Interorganizational alliances and the performance of firms ...faculty.haas.berkeley.edu/tstuart/publications/2000... · Strat. Mgmt. J., 21: 791–811 (2000) INTERORGANIZATIONAL ALLIANCES

Strategic Management JournalStrat. Mgmt. J.,21: 791–811 (2000)

INTERORGANIZATIONAL ALLIANCES AND THEPERFORMANCE OF FIRMS: A STUDY OF GROWTHAND INNOVATION RATES IN A HIGH-TECHNOLOGY INDUSTRY

TOBY E. STUART*Graduate School of Business, University of Chicago, Chicago, Illinois, U.S.A.

This paper investigates the relationship between intercorporate technology alliances and firmperformance. It argues that alliances are access relationships, and therefore that the advantageswhich a focal firm derives from a portfolio of strategic coalitions depend upon the resourceprofiles of its alliance partners. In particular, large firms and those that possess leading-edgetechnological resources are posited to be the most valuable associates. The paper also arguesthat alliances are both pathways for the exchange of resources and signals that convey socialstatus and recognition. Particularly when one of the firms in an alliance is a young or smallorganization or, more generally, an organization of equivocal quality, alliances can act asendorsements: they build public confidence in the value of an organization’s products andservices and thereby facilitate the firm’s efforts to attract customers and other corporatepartners. The findings from models of sales growth and innovation rates in a large sample ofsemiconductor producers confirm that organizations with large and innovative alliance partnersperform better than otherwise comparable firms that lack such partners. Consistent with thestatus-transfer arguments, the findings also demonstrate that young and small firms benefitmore from large and innovative strategic alliance partners than do old and large organizations.Copyright 2000 John Wiley & Sons, Ltd.

Owing to some recent empirical studies, we aregaining an understanding of the factors that com-pel firms to enter strategic alliances (e.g., Nohriaand Garcia-Pont, 1991; Gulati, 1995; Eisenhardtand Schoonhoven, 1996; Walker, Kogut, andShan, 1997). With few exceptions, explanationsof why firms establish alliances are directly linkedto presumptions about the benefits of alliances toparticipant firms. In light of the natural associ-ation between cause and consequence in purpo-sive theories of organizational behavior, however,it is surprising to find relatively few large-samplestudies that confirm widely-espoused assumptions

Key words: strategic alliances; firm performance; inter-organizational networks; innovation; patents* Correspondence to: Toby E. Stuart, Graduate School ofBusiness, University of Chicago, 1101 East 58th St., Chicago,IL 60637, U.S.A.

Received 12 January 1998Copyright 2000 John Wiley & Sons, Ltd. Final revision received 14 March 2000

that strategic alliances are advantageous for par-ticipant firms (exceptions include Hagedoorn andSchakenraad, 1994; Shan, Walker, and Kogut,1994; Powell, Koput and Smith-Doerr, 1996; andMitchell and Singh, 1996).

This article has two objectives. First, to offerconsultation regarding the conditions under whichstrategic alliancing is advantageous, it is firstnecessary to develop a nomothetic literature onthe effects of alliances on firm performance and,in particular, on the contingencies that bear uponthe alliance-performance link. Because I believethat there has been insufficient attention to theconnection between the value of alliances and thecharacteristics of the firms in the partnership (andthe interactions between partner characteristics),I will attempt to show in this study that theadvantage of a portfolio of alliances is determinednot so much by the portfolio’s size, but by the

792 T. E. Stuart

characteristics of the firms that a focal organi-zation is connected to. Second, theories of thefunctionality of alliances have devoted little atten-tion to one of the most significant and one ofthe easiest to obtain of the potential advantagesof intercorporate affiliations: under frequently-metconditions, alliances can significantly enhance (ordevalue) the reputation of one or both of theparticipant firms. In this study, I investigatewhether alliances with prominent partners upgradea focal firm’s reputation, which I infer from therelationship between attributes of a focal firm’salliance partners and its post-alliance performance(Rao, 1994; Wilson, 1985). These ideas are sub-jected to empirical scrutiny in a study of theeffect of horizontal technology alliances in thesemiconductor industry on two outcome variables:the rate of innovation and the rate of revenuegrowth of the firms in the industry.

Literature and theory

Firms establish alliances for many reasons(Gulati, 1998, offers a current review). Salientamong the incentives to collaborate is the possi-bility of bringing together complementary assetsowned by different organizations (Nohria andGarcia-Pont, 1991). For instance, two companiesmay establish an alliance when each one pos-sesses strength in a different stage in a product’svalue chain, such as when one firm has manufac-turing expertise and a second one controls adistribution channel. Second, firms may formcoalitions to defray costs and share risk whenthey undertake high-cost (capital- or development-intensive) projects or very speculative strategicinitiatives (Hagedoorn, 1993). It has also beensuggested that the resources acquired from analliance partner can facilitate a firm’s efforts toalter its competitive position (Kogut, 1988), aswell as that corporations in concentrated indus-tries utilize alliances to collude or to gain marketpower at the expense of other competitors (Pfefferand Nowak, 1976). Empirically, researchers haveobserved an association between the propensity toenter into alliances and a variety of organizationalattributes, including firm size, age, scope, andresources (Shan et al., 1994; Burgers, Hill, andKim, 1993).

Rather than investigate the antecedents of inter-corporate partnerships, I treat the formation of

Copyright 2000 John Wiley & Sons, Ltd. Strat. Mgmt. J.,21: 791–811 (2000)

alliances as exogenous in this study. I do so tofocus on the question of whether and under whatconditions firms that have a portfolio of strategicpartnerships outperform those that do not. How-ever, one area of the literature on alliance ante-cedents—the body of work suggesting that thepotential to learn from a strategic partner is animportant and increasing prevalent rationale foralliancing—directly informs the paper’s predic-tions. A number of recent papers have conceivedof alliances as instruments used by firms toacquire know-how and to learn new skills thatreside within other organizations (Hamel, 1991;Powell et al., 1996; Hagedoorn, 1993; Hagedoornand Schakenraad, 1994). There are likely tworeasons for this recent emphasis. First, attentionto the acquisition of know-how as an incentiveto collaborate is driven by the concentration ofalliance activity within particular sectors of theeconomy: high-technology industries are thearenas in which alliance activity has been mostintensive in the recent past (Hagedoorn, 1993).A second and related point is that the emphasison learning is both consistent and coincident withthe gaining influence of knowledge- or com-petence-based conceptions of the firm (e.g., Nel-son and Winter, 1982; Henderson andCockburn, 1994).

As a reason to enter alliances, the potential tolearn from partners highlights the fact thatalliances are, in the first instance,access relation-ships. Just as scholars of social networks haveobserved that social ties purvey access to infor-mation possessed by one’s contacts (Burt, 1992),so have alliance researchers recognized that stra-tegic coalitions can convey access to the resourcesor know-how possessed by one’s partners. Forexample, many of the R&D alliances betweenestablished pharmaceutical firms and dedicatedbiotechnology firms have been structured so thatthe pharmaceutical firm, in exchange for fundinga research project at its biotech partner, acquiresthe right to observe the development process ofthe biotechnology firm (of course, it also obtainsa claim to a large fraction of the revenue streamgenerated by the resultant discoveries). Similarly,many of the horizontal alliances in the semicon-ductor industry have been forged by corporationseager to acquire device or manufacturing tech-nology from their strategic partners.

If learning specifically and gaining access toresources more generally are the sources of the

Alliances and Firm Performance 793

advantage attained from alliances (in addition tobeing among the most compelling motives toenter them), then all potential alliance partnersare not of equal value. In this study, I positthat well-endowed firms—for instance, those thatpossess a large stock of technological resourcesor those with extensive market coverage—are thetypes of alliance partners that can produce thebest ex post results for their associates. Moreover,I suggest that attributes of a focal firm are likelyto interact with the characteristics of its alliancepartners to influence the relationship between itscollaborative activity and subsequent perform-ance.

Existing research has established that alliancesoften have positive effects on a number of differ-ent measures of corporate performance. Forexample, McConnell and Nantell (1985) showedthat the equity markets reward parent companies’share prices when they announce joint ventures.Baum and Oliver (1991) and Mitchell and Singh(1996) treated mortality as the performance vari-able and showed that alliances raised organi-zational survival rates. Uzzi (1996) showed thatapparel firms with strong ties to business groupsenjoyed improved life changes. Taking a differentapproach, Singh and Mitchell (1996) demon-strated that the mortality rate of a focal firmincreased when its strategic partner ceased oper-ations or established a new alliance with a differ-ent firm. Studying a sample of young firms inthe biotechnology industry, Powellet al. (1996)found that companies which had formed manyalliances experienced accelerated growth rates. Inone of the few studies that has investigatedwhether the configuration of exchange relationsaffected firm performance, Chung (1996) foundthat patterns of exchange relations betweeninvestment banks affected the volume of firms’security underwriting activity. Finally, Hagedoornand Schakenraad (1994) demonstrated a positiverelationship between entry into technologyalliances and innovation rates.

Although the evidence rests heavily on the sidethat alliances engender superior performance, theextent to which characteristics of a firm’s strategicpartners mediates the link between alliances andperformance remains a largely unexplored areaof research. I argue that because alliances areformed to achieve access to partner-firmresources, the benefit gained from a portfolio ofstrategic alliances is determined in part by attri-


butes of the partner firms that make up theportfolio. No doubt, it is the contingency betweenthe quality of a partner and the dividends of analliance that underlies the costly and time con-suming search and screening processes that manyfirms follow when selecting strategic partners.1

More generally, it is an axiom in the socialnetworks literature that the potential advantagesof a relationship depend upon the social andmaterial capital possessed by the contact (cf.Burt, 1992: chs. 1 and 2). However, despite theconsensus that firms enter strategic alliances toacquire know-how or other resources, there islittle large-sample research that has documentedthe importance of partner characteristics foralliance outcomes.

As a first step toward generating this kind ofevidence, it will be necessary to characterize thefirms in a focal industry according to their quality,skill level, or resources in domains of businessactivity that are critical for competitive successin that particular market. Firms at the apex of theranking in a domain possess the highest quality orgreatest amount of resources in that domain; forexample, they may have the foremost technologi-cal capabilities, leading-edge production tech-nology, the largest customer base, or a premierbrand image. Regardless of the empirical locale,the industry’s incumbents will be stratified interms of the quantity and quality of the resourcesin their possession. My first two hypotheses positassociations between the statuses of a focal firm’salliance partners in an industry’s key resourceshierarchies and the size of the change in the focalfirm’s post-alliance performance.2

The empirical locale for this paper is a high-technology industry and most of the alliances thatI study are technology-related (for example, joint

1 For instance, when Applied Materials was searching for aJapanese joint venture partner to support its initial entry intothe business of designing equipment for the production ofactive matrix liquid crystal displays (high resolution flat paneldisplays), it engaged McKinsey & Co. to evalute 130 Japanesecompanies before ultimately selecting Komatsu as its alliancepartner. This is just one example of the high search costsfirms often incur when they screen potential alliance partners.2 Of course, the most important domains for determiningcompetitive success will vary considerably across contexts. Inconsumer products industries such as packaged foods andover-the-counter medicines, key resources and skills may be,respectively, brand names and consumer marketing capabili-ties. In contrast, in high-technology industries they are likelyto be state-of-the-art manufacturing facilities and a leadingR&D organization.

794 T. E. Stuart

product development agreements). Given the set-ting and types of interfirm coalitions, I investigatehow the standing of alliance partners on twodimensions—degree of innovativeness and extentof market coverage—affect the post-alliance per-formance of the firms that are partnered withthem. With respect to the first dimension, iflearning is one of the primary motives foralliances as emphasized in the literature on coop-erative technology strategies, or even if learningis the unintended by-product of relationshipsestablished to serve quite different purposes, thenthe benefits of a portfolio of collaborativerelations will depend upon the technological com-petencies of the alliance partners that make upthe portfolio. Other factors held constant,alliances with the most skilled innovators are themost viable opportunities to learn new routinesand acquire advanced technical know-how.3

Because innovative firms possess the highestquality technological capabilities, I expect thatthe know-how acquired from highly innovativealliance partners should contribute to a firm’sability to develop new technology in a subsequentperiod. Therefore, I predict:

Hypothesis 1: The greater the technologicalcapabilities of a high-tech firm’s alliance part-ners, the higher the rate of innovation ofthat firm.

In addition to the chance to acquire technologi-cal know-how from a collaborator, technologyalliances may also represent present opportunitiesto enter new market segments and to servicenew customers (Mitchell and Singh, 1992). Thissuggests that revenue growth is another area ofperformance that may be affected by alliancesbecause strategic partners can facilitate entry intonew market niches (Kogut, 1988) and because

3 Although the context was quite different, Lin, Ensel, andVaughn (1981) developed an argument similar to this one.Noting that job seekers often obtain leads for employmentopportunities through social contacts (e.g., friends, friends offriends, and so forth), Linet al. found that the higher theoccupational prestige of the contact, the greater the probabilitythat the job seeker would obtain a high prestige position.Hence, at least in the job search process, ties to individualsat the top of the prestige (resource) hierarchy are morevaluable than ties to those in a lower rank. Indeed, it hasbeen widely observed that connections to high-prestige orotherwise accomplished actors provide access to informationthat may help to achieve a desired results.


the members of an alliance may gain businessfrom their collaborator’s customers, particularlywhen partnerships lead to jointly-developed prod-ucts. For instance, in the semiconductor industrystrategic alliances with platform sponsors histori-cally have been the only route by which firmscould gain entry into a proprietary technologystandard, such as a microprocessor architecture,that was controlled by a different organization(Kogut, Walker, and Kim, 1995; Wade, 1995).When a firm obtains a large alliance partner, therevenue potential of the association is likely tobe significant because the partner is tied into alarge revenue stream. This just reflects the com-mon wisdom that partnerships with the well-connected and well-endowed offer greater rewardsthan do alliances with business associates thatlack resource. Because they may provide accessto extensive distribution channels, long-standingcustomer relations with influential end users, ora widely-adopted technology platform, large stra-tegic partners are more valuable than smallassociates. I predict:

Hypothesis 2: The greater the revenues of ahigh-tech firm’s alliance partners, the higherthe rate of sales growth of that firm.

The rationale for the first two hypotheses isthat alliances are access relationships, and sostrategic ties with well-endowed partners are, onbalance, the most valuable associations. Inaddition to purveying access to resources suchas technological know-how and new customers,alliances often play a second but in no senseancillary role: they can elevate the reputations ofparticipant firms in the eyes of existing and poten-tial customers and the financial community(Podolny, 1994; Rao, 1994; Stuart, Hoang, andHybels, 1999). Moreover, because a good repu-tation is thought to be a rent-generating asset(Wilson, 1985; Fombrun and Shanley, 1990), stra-tegic alliances also affect firm performancethrough their influence on an organization’s repu-tation.

A corporate reputation is a set of attributesthat observers perceive to characterize a firm(Weigelt and Camerer, 1988). My assertion isthat intercorporate alliances convey status to afocal firm when its partners include large, highlyskilled or otherwise well-known organizations,particularly when the focal enterprise is itself


relatively unknown. Therefore, in addition to thetransfer of tangible and knowledge-basedresources, interfirm affiliations may convey socialstatus. To develop this argument, I build upon asociological literature asserting that the status ofan actor can be affected by the statuses of itsclose associates.

Sociologists have argued thatwhen there isuncertainty about the quality of someone or some-thing, evaluations of it are strongly influenced bythe social standing of the actors associated withit (Merton, 1968 [1973]; Podolny, 1994). Thereason that this dynamic may apply in alliancecontexts is that highly regarded organizations arelikely to meticulously evaluate a potential alliancepartner before entering into a collaborative ven-ture with it, and this evaluation acts as a certifi-cation of the quality of the partner. For threereasons, surviving the due diligence of a well-known organization serves as a signal (Spence,1974) of quality for the lesser-known of the firmsin an alliance. First, prominent organizations arelikely to be selective in their choice of strategicpartners in order to preserve their own repu-tations, which may be damaged if they transactwith low-quality or disreputable firms. Second,highly regarded organizations are likely to beperceived as reliable evaluators that are capableof discerning quality differences among potentialpartners. Third, prominent organizations typicallyhave many potential strategic partners, and there-fore their partners—by virtue of being selected—typically were deemed more desirable than anumber of alternates. For these reasons, a firm’simportant constituents (customers, the financialcommunity, the media) will view the gainingof a large or prestigious alliance partner as anendorsement of its quality (Stuart et al., 1999).

There are many documented instances of aproduct or firm of unknown quality gaining anenhanced image because of its association withprominent alters. For example, the rate of dif-fusion of new innovations and of new drugsappears to depend upon whether prominent actors(firms or physicians, respectively) have previouslyadopted the invention or medicine (Podolny andStuart, 1995; Burt, 1987). In other words, broaderperceptions of product quality have been shownto depend upon the reputation of those who haveadopted the product. Similarly, it has been shownthat when organizations are endorsed by insti-tutions such as licensing organizations or trade


associations, they gain an advantage in their sub-sequent attempts to acquire resources (Baum andOliver, 1991; Aldrich and Auster, 1986).

The effect on a firm’s reputation of acquiringwell-known alliance partners will be particularlysignificant if the firm occupies a precarious com-petitive position—for example, if it is young orsmall. The reason is that evaluators rely uponsignals of an organization’s abilities when theypossess few indicators to inform their inferencesabout the firm’s quality, and when they are unsureabout the firm’s future prospects. Potential cus-tomers, suppliers, employees, collaborators andinvestors tend to be knowledgeable about thereliability and ability of large and old enterprises(Stinchcombe, 1965; Hannan and Freeman, 1984),either because they have previously transactedwith them or because old and large organizationshave verifiable reputations. Similarly, it is typi-cally safe to assume that large and old organi-zations will continue to survive in one capacityor another. In contrast, much less is known aboutyoung and small firms and the future of theseorganizations is far from certain. Potential cus-tomers, suppliers, and employees—particularlythose that are risk averse—will be more confidentof the quality and more likely to transact with ayoung or small firm after it has been implicitlycertified by a prominent alliance partner.

Based on the preceding arguments, I expectthat when a young or small firm gains a well-regarded alliance partner, its reputation will beenhanced in the eyes of its public. Because time-varying perceptual data on the assessments of anorganization’s reputation by its constituents donot exist for most firms, my strategy for testinghypotheses about the effect of alliances on repu-tations is to follow the literature in assuming thatimprovements in a firm’s reputation will manifestas enhancements in its performance. Therefore, Itest for a relationship between alliance partnerprominence and improvements in a firm’s repu-tation by investigating how the former affects theperformance of an organization. Following theargument that intercorporate ties are status-enhancing only when there is uncertainly abouta firm’s true quality, I test for the effect ofalliances on reputations by investigating whetherlarge or innovative alliance partners have a partic-ularly strong effect on the performance of youngand small firms. If the effects of having well-known alliance partners are invariant across mea-

796 T. E. Stuart

sures of the uncertainty (age and size) of focalfirms, then I can reject the hypothesis that promi-nent affiliations enhance reputations. I predict:

Hypothesis 3: The greater the technologicalinnovativeness of a high-tech firm’s alliancepartners, the higher the rate of sales growthof that firm particularly if it is young or small.

Hypothesis 4: The combined sales volume ofa high tech firm’s alliance partners will havea more substantial effect on the rate of salesgrowth of that firmif it is young or small.4

Setting, sample, and data

The ideas I have exposited are best exploredempirically in the context of a large sample offirms drawn from a single industry. Limiting theanalysis to a single industry insures that thedimensions on which alliance partners are charac-terized will be of comparable importance. More-over, because the data requirements for testingthe hypotheses are quite high (the models dependupon time-series alliance, patent, and revenue datafor all of the firms in an industry), a multi-industry design was not practical.

For four reasons, I have chosen to draw thesample from the semiconductor industry. First, asnoted by Hagedoorn (1993) and others, companiesin the microelectronics industry have formedmany hundreds ofhorizontal strategic alliances(i.e., agreements between two firms in theindustry). There have been a sufficient numberof alliances in the industry to allow for a largesample study of the effects of horizontal allianceson corporate performance. Second, the industryhas been driven by innovation, meaning that thesurest path to commercial success has been todevelop new technologies (e.g., Tilton, 1971; Wil-son, Ashton, and Egan, 1980). Because techno-

4 The final two hypotheses are limited to sales growth becausestakeholders’ perception of a firm’s reputation are not likelyto influence its rate of innovation in the short term. Over alonger time period, however, a firm’s reputation will influenceits innovation rate by affecting its ability to recruit and retainhigh-quality human resources and to secure the funds andmarket position necessary to launch major innovation projects.Because my data span only a six year window, I am not ableto look in detail at processes that operate over long periodsof time (e.g., how interfirm relationships formed a number ofyears into the past affect a firm’s current rate of innovation).


logical innovation is viewed as a priority by mostof the firms in the industry, the industry is anappropriate context to explore the role of alliancesas a strategy for learning from and gaining accessto the technological capabilities of strategic part-ners (Hypothesis 1). Third, the microelectronicsindustry is well suited for this study because thefirms in it routinely patent their inventions. Thisis an important consideration because patent dataare necessary to operationalize a number of thevariables in the analysis. Finally, the industryconsists of a very heterogeneous population offirms, from small, dedicated producers to largeand diversified electronics conglomerates. There-fore, the industry offers ample variation for test-ing the hypotheses.

The sample that I have analyzed included thesemiconductor companies followed by Dataquest,a consultancy and information services firm, dur-ing the period from 1985 to 1991. Dataquestused information on product shipments to compilerevenue figures for a large number of semicon-ductor producers. Because sales volume is thedependent variable in the sales growth models(Hypotheses 2, 3, and 4), the sample was limitedto the set of organizations tracked by Dataquest.5

The one instance in which the Dataquest database was supplemented was for the small numberof captive semiconductor producers. Sales figuresfor the captive producers were available in theIntegrated Circuit Engineering Corp.’s annualSTATUSvolumes. Adding the captive producersto the Dataquest data, I built a sample of 150companies, although some of the firms in thesample were founded during the analysis periodand so are not represented in all years of thedata. These firms hailed from the U.S., Europe,Japan, and other Southeast Asian countries, andthe sample accounted for over 90 percent ofthe worldwide semiconductor production volumein 1991.

In addition to the sales figures, I required dataon the strategic alliances and the patents of thefirms in the industry to construct the variables

5 Many of the organizations in the sample participated inmultiple business lines (e.g., IBM, Siemens, and Hitachi) andmany were privately-owned. While corporate-level sales fig-ures could be ascertained from public sources for the publicly-traded firms in the sample, longitudinalsemiconductoronlysales volume data were quite difficult to obtain even for manyof the publicly-owned firms. For this reason, I had to relyupon the revenue data from Dataquest.


for the analysis. I recorded all publicly-reportedalliances formed between semiconductor pro-ducers during an 11 years period. To preserve theconsistency of the measures of partner attributes, Ichose to focus only on horizontal (intra-industry)alliances: the data excluded all partnershipsinvolving a semiconductor firm and a secondorganization outside of the industry, such as asoftware producer. The sources for the alliancedata included thePredicastsindexes (U.S., Eur-ope, and International), articles inLexis/Nexis,Infotrak, Electronic News, Electronic Buyer’sNews, Electronic Engineering Times, Electronics,Electronic Business, as well as company SECfilings. The data, consisting of more than 1600dyadic alliances, include five types of collabo-rative relationships: joint product developmentagreements, joint ventures, technology exchanges,licensing, and marketing agreements.6

I also collected all U.S. semiconductor patentsassigned to the firms in the sample. I chose touse domestic patents because the U.S. is theworld’s largest technology marketplace. Becausea firm must patent in a country to gain intellectualproperty protection in that geography, non-U.S.-based firms regularly patent in the U.S. (seeAlbert et al., 1991). To assemble the patent data,I first identified approximately 2400 distinct U.S.patent classes which contained semiconductorproduct, device, and design inventions. I thenretrieved the patents in these classes from theMicropatent 1994 Patent AbstractCD, whichincluded all U.S. patents issued between 1975and 1993. For each patent document, I recordedthree pieces of information that were necessary

6 I have estimated the models using two different criteria forincluding alliances. First, I estimated models that included alltypes of agreements in the computation of the alliance-basedindependent variables. Second, I restricted the alliance datato just three of the five types of agreements: joint ventures,joint product development agreements, and technologyexchanges. The reason to impose this screen is that thesethree forms are the more durable, more intensive, and oftenthe more strategically significant of the five types of alliances.The reported results are from the models including all partner-ships, but the findings are similar to those resulting whenonly the three more durable and intensive alliance types areused to compute the alliance-based covariates. I have chosento report the estimates from the models that include allalliances because the data on alliance type are occasionallymissing for partnerships between small and non-U.S.-basedfirms. Therefore, excluding alliances by type may introducesystematic measurement error by lowering the realizations onthe alliance-based variables for small firms and enterprisesheadquartered outside of the U.S.


to construct the variables for the analysis: thedate of application, the corporate assignee, andthe list of prior art (patent) citations. For the 150firms in the sample, I then constructed detailedfamily ownership trees using theDirectory ofCorporate Affiliations. These corporate ownershiprelations were used to assign subsidiaries’ patentsto their corporate parents.

While there has been some question about thereliability of patents as innovation indices (Levinet al., 1987), there is evidence that firms in thesemiconductor industry actively file for patents.As the strength of U.S. intellectual property pro-tection has increased and a number of the firmsin the industry have begun to appeal to the courtsto defend their intellectual property positions,7

semiconductor firms have raised the priority ofpatenting (Rivette, 1993). The proclivity of do-mestic and international firms to patent semicon-ductor technologies in the U.S. is evidenced bythe fact that the six firms that had received thegreatest number of U.S. patents in 1996—IBM,Motorola, NEC, Hitachi, Canon, and Mitsubishi—each had substantial semiconductor operations,and four of these firms were headquartered out-side of the U.S.

Data and variables

Characteristics of alliance partners

The four hypotheses posit relationships betweensummary attributes of a firm’s alliance partnersand its ex post performance, measured either asa rate of innovation or as a rate of sales growth.In particular, I have argued that large and techno-logically innovative firms are the alliance partnersthat will lead to the most substantial performance

7 In 1982 the Congress established the Court of Appeals forthe Federal Circuit in Washington DC specifically to hearpatent cases. The new court has fortified patent protectionin the semiconductor industry by consistently ruling againstchallenges to patent claims (Almeida and Rosenkopf, 1997).Adding to the strategic importance of patents, a number offirms in the industry, such as Intel and Texas Instruments,have been particularly vigilant in litigating perceivedviolations of their intellectual property. As the incidence ofpatent infringement suits in the industry has grown, patentingas a defensive strategy has become considerably moreimportant because a large and extensive patent portfolio helpsto defend against infringement charges. For example, whenDEC recently accused Intel of infringing on its earlymicroprocessing patents, Intel’s extensive patent portfolioenabled it to file a counter suit charging infringement by DEC.

798 T. E. Stuart

gains. Therefore, to test the hypotheses it isnecessary to derive measures of the inno-vativeness and the size of the firms in thesemiconductor industry so that each organi-zation’s alliance partners can be described onthose dimensions.

I have used patent citation data to constructinnovativeness scores for the sampled firms. Oneof the requirements of a patent application is tolist citations to all previously-granted patentswhich made technological claims similar to thoseclaimed in the application. This process is tanta-mount to mandating that patent applicants identifyand acknowledge the existing, patented inventionsthat are technologically nearest to their inventions.It is then the obligation of the patent examinerto verify that the list of references in the patentapplication, known as the ‘prior art,’ is complete.When a patent application is granted, the patentissues with the list of prior art citations, includingall citations added by the patent examiner.

Just as citations between journal articles revealthe transmission of ideas between papers, patentcitations trace technological ancestries. Centralnodes in the patent citation network (i.e., highlycited patents) therefore represent highly influentialinnovations. Patent citation data have been usedto measure the importance of inventions in theeconomics literature (Trajtenberg, 1990), theapplied technology literature (Albertet al., 1991)and work on the sociology of technology(Podolny and Stuart, 1995). Perhaps the mostdirect evidence of the validity of patent citationsas a measure of the quality of innovations comesfrom studies such as Albertet al. (1991), whichuncovers a very high correlation between thenumber of citations received by a set of patentsand the rankings by technical experts in the rel-evant field of the importance of these inventions(see also Carpenter, Narin, and Woolf, 1981).

Assuming that the most important patentedinventions are those that are highly cited in futurepatents, then the most innovative firms in anindustry are those that have developed a signifi-cant fraction of the highly-cited patents. Accord-ingly, I compute the innovativeness of a firmas a composite, citation-based measure of theimportance of the individual patents in its port-folio. Specifically, I define the innovativeness ofa given semiconductor firm (denoted asi) duringa particular time period (denoted ast) as theproportion of prior art citations included in the


universe of U.S. semiconductor patents appliedfor in year t that refer to patents that areassigned to firm i.8 I denote the innovativeness ofa firm i at a time t as dit. For each time periodin the data series, anNx1 vector dt contains theinnovativeness scores of theN firms in the sam-ple.9

With measures of the innovativeness and size(which I have operationalized as annual semicon-ductor sales volume) of all semiconductor firmsas well as a record of the alliance activity in theindustry, it is possible to construct time-varying,summary measures of the innovativeness and sizeof the semiconductor alliance partners of each ofthe firms in the sample. To compute these mea-sures, I first created a set of time-changingalliance matrices, labeledWt=[wijt]. The Wt areNxN (firm-by-firm) symmetrical matrices. Theelements of the alliance matrices (thewijt) aredefined as a positive value when theij th pair offirms had formed an alliance during periodt, andas ‘0’ if there was no alliance between firmsiand j in t.

The innovativeness (size) of the semiconductoralliance partners of each of the firms in thesample at periodt is the product of the alliancematrix at t at the corresponding vector of inno-vativeness (size) scores for the firms in the indus-try. Hence, I define the vectorspt (vt) as:pt = Wtdt, vt = Wtst where Wt are the binaryalliance matrices anddt (st) are the innovativeness(sales) vectors. Thereforept (vt) are time-changingNx1 vectors containing the summed innovativeness(size) scores for the alliance partners of each ofthe firms in the sample during each yeart. Hypoth-esis 1 predicts that firms which enjoy access totechnologically-advanced alliance partners willinnovate at a greater rate than otherwise compara-

8 I have also constructed a number of different permutationsof this variable by altering the treatment of time. For instance,I have computed innovativeness scores as future citations tocurrent-period patents, in addition to measuring innovativenessas current-year citations to previously-issued patents. Fortu-nately, all of the measures I have computed were correlatedabove 0.90.9 As an alternative to weighting each organization’s patentportfolio by future patent citations, one could just use a countof the number of patents received by a firm as an innovationindex. Conceptually, the difference between the two is thatthe raw patents count does not weight the quantity of patentsby the best available measure of their importance. However,empirically the two measures are often highly correlated(above 0.80 in the data for this paper), and the raw patentcount is much more expeditiously constructed.


ble firms that do not enjoy access to innovativeaffiliaties (in other words, a positive coefficient onthe pt variable in models of innovation rates).10

Hypothesis 2 predicts that firms with large alliancepartners will grow at a greater rate than otherwisecomparable firms that do not possess alliance part-ners with extensive market coverage (a positivecoefficient on vt in the sales growth models).Hypotheses 3 and 4 predict that possessing largeand innovative alliance partners, although ben-eficial for all firms, will have the greatest effecton the growth rates of young and small firms.

The one outstanding issue in the computationof the characteristics-of-alliance partner variablesconcerns the lag structure. Learning from anotherorganization and then integrating that knowledgeinto a firm’s own routines or technologies maytake time. Similarly, it requires time for analliance to lead to jointly-developed products andfor a focal organization to gain access to a collab-orator’s customer base or entry into new marketniches. Therefore, including only the alliancesformed during the prior year in models of current-year performance may not allow a sufficient inter-val of time for the benefits of a cooperative strat-egy to manifest in observable performance mea-sures. To address this issue, I have chosen todefine the alliance matrix for yeart to incorporateall alliance activity that had occurred during theprevious five-year period, that is duringt-1 to t-5.11

I experimented with two weighting schemes tomodify the influence of alliances that occurred inthe past (in addition to estimating models thatused no weights). First, I (linearly) depreciated

10 The specification ofpt (vt) and the models to be estimatedare similar to a general class of social influence models ofthe form: y = ρWy + xβ + e. In these models,y typicallyrefers to an attitude or opinion held by the actors in anetwork, W is often known as a structure matrix becauseeachwij measures the influence that actorj has on the opinionof actor i, and x is an nxk matrix of k covariates. In otherwords, the value ofyi is assumed to be influenced by aweighted combination of the opinions (yj) of other actors.11 The existing literature helps to define the lag structure.Pakes and Griliches (1984) modeled firms’ current-year patent-ing as a function of five lags of annual R&D spending. Theyfound that contemporaneous R&D spending and the 5th-yearlag were the two significant predictors of current patenting.They also estimated the mean R&D project gestation lag tobe 1.6 years (the time from when an R&D project is begununtil it first generates a revenue stream). Pakes and Schanker-man (1984), which reports estimated gestation lags by a fewmajor industry groupings, contends that the mean gestationlag in electronics is only 0.84 years. Based on these findings,I have chosen a five-year window to compute the alliance-based covariates. The window begins at a one year lag.


the contribution of older alliances to the summarymeasure of each firm’s alliance partners. I con-structed a measure such that alliances whichoccurred five years prior to the current yearreceived a weight of 0.2, those that were estab-lished four years ago received a weight of 0.4,and so on until the lagged year, which receiveda weight of 1.0. I then multiplied each alliancematrix by the corresponding size and inno-vativeness vectors. Conversely, based upon thelogic that effective interorganizational learningrequires the development of relationship-specificknowledge-sharing routines (Lane and Lubatkin,1998; Dyer and Singh, 1998), I also experimentedwith a weighting scheme that depreciated thecontribution of alliances which were formedwithin the past three years (by assigning agree-ments that were formed within the last three yearsa weight of 0.5, versus 1.0 for older agreements).The results were weakest in the latter weightingscheme, when more recent agreements wereassigned a reduced weight (although the coef-ficient magnitudes differed relatively slightlyacross the weighting schemes). The reportedresults are from the models in which the influenceof older alliances was linearly depreciated.

Estimation

Modeling innovation rates

The outcome variable in the test of the firsthypothesis, that organizations which possess tech-nologically advanced alliance partners innovate ata higher rate, is a count of the number of newsemiconductor patents applied for by each organi-zation in the sample in each year of the analysisperiod. This variable is bounded at zero, canassume only integer values, and consists of obser-vations on the same firms at multiple points intime. I have modeled the data using a randomeffects Poission estimator with a robust varianceestimator, (i.e., it does not assume within-firmobservational independence for the purpose ofcomputing standard errors). Poisson regressionassumes that the event count is drawn from thesingle parameter Poisson distribution, which canbe written as:

Pr(Yit = yit) =exp(-λit)λyitit

yit!(2)

800 T. E. Stuart

where the parameterλit represents the mean andthe variance of the event count. It is assumedthat In λit2 = β9xit1. Hypothesis 1 is tested byincluding as a regressorpit, the summed inno-vativeness score of a firm’s alliance partners, inthe patent rate models.12

In addition to the innovativeness of a firm’salliance partners, the patent rate models controlfor a number of firm characteristics. First, themodels include a raw count of the number oftechnology alliances formed by each firm duringthe previous five years. It may be that firmsthat have formed a greater number of technologyalliances innovate at a higher rate simply becausethe decision to enter alliances is a reflection ofa firm’s commitment to an innovation-focusedtechnology strategy (in other words, that the num-ber of alliances formed by the firm is a proxyfor an underlying and unobserved strategicdisposition). By controlling for the total numberof alliances, I am able to separate the effect ofcharacteristics of a firm’s alliance partners fromthe effect of the number of alliances the firm hasformed. Second, the models include the size ofeach firm measured as its annual semiconductorsales. Because larger firms typically possessgreater resources to invest in R&D, larger firmsare likely to innovate at a higher rate (Cohenand Levin, 1989, review the literature on therelationship between firm size and innovation).13

The models also include annual dummy vari-

12 Poisson regression assumes that the mean and variance ofthe event count are equal. Because this assumption is oftenviolated, I have also used a fixed-effects negative binomialestimator to fit the innovation rate models, which accommo-dates overdispersed data (I have implemented the maximumlikelihood estimator developed by Hausman, Hall, and Gril-iches, 1984, using Stata 5.0’s built in maximum likelihoodcapability). Effectively, this model conditions on the eventcount for each unit (firm) over the observation window, sowhen using this model I omit the occurrence-dependence term(a one-year-lagged count of the total number of patents issuedto each firm since 1975). I concentrate on the random effectsPoisson models because they allow for informative estimatesof the impact of time-invariant firm characteristics, such asthe firm nationality dummies. However, I do report the fullmodel using the fixed effects negative binomial to show thatthe results are robust to the estimator.13 I was unable to collect time series R&D spending data onthe privately-held, diversified, and non-U.S.-based firms in thedataset. However, among all of the dedicated (non-diversified)semiconductor producers in the Compustat data base (all firmsthat participated only in SIC 3674), the bivariate correlationbetween annual sales revenue and annual R&D expenditurewas 0.977. From this I infer that controlling for semiconductorsales volume is a close approximation to controlling forannual semiconductor R&D spending.


ables to account for time-changing factors, includ-ing macroeconomics conditions, that may haveaffected the industry as a whole. These dummiescontrol for omitted factors that have constanteffects on the organizations in the sample butvary over time. They also serve to capture anysecular trends in the incidence of patenting.Hence, the regression coefficients in all modelscan be interpreted as within-year effects.

Finally, to control for firm heterogeneity in thepropensity or ability to patent, I have includedin the patent rate models a variable that reflectshistorical differences across organizations in theirpatenting behavior: all models contain a count ofthe number of semiconductor patents granted toeach organization from 1975 until the year priorto the dependent variable. Including the numberof times that the focal event has previouslyoccurred for each firm is a common method ofcontrolling for unobserved heterogeneity(Heckman and Borjas, 1980). The occurrencedependence variable should control for the time-constant effects of unobserved factors (such asinterfirm differences in internal processes andincentive structures, as well as differences inunderlying innovation strategies) that producevariance in organizations’ abilities, opportunities,or dispositions to patent.

Modeling the rate of sales growth

Hypotheses 2 through 4 consider how character-istics of firms’ alliance partners affect their sub-sequent-period performance. As a measure of per-formance, I have chosen semiconductor salesvolume rather than an accounting-based measure.There are two reasons for this choice, the firsttheoretical and the second pragmatic. First, theexpected enhancement in reputation associatedwith gaining a highly regarded alliance partnershould manifest in revenue increases because riskaverse customers will be more willing to sourcefrom firms that have been endorsed by well-regarded organization. Therefore, the reputationarguments (Hypotheses 3 and 4) should leave adiscernible trail in changes in sales volume ifthey operate in this sample of firms. While it isalso probable that improvement in a firm’s repu-tation will create better financial performance bylowering the organization’s cost structure andincreasing the prices that the market will acceptfor its products (Podolny, 1993), these processes


may unfold over a longer period of time. I expectincreases in sales volume from gaining new cus-tomers to occur more quickly than changes inaccounting-based performance measures, andtherefore modeling sales growth permits a shorterlag structure in the statistical analyses. Second,because many of the firms in the data base werediversified into an array of end use products (e.g.,IBM, Siemens, Hitachi) and many others wereprivately held, I was unable to obtain accountingmeasures reflecting firms’ activities in thesemiconductor business for the majority of thefirms in the sample. However, I was able togather sales data for thesemiconductoroperationsof each of the firms in our sample, even whenthey were diversified or privately owned.

Following prior research (Barron, West, andHannan, 1994; Barnett and Carroll, 1987;Podolny, Stuart, and Hannan, 1996), I have mod-eled the sales of the firms in the sample withthe function:

Si,t+1 = Sita exp(π9xit)ε (3)

WhereSit is the sales of firmi at time t and xit

is a covariate matrix. Log transforming this powerfunction, equation (3) can be expressed as:

log(Si,t+1) = α log(Sit) + π9xit + ei,t+1 (4)

Equation 4 can be estimated using OLS. Thisapproach yields unbiased and efficient estimatesunder the standard linearity, homoscedasticity,and independence assumptions. However, the rawdata are a pooled cross-section time-series, andnot surprisingly there is evidence that the distur-bances in Equation 4 are autocorrelated. Becausethe autocorrelation appears to arise from persis-tent, within-firm effects (inter-temporally stablefeatures of firms that affect the growth process),I have estimated Equation 4 using a least squaresconstants estimator (Tuma and Hannan, 1984).Adding fixed effects for firms assumes that thecorrelation structure in the disturbance term canbe decomposed into a firm-specific effect anda residual term (µ) that is uncorrelated acrossobservations and is homoscedastic. It is importantto note that because of the firm dummy variablesin Equation 4, the estimated coefficients representwithin-firm effects.

In addition to lagged sales volume and firmdummy variables, the sales growth models


include time period effects (annual, calendar timedummy variables) and a raw count of the numberof alliances formed by each of the firms in thesample. As in the innovation rate models, theeffect of the characteristics of alliance partners isassessed after controlling for the number ofalliances that a firm had formed during the pre-vious five-year period. The models also includethe age of the firms in the sample, defined asthe number of years since founding for dedicatedsemiconductor producers and as the number ofyears since entry into the industry for diversifiedproducers. In addition to these variables, Hypoth-eses 2, 3, and 4 are tested by including thecombined revenues of each firm’s alliance part-ners, the innovativeness of its alliance partners,and a series of interaction terms.

Results

Table 1 recapitulates variable definitions and,when there are hypothesized relationships, thedirection of the predicted effects. Table 2 presentsa correlation matrix and descriptive statistics forthe variables in the innovation rate models, andTable 3 the results from the patent rate analysis.

Model 1 in Table 1 includes lagged firm sales,nationality dummy variables, annual periodeffects, and the lagged patent count as an unob-served heterogeneity control variable. Note thatall of the models in the paper are multiplicative,so the partial effect of a variable can be under-stood as a multiplier rate. In Model 1, the laggedpatent count has a positive and highly significanteffect on the patent rate. The level of semicon-ductor sales is also positive. The ‘Firm is Europe-an’ dummy is positive although not significant,and the ‘Firm is Japanese’ dummy is positiveand significant. Because the omitted nationalitycategory in the patent rate models is U.S.-basedfirms, the coefficient on the ‘Japan’ dummy indi-cates that Japanese semiconductor producers pat-ented at a higher rate in the U.S. than did U.S.-based companies, even after controlling for thesize of the organization and a firm’s proclivity topatent as indicated by the lagged patent count.14

14 It is often asserted that Japanese firms apply for patentsthat make (relatively) narrow claims for intellectual propertyprotection, while U.S. firms apply for fewer patents thatclaim broader property rights. Because ‘U.S.’ is the omittednationality in the patent rate models, this difference would

802 T. E. Stuart

Table 1. Definitions of variables appearing in patent rate and growth models

Variable name Variable description Expected effect

Lag of semiconductor Log of firms sales in semiconductorssalesTotal semiconductor Count of the number of semiconductor patents issued to thepatents firm since 1976Firm age Number of years since firm began operations in

semiconductorsNumber of technology Count of the number of strategic alliances formed by the firmalliances (SZijt) in the five previous years (t-5 to t-1)Firm is Japanese Dummy variable denoting that the firm is headquartered in

JapanFirm is European Dummy variable denoting that the firm is headquartered in

EuropeFirm is other Asia- Dummy variable denoting that the firm is headquartered inPacific Asia, but outside of Japan

Sales of partners(ΣVit) Sum of the semiconductor sales of the firm’s strategic Positivealliance partners

Age-by-sales of Interaction of the firm age with the sum of the sales of theNegativepartners firm’s alliance partnersFirm sales-by-sales Interaction of firm size with the sum of the sales of the Negativeof partners firm’s alliance partnersInnovativeness of Sum of the patent citations received by the firm’s alliance Positivepartners (Vit) partnersAge-by-innov. Interaction of firm age with the number of patent citations Negativeof partners received by the firm’s alliance partnersFirm sales-by-innov. Interaction of firm size with the number of patent citationsNegativeof partners received by the firm’s alliance partners

Notes: All variables are included as one-year lags. Not all variables appear in all models. For variables appearing in thepatent rate and sales growth models, the expected effect pertains to both models.

Turning to the alliance variables, Model 2includes a count of the total number of horizontaltechnology alliances formed by each firm duringthe previous five-year period, as well as the mea-sure of the innovativeness of alliance partners.First, the addition of the two variables substan-tially improves upon the fit of the baseline model(chi-squared[2]=23.67, p<0.0001; all subsequentmodels are also significant against the baseline

explain the positive coefficient on the ‘firm is Japanese’dummy variable in Table 2, although note that the effectpersists even when the lagged patent count is included in themodels. A number of supplemental analyses did suggest thatthis may be occurring; for example, I modeled the rate atwhich patents were cited by future patents as a function ofthe nationality of the firms that own the patents. Assumingthat patents which make broad claims will be more influential,this analysis should and did show that patents held by U.S.firms are cited at a higher rate than patents assigned toJapanese firms. Although the differences in citation rates couldbe caused by other factors, the scope of patent claims is aplausible explanation. Regardless, differences such as thesehighlight the importance of controlling for firm nationality.


model). The findings support the first hypothesis:the effect of the alliance count variable is notsignificantly different from zero, but the effect ofthe innovativeness of alliance partners is a highlysignificant predictor of the patent rate. Other vari-ables held constant, a one-standard deviationincrease in the innovativeness of a firm’s alliancepartners produces a 40 percent increase in itsinnovation rate (=exp[3.400∗0.098], or 1.395).Model 2 thus confirms Hypothesis 1: firms thatpossessed technologically advanced alliance part-ners innovated at a substantially greater rate thanthose that did not.15

15 In an influential article, Hamel, Doz, and Prahalad (1989)asserted that Japanese firms in particular have excelled atmanaging alliances from the standpoint of appropriating learn-ing from their collaborators. This is readily testable with thesedata; I have tested for this relationship by interacting the‘Firm-is-Japanese’ dummy with (a) the alliance count, and(b) the innovativeness-of-alliance-partners variables in theinnovation rate models. The coefficients on the interactionvariables were not significantly different from zero; in these


Not surprisingly, there is a high bivariate corre-lation between the total number of alliances anorganization has formed during the previous fiveyears and the summed innovativeness scores ofits alliance partners computed during that sameperiod. I have taken two additional steps to dem-onstrate that the innovativeness-of-partner findingis not driven by collinearity. First, in Model 3 Ihave included the summed innovativeness ofalliance partners while omitting the variabledesignating total number of alliances formed.Again, I find a positive and significant coefficienton the innovativeness-of-partners variable (alsonote that the standard error for the innovativenessvariable changes little when the alliance count isexcluded from the model). Second, in Model 4 Ihave defined a new variable that is the averageinnovativeness score computed over the set ofpartners in each firm’s alliance portfolio. Thisvariable is not highly correlated with the totalcount of alliance partners but still captures differ-ences between firms in the innovativeness of theirstrategic partners. Consistent with the Model 2findings, the Model 4 results show that anincrease in the mean innovativeness of alliancepartners positively multiplies a focal firm’s patentrate. Finally, Model 5 in Table 3 reports fixedeffects negative binomial estimates using theModel 2 covariate vector (see footnote 11 abovefor a brief description of the estimator). Compar-ing the Model 2 with the Model 5 findings dem-onstrates that the results are not at all sensitiveto the estimator. Additionally, Model 5 confirmsthat the results hold up when firm fixed effectsare included.

Moving on to the sales growth models to testHypotheses 2, 3, and 4, Table 4 reports withinfirm means, standard deviations, and correlationsfor the variables included in the sales growthmodels, and Table 5 reports the estimates fromthe fixed effects models of semiconductor sales.16

data, the evidence suggests that Japanese firms were no betterthan companies from other nations in terms of their abilitiesto translate alliances into higher innovation rates. I alsointeracted the U.S.-firm dummy with the alliance variables aswell as a dummy variable that designated all Pacific Rimfirms (consisting of Japanese, Taiwanese, and South Koreanfirms). The interaction terms in these additional models werealso not significant.16 In the correlation matrix reported in Table 3 a number ofthe interaction variables are reasonably highly inter-correlated(p<0.70). The high bivariate correlations are not a cause forconcern because the interaction variables were not si-multaneously entered into any of the estimated models.


The baseline Model 1 includes calendar timedummies, firm age, and firm size measured duringthe previous year, in addition to the fixed effects.It is not necessary to incorporate the nationalitydummy variables in the sales growth modelsbecause all time-invariant attributes of the firmsin the sample are captured by the firm-specificintercept adjustments. Of interest in Model 1 (andthroughout Table 5) is the fact that the coefficient(denotedα in Equations 3 and 4) on the laggedsales variable is substantially less than unity.Given the log-log specification, the implicationof this is that small firms have historically grownat a substantially higherrate than have largefirms in the semiconductor industry (a coefficientof ‘1’ would suggest that growth rates do notdepend upon starting size;α>1 would imply thatlarge firms grow at a higher rate than smallfirms).

Model 2 adds to the baseline the number ofalliances formed during the previous five yearsand the combined size of each firm’s alliancepartners. AnF-test shows that adding the twoalliance variables to the baseline model signifi-cantly improves the model’s fit (p<0.01; Models2 through 6 are all statistically significantimprovements over the baseline). Consistent withthe findings in the innovation rate models, thealliance count variable, although positive, is sta-tistically insignificant (as it is in each of themodels in Table 5). In support of Hypothesis 2,the positive and significant coefficient on the size-of-partners variable establishes that semiconductorfirms that had strategic alliances with large part-ners grew at a higher rate than did firms that didnot enjoy access to large partners. Moreover, themagnitude of the effect is substantial: a one-standard deviation increase in the sales of a firm’salliance partners leads to a 2.7 percent increasein its annual growth rate (=exp[0.0172∗1.52], or1.027).17

Turning to the final two hypotheses, the repu-tation arguments expressed in the third and fourthhypotheses assert that the effect of possessinglarge or innovative alliance partners will be great-

17 Because of the fixed effects specification, I interpret themagnitude of all of the growth model findings using within-firm standard deviation changes in a variable to computeimplied changes in a firm’s growth rate. A 2.7 percent annualincrease would translate into a 21 percent increase in size(relative to an otherwise comparable firm) over the fulltime series.

804 T. E. Stuart

Tab

le2.

Mea

ns,

stan

dard

devi

atio

ns,

and

corr

elat

ions

for

pate

ntra

tem

odel

s

Var

iabl

em

ean

std.

dev.

1)2)

3)4)

5)6)

7)

1)F

irmis

Japa

nese

0.13

80.

345

–2)

Firm

isA

sia-

Pac

ific

0.08

20.

275

20.

12–

3)F

irmis

Eur

opea

n0.

110

0.31

32

0.14

20.

11–

4)La

gof

sem

icon

duct

orsa

les

0.32

90.

728

0.39

20.

090.

05–

5)T

otal

num

ber

ofpa

tent

s0.

179

0.42

40.

172

0.12

0.09

0.69

–6)

Num

ber

ofte

chno

logy

allia

nces

5.47

90.

891

0.14

20.

090.

110.

600.

59–

(SZ

ijt)

7)In

nova

tiven

ess

ofal

lianc

epa

rtne

rs0.

066

0.09

70.

152

0.11

0.07

0.54

0.58

0.91

–(S

Pit)

8)M

ean

inno

vativ

enes

sof

part

ners

0.00

90.

011

0.11

20.

132

0.10

0.10

0.17

0.09

0.37

(SP

it/S

Zijt

)



Table 3. Determinants of the patent rate of semiconductor firms, 1986–1992

Variable Model 1 Model 2 Model 3 Model 4 Model 5

Lag of semiconductor sales 0.206∗ 0.142 0.1411 0.1269 0.1457∗

(0.103) (0.115) (0.115) (0.101) (0.063)Total number of patents 0.700∗ 0.651∗ 0.651∗ 0.777∗

(0.179) (0.202) (0.203) (0.162)Number of technology 20.005 0.0306∗ 20.0017alliances (SZijt) (0.014) (0.008) (0.009)Innovativeness of alliance 3.400∗ 3.373∗ 2.544∗partners (SPit) (1.136) (0.996) (0.729)Mean innovativeness alliance 18.094∗partners (SPit/SZijt) (5.104)Firm is Japanese 1.060∗ 1.015∗ 1.016∗ 1.041∗

(0.315) (0.273) (0.269) (0.281)Firm is Asia-Pacific 20.997 20.784 20.786 20.936

(1.053) (0.978) (0.978) (0.989)Firm is European 0.467 0.347 0.345 0.324

(0.446) (0.309) (0.311) (0.324)Year is 1987 0.127∗ 0.072 0.047 0.041 0.107

(0.058) (0.052) (0.057) (0.060) (0.086)Year is 1988 0.305∗ 0.117 0.116 0.135 0.299∗

(0.075) (0.076) (0.076) (0.084) (0.087)Year is 1989 0.129 0.128 20.063 20.084 0.249∗

(0.091) (0.096) (0.096) (0.099) (0.089)Year is 1990 20.054 20.169 20.201 20.26 0.205∗

(0.139) (0.134) (0.135) (0.142) (0.091)Year is 1991 20.849 20.983∗ 21.010∗ 21.043 20.425∗

(0.187) (0.175) (0.173) (0.189) (0.110)Constant 1.908∗ 1.775∗ 1.775∗ 1.605∗ 1.925∗

(0.196) (0.181) (0.181) (0.178) (0.138)Pearson Chi-Square 24524.21 17298.58 17289.89 17336.63Number of firms 150 150 150 150 150Number of firm years 825 825 825 825 825Log-likelihood 2187.76∗p<0.05

Notes: Models 1–4 use the random effects Poisson estimator; model 5 uses a fixed effects negative binomial estimator

est if a firm is young or small: for relativelyunknown organizations, a notable strategic partneris akin to a signal of its quality. To test thesehypotheses, I added a series of interaction effectsto the Model 2 covariate vector. The third modelincludes an interaction between the age of a focalfirm and the size of its alliance partners, and thefifth model contains an interaction between theage of a focal firm and the innovativeness of itsalliance partners. The coefficient on the age-by-size-of-partners interaction is negative and sig-nificant, indicating that possessing large alliancepartners increased the growth rate of youngerfirms more than it augmented the growth rateof older firms. The coefficient on the age-by-innovativeness-of-partners interaction in Model 5is also negative, showing that having highly inno-


vative alliance partners was a greater benefit toyoung than to old organizations. Both findingssupport the prediction that the value of coalitionswith large and innovative firms is greatest foryoung producers, probably because importantconstituents are uncertain about the quality andreliability of those organizations.

To demonstrate the magnitude of the age inter-actions, consider the differential impact of havinglarge or innovative alliance partners on firmgrowth rates assuming different levels of focalfirm age. For example, consider the typical firmat two different points in its life: when it is onestandard deviation below its mean age in the timeseries and when it is one standard deviation aboveits mean age (I will use the overall sample meanand the within-firm standard deviation of the age

806 T. E. Stuart

Table 4. Within-firm means, standard deviations, and correlations for sales growth models

Variable mean st. dev. 1) 2) 3) 4) 5) 6) 7) 8)

1) Lag of 4.390 0.499 –semiconductorsales (logged)

2) Firm age 18.343 1.938 0.45 –3) Number of 5.479 2.337 0.18 0.30 –

technologyalliances (SZijt)

4) Sales of alliance 5.387 1.523 0.13 0.26 0.25 –partners (SVit)

5) Innovativeness of 0.066 0.034 0.32 0.34 0.54 0.35 –alliance partners(SPit)

6) Interaction: 27.367 8.035 0.49 0.44 0.38 0.76 0.45 –lagged sales-by-(SVit)

7) Interaction: 0.419 0.233 0.35 0.31 0.60 0.23 0.86 0.43 –lagged sales-by-(SPijt)

8) Interaction: age- 118.470 40.402 0.17 0.32 0.37 0.72 0.35 0.83 0.34by-(SVit)

9) Interaction: age- 1.861 1.077 0.15 0.42 0.63 0.21 0.76 0.41 0.4 0.39by-(SPijt)

variable for this illustration). Now, assume thatthis firm has large alliance partners at both stagesof its life—suppose that the combined sizes ofits alliance partners rank at the 75th percentileof the size-of-partners distribution at both lifestages. When the firm is young (one standarddeviation below its mean age in the time series),the partial effect of having large alliance partnersleads to a predicted increase in the annual growthrate of 15.7 percent; when it is one standarddeviation above its mean age, however, havingthe same alliance partners leads to a predictedannual growth rate increase of only 9.6 percent.18

Assuming the same age conditions but substitut-ing the size-of-partners results (Model 3) withthe innovativeness-of-partners results (Model 5),the predictions are for a 9 percent increase in

18 The partial effect of the size-of-alliance-partners (Vit) inTable 5, Model 3 is: exp[Vit(0.054–0.002AGEit)]. Replacingthe variables with their assumed values, the young firm experi-ences a growth rate increase of 15.7 percent (=exp[6.88(0.054–0.002∗16.4)]) from having relatively large strategic alliancepartners. Without altering the value of the size-of-partnersvariable and changing only firm age, the older firm has apredicted growth rate increase of 9.6 percent(=exp[6.88(0.054–0.002∗20.3)]) from having relatively largealliance partners.


growth rate when the firm is younger and a 6percent increase when it is older.

The results for the interactions between thesize of a focal firm and the characteristics of itsalliance partners appear in Models 4 and 6 inTable 5. The coefficient on the focal-firm-size-by-size-of-alliance-partners interaction in Model4 is negative and significant, demonstrating thatlarge alliance partners had the greatest effect onthe growth rates of firms when they were small.In Model 6, the coefficient on the size of afocal firm interacted with the innovativeness ofits alliance partners is also negative, demonstrat-ing that having large alliance partners was agreater advantage for firms when they were small.Therefore, the identical patterns appear in the ageand the size interactions: the Model 4 and 6findings offer evidence that the value of havingwell-known strategic partners was greatest forsmall firms, just as the Model 3 and 5 findingsshowed that it was greatest for young firms.Calculations similar to those reported above forthe age interactions suggest even larger disparitiesin the advantage of possessing large and inno-vative alliance partners across different levels offocal firm size: the benefit of large alliance part-


Table 5. Fixed effects (OLS) estimates of growth rate of semiconductor firms, 1986–1992

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

Lag of semiconductor sales (lagged) 0.669∗ 0.683∗ 0.664∗ 0.752∗ 0.659∗ 0.692∗

(0.025) (0.025) (0.025) (0.033) (0.026) (0.025)Firm age 0.033∗ 0.030∗ 0.039∗ 0.029∗ 0.037∗ 0.033∗

(0.007) (0.008) (0.008) (0.008) (0.008) (0.008)Number of technology alliances 20.007 20.002 20.002 0.002 20.001(SZijt) (0.006) (0.006) (0.005) (0.007) (0.006)Sales of partners (SVit) 0.0172∗ 0.054∗ 0.063∗

(0.007) (0.012) (0.017)Age-by-sales of partners 20.002∗

(0.001)Lag sales-by-sales of partners 20.012∗

(0.004)Innovativeness of alliance partners 2.186∗ 3.666∗

(SPit) (0.713) (1.199)Age-by-innov. of partners 0.081∗

(0.02)Lagged sales-by-innov. of partners 20.557∗

(0.17)Year is 1987 20.042 20.018 20.022 20.022 20.019 20.021

(0.03) (0.035) (0.035) (0.035) (0.035) (0.035)Year is 1988 0.083∗ 0.106∗ 0.104∗ 0.101∗ 0.107∗ 0.107∗

(0.03) (0.033) (0.033) (0.033) (0.033) (0.033)Year is 1989 20.039 20.037 20.036 20.036 20.037 20.037

(0.03) (0.033) (0.032) (0.033) (0.033) (0.033)Year is 1990 0.029 0.015 0.021 0.021 0.018 0.018

(0.03) (0.033) (0.033) (0.032) (0.033) (0.033)Year is 1991 0.035 0.003 0.002 0.005 20.003 20.001

(0.03) (0.034) (0.034) (0.034) (0.035) (0.035)Constant 1.272∗ 0.966∗ 0.834∗ 0.684∗ 0.926∗ 0.869∗

(0.092) (0.126) (0.128) (0.152) (0.130) (0.132)R-squared (within) 0.6283 0.6367 0.6444 0.6414 0.6407 0.6434Number of firms 150 150 150 150 150 150Number of firm years 825 825 825 825 825 825∗P<0.05

Notes: models 1–4 use the random effects Poisson estimator; Model 5 uses a fixed effects negative binomial estimator

ners was much greater for small firms.Note finally that although young and small

firms may be expected to grow at high rates (infact, the baseline model in Table 5 demonstratesthis to have been the case in the chip industry),the strong support for Hypotheses 3 and 4 cannotbe attributed to the fact that young and smallsemiconductor producers grow at a higher ratethan larger and older producers. Although true,the differences in growth rates attributed to initialsizes and ages are already captured in all of theTable 5 models by the inclusion of the maineffects of age and size. Thus, the results fromthe models that include the partner characteristicsvariables interacted with focal firm attributes


(Models 3–6) convey consistent and convincingsupport for the hypothesized relationships.

Discussion and conclusions

This study has offered additional evidence toconfirm the prevalent assumption that strategicalliances can improve performance. However, inboth the patent rate and the sales growth rateanalyses, the results demonstrated that theimportant determinants of the strength of thealliance-performance link are the attribute profilesof the firms that an organization is affiliatedwith—not the mere fact that it is affiliated. In

808 T. E. Stuart

fact, in the analyses reported in this paper, thecount of the number of alliances formed provedto be an insignificant predictor in the models thatalso included measures of the size or inno-vativeness of a firm’s alliance partners (see alsoHagedoorn and Schakenraad, 1994). In short,technology alliances with large and innovativepartners improved baseline innovation and growthrates, but collaborations with small and techno-logically unsophisticated partners had an imma-terial effect on performance.

The results also at least suggest that alliancesare more than pathways for the exchange ofresources and know-how; they also can be signalsthat convey social status and recognition. TheTable 5 results suggest that alliances with well-known partners may fortify producers’ repu-tations, in addition to providing access toresources such as technological know-how andnew customers. Although I have been unable todirectly measure customer and investor percep-tions of the firms in the sample, the consistentperformance effects of the interactions betweenthe size and age of an organization and theprominence of its partners are in full accordancewith sociological arguments about the effects ofaffiliations on actors’ reputations. Particularlywhen one of the firms in an alliance is a youngor small organization or, more generally, anorganization of ambiguous quality, I believe thatalliances convey endorsements: they build publicconfidence in the value of an organization’s prod-ucts and services and facilitate the firm’s effortsto attract risk averse customers. In this sense,gaining an alliance partner signals a firm’s qual-ity. Not surprisingly, however, the value of analliance as an endorsement is also highly contin-gent upon the regard accorded to the partner firm:because large and innovative organizations arerecognized for their reliability and a track recordof prior accomplishments, the imprimatur implicitin an alliance with a large and innovative firmmay be a particularly valuable signal of theassociate’s quality. In contrast, alliances withsmall and insignificant firms apparently do littleto promote a focal organization’s social standing.Thus, both from a resource access and reputationstandpoint, large and innovative firms are likelyto be the most valuable associates.

Another implication of this study that meritsemphasis, particularly as it relates to the existingliterature on alliances, is that endorsements are


perhaps the easiest to obtain of the potentialbenefits of intercorporate partnerships. Theempirical work investigating the performance ofalliances has concluded that most partnerships failto achieve hoped-for goals (e.g., Harrigan, 1985).We know from a large body of research thatinterorganizational collaboration is fraught withthe potential for opportunistic behavior and isinherently difficult to manage. However, the find-ings of this study suggest that alliances can behighly advantageouseven when they fail toachieve the strategic objectives that led to theirformation. The reason for this is that a focalorganization’s reputation may be upgraded simplybecause it has survived the due diligence of aprominent strategic partner, particularly if thefocal organization is young or small. This advan-tage occurs regardless of whether or not theresource access benefits of an alliance materialize.

In conclusion, I would like to suggest a fewavenues for future research. First, the demon-stration that characteristics of an organization’sstrategic partners affect the benefits that it derivesfrom strategic coalitions has relevance for thelarge and active research on the antecedents ofalliance formation. The contingent value ofalliance partners suggests that it would beinformative to have studies of the alliance forma-tion process that also consider the characteristicsof strategic partners, rather than simply viewingthe formation of an alliance as a binary event(and therefore implicitly treating all partners asbeing of equal value). Thus, research on theorganization and industry-level conditions thatpredict firms’ propensities to enter alliances couldbegin to explore two-stage models, in which theoccurrence of the alliance is modeled and, con-ditioning on the occurrence, the attributes of stra-tegic partners are then explored in a second-stagemodel. For instance, using the data in this paper,one could first estimate the rate at which a firmenters alliances as a function of firms’ attributes,firms’ positions in the alliance network, or time-varying industry conditions, and then, in a secondmodel, explore the size and innovativeness of thefirm’s strategic partners. If it is true that theperformance consequences of alliances are tightlyconnected to the characteristics of a firm’s stra-tegic partners, it is important that the allianceantecedent literature begin to attend to the factorsthat promote the more valuable kinds of collabo-rations.


Second, we know little about the multifacetedrelationships between the characteristics ofalliance partners and the advantages of a cooper-ative strategy. For instance, at least from a learn-ing standpoint, Burt’s (1992) structural holesargument suggests that the addition of a non-redundant strategic partner, because it purveysaccess to new information, is likely to be morevaluable than the acquisition of a new partnerthat is similar in kind to an existing one. Thissuggests that a portfolio of alliances consistingof ties to organizations in a variety of differentmarket niches may be more valuable than anotherwise similar portfolio of alliances with firmsin the same or similar market niches. In additionto research on the returns to particular kind ofalliance network structures, it is also importantto understand how a focal firm’s characteristics,such as its level of absorptive capacity, conditionsthe returns it garners from occupying particularpositions in an industry’s alliance network. Moregenerally, a large number of partner attributes aswell as characteristics of the structural configur-ation of firms’ alliance networks are likely todetermine the magnitude of the advantage of acooperative strategy, both on their own and wheninteracted with focal-firm characteristics.

In a related vein, researchers have recentlybecome attentive to the dyadic conditions thatmust be present for interoganizational learning tooccur in the context of strategic alliances (e.g.,Lane and Lubatkin, 1998; Dyer and Singh, 1998;Stuart, 1998). One interesting question is theextent of dependence of the returns to technology-based alliancing on the absorptive capacity of thefirms in the partnership. This question could beexplored with many existing datasets and aresearch design that investigates the interactioneffects between partner characteristics, propertiesof the collaborating dyad, and proxies for focalfirm absorptive capacity on the firm-level inno-vation rate. For example, it would be informativeto know whether the level of focal firm R&Dspending interacts with the innovativeness of thefirm’s strategic partners in a model of patentingrates. It would be very valuable to have evidencethat informs the necessary conditions for inter-organizational learning to occur and the contin-gencies that bear upon the intercorporate knowl-edge transfer process.

Because technology alliances are accessrelationships, it is also likely that the pre-


partnership quality of the relationship betweentwo firms affects the gain in ex post performance.Much of the work on alliance antecedents thathas grown out of network theory has been keenlyinterested in how prior relationships betweenfirms affects the likelihood that they will collabo-rate in the future (Gulati, 1998, provides a com-prehensive review of this literature). For instance,network theorists have argued that existingalliance ties dictate the selection of collaborators.Because prior alliances convey first-hand infor-mation on the reliability and trustworthiness ofpotential partners, an established relationship witha particular partner reduces the risk and trans-action costs of a future partnership with thatorganization (Granovetter, 1985; Podolny, 1994;Gulati, 1995; Dyer, 1996). These ideas, whichhave been shown to influence the alliance forma-tion process, may also have implications for thedegree to which access is actually achieved inalliance contexts: if an intercorporate relationshipis rooted in a high degree of trust, mutual accessis a much more likely alliance outcome. In otherwords, relationship-level variables such as thedegree of trust between alliance partners areanother set of factors that influence the linkbetween alliances and firm performance.

ACKNOWLEDGEMENTS

I would like to thank the Graduate School ofBusiness at the University of Chicago for finan-cial support for this project.

REFERENCES

Albert MB, Avery D, Narin F, McAllister P. 1991.Direct validation of citation counts as indicators ofindustrially important patents.Research Policy20:251–259.

Aldrich H, Auster ER. 1986. Even dwarfs started small:liabilities of age and size and their strategic impli-cations. Research in Organizational Behavior8:165–198.

Almeida P, Rosenkopf L. 1997. Interfirm knowledgebuilding by semiconductor startups: The role ofalliances and mobility, working paper, The Whar-ton School.

Barnett WP, Carroll GR. 1987. Competition and mu-tualism among early telephone companies.Adminis-trative Science Quarterly32: 400–421.

Barron DN, West E, Hannan MT. 1994. A time togrow and a time to die: growth and mortality of

810 T. E. Stuart

credit unions in New York, 1914–1990.AmericanJournal of Sociology100: 381–422.

Baum JAC, Oliver C. 1991. Institutional linkages andorganizational mortality. Administrative ScienceQuarterly 36: 187–218.

Burgers WP, Hill CW, Kim WC. 1993. A theory ofglobal strategic alliances: The case of the globalauto industry.Strategic Management Journal14(6):419–432.

Burt RS. 1987. Social contagion and innovation:Cohesion versus structural equivalence.AmericanJournal of Sociology92: 1287–1335.

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

Carpenter MP, Narin F, Woolf P. 1981. Citation rates totechnologically important innovations.World PatentInformation 3: 160–163.

Chung S. 1996. Performance effects of cooperativestrategies among investment banking firms: a logli-near analysis of organizational exchange networks.Social Networks18: 121–148.

Cohen WM, Levin RC. 1989. Empirical studies ofinnovation and market structure. InHandbook ofIndustrial Organization, Schmalensee R, Willig RD(eds.). Elsevier: New York; 1060–1107.

Dyer JH. 1996. Effective interfirm collaboration: How firmsminimize transaction costs and maximize transactionvalue.Strategic Management Journal18(7): 535–556.

Dyer JH, Singh H. 1998. The relational view: cooperat-ive strategy and sources of interorganizational com-petitive advantage.Academy of Management Review23: 660–679.

Eisenhardt KM, Schoonhoven CB. 1996. Resource–based view of strategic alliance formation: strategicand social effects in entrepreneurial firms.Organiza-tion Science7(2): 136–150.

Fombrun C, Shanley M. 1990. What’s in a name?Reputation building and corporate strategy.Academyof Management Journal33(2): 233–258.

Granovetter M. 1985. Economic action and social struc-ture: A theory of embeddedness.American Journalof Sociology91: 481–510.

Gulati R. 1995. Social structure and alliance formationpatterns: a longitudinal analysis.AdministrativeScience Quarterly40(4): 619–652.

Gulati R. 1998. Alliances and networks.Strategic Man-agement Journal19(4): 293–317.

Hagedoorn J. 1993. Understanding the rationale ofstrategic technology partnering: interorganizationalmodes of cooperation and sectoral differences.Stra-tegic Management Journal14(5): 371–385.

Hagedoorn J, Schakenraad J. 1994. The effect of stra-tegic technology alliances on company performance.Strategic Management Journal15(4): 291–309.

Hamel G. 1991. Competition for competence and inter-partner learning within international strategicalliances. Strategic Management Journal, SummerSpecial Issue12: 83–103.

Hamel G, Doz YL, Prahalad CK. 1989. Collaboratewith your competitiors-and win.Harvard BusinessReview67(1): 133–139.


Hannan MT, Freeman J. 1984. Structural inertia andorganizational change.American Journal of Soci-ology 49: 149–164.

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

Hausman J, Hall BH, Griliches Z. 1984. Econometricmodels for count data with an application to thepatents-R&D relationship. Econometrica 52(4):909–938.

Heckman JJ, Borjas GJ. 1980. Does unemploymentcause future unemployment: Definitions, questions,and answers from a continuous time model of het-erogeneity and state dependence.Econometrica48:247–283.

Henderson R, Cockburn I. 1994. Measuring com-petence? Exploring firm effects in pharmaceuticalresearch. Strategic Management Journal, WinterSpecial Issue15: 63–84.

Kogut B. 1988. Joint ventures: Theoretical and empiri-cal perspectives.Strategic Management Journal9(4): 319–332.

Kogut B, Walker G, Kim D-J. 1995. Cooperation andentry induction as an extension of technologicalrivalry. Research Policy24: 77–95.

Lane PJ, Lubatkin M. 1998. Relative absorptivecapacity and interorganizational learning.StrategicManagement Journal19(5): 461–477.

Levin RC, Klevorick AK, Nelson RR, Winter SG.1987. Appropriating the returns from industrialresearch and development’Brookings Papers onEconomic Activity3: 783–820.

Lin N, Ensel WM, Vaughn JC. 1981. Social resourcesand strength of ties: structural factors in occupationalstatus attainment.American Sociological Review46:393–405.

McConnell JJ, Nantell TJ. 1985. Corporate combi-nations and common stock returns: the case of jointventures.Journal of FinanceXL (2): 519–536.

Merton RC. 1968 [1973]. The Matthew Effect inscience. InThe Sociology of Science, Storer NW(ed.). University of Chicago Press: Chicago, IL;439–459.

Mitchell W, Singh K. 1992. Incumbents use of pre-entry alliances before expansion into new technicalsubfields of an industry.Journal of EconomicBehavior & Organization18(3): 347–372.

Mitchell W, Singh K. 1996. Survival of businessesusing collaborative relationships to commercializecomplex goods. Strategic Management Journal17(3): 169–196.

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

Nohria N, Garcia-Pont C. 1991. Global strategic link-ages and industry structure.Strategic ManagementJournal, Summer Special Issue12: 105–124.

Pakes A, Griliches Z. 1984. Patents and R&D at thefirm level: a first look. InR&D, Patents, and Pro-ductivity, Griliches Z (ed.). The University ofChicago Press: Chicago, IL; 55–72.

Pakes A, Schankerman M. 1984. The rate of obsol-escence of patents, research gestation lags, and the


private rate of return to research resources. InR&D, Patents, and Productivity, Griliches Z (ed.). TheUniversity of Chicago Press: Chicago, IL; 73–88.

Pfeffer J, Nowak P. 1976. Joint ventures and inter-organizational interdependence. AdministrativeScience Quarterly21: 398–419.

Podolny JM. 1993. A status-based model of marketcompetition. American Journal of Sociology98(4):829–872.

Podolny JM. 1994. Market uncertainty and the socialcharacter of economic exchange.AdministrativeScience Quarterly39: 458–483.

Podolny JM, Stuart TE. 1995. A role-based ecology oftechnological change.American Journal of Sociology100(5): 1224–1260.

Podolny JM, Stuart TE, Hannan MT. 1996. Networks,knowledge, and niches: competition in the world-wide semiconductor industry, 1984–1991.AmericanJournal of Sociology102: 659–689.

Powell WW, Koput KW, Smith-Doerr L. 1996.Interorganizational collaboration and the locus ofinnovation: Networks of learning in biotechnology.Administrative Science Quarterly41(1): 116–145.

Rao H. 1994. The social construction of reputation:certification contests, legitimization, and the survivalof organizations in the American automobile indus-try: 1985–1912. Strategic Management Journal,Winter Special Issue15: 29–44.

Rivette KG. 1993. Corporate patent portfolios.Tech-nology Forecast. Price Waterhouse: New York.

Shan W, Walker G, Kogut B. 1994. Interfirmcooperation and startup innovation in the biotechnol-ogy industry.Strategic Management Journal15(5):387–394.

Singh K, Mitchell W. 1996. Precarious collaboration:business survival after partners shut down or formnew partnerships.Strategic Management Journal,Summer Special Issue17: 99–115.

Spence AM. 1974.Market Signaling. Harvard Univer-sity Press: Cambridge, MA.


Stinchcombe AL. 1965. Social structure and organi-zations. In Handbook of Organizations, March JG(ed.). Rand McNally: Chicago, IL.

Stuart TE. 1998. Network positions and propensitiesto collaborate: an investigation of strategic allianceformation in a high-technology industry.Adminis-trative Science Quarterly43(3): 668–698.

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

Tilton JE. 1971.International Diffusion of Technology:The Case of Semiconductors. The Brookings Insti-tute: Washington, DC.

Trajtenberg M. 1990. A penny for your quotes: patentcitations and the value of information.Rand Journalof Economics21(1): 172–187.

Tuma NB, Hannan MT. 1984.Social Dynamics: Modelsand Methods. Academic Press: Orlando, FL.

Uzzi B. 1996. The sources and consequences of embed-dedness for the economic performance of organi-zations: the network effect.American SociologicalReview61: 674–698.

Wade J. 1995. Dynamics of organizational communitiesand technological bandwagons: an empirical investi-gation of community evolution in the microprocessormarket. Strategic Management Journal, SummerSpecial Issue16: 111–133.

Walker G, Kogut B, Shan W. 1997. Social capital,structural holes and the formation of an industrynetwork. Organization Science8(2): 109–125.

Weigelt K, Camerer C. 1988. Reputation and corporatestrategy: a review of recent theory and applications.Strategic Management Journal9(5): 443–454.

Wilson R. 1985. Reputations in games and markets. InGame Theoretic Models of Bargaining, Roth AE(ed.). Cambridge University Press: New York; 65–84.

Wilson RW, Ashton PK, Egan TP. 1980.Innovation,Competition, and Government Policy in the Semicon-ductor Industry. D.C. Heath: Lexington, MA.

Documents

Interorganizational alliances and the performance of firms ...faculty.haas.berkeley.edu/tstuart/publications/2000... · Strat. Mgmt. J., 21: 791–811 (2000) INTERORGANIZATIONAL ALLIANCES