderson, 1986). Given the high uncertainty thatsurrounds this field, however, a high failure rateof new entrants is a common occurrence, whichis reflected in the theoretical notion of a liabil-ity of newness. Stinchcombe (1965) proposed thatyounger organizations experience a higher mor-tality rate because they lack accumulated produc-tion experience, well-developed internal processesand procedures, strong ties with customers and



suppliers, and experienced human resources. Evo-lutionary theorists posit that newer organizationsare prone to higher mortality, because their rou-tines have not yet developed to a point that theyserve as sufficient repositories of organizationalknowledge (Nelson and Winter, 1982; March,1988).

Alliance choice and ultimate success are influ-enced by the capabilities of each partner. Whenan incumbent technology firm allies with a newtechnology firm, the incumbent technology firmis likely to be the member of the pair that islarger and more powerful, and thus the legiti-macy of the incumbent technology (in this casepharmaceutical) partner is seldom questioned. Thelegitimacy and viability of the new technology(biotechnology) firm, conversely, may be muchmore uncertain. As Stinchcombe (1965) pointedout, development of trust and internal legitima-tion of the new firm’s management and businessmodel takes time, as does the acquisition of insti-tutional identity. Younger firms, such as the newtechnology firms in this study, are likely to havea lower level of legitimation than older ones. Inaddition, as firms age, they accumulate innovationand production experience, and are thus able tobuild on prior knowledge when discovering anddeveloping new products or when incorporatingnew external knowledge (Cohen and Levinthal,1990).

As incumbent technology firms decide withwhom to partner, they examine and assess thestrengths and weaknesses of an array of potentialnew technology partners. The age of the new tech-nology partner may be particularly instructive insuch assessment when the quality of the new tech-nology partner’s potential products and markets isnot well known, given its early stage of develop-ment. In this case, the new technology firm’s agetakes on an important role of signaling to othersthe quality of the firm (Sorensen and Stuart, 2000).

Processes of external legitimation also take time.Although an organization must have some minimallevel of legitimacy to be attractive as an alliancepartner, newer firms have comparatively weakerclaims on support and interest from potential part-ners. Newer, younger firms may be less prominentand may have more difficulty attracting resourcesor partners. As firms with a new technology (inthis case biotechnology companies) evolve, theygain credibility and legitimacy. They are also morelikely to be perceived by the broader community

of pharmaceutical firms that are interested in form-ing alliances as more reliable alliance partners.Older organizations tend to have a dense web ofexchanges, to affiliate with centers of power, andto have increased status. External actors may alsowait for an initial period of testing to be passedbefore making investments in exchange relationswith new organizations.

Moderating effects of new technology firm ageon alliance formation

A critical issue for the incumbent technology firmconcerns the choice of when to partner with a newtechnology firm. Ideally, the incumbent firm wouldchoose a new technology partner early in its devel-opment, before it becomes too established, toolegitimate, too attractive to others, and too power-ful. Choosing a partner early allows the incumbentfirm to lock in partnerships with the best new tech-nology firms before they begin to work with other,rival, incumbent firms. Waiting too long can createtwo problems: the window of opportunity for cap-italizing on the new set of ideas and technologiesmay have passed, or its value may be undermined;and rival incumbent firms may be able to capital-ize on their own opportunities to work with newtechnology partners—thus the potential competi-tive advantage presented by the new technologyfirm is ceded to a rival.

The age of the new technology firm is par-ticularly important in alliances formed to exploitcomplementarities across partners. Complementar-ity involves the creation of immediate value forthe combined entity. Speed of the alliance forma-tion process is key, and at a premium. Under theseconditions it is advantageous to form partnershipsearly with promising new technology partners.Forming a partnership early may also be in thebest interest of the new technology firm becausethis can pave the way to market access (Shan et al.,1994), and enhance the new venture’s legitimacyas the established partner’s reputation spills overthrough affiliation (Stuart et al., 1999). It is alsoclear, however, that younger new technology firmsare likely to have less power and influence, fol-lowing the legitimacy and institutional argumentsmade above. Early after founding, the new tech-nology firm may be in a relatively weaker posi-tion in the alliance compared to the larger, moreestablished incumbent firm. Differences in age andlegitimacy exacerbate power differences between



the two potential partners, with the relative powerdifferential inversely related to the new technologyfirm’s age. Because of the new technology firm’srelative youth and lack of legitimacy and influ-ence, it is much more likely that the incumbenttechnology partner will play a more proactive rolein setting the terms for the alliance in a manner thatsupports and facilitates its own business. The newtechnology firm, with less power, may be forcedto accept less attractive terms than if it had morepower relative to the incumbent technology firm.

Complementary technologies present a com-pelling and straightforward logic for entering intoalliances, particularly between new technology andincumbent technology firms (Teece, 1992). Estab-lished technology firms are likely to be particularlyvigilant and aggressive in uncovering new tech-nology firms that appropriately fit their existingset of complementary assets and value chain posi-tions (Kale, Dyer, and Singh, 2002). Because newtechnology firms may also find the prospect of analliance with an established technology firm com-pelling, it is likely that the new technology firmwill also be interested in such a venture. New tech-nology ventures may have no choice but to enter analliance if they desire market access, because for-ward integration is often difficult and both time andresource intensive (Rothaermel and Deeds, 2004).Added to this dynamic is the likely reality thatthe new technology firm has less power earlier inits development and thus is more willing to par-ticipate in an alliance initiated by the establishedtechnology partner. Given the importance of com-plementarities to the established technology firmtrying to take advantage of the scientific advancesof the new technology firm early in its develop-ment, it is likely that complementarities will drivealliances between young new technology firms andestablished technology firms.

Hypothesis 3: Incumbent and new technologyfirms are more likely to enter alliances basedon complementarities when the new technologyfirm is younger.

Choosing relatively young partners because oftheir potentially complementary competences cancreate a successful alliance, but the alliance isalso infused with uncertainty. As time passes, anincumbent technology firm finds it easier to assessthe performance of the potential new technologypartner and also to assess its ability to contribute

to the value created by a potential alliance. Asthe new technology firm ages, critical externalconstituencies such as investors, suppliers, andcustomers are better able to form judgments aboutthe quality of the firm.

The power, reputation, and status of the newtechnology firm increases as it ages, enhancingits attractiveness as a potential alliance partner.Because older new technology firms may be seenas more competent and more prominent, allianceswith them can offer a better opportunity to bene-fit from their competences and skills, while alsoleading to a greater likelihood of alliance suc-cess. In addition to the direct benefits accordedthrough access to superior capabilities, affiliationwith a longer-lived, more legitimate, and thus moreestablished new technology firm can increase theperceived quality of the incumbent technology firmin the minds of other external actors.

Allying with successful new technology ven-tures may signal that the established technologyfirm is making headway in adapting to the newtechnology (Hill and Rothaermel, 2003; Rothaer-mel and Hill, 2005). In this case, reputationspillovers would flow from the older, and gener-ally more successful new technology firm to theincumbent technology firm. For example, Lilly isseen as more likely to maintain its market lead-ership in insulin after it entered an alliance withthe biotechnology firm Genentech and obtainedexclusive rights to market and distribute a newbiotechnology-based human insulin drug(Humulin). One could argue that Genentechbestowed legitimacy and reputation effects uponLilly. The amount of power and influence that thenew technology firm has relative to the incum-bent technology firm is likely to increase overtime, which in turn may influence its dealingswith any potential partner. An alliance partnershipwith a more established new technology firm thathas overcome the liability of newness is likely torepresent less of an absolute difference in powerbetween the two partners. The new technology firmhas gained prominence, legitimacy, and influence,and is more likely to take a stronger role in decid-ing on its alliance partner and in setting the termsof the alliance.

Given the greater balance of power between thetwo potential partners, the new technology firmshould have a more active voice in any alliancedecision. It is likely that the new technology firm,



while interested in complementarities, is also inter-ested in similarities with the established firm. Tothe extent that both parties are similar, they mayfeel a greater comfort level, sense that they canwork together more successfully, and more suc-cessfully manage the integration of competencesand skills that must happen for the alliance to cre-ate value. Assuming a diminishment of the powerdifferential between new and incumbent technol-ogy firms that occurs as a byproduct of the newtechnology firm’s aging, we posit that these firmstend to ally with incumbent technology firms towhich they are similar.

Hypothesis 4: Incumbent and new technologyfirms are more likely to enter alliances basedon similarities when the new technology firm isolder.

METHODOLOGY

Research setting

To empirically test the role of dyad-level capabil-ities and their interaction with the age of the newtechnology firm in predicting alliance formation,we chose the global biopharmaceutical industry asour research setting. The global biopharmaceuticalindustry consists of pharmaceutical and biotech-nology firms such as Biogen and Chiron and tra-ditional pharmaceutical companies such as Pfizerand Merck that focus on human therapeutics.

We chose this particular research setting for anumber of reasons. The scientific foundation ofthe pharmaceutical companies is organic chem-istry, while the sciences underlying biotechnologyare molecular biology, immunology, and genetics,among others. The emergence of biotechnology isconsidered a radical process innovation in the waydrugs are discovered and developed (Stuart et al.,1999). This radical process innovation in turn cre-ated the need for pharmaceutical firms to collabo-rate with new biotechnology firms to aid in transi-tioning from old to new methods of drug discoveryand development. Conversely, new biotechnologyfirms also needed to collaborate with the phar-maceutical firms because the incumbent pharmafirms controlled access to the market for humanpharmaceuticals, in particular valuable and path-dependent capabilities in clinical trial management

and drug marketing and distribution (De Caro-lis, 2003). Not surprisingly, the commercializa-tion of biotechnology is characterized by extensiveinterfirm cooperation. The biotechnology industryexhibits high alliance intensity and accounts forabout 20 percent of the observed strategic alliancesin high-technology industries (Hagedoorn, 1993).Given the need of old and new technology firmsto collaborate and the ensuing high propensity ofalliance formation, the global biopharmaceuticalindustry seems an ideal setting to study antecedentsto alliance formation.

Sample and data

To create the sample for this study, we identifiedall pharmaceutical and biotechnology firms activein in-vivo human therapeutics listed in variousvolumes of BioScan, a publicly available indus-try directory.2 The biotech and pharma firms inthe sample are engaged in the discovery, develop-ment, and commercialization of human therapeu-tics that are placed inside the human body (in-vivo). Applying this industry demarcation reflectsthe uniqueness of the human in-vivo segment ofbiotechnology in terms of its economic importanceand potential, its regulatory environment, and itsconsumer market (Powell et al., 1996).

We examined factors that affect alliance forma-tion between a pharmaceutical and a biotechnologyfirm, with the unit of analysis being the dyad.We identified 59 incumbent pharmaceutical firmsand 548 new biotechnology firms engaged glob-ally in human in-vivo therapeutics (59 × 548 =32, 332 pharma–biotech dyads). The time framein which we chose to investigate alliance forma-tion between a pharmaceutical and biotechnologyfirm is the 4-year time window between 1998and 2001.3 We chose this specific time windowfor several reasons: it was characterized by sus-tained high alliance activity in the biopharmaceu-tical industry and was also far enough removedfrom the beginnings of commercialized biotechnol-ogy (marked by Genentech’s founding in 1976 andsuccessful initial public offering in 1980) to assessthe impact of firm- and dyad-level characteristics

2 BioScan provides comprehensive data about the worldwidebiotechnology industry. The data contained in BioScan arecumulative (each subsequent issue includes the information ofall prior versions), which enabled us to track alliance formationover time.3 The results remained robust to variations in the time window.



on alliance formation between biotechnology andpharmaceutical companies. In addition, this timewindow covered both a bull and a bear marketin public equity evaluations. This is salient in ourinvestigation since the condition of the equity mar-ket has been shown to impact alliance formationbetween biotechnology and pharmaceutical com-panies (Lerner, Shane, and Tsai, 2003). Finally,the time window captures recent alliance activityin this industry. To construct the network measuresand other independent variables, we drew on datadocumenting the alliance activity in this industrysince the emergence of biotechnology in 1973, andwere thus able to attenuate a problem of left cen-soring prevalent in prior alliance studies.

Our data sources are multiple issues of theBioScan industry directory and various databasesprovided by Recombinant Capital, an indepen-dent research firm specializing in the life sciences.BioScan and the recap database (by RecombinantCapital ) appear to be the two most comprehen-sive publicly available data sources documentingalliance activity in the global biopharmaceuticalindustry. Prior research addressing different ques-tions has relied on either BioScan or recap data,and thus their usefulness has been validated (Shanet al., 1994; Lane and Lubatkin, 1998; Lerneret al., 2003). To ensure accurateness and complete-ness of our data, we drew data from both BioScanand recap.4 In addition, we obtained patent datafrom the U.S. Patent and Trademark Office (U.S.PTO), and patent citations data from the NationalBureau of Economic Research (NBER) patentdatabase (Hall, Jaffe, and Trajtenberg, 2001).

Dependent variable: alliance formation

We focus on bilateral dyadic relationships that areformalized as strategic alliances between indepen-dent organizations. A focus on bilateral dyadicalliances is appropriate because the biopharma-ceutical industry is generally not characterized bygroup structures or alliance blocks. We consider all32,332 dyads at an equal a priori risk of entering analliance, because we do not believe that excludingcertain dyads is theoretically justifiable. Includingall possible dyads in the alliance opportunity risk

4 We found that these two data sources were quite consistentin their reporting. For example, the inter-source reliability wasgreater than 0.90 when documenting alliances.

set is the most conservative approach in assess-ing different variables’ effect on the probability ofalliance formation (Gulati, 1995a, 1995b). Table 1provides an overview on how each variable wasconstructed and depicts the expected signs basedon the hypotheses advanced. Below, we describeeach variable in detail.

We measured alliance formation in two ways.We focused on the event of alliance formationand the intensity of alliance activity in any givendyad. A similar approach was taken by Parket al. (2002) when studying alliance formation ofsemiconductor start-ups at the firm level, ratherthan at the dyad level. The event of alliance for-mation between an old and a new technologyfirm is proxied by the probability of an alliancebeing formed in any possible dyad opportunityset between a pharmaceutical and a biotechnol-ogy firm. This dependent variable is coded 1 ifthe given dyad formed an alliance, and 0 other-wise.

Clearly, not all alliances are equal in terms ofpartner involvement and commitment. The mostfrequent distinction made in the literature is bet-ween equity and non-equity alliances (Gulati,1995a). Non-equity alliances are contract-basedcooperative agreements, whereas equity alliancesare based on taking an equity stake in a partner,exchanging equity, or setting up a third organiza-tion as a joint venture. As a consequence, non-equity alliances are much more frequent, althoughequity alliances are considered to be stronger ties.To assess problems of unobserved heterogeneitythat can arise when including different alliancetypes as indicators of alliance formation, we alsoapplied the formation of non-equity alliances as asecond dependent variable. This variable is coded1 if the given dyad formed a non-equity alliance,and 0 otherwise.

We proxied the alliance intensity in a given dyadby a count of the total number of alliances formedwithin a pharmaceutical and a biotechnology dyad.Parallel to the distinction between all alliances andnon-equity alliances when assessing the probabil-ity of alliance formation in a given dyad, we alsoassessed alliance intensity by both the total num-ber of all alliances and the total number of non-equity alliances formed. During the 4-year timewindow between 1998 and 2001, 508 dyads hadan alliance event, and a total of 580 alliances wereformed.



Tabl

e1.

Defi

nitio

nof

vari

able

san

dhy

poth

esis

pred

ictio

n

Dep

ende

ntva

riab

les

Defi

nitio

n

Alli

ance

form

atio

n1

=ph

arm

a–bi

otec

hdy

adi

form

edan

allia

nce;

0ot

herw

ise

Alli

ance

form

atio

n(n

on-e

quity

)1

=ph

arm

a–bi

otec

hdy

adi

form

eda

non-

equi

tyal

lianc

e;0

othe

rwis

eA

llian

cein

tens

ityTo

tal

num

ber

ofal

lianc

esfo

rmed

inph

arm

a–bi

otec

hdy

adi

Alli

ance

inte

nsity

(non

-equ

ity)

Tota

lnu

mbe

rof

non-

equi

tyal

lianc

esfo

rmed

inph

arm

a–bi

otec

hdy

adi

Inde

pend

ent

vari

able

sH

ypot

hesi

sPr

edic

tion

Dya

dic

com

plem

enta

rity

Non

-ove

rlap

ping

nich

esas

prox

yfo

rst

rate

gic

inte

rdep

ende

nce

(Gul

ati,

1995

b;C

hung

etal

.,20

00)

Cou

ntof

each

phar

ma-

biot

ech

dyad

i’s

non-

over

lap

inbi

otec

hnol

ogy

subfi

elds

H1

Posi

tive

Bio

tech

firm

’sco

mpe

tenc

ein

drug

deve

lopm

ent

com

bine

dw

ithph

arm

afir

m’s

com

pete

nce

inm

arke

ting

and

dist

ribu

tion

(com

plem

enta

rity

inde

x)

((D

rugs

inde

velo

pmen

t)bi

otec

h i

Mea

n(D

rugs

inde

velo

pmen

t)bi

otec

h i...n

) +(

(SM

&A

) pha

rma i

Mea

n(S

M&

A) p

harm

a i...n

)H

1Po

sitiv

e

Dya

dic

sim

ilar

ity

Pate

ntcr

oss-

cita

tion

rate

s(M

ower

yet

al.,

1996

,19

98)

( Cita

tions

tobi

otec

hfir

mi’

spa

tent

sin

phar

ma

firm

j’s

pate

nts

Tota

lci

tatio

nsin

phar

ma

firm

j’s

pate

nts

)+( C

itatio

nsto

phar

ma

firm

j’s

pate

nts

inbi

otec

hfir

mi’

spa

tent

sTo

tal

cita

tions

inbi

otec

hfir

mi’

spa

tent

s

)H

2Po

sitiv

e

Pate

ntco

mm

onci

tatio

nra

tes

(Mow

ery

etal

.,19

98)

( Cita

tions

inbi

otec

hfir

mi’

spa

tent

sto

pate

nts

cite

din

phar

ma

firm

j’s

pate

nts

Tota

lci

tatio

nsin

biot

ech

firm

i’s

pate

nts

)+( C

itatio

nsin

phar

ma

firm

j’s

pate

nts

topa

tent

sci

ted

inbi

otec

hfir

mi’

spa

tent

sTo

tal

cita

tions

inph

arm

afir

mj

’spa

tent

s

)H

2Po

sitiv

e

Sim

ilari

tyin

pate

ntin

gpr

open

sity

Max

imum

Val

ue−

∣ ∣ ∣ ∣((P

aten

ts) b

iote

chi

Mea

n(P

aten

ts) b

iote

chi...n

) −(

(Pat

ents

) pha

rma i

Mea

n(P

aten

ts) p

harm

a i...n

)∣ ∣ ∣ ∣H

2Po

sitiv

e

Dya

dic

and

firm

-lev

elin

tera

ctio

nsD

yadi

cco

mpl

emen

tari

ty×

Age

biot

ech

firm

iH

3N

egat

ive

Dya

dic

sim

ilar

ity

×A

gebi

otec

hfir

mi

H4

Posi

tive



Tabl

e1.

(Con

tinu

ed)

Mod

erat

ing

vari

able

Defi

nitio

n

Bio

tech

Firm

Age

Age

inye

ars

sinc

efo

undi

ngof

biot

echn

olog

yfir

mi

Con

trol

vari

able

sG

eogr

aphi

cZ

one

1=

phar

ma–

biot

ech

dyad

iis

loca

ted

inth

esa

me

geog

raph

iczo

ne,

0ot

herw

ise

Tim

eE

laps

edT

ime

elap

sed

inm

onth

ssi

nce

late

stal

lianc

eD

irec

tT

ies

Cou

ntof

each

phar

ma–

biot

ech

dyad

i’s

tota

lnu

mbe

rof

prio

rdi

rect

ties

Indi

rect

Tie

sC

ount

ofea

chph

arm

a–bi

otec

hdy

adi’

sto

tal

num

ber

ofin

dire

cttie

sat

degr

eedi

stan

cetw

oC

entr

ality

Bio

tech

Tota

lnu

mbe

rof

allia

nces

the

biot

ech

firm

has

ente

red

with

inth

ene

twor

k(d

egre

ece

ntra

lity)

Cen

tral

ityPh

arm

aTo

tal

num

ber

ofal

lianc

esth

eph

arm

afir

mha

sen

tere

dw

ithin

the

netw

ork

(deg

ree

cent

ralit

y)Fi

rmSi

zeB

iote

chC

ount

num

ber

ofem

ploy

ees

inbi

otec

hfir

mi

Firm

Size

Phar

ma

Cou

ntnu

mbe

rof

empl

oyee

sin

phar

ma

firm

jPu

blic

Bio

tech

1=b

iote

chfir

mi

ispu

blic

,0

othe

rwis

ePu

blic

Phar

ma

1=

phar

ma

firm

jis

pubi

c,0

othe

rwis

ePa

tent

sB

iote

chC

ount

num

ber

ofpa

tent

sas

sign

edto

biot

ech

firm

iPa

tent

sPh

arm

aC

ount

num

ber

ofpa

tent

sas

sign

edto

phar

ma

firm

j



Independent variables: complementarities andsimilarities

We employ multiple measures for complementar-ities and similarities, utilizing measures used inpast research and also developing new measures.We suggest that using multiple proxies enablesresearchers to detect which dimension is morecritical when predicting alliance formation. Thispoint is especially salient in this study becausewe develop more broad-based capability measures,while some of the measures employed in priorresearch tend to be more fine grained, especiallywhen comparing patent measures. In the regressionanalysis, we are thus able to juxtapose broad-basedand fine-grained types of capability relatedness toassess their respective merit in predicting allianceformation.5

Non-overlapping niches

We employed two different measures for a dyad’slevel of complementarities, with the first onederived from prior research. Following Gulati(1995a) and Chung et al. (2000) we used a count ofeach dyad’s non-overlapping niches as a proxy forits strategic interdependence. Population ecologyposits that firms competing in the same organiza-tional niche possess similar resources and capabil-ities (Hannan and Freeman, 1977), which makes apotential resource combination through an allianceredundant and less complementary. On the otherhand, firms active in non-overlapping niches arelimited in their direct competition, which in turnenhances their potential strategic interdependence(Gulati, 1995b; Chung et al., 2000). The biophar-maceutical industry is characterized by a mul-titude of technological trajectories (Dosi, 1982),which can be represented as different organiza-tional niches.

We developed the proxy for non-overlappingniches in four steps. First, we coded each firm’sparticipation/non-participation in different biotech-nology subfields. This resulted in 133 biotechnol-ogy fields across all firms in the sample. Sec-ond, because the descriptions of a firm’s subfieldsare highly technical, we solicited the help of twoindustry experts (one a doctor of pharmacology,

5 Unless specific time windows are indicated, all independentvariables were assessed at t − 1.

the other a research/laboratory scientist), to inde-pendently collapse the 133 different specific cat-egories into broader areas. This resulted in 54distinct biotechnology subfields, with an interraterreliability of 0.99.6 In a third step, we calculatedthe absolute overlap in the 54 technology subfieldsfor each biotech–pharma dyad. This measure rep-resents a count of the overlapping biotechnologysubfields that each biotech–pharma dyad competedin. In a final step, we subtracted this value for eachpharma–biotech dyad from the maximum levelof overlap (54 technology subfields) to obtain themeasure for non-overlapping niches.7

Complementarity index

Prior research has argued that the emergence ofbiotechnology is competence destroying for theupstream R&D competences of incumbent phar-maceutical firms, but competence enhancing fortheir downstream commercialization competences(Powell et al., 1996; Rothaermel and Hill, 2005).Subsequently, access to mutually complementaryassets has been invoked in explaining alliancesbetween biotechnology and pharmaceutical com-panies (Teece, 1992; Rothaermel, 2001). In thesealliances, gains can accrue due to economies ofspecialization as the biotechnology firm focuseson the upstream value chain activities of drugdiscovery and development, and the pharmaceu-tical firm focuses on the downstream value chainactivities like regulatory approval, drug distribu-tion, and marketing. Based on these conceptualarguments, we constructed a measure to proxy theextent of complementarities for each biotechnol-ogy–pharmaceutical dyad. We combined a bio-technology firm’s competence in drug develop-ment with a pharmaceutical firm’s competence indrug commercialization.

We proxied a biotechnology firm’s competencein drug development by the number of biotech-nology drugs the firm had in development. Wecounted only drugs that have entered clinical trialsas being products in development since these prod-ucts have overcome a major hurdle toward success-ful commercialization in the product developmentprocess (De Carolis and Deeds, 1999; Rothaermel

6 A complete list of the 54 biotechnology subfields is availablefrom the first author upon request.7 As a robustness check, we also employed the measure based onthe total number of non-overlapping subfields (133). The resultsremained robust.



and Deeds, 2006). It is important to note that thedrug product development process is beset withhigh uncertainty, with only 1 percent of moleculesscreened making it into clinical trials (Ernst &Young Biotechnology Reports).

Following prior research, we proxied a phar-maceutical firm’s competence in commercializingnew drugs by its expenses allocated to selling,marketing, and administrative activities (SM&A)(Chatterjee and Wernerfelt, 1991; De Carolis,2003). The vast majority of a pharmaceutical com-pany’s SM&A expenses are devoted to managingnew drugs through the regulatory process and dis-tributing them through sales forces (‘detail peo-ple’) that are frequently 15,000 people strong.While the downstream complementary assets nec-essary to commercialize new drugs are specializedresources, and thus partly explain the bargainingpower of incumbent pharmaceutical firms vis-a-vis new biotechnology firms (Teece, 1986), thesedownstream assets also contain a generic compo-nent in the sense that the regulatory process anddrug distribution are more or less identical regard-less of whether the drug is based on biotechnol-ogy (large molecules) or organic chemistry (smallmolecules). This in turn makes this competencefungible across different drug commercializationprojects (Hoang and Rothaermel, 2005). Consis-tent with past research (Chatterjee and Wernerfelt,1991; De Carolis, 2003), we argue that a phar-maceutical company’s SM&A expenses, holdingeverything else constant including firm size, area reasonable proxy for its strength in the down-stream complementary assets needed to commer-cialize new drugs. In support of this notion, priorresearch has identified the pharmaceutical industryas an industry where sales and marketing expen-ditures, besides R&D, are a prime competitiveweapon (Matraves, 1999; De Carolis, 2003).

To proxy each dyad’s level of complementari-ties, we summed the centered ratios of the biotech-nology firm’s drugs in development and the phar-maceutical firm’s SM&A expenses. The centeredratios were developed as follows: first, we countedthe number of drugs the biotechnology firm hadin development. We then centered (or normal-ized) the firm’s number of drugs in developmentaround the mean value for all biotechnology firms(Cohen et al., 2003). A ratio of 1.0 implies that thebiotechnology firm under consideration has exactlyas many biotechnology drugs in development asthe average biotechnology firm; a ratio of 1.2

implies that the firm has 20 percent more drugs indevelopment than the average firm; whereas a ratioof 0.4 implies that the firm has 60 percent fewerdrugs in development in comparison to the aver-age firm. Likewise, we measured a pharmaceuticalfirm’s competence in marketing and distributionby centering the firm’s SM&A expenses aroundthe SM&A mean for all pharmaceutical firms. Thenotion behind complementarities driving allianceformation is that the partners strive to combineassets in which they possess a comparative advan-tage. The sum of these two different competencesis a reasonable proxy for the magnitude of thepotential synergies realizable in an alliance.

Similarities

We constructed three different measures to proxya dyad’s level of similarity, with two derivedfrom prior research. To develop proxies for dyadicsimilarity, we focused on patenting measures toreflect a firm’s innovative competence (Shan et al.,1994; Stuart, 2000). Focusing on patenting mea-sures as proxies for similarities is appropriate,because a firm’s innovativeness has been shownto be a crucial competence in the pharmaceuticalindustry (Matraves, 1999). This industry is char-acterized by a winner-take-all scenario (Arthur,1989), in which innovative firms create tempo-rary monopolies based on proprietary drugs pro-tected by patents, directly affecting firm perfor-mance (Roberts, 1999; De Carolis, 2003). More-over, patents are a more appropriate measure fora firm’s innovative capabilities than, for exam-ple, R&D expenditures, because patents moreaccurately capture the output of R&D activities(Griliches, 1990), which directly underlie a firm’stechnological capabilities (Mowery et al., 1996;De Carolis, 2003).

Following Mowery et al. (1996, 1998), weemploy patent cross-citation and patent commoncitation rates. Each patent granted by the U.S.Patent and Trademark Office contains a sectionin which all prior patents are listed on which thecurrent patent draws. Prior patent citations can beviewed as ‘technological fossils’ representing theintellectual lineage of new patents. In this sense,patent citations are somewhat comparable to ref-erences in academic research. The firms applyingfor a patent have an incentive to be forthcoming inproviding a complete a list as possible of all ‘prior



art’ not only to reduce the probability of interfer-ence being declared during the processing of thepatent application, but also to clearly establish thescope of the patent under evaluation, because onlynovel aspects of the invention result in legally pro-tected property rights. Before granting the patent, apatent examiner verifies the provided list of refer-ences to prior art, and thus serves as an institutionalsafeguard for the integrity of the citation process.

Given the importance of continued innovationin the biopharmaceutical industry (Roberts, 1999;Deeds, De Carolis, and Coombs, 2000; De Carolis,2003), one area of similarities we focus on is thetechnological similarity between a pharmaceuticaland a biotechnology firm. Our first proxy for tech-nological similarity, patent cross-citations, mea-sures the extent to which the pharma–biotech pairin each dyad cites each other’s patents (Moweryet al., 1996, 1998). This dyad-level measure pro-vides a proxy of how important the pharmaceuticalfirm’s patents are among the biotech’s externaltechnology portfolio, and vice versa. The secondproxy for technological similarity, patent commoncitations, measures the degree to which the pharmaand biotech firm in a dyad draw from the sameexternal technology base (Mowery et al., 1998).Dyads with higher cross-citation or common cita-tion rates exhibit higher technological overlap, andthus are more likely to form alliances.

To construct the patent cross-citation andcommon citation measures, we used the NBERpatent database (Hall et al., 2001), which containsdetailed and fine-grained information on all utilitypatents granted by the U.S. PTO between 1963and 1999. The database contains informationon 175,115 companies, 2,923,922 patents, and16,522,438 patent citations.8 We constructed thepatent cross-citation and patent common citationrates for the sample firms between 1994 and1997. We chose a 4-year window for the patentingproxies, because it attenuates fluctuations inany given year, while at the same time it isshort enough to have a reasonable influence onsubsequent alliance formation. Further, using a 4-year time window is consistent with prior researchattempting to proxy innovative capabilities (Stuart

8 Hall et al. (2001: 8, footnote 4) explain that ‘in addition toutility patents, there are three other minor patent categories:Design, Reissue, and Plant patents. The overwhelming majorityare utility patents: in 1999 the number of utility patents grantedreached 153,493, versus 14,732 for Design patents, 448 Reissue,and 421 Plant.’

and Podolny, 1996; Ahuja, 2000). We focusedon patents obtained in the United States becauseit represents worldwide the largest market fortherapeutics, and thus firms generally patent firstin the United States before patenting in any othercountry (Albert et al., 1991). Moreover, firmsactive in biotechnology have a strong incentiveto patent, because intellectual property protectionhas been held up consistently in court and is thusconsidered to be quite strong (Levin et al., 1987).

During the 1994–97 time period, the pharmaand biotech firms in our sample were granted atotal of 17,565 patents, with pharma firms beinggranted 13,949 patents and the biotech firms 3,616patents. These patents contained a total of 116,234patent citations, with 89,819 patent citations inthe pharma patents and 26,415 patent citationsin the biotech patents. During the 1994–97 timeperiod, the 59 pharmaceutical companies and the548 biotechnology firms in the sample cited eachother’s patents 751 times (cross-citation rate). Dur-ing the same time period, the sample firms cited thesame patents 5,460 times (common citation rate).

Both patent cross-citations and patent commoncitations are fine-grained and accepted indicatorsof science and technology relatedness. Our thirddyadic similarity proxy focuses on similaritiesin the absolute level of a firm’s innovativenessby measuring the patenting propensity of eachpharma–biotech dyad. First, we centered eachbiotechnology and pharmaceutical firm’s patent-ing propensity by their respective industry aver-age in the same manner in which we createdthe centered ratios for the complementarity indexdescribed above. Next, for each dyad, we createdits patenting distance measure by taking the abso-lute difference of the centered patenting ratios.This procedure implies that a distance score of zeroreflects complete similarity in patenting propensitybetween the two partners. Finally, we transformedthis patenting distance measure by subtracting itfrom its maximum value so that the resulting mea-sure represents similarity instead of distance.

Moderating variable

New technology firm age

We used the biotechnology firm’s age since found-ing to proxy new technology firm age (BiotechFirm Age). All new technology firms in this sam-ple are fully dedicated biotechnology firms, and



thus age is measured beginning at incorporation(Sorensen and Stuart, 2000). While biotech firmage is the moderating variable of this study, thedirect effect of firm age is included in all regressionmodels. This approach makes moderated regres-sion a conservative method for examining interac-tion effects, because the interaction terms are testedfor significance after all lower-order effects havebeen entered into the regression equation (Jaccard,Wan, and Turrisi, 1990). This enables us to assessthe interaction effects between biotech firm ageand complementarities or similarities on allianceformation, above and beyond the direct effect ofbiotech firm age (Cohen et al., 2003). The averagebiotechnology firm in the sample is 15 years old.

Control variables

We included several variables to control for alter-native factors that may influence the propen-sity of a biotech–pharma pair to enter into analliance. Some of our control variables, like directand indirect ties, were key independent variablesin prior research (Gulati, 1995a, 1995b; Ahuja,2000; Chung et al., 2000; Stuart, 2000; Rosenkopfet al., 2001). Aside from dyadic variables, we alsoincluded firm-level variables to control for firm-level heterogeneity.

Geographic zone

The biotechnology and pharmaceutical firms inthe sample are globally dispersed. This impliesthat the firms are exposed to different institutionalframeworks and cultures, which should affect theirtendency to form alliances. Firms tend to be lesslikely to form alliances across cultural and insti-tutional boundaries. Hagedoorn (2002), for exam-ple, studied international alliances across the TriadNorth America, Europe, and Asia, and has foundthat the companies in most high-tech industriesprefer to partner with firms in their same geo-graphic zone. To control for this effect, we dividedthe firms in our global sample into three geo-graphic zones (North America, Europe, and Asia)based on the location of their headquarters. Thesethree geographic zones capture the economic cen-ters of the world in high technology and accountfor almost all alliance activity (Hagedoorn, 2002).Subsequently, we constructed an indicator variablefor each dyad coded 1 if both firms were located inthe same geographic zone, and 0 otherwise. About

one third of all biotech–pharma pairs were locatedin the same geographic zone.

Time elapsed

Researchers have long argued that organizationsare governed by routines (Cyert and March, 1963).Others have found evidence for the notion thatorganizations have short memories and thus arelikely to repeat activities that they had undertakenin the recent past, with the likelihood of suchan action taking place decreasing as more timeelapses (Amburgey, Kelly, and Barnett, 1993).Forming alliances can be viewed as an exampleof firm routines, with the likelihood of a firmactivating this routine decreasing with the timeelapsed since latest use. Recent empirical work hasfound that the relationship between time elapsedand future alliance formation was characterizedby diminishing returns (Gulati, 1995b). To controlfor this temporal dynamic, we included a variablecapturing the time elapsed in months since latestalliance for each dyad and its squared term.

Direct and indirect ties

Social networks have been found to be reliablepredictors of alliance formation in prior stud-ies across different industries and time frames(Podolny, 1994; Gulati, 1995b; Eisenhardt andSchoonhoven, 1996; Ahuja, 2000; Chung et al.,2000). Social networks are conduits of informationabout the reliability and competence of potentialalliance partners. Direct ties can create dyadic-specific alliance routines, capabilities, and repu-tation spillovers, while indirect ties provide accessto information.

We controlled for a firm’s social embeddednessthrough its direct and indirect ties. We proxieddirect ties, indicating the level of partner-specificprior alliance experience in each dyad (Zollo,Reuer, and Singh, 2002; Hoang and Rothaermel,2005), by the total number of prior alliances foreach pair (Gulati, 1995a, 1995b; Ahuja, 2000).For dyads without prior direct ties, we assessedtheir level of indirect ties through a count numberof common partners shared (Powell et al., 1996).Here we drew on the complete network in thebiotechnology field, including research universi-ties, nonprofit research organizations, and govern-ment agencies—such as the National Institutes ofHealth—and counted the number of indirect ties



at degree distance two. For example, the pharma-ceutical company Lilly was not directly tied tothe biotechnology firm Apollon, but the two firmsin this dyad were indirectly tied to one anotherbecause both firms had a direct tie to the biotech-nology firm Biogen. We argue that Biogen canserve as a bridge between Lilly and Apollon andthus facilitate future alliance formation. Likewise,research universities and nonprofit research insti-tutions can also serve as bridges between uncon-nected pharmaceutical and biotechnology firms.

Given the relatively short time span since theemergence of biotechnology and the longevity ofalliances in this industry (successful new productdevelopment and commercialization can take up to15 years and patents provide a 20-year protection),we considered all prior direct and indirect ties, con-sistent with prior research (Gulati, 1995a, 1995b;Singh, 1997). A total of 531 biotech–pharmadyads already had a prior alliance before weassessed future alliance formation. The biotech–pharma dyads that did not have a prior alliancewere linked indirectly to one another through 2,472indirect ties at degree distance two.

Network centrality

We proxied each firm’s embeddedness in theindustry network through a firm’s degree central-ity. We constructed the industry network for thebiotechnology and pharmaceutical firms to assesseach firm’s level of embeddedness. We proxieda firm’s centrality in the network by the totalnumber of ties the firm had entered within thepharma–biotech network. We used this procedurefor both the biotechnology and the pharmaceuticalfirm to construct a firm-level network centralitymeasure for new and old technology firms (Cen-trality Biotech and Centrality Pharma). Employ-ing a count number of all alliances entered intowithin the local network as a proxy for firm net-work centrality is consistent with prior research(Powell et al., 1996; Ahuja, 2000). When con-structing the network centrality measures, we werecareful not to include direct ties between twodyad partners to ensure the independence of thenetwork centrality and direct ties measures, andto reduce the threat of finding spurious resultsbased on an inflated network centrality measure.To assess a potential diminishing returns effect,we included the linear and squared term of net-work centrality. Within the local biotech–pharma

network, the average biotech firm had entered atleast one alliance, while the average pharma firmhad entered more than nine alliances.

Firm size

A critical control variable in isolating the dyadiceffect of complementarities and similarities onalliance formation is firm size. Controlling for firmsize helps to avoid finding spurious age effectsdue to the expected positive correlation betweenfirm age and firm size, and thus controlling forfirm size isolates the firm age effect and reducesthe threat of unobserved heterogeneity (Barron,West, and Hannan, 1994). Controlling for newtechnology firm size is indicated when assessinghow the new technology firm’s age interacts withcomplementarities and similarities in predictingalliance formation. A similar approach was takenby Sorensen and Stuart (2000) when assessingthe effect of organizational aging on firm-levelinnovation.

We proxied firm size through the number ofemployees for each biotechnology firm (Firm SizeBiotech). Using the number of employees as aproxy of firm size is the preferred measure inthis industry, because many biotechnology firmsdo not yet have any positive revenues that wouldallow the use of more traditional size measures likemarket share. Moreover, the assets of dedicatedbiotechnology firms are largely intangible. Accord-ingly, the number of employees as proxy forbiotechnology firm size has been used in a num-ber of prior studies (Powell et al., 1996; Sorensenand Stuart, 2000; Rothaermel and Deeds, 2004).In parallel, we proxied for the size of the phar-maceutical companies through the number of itsemployees (Firm Size Pharma). The difference insize between the two different partners is striking:the average biotechnology firm had 299 employ-ees, while the average pharmaceutical companywas more than 80 times larger, with approximately25,000 employees.

Ownership status

We controlled for whether the firms were in pub-lic or in private ownership. Public biotechnologyfirms might have more similar business processesand procedures compared to pharmaceutical com-panies than private biotechnology firms have. Wealso controlled whether the ownership status of



the pharmaceutical company might influence itspropensity to enter alliances. To control for own-ership effects, we included a firm-level dummyvariable that takes the value 1 if the firm is pub-lic, and 0 otherwise (Public Biotech and PublicPharma).

Patenting

To test Hypothesis 2, we developed the dyadic sim-ilarity measures of patent cross-citations, patentcommon citations, and patenting propensity. Tofurther isolate the effects of these patent-basedsimilarity constructs on alliance formation, andthus to reduce the threat of unobserved hetero-geneity, we additionally controlled for firm-levelpatenting effects. Through the inclusion of thefirm-level patent propensities, we were able toassess the unique effect of the dyad-level patentconstructs above and beyond firm-level patent-ing. We included a count number of the totalpatents assigned between 1994 and 1997 to thebiotechnology and pharmaceutical firms in oursample (Patents Biotech and Patents Pharma).The average biotechnology firm was granted sevenpatents, while the average pharmaceutical com-pany obtained 236 patents.

Estimation method

Because we focus on the event of alliance for-mation and the intensity of interfirm cooperationas two separate dependent variables, we neededto employ two different econometrical models(Greene, 1997). The first dependent variable mea-sures the event of alliance formation and is binaryin nature (0, 1), and thus for estimation we useda logit regression. The outcome variable, Y , isthe probability of alliance formation/non-formationbased on a nonlinear function with two outcomes.The logit model is estimated with a maximum like-lihood procedure with the following specification:

ln

(Y

1 − Y

)= α +

∑βj Xij

where Xij is a vector of independent variables.The second dependent variable proxies the alli-

ance intensity in each dyad and is measuredthrough a count number of the total number ofalliances formed between each dyadic pair. Herethe dependent variable is a non-negative integer

count with a limited range, and a negative bino-mial regression is the preferred estimation tech-nique (Greene, 1997). A negative binomial regres-sion model relaxes the restrictive assumption ofmean and variance equality inherent in the Pois-son model and accounts for omitted variable bias,while estimating heterogeneity. We applied the fol-lowing specification:

P(ni/ε) = e−λi exp(ε)λni

i /ni!

where n is a non-negative integer count variablecapturing the alliance intensity in each dyad, andthus P(ni/ε) indicates the probability that dyad i

will form n alliances.9

RESULTS

All of the bivariate correlations between the inde-pendent variables fall below the 0.70 threshold,thus indicating acceptable discriminant validity(Cohen et al., 2003). The bivariate correlationsamong the five variables to proxy complementar-ities and similarities are quite low, with the high-est being between the complementarity index andpatenting propensity (r = −0.29). It is importantto note that the bivariate correlations between thebroad-based capability measures (non-overlappingniches, complementarity index, patenting propen-sity), and the fine-grained technology relatednessmeasures (patent cross-citation and patent commoncitation rates) are all less than 0.092, and thusreflective of low or non-existing bivariate corre-lations (range: −0.007 ≤ r ≤ 0.091). This obser-vation becomes pertinent when later assessing theutility of fine-grained vs. more broad-based capa-bilities in predicting alliance formation.10 Table 2depicts the descriptive statistics and the bivariatecorrelation matrix.

9 To interpret the results in a meaningful manner, we standard-ized all independent variables before entering them into thevarious regression models. The variables contained in the cross-products of the interaction terms to test the moderating hypothe-ses were standardized before creating the respective cross-products. Standardizing the independent variables improves therobustness of the analysis without degrading the quality of thedata or affecting the level of statistical findings. While this proce-dure allows the researcher to directly compare beta coefficients,the level of significance is not affected (Cohen et al., 2003).10 We are indebted to the anonymous reviewers for suggesting toemploy existing fine-grained measures of science and technologyrelatedness, and to compare them to the more broad-basedcapability measures developed for this study.



We assess the event of alliance formation andthe intensity of allying in any given biotech–pharma dyad. In Models 1–3 we assess the prob-ability of alliance formation, while in Models4–6 we evaluate the probability for the forma-tion of non-equity alliances to correct for potentialunobserved heterogeneity due to different alliancetypes. In parallel, in Models 7–9 we assess allianceintensity, while in Models 10–12 we evaluate thealliance intensity for non-equity alliances. We useda hierarchical regression approach. The first modelin each of the four blocks (Models 1, 4, 7, and 10)contains the control variables only and serves as abaseline model. In subsequent models, we enteredthe variables for complementarity and similarity(Models 2, 5, 8, and 11). In the final model ofeach regression block we added the hypothesizedinteraction effects (Models 3, 6, 9, and 12). In thefour regression blocks, each model represents asignificant improvement over the baseline model(at p < 0.01 or smaller). Tables 3 and 4 presentthe regression results.

Direct effect hypotheses

In Hypothesis 1 we postulated that complementar-ities between old technology and new technologyfirms increase the probability of alliance forma-tion, while in Hypothesis 2 we argued that simi-larities enhance the chance of future allying. Wefind support for the two hypotheses across all fourregression blocks when focusing on a dyad’s com-plementarity index, patent cross-citation rates, andoverall patenting propensity. When comparing theeffects of the two complementarity measures onalliance formation, we find that the dyadic com-plementarity index, based on the biotech firm’scompetences in drug development and the pharmafirm’s competences in marketing and distribution,is positive and significant in all four direct effectsmodels at p < 0.01 or smaller (Models 2, 5, 8,and 11);11 non-overlapping niches are not signifi-cant in predicting alliance formation. When assess-ing the effects of the three similarity measures

11 These results cannot be attributed reasonably to outliers, sincethe coefficient of variance (= S.D./mean) is only 0.75, andthus indicates that overdispersion is not a problem. This isimportant because it highlights the notion that the positive effectof the complementarity index on alliance formation is driven byboth the new and old technology firms scoring high on theirrespective dimensions proxying for upstream and downstreamcompetences, rather than by outliers of one firm in the dyad.

on alliance formation, we find that both a dyad’spatent cross-citation rate as well as its overallpatenting propensity are positive and significant inpredicting alliance formation (p < 0.05 in Models2, 5, and 11, and p < 0.10 in Model 8, for bothvariables). Contrary to our expectations, a dyad’spatent common citation rate, however, is negativeand significant, and thus reduces the probability ofalliance formation (p < 0.05 in Models 2, 5, and11).

To illustrate the findings, we use the coefficientsfrom Model 2, the fully specified direct effectsestimation, to evaluate the effect of the differ-ent complementarity and similarity measures onthe probability of alliance formation. Holding allother variables constant, a dyad’s complementar-ity index multiplies the rate of alliance formationby a factor of 1.160 (exp(β), here exp(0.1482)).When considering the similarity measures, we findthat a dyad’s cross-citation rate multiplies the rateof alliance formation by a factor of 1.042, and adyad’s patenting propensity by a factor of 1.096.In contrast, a dyad’s patent common citation ratereduces the rate of alliance formation by a factorof 0.879.

Taken together, the results suggest that thebroad-based capability measures developed for thisstudy, the complementarity index and the over-all patenting propensity, perform well in predict-ing alliance formation across different dependentvariables and estimation procedures. Both dyadiccomplementarity across different competences inthe value chain and a similar overall orientationtowards innovation predict the event of allianceformation as well as the intensity of allying in thebiotech–pharma dyads under investigation. On theother hand, only a dyad’s patent cross-citation ratesbehave as expected, while a dyad’s patent com-mon citations behave opposite to the hypothesizeddirection.

Interaction hypotheses

In the interaction hypotheses we predicted thatincumbent and new technology firms are morelikely to enter an alliance based on complemen-tarities when the new technology firm is younger(Hypothesis 3), while alliance formation based onsimilarities is more likely when the new technologyfirm is older (Hypothesis 4). When applying adyad’s complementarity index, we find consistentsupport for Hypothesis 3 across the four models



Tabl

e2.

Des

crip

tive

stat

istic

san

dbi

vari

ate

corr

elat

ion

mat

rix

Mea

nS.

D.

1.2.

3.4.

5.6.

7.8.

9.10

.11

.12

.13

.14

.15

.16

.17

.18

.19

.20

.21

.

1.A

llian

cefo

rmat

ion

(%)

1.57

112

.436

2.A

llian

cefo

rmat

ion

(non

-equ

ity)

(%)

1.54

612

.339

0.99

2

3.A

llian

cein

tens

ity(%

)1.

794

15.4

680.

701

0.70

5

4.A

llian

cein

tens

ity(n

on-e

quity

)(%

)1.

775

15.4

110.

912

0.91

90.

778

5.G

eogr

aphi

czo

ne0.

340

0.47

40.

035

0.03

30.

016

0.02

66.

Tim

eel

apse

d0.

589

6.07

90.

084

0.08

50.

087

0.10

80.

021

7.D

irec

ttie

s0.

024

0.21

90.

091

0.09

00.

088

0.12

60.

025

0.69

58.

Indi

rect

ties

0.07

60.

305

0.06

90.

069

0.04

30.

059

0.02

0−0

.006

−0.0

029.

Cen

tral

itybi

otec

h0.

998

1.47

30.

081

0.07

90.

076

0.07

6−0

.006

0.15

70.

174

0.30

2

10.

Cen

tral

ityph

arm

a9.

220

8.45

60.

099

0.09

90.

067

0.09

00.

284

0.09

10.

100

0.14

30.

000

11.

Firm

size

biot

ech

299.

0698

9.18

0.03

10.

031

0.03

30.

033

−0.0

020.

041

0.02

80.

106

0.12

60.

000

12.

Firm

size

phar

ma

2467

6.71

2575

8.27

0.07

10.

071

0.06

30.

065

0.21

40.

049

0.05

60.

057

0.00

00.

532

0.00

0

13.

Publ

icbi

otec

h0.

578

0.49

40.

052

0.05

20.

051

0.05

00.

001

0.05

40.

057

0.10

10.

323

0.00

00.

092

0.00

014

.Pu

blic

phar

ma

0.81

40.

389

0.01

60.

016

0.01

40.

018

0.04

90.

014

0.01

70.

009

0.00

00.

090

0.00

00.

191

0.00

015

.Pa

tent

sbi

otec

h6.

599

19.6

740.

059

0.05

70.

069

0.06

80.

001

0.05

80.

059

0.14

10.

257

0.00

00.

393

0.00

00.

074

0.00

016

.Pa

tent

sph

arm

a23

6.44

285.

540.

021

0.01

90.

025

0.01

40.

219

0.02

30.

022

0.02

50.

000

0.26

40.

000

0.51

60.

000

−0.0

220.

000

17.

Bio

tech

firm

age

15.1

355.

784

0.01

20.

011

0.01

70.

016

0.00

10.

034

0.01

70.

037

0.07

10.

000

0.15

20.

000

0.16

30.

000

0.10

30.

000

18.

Non

-ove

r-la

ppin

gni

ches

51.6

911.

884

−0.0

32−0

.031

−0.0

30−0

.032

−0.0

22−0

.052

−0.0

47−0

.063

−0.1

00−0

.130

−0.1

62−0

.106

0.02

9−0

.125

−0.1

12−0

.034

−0.0

92

19.

Com

plem

en-

tari

tyin

dex

2.01

51.

518

0.09

30.

093

0.08

10.

087

0.19

20.

085

0.08

70.

155

0.27

10.

391

0.10

10.

535

0.24

6−0

.030

0.15

00.

334

0.04

3−0

.144

20.

Pate

ntcr

oss-

cita

tion

rate

(%)

0.06

80.

953

0.02

10.

021

0.01

40.

018

0.04

40.

017

0.01

70.

034

0.02

00.

047

0.01

10.

018

0.01

60.

016

0.04

20.

027

0.01

6−0

.018

0.03

7

21.

Pate

ntco

mm

onci

tatio

nra

te(%

)2.

540

10.1

770.

000

0.00

00.

006

0.00

0−0

.012

0.02

40.

026

0.05

20.

121

0.00

10.

012

−0.0

150.

096

−0.0

420.

100

0.01

1−0

.003

−0.0

070.

091

0.05

4

22.

Pate

ntin

gpr

open

sity

23.0

262.

172

−0.0

43−0

.041

−0.0

53−0

.045

−0.0

91−0

.069

−0.0

72−0

.144

−0.2

49−0

.087

−0.4

03−0

.212

−0.0

580.

015

−0.6

00−0

.424

−0.0

950.

170

−0.2

90−0

.013

−0.0

67

N=

32,33

2ph

arm

a-bi

otec

hdy

ads



Table 3. Logistic regression estimates of dyad-level alliance formation

Alliance formation Alliance formation (non-equity)

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6

Constant −4.4309∗∗∗ −4.4704∗∗∗ −4.4780∗∗∗ −4.4643∗∗∗ −4.5049∗∗∗ −4.5114∗∗∗

(0.0727) (0.0746) (0.0751) (0.0737) (0.0756) (0.076)Geographic zone 0.0824∗ 0.0663† 0.0647† 0.0718† 0.0556 0.0539

(0.0469) (0.0472) (0.0473) (0.0473) (0.0477) (0.0477)Time elapsed 0.1865∗∗ 0.1782∗∗ 0.1839∗∗ 0.1910∗∗ 0.1826∗∗ 0.1895∗∗

(0.0645) (0.0648) (0.0647) (0.0646) (0.0649) (0.0648)(Time elapsed)2 −0.0079∗ −0.0075∗ −0.0077∗ −0.0080∗ −0.0076∗ −0.0078∗

(0.0037) (0.0037) (0.0037) (0.0037) (0.0037) (0.0037)Direct ties 0.0438† 0.0495∗ 0.0475∗ 0.0425† 0.0482∗ 0.0459∗

(0.0273) (0.0272) (0.0273) (0.0275) (0.0274) (0.0275)Indirect ties 0.1766∗∗∗ 0.1762∗∗∗ 0.1796∗∗∗ 0.1774∗∗∗ 0.1768∗∗∗ 0.1807∗∗∗

(0.0526) (0.0527) (0.0528) (0.0528) (0.0529) (0.0529)(Indirect ties)2 −0.0130∗ −0.0127∗ −0.0127∗ −0.0126∗ −0.0123∗ −0.0124∗

(0.0074) (0.0075) (0.0075) (0.0074) (0.0075) (0.0075)Centrality biotech 0.3379∗∗∗ 0.2985∗∗∗ 0.3147∗∗∗ 0.3473∗∗∗ 0.3069∗∗∗ 0.3255∗∗∗

(0.0685) (0.0704) (0.0715) (0.0693) (0.0712) (0.0725)(Centrality biotech)2 −0.0440∗∗ −0.0322∗ −0.0347∗ −0.0484∗∗ −0.0361∗ −0.0401∗

(0.0176) (0.0188) (0.0199) (0.0179) (0.0192) (0.0205)Centrality pharma 0.7092∗∗∗ 0.6696∗∗∗ 0.6678∗∗∗ 0.6936∗∗∗ 0.6541∗∗∗ 0.6521∗∗∗

(0.0895) (0.0905) (0.0906) (0.0904) (0.0913) (0.0915)(Centrality pharma)2 −0.1436∗∗ −0.1252∗∗ −0.1246∗∗ −0.1235∗∗ −0.1043∗ −0.1037∗

(0.0497) (0.0503) (0.0503) (0.0499) (0.0505) (0.0506)Firm size biotech 0.0127 0.0317 0.0407 0.0192 0.0383 0.0477

(0.0380) (0.0418) (0.0413) (0.0378) (0.0416) (0.0412)Firm size pharma 0.1924∗∗∗ 0.1099∗ 0.1029∗ 0.2056∗∗∗ 0.1221∗ 0.1151∗

(0.0538) (0.0601) (0.0603) (0.0543) (0.0607) (0.0609)Public biotech 0.3374∗∗∗ 0.3199∗∗∗ 0.3116∗∗∗ 0.3307∗∗∗ 0.3115∗∗∗ 0.3032∗∗∗

(0.0586) (0.0602) (0.0601) (0.0589) (0.0605) (0.0604)Public pharma 0.0054 0.0172 0.0175 −0.0135 −0.0006 −0.0003

(0.0569) (0.0585) (0.0585) (0.057) (0.0585) (0.0586)Patents biotech 0.1424∗∗∗ 0.1648∗∗∗ 0.1654∗∗∗ 0.1384∗∗∗ 0.1603∗∗∗ 0.1605∗∗∗

(0.0284) (0.0287) (0.0287) (0.0289) (0.0293) (0.0293)Patents pharma −0.0689† −0.0436 −0.0324 −0.0865∗ −0.0619 −0.0476

(0.0502) (0.0526) (0.0551) (0.051) (0.0534) (0.0559)Biotech firm age −0.0286 −0.0386 0.0187 −0.0302 −0.0398 0.0188

(0.0503) (0.0506) (0.0553) (0.0507) (0.0510) (0.0557)Dyadic complementaritiesNon-overlapping niches −0.0151 −0.0200 −0.0084 −0.0134

(0.0449) (0.0464) (0.0454) (0.0467)Complementarity index 0.1482∗∗ 0.1654∗∗∗ 0.1510∗∗ 0.1681∗∗∗

(0.0496) (0.0502) (0.0499) (0.0506)Dyadic similaritiesPatent cross-citation rate 0.0408∗ 0.0398∗ 0.0419∗ 0.0411∗

(0.0228) (0.0238) (0.0226) (0.0237)Patent common citation rate −0.1293∗ −0.1349∗ −0.1230∗ −0.1280∗

(0.0599) (0.0616) (0.0592) (0.0610)Patenting propensity 0.0919∗ 0.1139∗ 0.0911∗ 0.1209∗

(0.0467) (0.0612) (0.0473) (0.0621)Dyadic and firm-level interactionsNon-overlapping niches 0.0072 0.009

× Biotech firm age (0.0433) (0.0438)Complementarity index −0.1310∗∗ −0.1294∗∗

× Biotech firm age (0.0458) (0.0461)Patent cross-citation rate 0.0002 0.0010

× Biotech firm age (0.0302) (0.0302)



Table 3. (Continued )

Alliance formation Alliance formation (non-equity)


Patent common citation rate × −0.0056 0.0009Biotech firm age (0.1037) (0.1029)

Patenting propensity × Biotech −0.0406 −0.0491firm age (0.0425) (0.0432)

Log-likelihood −2325.17 −2316.01 −2311.30 −2294.24 −2285.28 −2280.65Likelihood ratio test (χ 2) 577.42∗∗∗ 595.74∗∗∗ 605.16∗∗∗ 572.95∗∗∗ 590.87∗∗∗ 600.14∗∗∗

Improvement over base model(�χ 2)

18.32∗∗ 27.74∗∗ 17.92∗∗ 27.19∗∗

N 32,332 32,332 32,332 32,332 32,332 32,332

†p < 0.10; ∗ p < 0.05; ∗∗ p < 0.01; ∗∗∗ p < 0.001; standard errors in parentheses.

Table 4. Negative binomial regression estimates of dyad-level alliance intensity

Alliance intensity Alliance intensity (non-equity)


Constant −4.2830∗∗∗ −4.3078∗∗∗ −4.3211∗∗∗ −4.3545∗∗∗ −4.3991∗∗∗ −4.4149∗∗∗

(0.0687) (0.0703) (0.0709) (0.0692) (0.0712) (0.0719)Geographic zone −0.0226 −0.0372 −0.0378 0.0486 0.0317 0.0299

(0.0442) (0.0446) (0.0447) (0.0444) (0.0448) (0.0449)Time elapsed 0.1877∗∗∗ 0.1792∗∗ 0.1819∗∗ 0.1933∗∗∗ 0.1850∗∗∗ 0.1889∗∗∗

(0.0604) (0.0608) (0.0605) (0.0559) (0.0564) (0.0563)(Time elapsed)2 −0.0070∗ −0.0068∗ −0.0067∗ −0.0076∗∗ −0.0073∗ −0.0073∗∗

(0.0033) (0.0034) (0.0033) (0.0031) (0.0032) (0.0031)Direct ties 0.0174 0.0231 0.0224 0.0525∗∗ 0.0579∗∗ 0.0560∗∗

(0.0245) (0.0246) (0.0247) (0.0209) (0.0208) (0.0209)Indirect ties 0.1579∗∗ 0.1554∗∗ 0.1578∗∗ 0.1805∗∗∗ 0.1792∗∗∗ 0.1832∗∗∗

(0.0589) (0.0590) (0.0590) (0.0514) (0.0516) (0.0518)(Indirect ties)2 −0.0202∗ −0.0199∗ −0.0194∗ −0.0142∗ −0.0138∗ −0.0137∗

(0.0101) (0.0101) (0.0101) (0.0075) (0.0075) (0.0076)Centrality biotech 0.3301∗∗∗ 0.2929∗∗∗ 0.3049∗∗∗ 0.3068∗∗∗ 0.2617∗∗∗ 0.2754∗∗∗

(0.0633) (0.0651) (0.0660) (0.0651) (0.0669) (0.0679)(Centrality biotech)2 −0.0451∗∗ −0.0362∗ −0.0352∗ −0.0525∗∗∗ −0.0387∗ −0.0381∗

(0.0160) (0.0172) (0.0181) (0.0166) (0.0178) (0.0188)Centrality pharma 0.3854∗∗∗ 0.3498∗∗∗ 0.3480∗∗∗ 0.6337∗∗∗ 0.5948∗∗∗ 0.5942∗∗∗

(0.0765) (0.0773) (0.0774) (0.0848) (0.0857) (0.0858)(Centrality pharma)2 −0.0611† −0.0494 −0.0492 −0.1085∗ −0.0895∗ −0.0895∗

(0.0460) (0.0466) (0.0467) (0.0468) (0.0473) (0.0474)Firm size biotech 0.0075 0.0176 0.0257 0.0107 0.0326 0.0405

(0.0347) (0.0391) (0.0385) (0.0351) (0.0391) (0.0384)Firm size pharma 0.2642∗∗∗ 0.2021∗∗∗ 0.1979∗∗∗ 0.1991∗∗∗ 0.1092∗ 0.1027∗

(0.0472) (0.0524) (0.0525) (0.0514) (0.0575) (0.0577)Public biotech 0.3825∗∗∗ 0.3657∗∗∗ 0.3577∗∗∗ 0.3685∗∗∗ 0.3497∗∗∗ 0.3416∗∗∗

(0.0562) (0.0577) (0.0577) (0.0564) (0.0580) (0.0579)Public pharma 0.0111 0.0205 0.0212 0.0180 0.0300 0.0299

(0.0517) (0.0528) (0.0529) (0.0544) (0.0558) (0.0558)Patents biotech 0.1598∗∗∗ 0.1718∗∗∗ 0.1736∗∗∗ 0.1528∗∗∗ 0.1740∗∗∗ 0.1759∗∗∗

(0.0237) (0.0247) (0.0246) (0.0246) (0.0249) (0.0249)Patents pharma 0.0047 0.0261 0.0305 −0.1140∗ −0.0856∗ −0.0795†

(0.0442) (0.0464) (0.0483) (0.0497) (0.0517) (0.0537)Biotech firm age 0.0166 0.0082 0.0667† 0.0149 0.0059 0.0792†

(0.0461) (0.0465) (0.0488) (0.0467) (0.0469) (0.0506)




Alliance intensity Alliance intensity (non-equity)


Dyadic complementaritiesNon-overlapping niches −0.0319 −0.0402 −0.0129 −0.0249

(0.0424) (0.0439) (0.0426) (0.0440)Complementarity index 0.1225∗∗ 0.1422∗∗ 0.1589∗∗∗ 0.1809∗∗∗

(0.0466) (0.0471) (0.0467) (0.0473)Dyadic similaritiesPatent cross-citation rate 0.0349† 0.0299 0.0420∗ 0.0419∗

(0.0247) (0.0269) (0.0220) (0.0228)Patent common citation rate −0.0430 −0.0408 −0.1310∗ −0.1337∗

(0.0438) (0.0450) (0.0576) (0.0591)Patenting propensity 0.0644† 0.0676 0.0972∗ 0.1035∗

(0.0415) (0.0533) (0.0432) (0.0561)Dyadic and firm-level interactionsNon-overlapping niches 0.0164 0.0267

× Biotech firm age (0.0405) (0.5041)Complementarity index −0.1334∗∗∗ −0.1508∗∗∗

× Biotech firm age (0.0418) (0.0428)Patent cross-citation rate −0.0169 0.0019

× Biotech firm age (0.0329) (0.0294)Patent common citation rate 0.0574 0.0521

× Biotech firm age (0.0742) (0.0961)Patenting propensity × Biotech −0.0182 −0.0273

firm age (0.0373) (0.0399)Log-likelihood −2655.371 −2649.81 −2643.48 −2558.33 −2547.22 −2539.57Likelihood ratio test (χ 2) 523.71∗∗∗ 534.83∗∗∗ 547.50∗∗∗ 669.26∗∗∗ 691.49∗∗∗ 706.78∗∗∗

Improvement over base model(�χ 2)

11.12∗ 23.79∗∗ 22.23∗∗∗ 37.52∗∗∗

N 32,332 32,332 32,332 32,332 32,332 32,332


assessing the interaction effects (Models 3, 6, 9,and 12). In each of these models, the interactionbetween a dyad’s complementarity index and theage of the biotechnology firm is negative and sig-nificant (p < 0.01 in Models 3 and 6, and p <

0.001 in Models 9 and 12), implying that the rateand intensity of alliance formation motivated bycomplementarities decrease as the new technol-ogy firm ages. Holding all else constant, whenthe biotechnology firm has reached the mean ageof the sample firms (15.135 years), alliance for-mation between an incumbent pharmaceutical anda new biotechnology firm based on complemen-tarities is reduced by a multiplier factor of 0.138(exp(−0.131 × 15.135) in Model 3); one standarddeviation below the mean of biotech firm age(9.351 years) it is reduced by a factor of 0.294; andone standard deviation above the mean of biotech-nology firm age (20.919 years) it is reduced by afactor of 0.065. Figure 1 graphically depicts how

the odds of forming an alliance based on comple-mentarities between a pharma and a biotech firmdecreases non-monotonically with the age of thenew biotechnology firm. Very young biotechnol-ogy firms are highly likely to enter an alliancewith a pharmaceutical firm based on complemen-tarities; however, this effect wanes drastically asthe biotechnology firm ages, thus providing visualsupport for Hypothesis 3.

The regression results also reveal that none ofthe other interaction effects reach statistical signif-icance. Thus, we fail to find support for Hypothe-sis 4.

Effects of control variables

The results for the control variables are consistentwith the key findings of prior studies conductedin different industries and at different time frames(Gulati, 1995a, 1995b; Ahuja, 2000; Chung et al.,



0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

3.567−2s.d.

9.351−1s.d.

15.135mean

20.919+1s.d.

26.703+2s.d.

Biotech firm age

Od

ds

rati

o a

llian

ce f

orm

atio

ncomplementarities × biotech firm age

Figure 1. Moderating effect of biotechnology firm ageon the relationship between dyadic complementarities and

alliance formation

2000; Stuart, 2000; Rosenkopf et al., 2001). Whenconsidering the probability of alliance formation,the regression models reveal a (marginally) signif-icant relationship for location—new and old tech-nology firms are more likely to enter an alliance ifthey are located in the same geographic zone.

We also find that the relationship between timeelapsed and future alliance formation appears tobe characterized by diminishing returns based onthe significance of the linear (positive sign) andsquare term (negative sign) of time elapsed sincelast alliance in dyads with prior ties. The num-ber of prior direct ties has a positive impact onfuture alliance formation when considering theprobability of alliance formation and the intensityof allying using non-equity alliances.12 Prior tiesenable firms to be familiar with one another and tobuild trust (Gulati, 1995a). The results also revealthat the number of prior indirect ties that a firmpair shares is related to future alliance formationand alliance intensity in a non-monotonic manner.The marginal informational value about a poten-tial partner appears to decrease after more than acouple of indirect ties.

At the firm level, we find that centrality in thelocal network, for both the biotech and the pharmafirm, is related to alliance formation and allianceintensity in a non-monotonic manner. Both the

12 We also tested for a non-monotonic effect of direct ties onalliance formation and alliance intensity through inclusion of thesquared term of direct ties. We found no support for diminishingreturns to prior allying.

smaller biotechnology firms and the larger pharma-ceutical companies may be exposed to a potentialoverembeddedness in the local network. Consis-tently, we also find that the larger the pharma-ceutical company, the more likely it is to enteralliances with biotechnology firms. Pharmaceuticalfirms like Merck or Pfizer that compete in the pro-prietary drug development arena tend to be larger,and thus have more at stake when confronted witha discontinuity like biotechnology. Results of thecontrol variables also reveal that public biotechnol-ogy firms are more active in forming alliances withpharmaceutical companies. At the dyad level, wefound that similarities in patent cross-citation ratesand patenting propensity increased the probabilityof alliance formation, which provided supportedfor Hypothesis 2. This result held while controllingfor firm-level heterogeneity in patenting propen-sity. We found that biotech firms that patent moreare more likely to form alliances.

Post hoc analysis

Biotechnology firm age is the moderator vari-able of this study, and we suggested that agingenhances a biotechnology firm’s credibility, legiti-macy, and power. These theoretical constructs canarguably also be captured by alternative measuressuch as whether the biotechnology firm is publicor through biotechnology firm size.13 Preliminaryevidence for this notion can be found in the posi-tive and significant bivariate correlations betweenbiotechnology firm age and biotechnology firmpublic (r = 0.163; p < 0.01) and between biotech-nology firm age and biotechnology firm size (r =0.152; p < 0.01). A biotechnology firm that hassuccessfully completed an initial public offeringhas transformed itself from a privately ownedentrepreneurial venture into a publicly owned com-pany, increasing its standing vis-a-vis potentialalliance partners. Likewise, the power differentialbetween biotech and pharma firms decreases as thebiotechnology firm grows in size. Thus, one wouldexpect that biotechnology firms that are privatelyheld or are smaller in size are more likely to enteralliances with pharma firms based on complemen-tarities (Hypothesis 3), while biotechnology firmsthat are public or are larger in size would be more

13 We thank an anonymous reviewer for suggesting these alter-native model specifications.



likely to enter alliances with pharma firms basedon similarities (Hypothesis 4).

Models 13–16 in Table 5 display the resultswhen using public firm status (biotech firm pub-lic = 1) as a moderator. Model 16 indicates thatthe interaction between a dyad’s complementarityindex and public biotech is negative and signifi-cant, thus providing some support for Hypothesis3. Models 13–15 reveal a positive and significantinteraction between a dyad’s patenting propensityand biotech firm public (p < 0.05 or smaller),thus providing support for Hypothesis 4. Takentogether, when applying public status as moder-ator of the relationship between complementari-ties, similarities, and alliance formation, we foundsupport for both interaction hypotheses advancedabove. When applying biotechnology firm size asmoderator, we found some support for Hypothesis3: the interaction between a dyad’s complemen-tarity index and biotech firm size was negativeand significant (p < 0.05) when predicting theformation of non-equity alliances.14 In sum, theresults for the interaction hypotheses are robust,and even provide support for Hypothesis 4. Asin the regression models presented earlier, it isthe more broad-based complementarity and sim-ilarity measures that reach statistical significancewhen interacting them with biotech firm public orbiotech firm size, and not the more fine-grainedscience and technology indicators.

Our theoretical assumption, which is borne outby the empirical findings of our study, is thatboth complementarities and similarities betweenbiotechnology and pharmaceutical firms lead toa greater probability of alliance formation. Wecannot definitively determine from our data, how-ever, the specific motivations that lead individ-ual companies to form alliances. To address thisshortcoming, we performed some additional qual-itative analyses.15 First, we analyzed the top-tenselling biotechnology drugs at the end of ourstudy period (2001) (Standard & Poor’s, 2002).The number one (Procrit, $3,430 million), numberthree (Intron A, $1,447 million), and number five(Humulin, $1,061 million) best-selling drugs wereall commercialized through an alliance formed bya biotechnology company and a pharmaceutical

14 These results are available upon request from the first author.15 We thank an anonymous reviewer for suggesting additionalqualitative analyses.

company.16 Moreover, each of these three allianceswas based on complementarities, because in eachcase the new biotechnology firm developed thedrug, while the existing pharmaceutical companycommercialized the new drug. This provides addi-tional qualitative support for Hypothesis 1.

To assess whether the motivations of these com-plementarity alliances were in line with Hypoth-esis 3, suggesting that complementarity alliancesare more likely to form when the biotechnologyfirm is younger, we assessed the age of the threebiotechnology firms when forming these alliances.Amgen was founded in 1980, and formed the Pro-crit alliance with Johnson & Johnson in 1985. Bio-gen was founded in 1979, and entered the Intron Aalliance with Schering-Plough in 1979. Genentechwas founded in 1976, and the entered the Humulinalliance with Lilly in 1978. All three biotechnol-ogy firms formed these complementarity alliancesvery early in their lives, with their average ageat alliance formation being 2.33 years. This aver-age age lies more than two standard deviationsbelow the mean age of the biotechnology firms(15.135 years) in this sample (see Figure 1). Thisprovides qualitative support for Hypothesis 3, sug-gesting that complementarity alliances are indeedformed when biotechnology firms are younger.

In addition, we drew a random sample of 58alliances (10% of all alliances formed). We thenhad these alliances independently coded for theage of the biotechnology firm at alliance forma-tion as well as the various complementarity andsimilarity measures described above. A researchassistant then tracked newspaper articles describ-ing each alliance from Lexis-Nexis, content codingeach article. Two different indicator variables wereapplied: 1 = complementarities are mentioned asmotivation for alliance formation; 1 = similari-ties are mentioned as motivation for alliance for-mation. A second researcher then independentlycoded a subsample of these articles, and we foundthe interrater reliability to be satisfactory. We thenran the bivariate correlations among these vari-ables, and found that the age of the biotech-nology firm at alliance formation was negativelycorrelated with the number of non-overlappingniches, a complementarity measure (r = −0.392,p < 0.01), and positively correlated with patent-ing propensity (r = 0.289, p < 0.01), a similarity

16 The alliance partners are: Amgen–Johnson & Johnson; Bio-gen–Schering-Plough; Genentech–Lilly.



Table 5. Binary logit and negative binomial regression estimates of dyad-level formation and alliance intensity

Alliance formation Alliance intensity

Model 13 Model 14(non-equity)


Constant −4.8873∗∗∗ −4.9122∗∗∗ −4.7847∗∗∗ −4.8658∗∗∗

(0.1242) (0.1247) (0.1145) (0.1184)Geographic zone 0.0639† 0.0531 −0.0419 0.0252

(0.0473) (0.0477) (0.0447) (0.0449)Time elapsed 0.1763∗∗ 0.1805∗∗ 0.1718∗∗ 0.1760∗∗∗

(0.0647) (0.0649) (0.0606) (0.0562)(Time elapsed)2 −0.0074∗ −0.0075∗ −0.0064∗ −0.0069∗

(0.0037) (0.0037) (0.0033) (0.0031)Direct ties 0.0503∗ 0.0491∗ 0.0262 0.0609∗∗

(0.0271) (0.0273) (0.0245) (0.0207)Indirect ties 0.1736∗∗∗ 0.1741∗∗∗ 0.1507∗∗ 0.1728∗∗∗

(0.0527) (0.0529) (0.0592) (0.0516)(Indirect ties)2 −0.0122† −0.0118† −0.0189∗ −0.0126∗

(0.0075) (0.0075) (0.0102) (0.0075)Centrality biotech 0.2959∗∗∗ 0.3037∗∗∗ 0.2845∗∗∗ 0.2517∗∗∗

(0.0705) (0.0713) (0.0650) (0.0668)(Centrality biotech)2 −0.0295† −0.0331∗ −0.0296∗ −0.0307∗

(0.0188) (0.0192) (0.0171) (0.0177)Centrality pharma 0.6672∗∗∗ 0.6518∗∗∗ 0.3464∗∗∗ 0.5922∗∗∗

(0.0905) (0.0914) (0.0773) (0.0857)(Centrality pharma)2 −0.1235∗∗ −0.1025∗ −0.0458 −0.0854∗

(0.0503) (0.0506) (0.0466) (0.0473)Firm size biotech 0.0373 0.0441 0.0291 0.0458

(0.0415) (0.0413) (0.0384) (0.0383)Firm size pharma 0.1016∗ 0.1137∗ 0.1907∗∗∗ 0.0922†

(0.0605) (0.0611) (0.0526) (0.0579)Public biotech 0.6940∗∗∗ 0.6765∗∗∗ 0.7892∗∗∗ 0.7641∗∗∗

(0.1330) (0.1334) (0.1218) (0.1263)Public pharma 0.0194 0.0016 0.0243 0.0347

(0.0586) (0.0587) (0.0529) (0.0559)Patents biotech 0.1697∗∗∗ 0.1654∗∗∗ 0.1733∗∗∗ 0.1781∗∗∗

(0.0299) (0.0305) (0.0253) (0.0259)Patents pharma −0.0584 −0.0773† 0.0085 −0.1075∗

(0.0532) (0.0539) (0.0463) (0.0516)Biotech firm age −0.0388 −0.0402 0.0044 0.0016

(0.0507) (0.0512) (0.0467) (0.0473)Dyadic complementaritiesNon-overlapping niches 0.0473 0.0499 0.0359 0.0595

(0.0889) (0.0892) (0.0840) (0.0845)Complementarity index 0.2248∗ 0.2250∗ 0.2020∗ 0.2788∗∗

(0.1147) (0.1152) (0.1091) (0.1097)Dyadic similaritiesPatent cross-citation rate 0.0415 0.0423 0.0380 0.0378

(0.0356) (0.0353) (0.0352) (0.0360)Patent common citation rate −0.3442 −0.3345 −0.0604 −0.2194

(0.2731) (0.2702) (0.1457) (0.2157)Patenting propensity −0.0765 −0.0839 −0.1839∗ −0.1887

(0.1042) (0.1042) (0.0836) (0.0847)Dyadic and firm-level interactionsNon-overlapping niches × Public biotech −0.0779 −0.0725 −0.0826 −0.0891

(0.1008) (0.1012) (0.0951) (0.0956)Complementarity index × Public biotech −0.0808 −0.0774 −0.0768 −0.1200∗∗∗

(0.1162) (0.1167) (0.1101) (0.1107)Patent cross-citation rate × Public biotech 0.0029 0.0035 −0.0055 0.0094

(0.0473) (−0.047) (0.0489) (0.0467)




Alliance formation Alliance intensity



Patent common citation rate × Public biotech 0.2292 0.2256 0.0182 0.0925(0.2791) (0.2762) (0.1526) (0.2234)

Patenting propensity × Public biotech 0.1925∗ 0.2004∗ 0.2862∗∗∗ 0.3332(0.1083) (0.1084) (0.0881) (0.0904)

Log-likelihood −2313.862 −2283.08 −2644.57 −2540.42Likelihood ratio test (χ 2) 600.04∗∗∗ 595.27∗∗∗ 545.32∗∗∗ 705.09∗∗∗

N 32,332 32,332 32,332 32,332


measures. Here again, it is the broad-based capabil-ity measures that reach significance. These corre-lations are also in line with our expectations basedon Hypotheses 1 and 2. Additionally, the com-plementarity and similarity proxies derived fromthe coding of the newspaper articles revealed thatthese two dimensions were negatively correlated(r < −0.346, p < 0.01). While these qualitativedata are far from conclusive, they lend support toour deductively derived theoretical model.

DISCUSSION

Past research on alliance formation has noted crit-ical differences between the complementary com-petences, skills, and capabilities that each partnerbrings to an alliance, and it has addressed theneed for some level of institutional, structural,or social similarity to lower the risk and createtrust in alliance building. Prior work has paid lessattention, however, to the manner in which bothcomplementarities and similarities across potentialpartners interact with firm-level characteristics toaffect partner choice.

Whereas most past research explaining whopartners with whom has focused on horizontalalliances between established firms, our studyexamines the antecedents of alliances betweenincumbent firms founded under an old technologyand firms founded under a new technology. Sucha focus is particularly critical given the impor-tance of alliances as a mechanism for establishedfirms to adapt to technological change. Moreover,prior work has not considered that the probabil-ity of alliance formation might be contingent uponhow dyadic constructs such as complementarity

and similarity may interact with firm-specific fac-tors. Our research addresses this issue by arguingthat alliance formation between an old and a newtechnology firm motivated by complementaritiesis more likely when the new technology firm isyounger, while alliance formation driven by sim-ilarities is more likely when the new technologyfirm is older. Thus, we argue that firm age of thenew technology partner moderates the impact ofcomplementarities in a negative fashion, while itmoderates the impact of similarities in a positivefashion. Our results provide support for the notionthat the age of the new technology firm mod-erates the relationship between complementaritiesand alliance formation in a negative fashion.

We did not find support for a positive interac-tion effect of firm age on the relationship betweensimilarities and alliance formation. In post hocanalyses, we did, however, find support for thishypothesis when operationalizing a biotechnologyfirm’s enhanced legitimacy, credibility, and powerby its public ownership status. Here, the positiveeffect of dyadic similarities on alliance formationand alliance intensity was reinforced when thebiotechnology firm was public. Taken together, ourresults indicate that the effects of complementar-ities and similarities on alliance formation appearto be contingent upon certain firm characteristicsof the new technology venture.

Our findings point to two sets of causal factorsthat directly affect alliance formation in specificways. First, complementarities in skills, capabili-ties and competences create opportunities for firmsto seek out partners that can provide valuable orga-nizational resources not found in the focal firm.Second, the search for alliance partners that havedifferent capabilities is tempered by a preference to



affiliate with firms that share similarities in termsof firm characteristics. The challenge of managingalliance relationships focuses on maintaining thisbalance between differences and similarities—theyin and yang of successful alliance formation.

The additional contribution of this research isin the examination of specific qualities of the newtechnology partner. We identified the problems oflegitimacy, credibility, and power, all linked tothe liability of newness and the age of the firm.We argued that the new technology firm wouldgain credibility, legitimacy, and market power asit grew older and became more established. This,in turn, would permit it to have a greater sayin the terms of the alliance with its more estab-lished potential partner. Alliances involving com-plementarities can focus on quick value creationthrough combining new technology firms that arejust beginning (young firms) and more established,incumbent technology partners. Thus, age of thenew technology firm has an inverse moderatingeffect on the relationship of complementaritiesbetween the established technology firm and thenew technology firm. Younger new technologyfirms that have complementarities with establishedtechnology firms are even more likely to formalliances.

As new technology firms age, they become moreindependent and established. They are also likelyto become more powerful. These (older) new tech-nology firms can afford to be more selective inchoosing alliance partners, and may be more inter-ested in partners to which they are similar and withwhom they feel a comfort level. In this case, sim-ilarities between the established technology firmand the new technology firm may be more likely todrive alliance formation when the new technologyfirm is older and more established. The age of thenew technology firm might play an opposing roleon complementarities and similarities—youngernew technology firms create alliances with estab-lished firms that are complementary, whereas oldernew technology firms might create alliances withfirms that are more similar.

The results demonstrated support for the hypoth-esized effects of complementarities and similar-ities, and some support for interactive effectsof the new technology firm age in influencingalliance behavior. The results for the complemen-tary hypothesis are based upon a novel approach toassess the potential of synergistic gains at the dyadlevel. The central perspective of these arguments is

that each partner has certain areas of strength thatmay compensate for the weaknesses of their poten-tial alliance partner. The results of our researchindicate that alliance formation is enhanced whennew technology partners focus on drug discoveryand development, while incumbents offer down-stream expertise as well as marketing and distri-bution systems for the commercialization of newproducts.

Somewhat surprisingly, the second measure em-ployed to proxy complementarities—non-overl-apping niches—was not significant in predictingalliance formation. One explanation for why priorresearch found significance for non-overlappingniches predicting alliance formation (Gulati,1995b; Chung et al., 2000) while we did notcould be that the prior work focused on horizon-tal alliances between established firms in moremature industries (e.g., new materials, industrialautomation, automotive products, and investmentbanking), while we focused on vertical alliancesbetween new and old technology firms in anemerging science-driven industry. Another reasonmight be that proxying complementarities basedon non-overlapping niches is not a suitable mea-sure, as pointed out recently by Gimeno (2004).Firms can occupy non-overlapping niches with-out being complementary to one another. Thisis the case when firms occupy dissimilar nichesthat are not complementary. Thus, while occupy-ing non-overlapping niches might be a necessarycondition for complementarity, it is not sufficient.Taken together, it appears that in newly emerginghigh-tech industries the level of (vertical) strategicinterdependence between potential alliance part-ners can be effectively captured by measuring eachpartner’s capability in complementary value chainactivities.

Similarities between potential partner firms alsoappear to promote alliance formation. In this study,we focused on similarities arising from techno-logical overlap in patent cross-citation and patentcommon citation rates as well as from a firm’sgeneral orientation toward technology and innova-tion that may imply similarities in terms of overallstrategic approach. Mowery et al. (1998) were thefirst to use patent cross-citation and patent com-mon citation rates as dyad-level proxies for tech-nological similarities, albeit in a smaller sample ofhorizontal alliances among established firms overa 2-year period in the mid 1980s. They found



that both patent cross-citation and patent com-mon citation rates predicted alliance formation. Inline with their result, we too find that a dyad’scross-citation rate consistently predicted allianceformation. Arguably, a dyad’s patent cross-citationrate is a strong indicator not only of how famil-iar each partner is with one another, but alsohow similar they are in their technological over-lap. Moreover, a dyad’s overall patenting propen-sity, reflective of each partner firm’s innovativecompetence, consistently predicted alliance forma-tion. Taken together, we found that similaritiesalong these dimensions enhance the probability ofalliance formation. Some of our control variablesalso lend support to the notion that similaritiesattract. For example, new and established technol-ogy firms that are located closer geographically aremore likely to cooperate. These firms are exposedto a similar cultural and institutional environment,which appears to facilitate interfirm cooperation.

Contrary to our expectations, however, we founda dyad’s patent common citation rate had a signif-icant negative effect on alliance formation. Whywould firms that are similar in their technologi-cal overlap through drawing on the same externaltechnology pool be less likely to form an alliance?We speculate that in this industry setting comple-mentarities are a powerful predictor of alliance for-mation, and thus the complementarity effect maycrowd out the similarity effect based on patentcommon citation rates. Firms that draw on thesame external technological pool as proxied bypatent common citation rates may not exhibit thethreshold level of complementarities that is neces-sary to initiate the search for an alliance partnerin the first place. While Mowery et al. (1998: 519)argue that patent cross-citation and patent commoncitation rates can be seen as ‘fairly close substitutesin terms of their performance as measures of tech-nological overlap,’ an important distinction to thepatent common citation rate is that dyads that scorehigh on patent cross-citation rates can still drawon very different external technological pools, thusenabling complementarities to emerge that propelthem to form alliances. This effect seems to beparticularly pronounced when studying alliancesbetween old and new technology firms in emergingindustries.

We controlled for several of the variables thatwere the focus of attention in prior studies (e.g.,Podolny, 1994; Gulati, 1995a, 1995b; Ahuja, 2000;Chung et al., 2000). Here, most of our results are

broadly consistent with this prior work conductedin different industries with different time frames,and focusing predominantly on horizontal alliancesbetween established firms. For example, we foundthat the relationship between time elapsed sincelast and subsequent alliance was non-monotonic.A similar relationship was found for indirect ties.On the other hand, we found that a dyad’s num-ber of direct ties is related to future alliances in apositive linear fashion. Perhaps our research set-ting is somewhat unique in a way that prior directties do not exhibit the dampening effect they havefound to have in other industries. To make moresense of this finding, we interviewed Anton Gueth,former director of the Office of Alliance Manage-ment at Lilly. He stated that multiple direct tieswith one partner build an ecosystem of alliances,which in turn is less likely to collapse in com-parison to stand-alone binary alliances. This mayexplain why we find a positive and linear effect ofdirect ties on alliance formation.

This study also makes methodological contribu-tions. While we build on prior research in devel-oping some of the complementarity and similaritymeasures, we also create new measures for eachcategory. The five different measures employedto proxy dyadic complementarities and similaritiescan be distinguished in regard to how fine grainedthey are. One group of these dyadic measures (non-overlapping niches, complementarity index, andpatenting propensity) are derived from broad-basedorganizational capabilities held by each dyad part-ner, while the second group (patent cross-citationand patent common citation rates) are more fine-grained proxies of a dyad’s science and technologyrelatedness. In evaluating the effectiveness of thedifferent types of proxies in predicting allianceformation, two observations can be made. First,among the broad-based capabilities it is a dyad’scomplementarity index and patenting propensity,both developed for this study, that consistently pre-dict alliance formation as hypothesized. Amongthe more fine-grained measures of similarities intechnological orientation, it is a dyad’s cross-citation rate that behaves in predicting allianceformation as expected, while a dyad’s commoncitation rate behaves opposite to the hypothesizedeffect. The second, and perhaps even more interest-ing insight, emerges through the comparison acrossgroups. When evaluating the results for the directeffects, the more broad-based capability measuresappear at least as effective, if not more so, than the



more fine-grained measures in predicting allianceformation. In addition, the more broad-based capa-bility measures appear to be more effective thanthe more fine-grained proxies of technological sim-ilarities in supporting the hypothesized moderatingeffects (regardless of interacting them with firmage, firm size, or public status) on the relation-ship between complementarities, similarities, andalliance formation.

Limitations and future research

Clearly, we focus only on a subsegment of thebiopharmaceutical alliances, i.e., alliances betweenlarge pharmaceutical companies and small biotech-nology firms. While such a sampling could intro-duce a sample selection bias, it is also importantto recognize that recent empirical research foundthat linkages between pharmaceutical and biotech-nology firms account for up to 80 percent of allalliances observed in this industrial sector (Hage-doorn and Roijakkers, 2002). Moreover, we drewon the complete network comprising pharma andbiotech firms as well as universities and othernonprofit research organizations when developingour indirect tie measure. We suggest that a focuson biotech–pharma dyads is useful when attempt-ing to understand alliance formation between newtechnology firms (often new entrants) and old tech-nology firms (incumbents). A focus on new and oldtechnology firms is particularly salient since recenttheoretical work has suggested that incumbentsmay use alliances with new entrants to adapt toradical innovations (Teece, 1992; Hill and Rothaer-mel, 2003).

Moreover, while focusing on a single indus-try may limit the generalizability of the findingsto some degree, it enhances its internal valid-ity since such an approach controls for exoge-nous industry factors (Stuart, 2000). Given thatbiotechnology and pharmaceutical firms competein relatively risky technologies, there may begreater uncertainty concerning the future directionof these firms, which may in turn increase the riskof alliance failure relative to less technologicallyintense industries, and thus limit the generalizabil-ity of our findings. Furthermore, although we col-lected information on global players in these indus-tries, other countries outside the United States havesomewhat different institutional or market frame-works for supporting the creation of alliances,which may influence the formation of partnerships.

A better understanding of appropriate matchesbetween complementarities and similarities amongpotential alliance partners can serve to more closelyintegrate current work in the areas of strategicmanagement and the management of technology.Future research should investigate in more detailthe interplay between the specific skills and capa-bilities of potential partner firms, and how theseskills and capabilities complement each other andhold the possibility of creating value. A morethorough examination of the organizational andstructural processes that bring firms more closelytogether and encourage alliance formation is alsorequired. Further study of the question of howsimilarities and differences between firms impactthe formation of partnerships can provide impor-tant insights into alliances as creators of economicvalue and relational rents (Zajac and Olsen, 1993;Dyer and Singh, 1998), and in particular howleveraging complementarities and similarities caninfluence alliance performance. The results of ourstudy support the perspective that future researchon alliance formation needs to take into accounthow dyadic-level constructs interact with firm-level variables, and thus to explicitly consider amultilevel dimension.

Managerial implications

It appears that what managers seem to considerare the broader types of capability relatednesswhen deciding whether or not to form an alliance.While fine-grained proxies of science and tech-nology relatedness (like a dyad’s patent cross-citation rate) appear to be a reliable predictor ofalliance formation, complementarities across thevalue chain and similarities in overall innovationstrategy seem to be foremost on the minds ofmanagers when forming (vertical) alliances. Thesefindings tie this research to theoretical work ondynamic capabilities, which argues that the capa-bility to form the ‘right’ alliances is an importantcompetence that allows firms not only to adaptto changing markets and technologies, but also tocreate market change that favors their competitivestrengths (Teece, Pisano, and Shuen, 1997; Eisen-hardt and Martin, 2000). We speculate that broad-based capabilities highlighted in this study can notonly provide important guideposts to search foralliance partners, but may also lay the foundationof successful alliances.



ACKNOWLEDGEMENTS

We thank Co-Editor Edward Zajac, the anonymousreviewers, Federico Aime, Roger Calantone, LacyFitzpatrick (formerly at ICOS), Stuart Graham,Anton Gueth (former director, Office of AllianceManagement at Lilly), Ranjay Gulati, PrashantKale, Richard Makadok, Trey Powell (Doctorof Pharmacology, formerly at Immunex), FrankSchultz, Habir Singh, Robert Wiseman, and theseminar participants at Dartmouth College, EmoryUniversity, Georgia Institute of Technology, OhioState University, University of Texas at Austin,and at the 2002 BYU-University of Utah WinterStrategy Conference, for helpful suggestions andcomments on earlier versions of this paper. We areparticularly indebted to Brian Silverman for gen-erously sharing his deep expertise concerning thepatent common citation and patent cross-citationmeasures. A prior version of this paper was pre-sented at the 2002 Academy of Management Meet-ing, where we thank the anonymous BPS review-ers for invaluable input. We thank Mark Edwardsof Recombinant Capital for making their variousdatabases available to us. We thank Shanti Agung,Jinsoo Lee, and Gavin Mills for research assis-tance, and Deborah Gray and Suzanne Rumsey foreditorial assistance.

Frank Rothaermel gratefully acknowledges sup-port for this research from the National ScienceFoundation (CAREER Award, NSF#0545544) andthe Sloan Foundation (Industry Studies Fellow-ship). Frank Rothaermel is an Affiliate of the SloanBiotechnology Industry Center at the University ofMaryland. All opinions expressed as well as allerrors and omissions are entirely the authors’.

REFERENCES

Ahuja G. 2000. The duality of collaboration: inducementsand opportunities in the formation of interfirmlinkages. Strategic Management Journal , MarchSpecial Issue 21: 317–343.

Albert MB, Avery D, Narin F, McAllister P. 1991.Direct validation of citation counts as indicators ofindustrially important patents. Research Policy 20:251–259.

Amburgey TL, Kelly D, Barnett WP. 1993. Resetting theclock: the dynamics of organizational change andfailure. Administrative Science Quarterly 38: 51–73.

Arora A, Gambardella A. 1990. Complementarity andexternal linkages: the strategies of large firms inbiotechnology. Journal of Industrial Economics 4:361–379.

Arthur WB. 1989. Competing technologies, increasingreturns, and lock-in by historic events. EconomicJournal 99: 116–131.

Barron DN, West E, Hannan MT. 1994. A time to growand a time to die: growth and mortality of credit unionsin New York City. American Journal of Sociology 100:381–421.

Baum JAC, Calabrese T, Silverman BS. 2000. Don’tgo it alone: alliance network composition andstartups’ performance in Canadian biotechnology.Strategic Management Journal , March Special Issue21: 267–294.

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

BioScan. Diverse years. The Worldwide Biotech IndustryReporting Service. Thomson, American HealthConsultants: Atlanta, GA.

Burgers WP, Hill CWL, Kim WC. 1993. A theory ofglobal strategic alliances: the case of the globalauto industry. Strategic Management Journal 14(6):419–432.

Chatterjee S, Wernerfelt B. 1991. The link betweenresources and type of diversification: theory andevidence. Strategic Management Journal 12(1):33–48.

Chung S, Singh H, Lee K. 2000. Complementarity, statussimilarity and social capital as drivers of allianceformation. Strategic Management Journal 21(1):1–22.

Cohen P, Cohen J, West SG, Aiken LS. 2003. AppliedMultiple Regression/Correlation Analysis for theBehavioral Sciences (3rd edn). Erlbaum: Hillsdale, NJ.

Cohen WM, Levinthal DA. 1990. Absorptive capacity:a new perspective on learning and innovation.Administrative Science Quarterly 35: 128–152.

Cool KO, Schendel D. 1987. Strategic group formationand performance: the case of the U.S. pharmaceuticalindustry, 1963–1982. Management Science 33(9):1102–1124.

Cool KO, Schendel D. 1988. Performance differencesamong strategic group members. Strategic Manage-ment Journal 9(3): 207–223.

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

De Carolis DM. 2003. Competencies and imitabilityin the pharmaceutical industry: an analysis oftheir relationship with firm performance. Journal ofManagement 11: 27–50.

De Carolis DM, Deeds DL. 1999. The impact ofstocks and flows of organizational knowledge onfirm performance: an empirical investigation ofthe biotechnology industry. Strategic ManagementJournal 20(10): 953–968.

Deeds DL, De Carolis D, Coombs J. 2000. Dynamiccapabilities and new product development in hightechnology ventures: an empirical analysis of newbiotechnology firms. Journal of Business Venturing15: 211–229.

Dosi G. 1982. Technological paradigms and technologicaltrajectories. Research Policy 11: 147–162.



Dyer JH, Singh H. 1998. The relational view: cooperativestrategy and sources of interorganizational competitiveadvantage. Academy of Management Review 23:660–679.

Eisenhardt KM, Martin JA. 2000. Dynamic capabilities:what are they? Strategic Management Journal ,October–November Special Issue 21: 1105–1121.

Eisenhardt KM, Schoonhoven CB. 1996. Resource-basedview of strategic alliance formation. OrganizationScience 7: 136–150.

Ernst & Young Biotechnology Annual Industry Reports .Diverse years. Ernst & Young: Palo Alto, CA.

Garcia-Pont C, Nohria N. 2002. Local versus globalmimetism: the dynamics of alliance formation in theautomobile industry. Strategic Management Journal23(4): 307–321.

Gimeno J. 2004. Competition within and betweennetworks: the contingent effect of competitiveembeddedness on alliance formation. Academy ofManagement Journal 47: 820–842.

Greene WH. 1997. Econometric Analysis . Prentice-Hall:Upper Saddle River, NJ.

Griliches Z. 1990. Patent statistics as economic indica-tors: a survey. Journal of Economic Literature 28:1661–1707.

Gulati R. 1995a. Does familiarity breed trust? Theimplications of repeated ties for contractual choicein alliances. Academy of Management Journal 38:85–112.

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

Hagedoorn J. 1993. Understanding the rationale ofstrategic technology partnering: interorganizationalmodes of cooperation and sectoral differences.Strategic Management Journal 14(5): 371–385.

Hagedoorn J. 2002. Inter-firm R&D partnerships: anoverview of major trends and patterns since 1960.Research Policy 31: 477–492.

Hagedoorn J, Roijakkers N. 2002. Small entrepreneurialfirms and large companies in inter-firm R&Dnetworks: the international biotechnology industry. InStrategic Entrepreneurship: Creating a New Mindset ,Hitt MA, Ireland RD, Camp SM, Sexton DL (eds).Blackwell: Oxford, U.K.; 223–252.

Hall BH, Jaffe AB, Trajtenberg M. 2001. The NBERpatent citations data file: lessons, insights andmethodological tools. NBER Working Paper 8498.Cambridge, MA.

Hamel G, Doz YL, Prahalad CK. 1989. Collaborate withyour competitors—and win. Harvard Business Review67: 133–139.

Hannan MT, Freeman J. 1977. The population ecologyof organizations. American Journal of Sociology 83:929–984.

Harrigan KR. 1985. Strategies for Joint Ventures.Lexington Books: Lexington, MA.

Hennart JF. 1991. The transaction costs theory of jointventures: an empirical study of Japanese subsidiariesin the United States. Management Science 37:483–497.

Hill CWL, Rothaermel FT. 2003. The performance ofincumbent firms in the face of radical technologicalinnovation. Academy of Management Review 28:257–274.

Hoang H, Rothaermel FT. 2005. The effect of generaland partner-specific alliance experience on joint R&Dproject performance. Academy of Management Journal48: 332–345.

Jaccard J, Wan CK, Turrisi R. 1990. The detection andinterpretation of interaction effects between contin-uous variables in multiple regression. MultivariateBehavioral Research 25: 467–478.

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

Kale P, Singh H, Perlmutter H. 2000. Learning andprotection of proprietary assets in strategic alliances:building relational capital. Strategic ManagementJournal , March Special Issue 21: 217–237.

Kogut B. 1991. Joint ventures and the option to expandand acquire. Management Science 37: 19–33.

Lane PJ, Lubatkin M. 1998. Relative absorptive capac-ity and interorganizational learning. Strategic Manage-ment Journal 19(5): 461–477.

Lerner J, Shane H, Tsai A. 2003. Do equity financingcycles matter? Evidence from biotechnology alliances.Journal of Financial Economics 67: 411–446.

Levin RC, Klevorick AK, Nelson RR, Winter SG. 1987.Appropriating the returns from industrial research anddevelopment. Brookings Papers on Economic Activity3: 783–820.

Lorange P, Roos J. 1992. Strategic Alliances. Blackwell:Cambridge, MA.

March JG. 1988. Decisions in Organizations. Blackwell:New York.

Matraves C. 1999. Market structure, R&D and advertis-ing in the pharmaceutical industry. Journal of Indus-trial Economics 42: 169–194.

Mowery DC, Oxley JE, Silverman BS. 1996. Strategicalliances and interfirm knowledge transfer. StrategicManagement Journal , Winter Special Issue 17:77–91.

Mowery DC, Oxley JE, Silverman BS. 1998. Technolog-ical overlap and interfirm cooperation: implications forthe resource-based view of the firm. Research Policy27: 507–523.

Nelson RR, Winter SG. 1982. An Evolutionary Theoryof Economic Change. Belknap Press of HarvardUniversity: Cambridge, MA.

Nohria N, Garcia-Pont C. 1991. Global strategic linkagesand industry structure. Strategic Management Journal ,Summer Special Issue 12: 105–124.

Ohmae K. 1989. The global logic of strategic alliances.Harvard Business Review 67: 143–154.

Oliver C. 1990. Determinants of interorganizationalrelationships: integration and future directions.Academy of Management Review 15: 241–265.

Park SH, Chen R, Gallagher S. 2002. Firm resources asmoderators of the relationship between market growthand strategic alliances in semiconductor start-ups.Academy of Management Journal 45: 527–545.



Podolny JM. 1994. Market uncertainty and socialcharacter of economic exchange. AdministrativeScience Quarterly 39: 458–483.

Porter ME, Fuller MB. 1986. Coalitions and global strat-egy. In Competition in Global Industry , Porter ME(ed.). Harvard Business School Press: Boston, MA;315–344.

Powell WW, Koput KW, Smith-Doerr L. 1996. Interor-ganizational collaboration and the locus of innovation:networks of learning in biotechnology. AdministrativeScience Quarterly 41: 116–145.

Roberts P. 1999. Product innovation, product-marketcompetition and persistent profitability in theU.S. pharmaceutical industry. Strategic ManagementJournal 20(7): 655–670.

Rosenkopf L, Metiu A, George VP. 2001. From thebottom up? Technical committee activity and allianceformation. Administrative Science Quarterly 46:748–772.

Rothaermel FT. 2001. Incumbent’s advantage throughexploiting complementary assets via interfirm cooper-ation. Strategic Management Journal , June–July Spe-cial Issue 22: 687–699.

Rothaermel FT, Deeds DL. 2004. Exploration andexploitation alliances in biotechnology: a systemof new product development. Strategic ManagementJournal 25(5): 201–221.

Rothaermel FT, Deeds DL. 2006. Alliance type, allianceexperience, and alliance management capabilityin high-technology ventures. Journal of BusinessVenturing 21: 429–460.

Rothaermel FT, Hill CWL. 2005. Technological disconti-nuities and complementary assets: a longitudinal studyof industry and firm performance. Organization Sci-ence 16: 52–70.

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

Singh K. 1997. The impact of technological complexityand interfirm cooperation on business survival.Academy of Management Journal 40: 339–367.

Sorensen JB, Stuart TE. 2000. Aging, obsolescence,

and organizational innovation. Administrative ScienceQuarterly 45: 81–112.

Standard & Poor’s Biotechnology Industry Survey . May2002. McGraw-Hill: New York.

Stinchcombe AL. 1965. Social structure in organizations.In Handbook of Organizations , March JG (ed.). RandMcNally: Chicago, IL; 142–193.

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

Stuart TE, Hoang H, Hybels RC. 1999. Interorgani-zational endorsements and the performance ofentrepreneurial ventures. Administrative Science Quar-terly 44: 315–349.

Stuart TE, Podolny JM. 1996. Local search and theevolution of technological capabilities. StrategicManagement Journal , Summer Special Issue 17:21–38.

Teece DJ. 1986. Profiting from technological innovation:implications for integration, collaboration, licensingand public policy. Research Policy 15: 285–305.

Teece DJ. 1992. Competition, cooperation, and innova-tion: organizational arrangements for regimes of rapidtechnological progress. Journal of Economic Behaviorand Organization 18(1): 1–25.

Teece DJ, Pisano G, Shuen A. 1997. Dynamic capabili-ties and strategic management. Strategic ManagementJournal 18(7): 509–533.

Tushman ML, Anderson P. 1986. Technological discon-tinuities and organizational environments. Administra-tive Science Quarterly 31: 439–465.

Williamson OE. 1985. The Economic Institutions ofCapitalism. Free Press: New York.

Zajac EJ, Olsen CP. 1993. From transaction cost totransactional value analysis: implications for thestudy of interorganizational strategies. Journal ofManagement Studies 30: 131–145.

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


OLD TECHNOLOGY MEETS NEW TECHNOLOGY: COMPLEMENTARITIES

Documents

Transcript of OLD TECHNOLOGY MEETS NEW TECHNOLOGY: COMPLEMENTARITIES