ArticlePDF Available

Value creation in university-firm research collaborations: A matching approach

December 2012
Strategic Management Journal 34(6)

December 2012
34(6)

DOI:10.2139/ssrn.1351904

Authors:

HEC Paris

University-based technological opportunities are often exploited through joint corporate and academic entrepreneurship activities such as university-industry research collaborations. This paper explores the partner attributes that drive the matching of academic scientists and firms involved in these relationships. The paper models the formation of firm-faculty partnership as an endogenous selection process driven by synergy between partners’ knowledge-creation capabilities. The main findings indicate that faculty-firm matching is multidimensional: firms and scientists complement each other in publishing capabilities but substitute each other in patenting skills. Furthermore, firms and scientists with specialized knowledge create more value by teaming with more knowledge-diversified partners. The paper contributes to the literature on university-industry knowledge transfer, and more generally, to the literature on alliance formation.

Variable Definition

…

Descriptive Statistics and Correlations

…

Figures - uploaded by Denisa Mindruta

Content may be subject to copyright.

Content uploaded by Denisa Mindruta

Content may be subject to copyright.

Electronic copy available at: http://ssrn.com/abstract=1351904

Value Creation in University−Firm Research Collaborations:

A Matching Approach

Denisa Mindruta

Assistant Professor of Strategy

HEC-Paris School of Management

1 rue de la Liberation

Jouy-en-Josas, 78350 France

mindruta@hec.fr

This version: February 2012

ABSTRACT

University-based technological opportunities are often exploited through joint corporate and

academic entrepreneurship activities such as university−industry research collaborations.

This paper explores the partner attributes that drive the matching of academic scientists and

firms involved in these relationships. The paper models the formation of firm−faculty

partnership as an endogenous selection process driven by synergy between partners’

knowledge-creation capabilities. The main findings indicate that faculty−firm matching is

multidimensional: firms and scientists complement each other in publishing capabilities but

substitute each other in patenting skills. Furthermore, firms and scientists with specialized

knowledge create more value by teaming with more knowledge-diversified partners. The

paper contributes to the literature on university−industry knowledge transfer, and more

generally, to the literature on alliance formation.

Keywords: Matching, Complementarity, Endogeneity, Value Creation,

university−industry alliances

Electronic copy available at: http://ssrn.com/abstract=1351904

INTRODUCTION

Universities have always been an important source of new knowledge, and long-standing

evidence shows that academic research has a significant impact on the productivity of private

sector R&D (e.g., Cohen, Nelson, and Walsh, 2002). Spillovers from published academic

research and informal collaboration are influential channels of knowledge transfer, but recent

studies suggest that firms also seek out new scientific knowledge and technological

opportunities by engaging in corporate entrepreneurship activities such as licensing academic

inventions and collaborative research (Thursby and Thursby, 2002). Of particular interest for

the management field is evidence showing that firms with direct research ties to universities

significantly increase their innovative performance (e.g., Cockburn and Henderson, 1998;

Zucker, Darby, and Armstrong, 2002; Belderbos, Carree, and Lokshin, 2004; Fabrizio, 2009).

While the economic benefits of collaboration are well documented in the literature, the

sources of value creation in university−industry research alliances remain under-explored.

Much of the empirical work on alliances performance links an innovation output to an

indicator measuring the presence of these relationships, while a few studies go deeper to

examine the outcome of collaboration in relation to partner characteristics. This work

typically regresses performance on the observed attributes of firms, scientists, or universities.

However, this methodology is valid only when (1) all attributes of the partners that influence

performance are included in the model and measured without errors, or when (2) alliance

formation is a random process. The first condition is hardly ever satisfied in empirical

settings. In strategy research, where great emphasis is placed on strategic, i.e., non-random,

choices (Shaver, 1998), the second condition is particularly problematic.

At the core of this study is the idea that partnership formation is a strategic decision. If

partner choice is guided by preferences about whom to ally with, and motivated by the

expectation that collaboration with certain partners will lead to superior performance, the

Electronic copy available at: http://ssrn.com/abstract=1351904

process of partner selection is neither random nor independent of performance considerations.

Put succinctly, if there are incentives whereby certain types of partners end up allying (i.e.,

they ‘match’ according to certain preferences), then partner attributes must be endogenous

choice variables in regressing alliance performance on the characteristics of the partners. As a

result, empirical models that do not account for the partner selection process will produce

biased estimates and even skewed normative conclusions. These econometric concerns have

been acknowledged in the alliance literature, but matching as a cause of endogeneity remains

poorly understood.

To illustrate the importance of matching and its implications, consider Zucker et al.’s

(1998, 2002) finding that partnerships with ‘star’ university scientists have a higher impact on

firms’ innovation performance than partnerships with non-star faculty. By concluding that the

differential productivity of firm−star alliances is due to the added value professors contribute

to the innovation process, these studies suggest that firms aspiring to increase performance

should pursue research ties with top faculty. This conclusion raises several questions. Firms

may prefer teaming-up with ‘stars’, but are such collaborations equally attractive to

scientists? Could scientists be equally productive no matter what firm or firms they

collaborate with? Further, if success is a result of partnering with ‘stars’, why would firms

partner with ‘non-star’ faculty?

The existing literature has also accumulated compelling evidence that scientists value

firms based on multiple factors, such as the strength of internal research, technological

expertise, and organizational culture. As this paper will show, if scientists care about the

quality of the firm, and firms care about scientists’ research excellence, the interaction of

these preferences will lead to a sorting of partners across alliances, causing ‘better’ scientists

to work with ‘better’ firms. The endogeneity arises when top faculty team with firms whose

qualities reinforce their expertise and effort in innovation (i.e., scientists and firms match on

complementary attributes) but the regression model does not include all relevant firm−quality

dimensions. The error term then captures the effect of the omitted firm attributes and

becomes correlated with faculty research status, overestimating star scientists’ contribution to

innovation. Conversely, when partners match on attributes that act as substitutes in

innovation, the omitted attributes of one party will bias downward the estimates of the

observed characteristics of the other party.1

The above example illustrates a large class of situations in which partner choice affects

performance estimates. Researchers typically attempt to control for the confounding effects

of partner selection through instrumental variables or experimental design. In contrast, this

paper argues that faculty−firm partnering reveals important information about the sources of

value creation driving cross-organizational alliances and, ultimately, their differential success

in innovation. It introduces a theoretical model and a methodology that examine alliance

formation as an endogenous matching process in which firms and scientists strategically

select each other based on attributes that are important for knowledge-creation. The model

predicts positive (i.e., top-down) sorting of scientists and firms on attributes that are

complements in the innovation process, and negative sorting on attributes that are substitutes.

I apply this approach to a proprietary database that contains the complete set of 447

discovery-type collaborations between professors at a top U.S. medical school and their

industry partners during 1995-2004. Employing recent developments in econometrics, I

estimate the model using a maximum score estimator (Fox, 2010). Parameters are identified

based on mathematical inequalities that compare the output created by observed and

1 In general, the bias depends on the relationship between the included and omitted variables and the impact of

the omitted variables on performance. In the example discussed here, the omitted variables measuring firms of

higher quality have, most likely, a positive effect on innovation. Thus, the direction of the bias rests on the

relationship between the omitted firm variables and scientists’ status. In the case of complementarity between

firms and top scientists, the omitted attributes of the firms will be positively related with scientists’ status,

leading to an upward bias impact of ‘stars’ on performance. In the case of substitution, attributes of the partners

will be negatively related, and the impact of ‘stars’ on performance will be biased downward.

counterfactual matches. These inequalities are derived from the equilibrium outcome of the

matching model, which predicts that when partner choice occurs under rivalry constraints, the

observed matches should create (weakly) higher value than counterfactual partnerships. The

results show that firm−faculty research alliances create more value when partners

complement each other in publishing capabilities and substitute each other for lack of

patenting capabilities. There is also evidence of complementarity between knowledge-

diversified and knowledge-specialized partners.

This paper extends the prior literature in several ways. The theoretical approach, rooted in

the management theories that recognize heterogeneity in capabilities as a central feature of

firms (e.g., Helfat et al., 2007), builds upon the conceptualization of two-sided matching

markets to include strategic aspects of partner choice (Mortensen, 1988). Unlike previous

work that focused on firms or universities, this paper considers alliance formation from the

perspective of both partners. Furthermore, the analysis of partner choice goes beyond

considerations of mutual attractiveness at the dyad-level and takes into account the

competition between participants on the same side of the market to ally with the most

desirable exchange partners on the other side of the market.

This approach generalizes previous findings and explains empirical regularities that other

models fail to clarify. While prior literature has emphasized the value-added contribution of

star faculty to the innovativeness of partner firms, this paper draws attention to partner

sorting in the market for research. The multidimensional sorting model captures the

differences between participants’ preferences and ability to attract partners more than other

simpler, univariate models. Within this perspective, firm collaboration with non-star faculty

is not a simple story of top-down sorting but a more nuanced reflection of heterogeneity in

preferences and synergy types (complementarity vs. substitution) in innovation.2

THEORETICAL BACKGROUND

This paper examines how new knowledge is created when actors deliberately form new ties

across organizations or sectors. Thus, although linkages between universities and industry

can take multiple forms, here I focus specifically on formal research collaboration between

university scientists and firms intended for scientific discovery.

Main Concepts

To key concepts are complementarity and substitutability. The mathematical definitions of

these terms stipulate that two inputs are complements in production if the cross-partial

derivative of the production function with respect to the two inputs is positive, and substitutes

if the cross-partial derivative is negative.3

For this study, the research alliance ‘production function’ can be framed as a knowledge

output whose inputs are partner attributes. Indeed, the objective of a research alliance is to

create value (in the form of new knowledge or cross-learning among partners) beyond what

scientists and firms would yield separately. The joint output depends on the combination of

partner attributes, which may include strengths building upon strengths or the strengths of

one partner substituting for the other partner’s shortcomings. Conceptualizing the production

function in this way implies that two attributes are complements in knowledge creation if

having a higher level of one raises the return from having a higher level of the other.

2 For example, the finding of negative sorting on patenting skills predicts, ceteris paribus, that partnerships

between firms with higher patenting activity and ‘non-star’ patenting scientists (i.e. scientists with fewer or no

patents) are better matches than alliances between partners that both patent a lot. Thus, these partnerships are not

the result of firms not having access to ‘star patenting’ scientists, but they reflect preferences for partner choice

on dimensions that are substitutes in innovation.

3 These terms have been used with many other meanings in the strategy literature (e.g., complementarity has

been equated with compatibility, non- similarity, or non-overlap), but this paper employs the formal economic

definition of these concepts. The formal definition has been applied to the study of organizational decisions

related to firm boundaries (Parmigiani and Mitchell, 2009), the relationship among various organizational

practices (Milgrom and Roberts, 1995), and among firm internal resources and R&D alliances (Arora and

Gambardella, 1994; Cassiman and Veugelers, 2006).

Conversely, two attributes are substitutes if having a higher level of one decreases the

marginal value of having a higher level of the other.

The term market for research refers collectively to all transactions in which professors

and firms agree upon joint research activities leading to the creation of new knowledge and

technologies, without changing organizational boundaries or asset ownership. This study also

advances the idea that the market for research is a two-sided matching market. Two-sidedness

indicates that agents involved in a transaction belong to one of two disjoint sets (here, firms

and universities). By contrast, in a typical commodity market, an agent might be a buyer at

some price and a seller at another price. Matching reflects the bilateral nature of exchange.

For example, in labor markets, an employee works for a firm, and the firm employs the

worker. As Roth and Sotomayor (1990:1) explain, this is in contrast with the markets for

goods, ‘in which someone may come to market with a truck full of wheat, and return home

with a new tractor, even though the buyer of wheat doesn’t sell tractors, and the seller of

tractors didn’t buy any wheat.’ The concept of a two-sided matching market captures the

theoretical underpinnings of the markets involving exchanges of heterogeneous and

indivisible goods (Mortensen, 1988). In this regard, university−industry partnerships are a

typical example. The inputs each partner brings to the relationship are highly differentiated:

the expertise of university scientists is not a commodity, and neither are the capabilities of

individual firms. Moreover, in this particular context, an exchange involves an indivisible

bundle of distinctive characteristics that parties on one side of the market offer to parties on

the other side. These two aspects−heterogeneity and indivisibility−lead to an important

theoretical distinction. While the identity of agents who trade in commodity markets does not

matter, it is important to understand whether agents in matching markets trade with the right

partners. The critical question in these markets is who trades with whom.

A matching model of alliance formation

The following three features of university−industry alliances define the matching process:

(1) University-industry research alliances are voluntary relationships that form with the

expectation of mutual gains. One of the professors interviewed for this paper described his

decision to collaborate with industry in the following way:

[…] because I only have so much time, I’m not going to spend my time on something

that I don’t think is going to be productive. And I certainly will not have students and

people in my group spend time on something that’s not productive. So, I think that’s

where it comes down to. There is the mutual interest and then there have to be mutual

benefits. And, if those aren’t there, then there is no reason to do it.

Likewise, as Stephan notes: ‘firms can ill afford to fund research that has little promise of

eventually relating to the company’s objectives’ (1996: 1211). Most commonly, research

projects involving university partners go through a competitive selection process along with

internal projects. Thus, it is reasonable to believe that only projects with justifiable benefits to

the firm will be approved.

(2) The value of innovation generated through collaboration is determined, at least in

expectation, by the identity of the university scientists and firms involved in a relationship.

Consequently, both professors and firms have preferences about with whom they collaborate.

In discovery-type collaborations, preferences are driven by the belief that joint work with

partners of a certain type (such as those with high research productivity and expertise in a

particular scientific field) will have a higher probability of generating new knowledge.

(3) Alliance partners are restricted in the number of collaborations they can undertake at

one time. On the academic side, time constraints are one of the main reasons for a careful

screening of the number and quality of industry partners. Moreover, research contracts often

impose confidentiality conditions that prevent scientists from collaborating with multiple

companies simultaneously (Kenney, 1986). Firms have a better ability to manage multiple

contracts with external parties, up to the point where over-embeddedness becomes a liability

(Uzzi, 1996). The management of alliances is a complex process facing numerous challenges,

such as managing coordination among partners and creating an appropriate organizational

structure for inter-partner learning and knowledge transfer. These constraints limit the

number of simultaneous alliances firms can pursue.

Taken together, preferences for potential partners and restrictions on the number of

collaborations suggest that, in the market for research, agents on each side of the market will

compete to ally with the most desirable partner on the other side of the market. While prior

literature has examined inter-firm competition in innovation races, and scientists’ competition

for funding and academic prestige (Stern, 1995), firms’ (respectively, scientists’) competition

for teaming up with the ‘best’ alliance partner has been largely ignored. This type of rivalry is

extremely important in matching markets, where ‘transactions’ involve differentiated bundles

of indivisible resources in limited supply.

From a theory standpoint, firm−scientist alliance formation belongs to a more general

class of models (known as assignment games) that address the formation of voluntary

partnerships among complementary agents under conditions of rivalry (Mortensen, 1988).4

Economists have typically studied matching in marriage and labor markets. More recently,

scholars have begun to apply these models to the study of various business relationships,

including those between firms and IPO underwriters (Fernando, Gatchev, and Spindt, 2005),

start-ups and VC investors (Sorensen, 2007), buyer-supplier contracting (Fox, 2009; Chatain,

2010), or the assignment of resources to firms (Lippman and Rumelt, 2003).

4 An outcome in assignment games consists of an assignment (i.e., the identity of the matching parties) and a

vector of payoffs (i.e., a division of surplus between the members of a matched pair). The equilibrium concept is

that of pairwise stability, a notion that captures the idea that none of the partners to a match have an incentive to

separate and team with other partners. Optimality means that un-matched agents could not form a partnership

and make one better off without making the other worse off. Assignment games study questions such as which

partnerships will form given the agents’ preferences and the rules governing the market. These models are

appropriate for studying the ex ante process of partnership formation. In contrast, prisoners’ dilemma and

assurance games are appropriate for studying the ex post dynamics of alliances, where partners decide how

much to contribute to joint activity (Agarwal, Croson, and Mahoney, 2010).

Matching theory explains what happens when all features of alliance formation−

voluntary collaboration, two-sided decision-making, and competition for better partners−are

considered in interaction. From this perspective, partnership formation is a two-sided process.

Each agent on one side of the market evaluates, and is simultaneously evaluated by, potential

partners on the other side of the market. At the dyad level, and in isolation from the rest of

the market, the decision rule is simple: a partnership occurs when both partners expect that

collaborating will create more value than remaining single. However, under rivalry for

partners and preferences on both sides, synergy at the dyad level is no longer a sufficient

condition for an alliance if any of the partners could create a higher surplus with another

agent on the other side of the market. At the market level, dyad-level decisions interact to

constrain each other. On the one hand, the likelihood that an agent (firm or scientist) teams

with his or her preferred partner is influenced by the existence of other agents on the same

side of the market wishing to ally with the same partner. On the other hand, the likelihood of

teaming with the preferred partner is also influenced by the alternative opportunities the

partner might have to forego by entering a particular deal. In short, the decision of two

individual agents to collaborate depends not only upon their preferences, but also on their

effective choice set, which is constrained by the decisions made by all other agents in the

market. Ultimately, it is the interaction of preferences and competition at the market level that

leads to the final matching.

The core theoretical properties of matching, in which the attributes of the partners are

complements or substitutes in the match production function, were introduced to the literature

by Becker (1973).5 Becker’s results lead to the following prediction: Ceteris paribus, if two

inputs or attributes (capabilities of a scientist and a firm, respectively) are complements in

5 Becker considers the unidimensional case (one attribute on each side) and shows that if two inputs/attributes

are complements (substitutes) in production, then positive (negative) assortative matching is stable and optimal.

knowledge-production, in equilibrium, firms and scientists scoring highest on these

dimensions will work together, leaving firms and scientists scoring second to select each

other, and so forth until the least-endowed agents select each other. Conversely, if two

attributes are substitutes, then firms (scientists) scoring high on one attribute will match with

scientists (firms) scoring low on the other attribute.6

The set of theorems proven by Becker (1973) constitute the basis for the empirical

matching model. Drawing from innovation literature, I examine the matching of scientists

and firms along three key dimensions in collaborative research: publishing capability,

patenting capability, and the degree of knowledge specialization (focused versus broad).

Hypotheses

As most fields of modern science, biomedical research considered in this study is

simultaneously motivated by scientific interest and considerations of usefulness and

applications (Stephan, 2010). In this context (typically referred to as Pasteur’s Quadrant)

discoveries have dual scientific and applied/commercial value, providing opportunities for

inventors to disclose their ideas through scientific publications, patented inventions, or both

(Murray and Stern, 2007). As I discuss below, the choice of disclosure regime sends differing

signals about inventors’ research abilities and motivations (Hicks, 1995; Anton and Yao,

2003), and plays an important role in facilitating knowledge exchanges among academics and

firms (Hellmann, 2007; Gans and Stern, 2010).

The norms of ‘Open Science’ reward priority in advancing novel ideas and provide

incentives for full disclosure of scientific findings in the public domain. The scientific merit

of a discovery is decided by the research community through the peer-review process and

citations. Publishing is the hallmark of academia, but articles published by corporations have

6 Some partnerships might look ‘suboptimal’ from the perspective of the dyad, such as, for example, alliances

between two less-endowed agents (as in the case of positive assortative matching), or alliances between one

highly endowed agent and one less-endowed agent (as in the case of negative assortative matching). However,

these partnerships are ‘optimal’ when all linkages in the market are taken into account.

increased significantly in the last decades (Stephan, 2010). A salient motivation for corporate

publishing is firms’ desire to participate in knowledge exchanges within the broader scientific

community (Hicks, 1995; Henderson and Cockburn, 1998). These publications serve two key

roles. First, they indicate a firm’s willingness to abide by the norms of reciprocation and full

information disclosure (Murray, 2011). Second, they signal a firm’s research quality

(including the existence of tacit knowledge and unpublishable resources) and thus, provide

the necessary ‘credentials’ for the firm to find exchange partners in the scientific community

(Hicks, 1995).

The norms of ‘Commercial Science’ reward priority in producing ideas that are novel,

useful and non-obvious. The choice of patent disclosure regime indicates that the inventor

seeks to secure monopoly rights on ideas and will follow the logic of commercial practices in

exchanges involving these ideas (Murray, 2011). Like publishing, patenting has a strong

signaling value (Anton and Yao, 2003) and is practiced both in the industry and academia.

Patents signal to other market participants the strength of a firm’s proprietary knowledge and

technical expertise in a field (Ahuja, 2000). In academia, patents indicate that a scientist’s

ideas have useful marketable applications, and are often perceived as a ‘symbol of

commercial savvy’ (Murray, 2011).

Building on this literature, I refer to publishing and patenting capabilities to denote the

ways firms and university scientists deploy their knowledge capital to signal expertise and

preferences for the ‘open’, respectively, ‘commercial science’ regime.7 8 By patenting or

7 Patenting and publishing take different roles in different markets (labor, capital, product, or ideas market) and

the variation in patent and publication records of firms and faculty reflects more than differences in capabilities.

For example, firms might patent aggressively when they engage in ‘patent blocking’ and patent races, or seek to

improve their bargaining power (e.g. Hall and Ziedonis, 2001).In academia, patenting might be driven by

strategic and reputational calculations, as well as social influences (Murray, 2011). I focus on knowledge-related

interpretations because participants in the market for research use observable patent and publication outputs to

make inferences about potential partners’ expertise.

8 To my knowledge, Murray and co-authors (e.g. Gans, Murray and Stern (2008) and Murray (2011)) were the

first to systematically use the terms ‘open’ vs. ‘commercial science’ to emphasize that the main difference

publishing in a narrower or a broader territory of the knowledge space, firms and scientists

also convey the breadth of their knowledge base. Specialization implies that an innovator’s

knowledge is concentrated in a narrow territory of the knowledge space, and diversity or

‘breadth’ of knowledge indicates familiarity with a broad range of scientific or technological

components.

Patenting and publishing capabilities. A number of findings in the innovation literature are

consistent with the idea that corporate publishing is driven by firms’ desire to participate in

networks of scientific knowledge. In particular, firms that provide ‘pro-publication’

incentives and promote scientists based on their scientific standing are more likely to

collaborate with universities and to rely on academic research in their innovation activities

(Cockburn and Henderson, 1998). In contrast with prior studies that looked exclusively at the

relationship between corporate publishing and firms’ propensity and degree of collaboration

with universities, I focus here on the quality of firms and faculty involved in these

relationships. Specifically, if publications signal scientific merit, I propose to examine the

extent to which firm−faculty alliances are driven by complementarity between partners’

publishing capabilities, such that firms and faculty select each other assortatively based on

their ability to generate high-quality scientific work. Survey studies reinforce the idea that

firms aim to partner with professors with outstanding scientific record (Blumenthal et al.,

1986). Particularly in basic research collaboration, firms perceive professors’ research quality

as being more important than their geographic proximity or other factors (Mansfield and Lee,

1996; Audretsch and Stephan, 1996). Preliminary interviews for this study also indicated that

faculty value firms with superior records of publication. Thus, I propose to test:

Hypothesis 1: Ceteris paribus, firms’ and university scientists’ publication

capabilities are complements in value creation.

between publishing and patenting lies neither in the type of knowledge disclosed (basic vs. applied), nor in the

locus of production (academia vs. industry), but rather in the institutional norms of the two disclosure regimes.

A long line of research in the innovation literature views academic science and

technological innovation as complementary and co-evolving. Numerous examples of key

discoveries corroborate the idea that technological development happens faster in areas with

strong science, while scientific progress emerges as a response to the challenges posed by

new technologies (Nelson, 2004). Furthermore, large-scale empirical studies done at high-

levels of aggregation have found a complementary relationship between public and industrial

R&D (David, Hall, and Toole, 2004). Scholars have also argued that due to the research cost

structure and the nature of incentives in universities and firms, it is more appropriate to have

the earlier stages of a research process done by universities, and the later, more applied and

more commercially viable innovation stages, in the industry (Aghion, Dewatripont, and Stein,

2008; Lacetera, 2009). These arguments suggest that alliances between faculty and firms are

driven by partners’ motivation to build on each other’s strengths in basic and applied

research. Thus, professors with a better record of scientific publications should be more

productive when teaming with firms with a better record of patenting, and vice-versa. Very

few studies exist at the alliance level, but those that do tend to support this perspective.

Orsenigo (1989) argues that, in the early years of the biotechnology industry, prominent

scientists preferred to collaborate with firms with higher number of patents. Studying a

broader range of industries in the mid 1990s, Mansfield and Lee (1996) found that

university−industry collaboration was driven by the synergy between professors’ research

and firms’ efforts to create new products and processes. I therefore propose the following

hypothesis:

Hypothesis 2: Ceteris paribus, firms’ patenting capabilities and university

scientists’ publication capabilities are complements in value creation.

Prior literature tells us little about whether the relationship between the patenting

capabilities of faculty and firms are complements or substitutes in innovation. Recent work

has identified a variety of individual and contextual factors that prompt professors to disclose

inventions and become further involved in their commercial development (Bercovitz and

Feldman, 2008; Stuart and Ding, 2006). However, although university professors might

occasionally disclose inventions in an attempt to seize immediate opportunities (Azoulay,

Ding, and Stuart, 2007), they need to build a track record of disclosures to credibly signal

expertise in converting scientific ideas to useful, marketable applications. For example,

Elfenbein (2007) showed that professors with a higher patenting reputation were more likely

to have their inventions licensed (although not necessarily for higher proceeds) than first-time

patentees.

A question of interest here is which firms are better positioned to harness professors’

patenting expertise? Two opposite scenarios are conceivable. On the one hand, firms with

superior patenting capabilities could contribute with technological skills and resources,

accelerating scientists’ inventions to commercialization. If scientists interested in the

commercial prospects of their ideas find collaboration with these firms valuable, we should

observe the following relationship:

Hypothesis 3a: Ceteris paribus, patenting capabilities of firms and scientists are

complements in value creation.

On the other hand, firms with a better in-house stock of patents also face higher

opportunity costs from working on inventions generated externally, particularly when those

inventions are in early stages, as is typical with inventions originating from universities

(Jensen and Thursby, 2001). On the contrary, firms with relatively lower technical capital

have lower opportunity costs from working on external ideas, and are more willing to take the

risks and exert the effort necessary to develop early-stage inventions (Lowe, 2001). I argue

that these aspects might be important for university scientists interested in the commercial

prospects of their ideas. The more technologically astute the scientist is, the more likely it is

that the research done with external partners is close to the scientist’s core agenda, and the

more likely it is that the project’s success depends more on the scientist’s knowledge than on

his partner’s patenting skills. Under this scenario, we should observe collaboration between

professors with patenting expertise and firms with a shorter history of patenting.

A substitution relationship between partners’ patenting capabilities is only credible if the

parallel story is credible, too: that scientists with lower or no patenting skills will team-up

with firms with higher patenting capabilities. Indeed, for a professor with little experience in

capitalizing on the technological opportunities of scientific research, partnering with a firm

with higher expertise in patenting makes it possible to learn how early-stage ideas develop

into marketable innovations. Conversely, firms with an established patenting record are well

positioned to utilize scientific information towards more applied goals (Arora and

Gambardella, 1994). By deciding to do joint research with universities, these firms

effectively ‘buy an option’ in early stage research ideas that might later be commercializable.

For these firms with high patenting capabilities, it is less important that a prospective faculty

partner has patenting expertise. Thus,

Hypothesis 3b: Ceteris paribus, firms’ and university scientists’ patenting

capabilities are substitutes in value creation.

Knowledge breadth and specialization. Knowledge attributes−breadth and

specialization−represent another set of potential sources of complementarities in research.

From the perspective of a party endowed with a diversified knowledge base, there is value in

partnering with a specialized external collaborator who can fill particular knowledge gaps.

Equally, a specialized agent will seek to ally with a more diversified external partner who

will envision new applications of the agent’s knowledge. While it is difficult to construct

irrefutable arguments against alternative combinations (between two broad or two specialized

partners, for example), numerous arguments suggest such combinations are less likely to

systematically drive the formation of firm−scientist alliances. The rationale behind this

assertion stems from the mechanisms through which breadth and specialization contribute to

the discovery process.

Consider firms with a diversified knowledge base. Their innovation performance stems

from their ability to capitalize on economies of scope from internal know-how and external

knowledge spillovers (Henderson and Cockburn, 1996). However, capabilities and resource

allocation priorities that enhance the development of breadth tend to conflict with those

required to develop depth (Wang and Tunzelmann, 2000). Diversification enables firms to

absorb and recombine outside know-how. Thus, firms keep up with the latest advances in a

field through alliances with specialized suppliers rather than through internal development

(Brusoni, Prencipe, and Pavitt, 2001).

Likewise, for a specialized university scientist (i.e., an ‘expert’ in a field), teaming with a

firm with a broader knowledge base constitutes a productive opportunity. Because diversified

firms already have mechanisms to internalize focused scientific information, professors’

interactions with diversified firms will require less effort. More importantly, formal work on

downsides of specialization has shown that reliance on multi-disciplinary teamwork helps

innovators with narrower expertise overcome the limitations of specialization (Jones, 2009).

Thus, exposure to diverse knowledge through teamwork or alliances tends to benefit

specialized inventors. However, assembling teams and capitalizing on diverse expertise can

be compromised by communication and coordination costs (Dahlin, Weingart, and Hinds,

2005). Faced with the trade-off between forming a partnership with multiple specialized

partners and a firm with a diverse array of knowledge, narrowly specialized scientists more

likely prefer the latter. These arguments imply cross-fertilization in innovation between

specialized scientists and firms with a diverse knowledge base. Hence:

Hypothesis 4a: Ceteris paribus, a complementarity relationship exists between firms

with a diversified knowledge base and university scientists with

focused expertise.

The complementarity between specialized firms and faculty with broad expertise becomes

evident when we examine these agents’ motivations for entering a research alliance. To

reduce the risk of technological exhaustion, specialized firms must pursue various boundary-

spanning activities (Rosenkopf and Almeida, 2003). Prior work has shown that building

alliances and relying on university research are methods of exploration (Bercovitz and

Feldman, 2007). Here, I propose that specialized firms have higher chances of successful

exploration when they partner with scientists whose expertise is broad rather than narrowly

specialized. This argument builds on prior research showing that scientific knowledge

increases the rate of invention by: (a) helping innovators identify useful directions of search,

(b) preventing them from wasting time with dead-ends, and (c) encouraging them in the face

of failure (Fleming and Sorenson, 2004). The returns from working with broader partners will

be higher when at least the first two mechanisms are considered. A partner with broader

expertise provides immediate access to multiple potentially relevant pieces of information.

This enables firms to become aware of possible linkages among various knowledge domains

and helps them more rapidly identify ways to leverage their knowledge. In addition, broader

scientists can deploy a larger set of information-processing filters, which in turn reduce the

uncertainty around possible applications of firms’ knowledge and eliminate inefficient

experimentation (Nelson, 2004).

I also conjecture that broader scientists are attracted to collaborations with specialist firms

who could enrich the scientist’s understanding of a particular scientific area. Working with

diversified firms is less likely to bring higher benefits to these scientists. Diversified firms

and scientists excel at recombination and new knowledge integration, and, if one alliance

partner possesses such skills, it is less necessary for the other partner to contribute the same

skill set. As noted before, potential benefits from increasing expertise diversity are subject to

diminishing returns. Thus, I propose:

Hypothesis 4b: Ceteris paribus, a complementarity relationship exists between

specialized firms and university scientists with broader expertise.

DATA AND METHOD

Because research contracts between firms and scientists are typically considered confidential

by both parties, prior researchers have used co-authorship as a proxy for collaboration.

However, many joint contracts do not result in co-authorship; co-authorship may also be a

result of informal ties between scientists. I overcame these problems by using first-hand

information from the Grants and Contracts Office of a top-ranked, private U.S. medical

school located on the East Coast. The sample contains the full set of faculty−industry

research contracts from 1995 (the earliest date for which this information was available) until

2004. The office released the following information: the name of the principal investigator,

the name of the firm, the project title, the start and end date of the contract, and the award

(contract) money. I supplemented this primary information with numerous secondary sources

detailed in the ‘variables’ section.

This empirical context is suitable for studying research collaboration between faculty and

firms for multiple reasons. Medical schools have a long history of collaboration with

industry, and research done in medical domains has been shown to have a strong impact on

industrial R&D (Cohen et al., 2002). Professors in medical schools patent more and are more

involved in technology transfer than professors in other fields equally important for the

private sector. The medical school studied here began collecting systematic data on research

contracts much earlier than other universities, allowing me to study a larger sample of

collaborations over a longer time span. Furthermore, all contracts in the sample reflect a clear

intent to generate innovation (i.e., they are referred to as discovery-type collaborations).

The sample consists of 447 contracts between 238 firms and 217 professors. The number

of contracts per year varied from 32 to 54. Almost 25 percent of the scientists and firms

collaborated with more than one partner during the observed decade, but there were no

instances of multiple scientists (or firms) working simultaneously with a firm (scientist) on

the same project. This data aspect implies a ‘one-to-one’ matching of firms and scientists.9

Variables

I collected firms’ stock of publications during a ten-year window (up to the collaboration

date) and scientists’ life-time publications from the ISI Science Citation Index. I chose a ten-

year window to allow a sufficiently long time for firms’ publication output to translate into

reputational effects. I weighted all publications by the number of citations they received in

the Web of Science database (as of April 2010). As is typical in innovation studies, I applied

a yearly depreciation rate of 10 percent.

I collected firms’ and scientists’ patents over a six-year window prior to the contract time.

I controlled for the variation in patent quality by weighting patents by citations received as of

June 2010, with a 10 percent annual depreciation rate.10

I measured the breadth of a firm’s knowledge base using an entropy index of

diversification based on the distribution of a firm’s patents in technology domains defined by

the inventive IPCR patent classes. The formula for this index is: ))/1ln(*(



ii ff , where N

represents the total number of technological domains (here, inventive IPCR) in which a firm

has been patenting and fi represents the fraction of patents in ith category. The index takes

higher values as firm knowledge becomes more diversified.

9Both the theory and the estimation method can be used to study ‘one-to-many’ and ‘many-to-many’ matching.

The assumption here is that projects are independent. Thus, treating alliance formation as ‘one-to-one matching’

is a better model for the current data and not the result of theoretical or methodological constraints. In contrast, a

‘one-to-many matching’ would imply, for example, that one firm collaborates with multiple scientists on the

same project. In this case, the choice of a particular professor is dependent on the choice of other faculty, and

the production function needs to be modified to include this aspect. There are no instances of such

collaborations in my sample. For a typical example of ‘many-to-many matching’, see Fox (2008).

10 Corporate publications and patents were aggregated at parent level after carefully considering the corporate

structure of the firm during the relevant time window. I took into account mergers and acquisitions, divestitures,

and name changes. I also performed several sensitivity analyses, and the final results are very similar even if I

use an alternative time window for patents (e.g., five or four years) or publications (e.g., nine or eight years).

The knowledge-diversification index of the academic scientists is similarly defined,

but technological domains were replaced with scientific domains. I identified the scientific

domains from the MEDLINE database of the National Library of Medicine (NLM). NLM

specialists classify each article according to a scheme known as the Medical Subject

Headings (MESH) thesaurus. The classification of articles based on the MESH thesaurus

insures consistency in classifying scientists’ expertise. The MESH thesaurus has a

hierarchical structure composed of descriptors (main headings), qualifiers (subheadings), and

Supplementary Concept Records. Descriptors indicate the main themes discussed in the

article, and qualifiers indicate a specific aspect of the descriptor. I treated a combination of

MESH descriptors and MESH qualifiers as constituting a scientific domain.11

Controls I included several control variables that might influence partner choice and the

matching of firms and scientists. Firm size (measured by the number of employees), as an

indicator of resources, and firm age (years since founding), as an indicator of experience, are

likely to play a role in alliance formation. I also included the size of the lab (measured by the

number of people associated with the principal investigator’s lab) and academic age (the

number of years since the scientist obtained her highest degree). I controlled for scientists’

academic titles by adding dummy variables for associate professor, full professor, and

chairman/chief (the excluded category was assistant professor). In a robustness checks

analysis, I also included a dummy variable for biotechnology firms (the excluded categories

were pharmaceutical firms and a small number of medical device firms). Table 1 describes

variables and their sources. Table 2 explains the relationship between variables and

hypotheses.

11 For example, ‘Coronary vessels’ and ‘Cardiac surgical procedures’ are descriptors, while ‘Injuries’ and

‘Adverse effects’ are subheadings. Accordingly, the combinations ‘Coronary vessels/injuries’ and ‘Cardiac

Surgical Procedures / adverse effects’ represent two different subject areas. I eliminated all MESH categories

that do not reflect substantive scientific information (e.g. the type of research support, the publication format,

the geographic location, etc.) The complete list of descriptors can be accessed from the NLM.

[Insert Table 1 here]

[Insert Table 2 here]

Model specification and estimation

The challenge in estimating a matching model arises from the fact that partner choice

depends on all other relationships formed in the market. Thus, there are as many endogenous

variables as the observed matches. When alliances are treated as independent decisions, as in

standard discrete-choice models, the choice probability that an agent chooses a particular

partner can be factored into a product of low-dimensional integrals. In a matching situation,

the choice probability is a multidimensional integral over the joint density of error terms. The

dimension of this integral is equal to all possible arrangements of pairings in the market. For

example, in one-to-one matching involving 100 agents on each side, the dimension is 100! =

9.33 * 1015. Integrating over such a high-dimensional integral is currently not feasible.

To overcome these problems, I estimated the model using a semi-parametric estimator

developed by Fox (2010), which generalizes Becker (1973)’s unidimensional analysis to the

multidimensional case. The estimator is semi-parametric in the sense that it requires the

specification of a production function but does not impose a distribution on the error terms.

The method relies on a property known as the ‘local production maximization’, which Fox

(2010) proved is satisfied whenever the observed matches are the equilibrium outcome of an

assignment game. The local production maximization condition states that the output created

by any two observed matches is greater than the output created by counterfactual matches

formed from an exchange of partners. 12

12Although the existence and the uniqueness of the equilibrium in one-to-one matching games are well known, a

general analytical solution for characterizing the matching of agents with multidimensional attributes does not

exist in the literature. However, because the local production maximization is a necessary condition, the

estimator does not require computing an equilibrium solution to a matching game (Fox, 2010).

To illustrate how the method works, I denote by (F, S) a firm−scientist pair in the sample,

and by g (F, S |β) the output (or surplus) created by the pair working together. The

production function that describes the match output takes the following form:

g (F, S |β)= β [XF *YS] + εSF (1)

where XF represents the vector of firms’ characteristics, YS the vector of scientists’

characteristics, εSF is the error term, and β is the vector of parameters to be estimated.

I consider that each of the 10 years in the sample represents a separate matching market.

For any two observed collaborations in a market, (Fi, Si) and (Fj, Sj), the corresponding

match output is g (Fi, Si |β) and, respectively, g (Fj, Sj |β). The counterfactual pairings are

obtained by switching partners such that Fi pairs with Sj to generate g (Fi, Sj |β) and Fj pairs

with Si to generate g (Fj, Si |β). The local production maximization implies that the following

inequality is true:

g (Fi, Si |β) +g (Fj, Sj |β) > g (Fi, Sj |β) +g (Fj, Si |β) (2)

The estimation procedure requires checking the local production maximization inequality

(1) for all 9,996 possible combinations of firm-scientist in the sample and the corresponding

counterfactuals. The parameters β are estimated by choosing the values that predict the

highest number of inequalities. Formally, the estimator maximizes the following objective

function:

Q(β) = * ,

where 1[ . ] denotes an indicator function which takes a value of 1 if the expression between

parentheses is true and 0 otherwise, Nh is the number of pairs in each individual market, and

h is a market index that here takes values from 1 to 10. 13

13 Note that the production function g (F, S |β) and the objective function Q(β) contain only interaction terms.

Production levels are also affected by the ‘main’ effects of XF and YS. However, the main effects cancel out in

the local maximization inequality (2). This property is consistent with the theoretical model, which asserts that

Each time the inequality (2) holds for a trial guess of the vector of parameters β, the score

of correct prediction increases by 1. The vector of parameters that yields the highest score of

correct predictions Q(β) provides a consistent estimator of the parameters β (Fox, 2010).

Note that adding a constant to the match output or multiplying it by a positive constant

will not affect the inequality (2). To take account of this aspect, the scale of the production

function must be normalized. I normalized the scale of the match output by setting to |1| the

coefficient β0 of the interaction term between two control variables: the size of the firm and

the lab size. 14 As Stephan (2010) points out, biomedical research relies increasingly on

sophisticated equipment and expensive material, which makes access to resources a necessary

condition for doing research in this field. In this context, a partner of larger size is more

valuable because it provides the possibility to undertake more complex research projects at

lower costs. In addition, a larger partner gives access to a broader network of people. I

conjectured that these aspects are important on both sides of the market. Thus, I chose the

interaction term |1|*FirmSize * LabSize to play the role of a ‘benchmark’ relationship,

relative to which I assessed the magnitude of the other relationships in the model. Following

a standard approach in the literature (e.g. Fox, 2009; Yang, Shi, Goldfarb, 2009), I ran

separate analyses for both +1 and −1 and I chose the sign that produced a higher value for the

objective function.

Hypothesis testing requires sub-sampling, a procedure that involves random sampling

without replacement and generating 95% confidence intervals for the estimated coefficients.

The reported confidence intervals were obtained from 200 random samples. The appendix

offers a more detailed description of the method.

matching is driven entirely by the interaction between partners’ attributes, while the ‘un-interacted’

characteristics cancel out because all potential partners value them equally.

14 All choice models impose a scale normalization of the utility function. Typically, the variance of the error

terms is set to a number chosen for convenience (e.g. the variance of the error terms is set to π2/ 6 in logit

models and to 1 in probit models). A standard way to normalize the scale in a maximum score estimator is to set

one coefficient to be equal to |1|.

RESULTS AND DISCUSSION

Table 3 provides the descriptive statistics and the correlation matrix. All of the key variables

have a substantial amount of variation. While prior research has focused on the contribution

of public research for start-ups and large, established firms (see Cohen et al., 2002), current

data show that university−industry research collaborations involve firms that vary

considerably in age, size, and publication and patenting activity. On average, firms in the

sample published 230 publications and applied for 82 patents every year.15 Faculty published

on average 3.8 articles yearly and 70 articles over a career. Faculty in the top decile of the

distribution had more than seven patents, and 42 percent of contracts involved a scientist with

at least one patent application. The sample mean is equal to two faculty patents.

[Insert Table 3 here]

Because the interpretation of results is not meaningful when variables have different units

of measurement and differ greatly in scale magnitude, I standardized the variables using the

z-score transformation. Table 4 presents the estimated sign of the baseline and the estimated

values of β coefficients and their 95% confidence intervals.

[Insert Table 4 here]

Model 1 focused on the relationships predicted in H1-H4, controlling for several other

determinants of matching. As the correlation matrix showed, larger firms tended to generate

patents and publications of better quality. Thus, I controlled for potential confounding effects

of firm size by including the interaction terms between firm size and scientists’ capabilities.

Likewise, I controlled for lab size effects by interacting this variable with firms’ capabilities.

Because experience and seniority in an industry/ field are likely to matter in alliance

formation, I added in the model the interaction between firm age and academic age.

15 To make the comparison easier, I discuss here yearly averages, while Table 3 reports the statistics for the

stock of patents and publications as described in Table 1.

However, the variable academic age might reflect both the experience and the academic

status of the scientist. To disentangle these effects, I included separately the interaction

between firm age and scientists’ title. Lastly, to separate the effects of firm age from those of

firm size, I also interacted firm size with scientists’ age and title variables.

The results of the multivariate matching analysis strongly corroborate the relationships

hypothesized in H1, H3b, H4a, and H4b. As predicted in H1, firms’ and scientists’ publishing

capabilities are complements in innovation. As conjectured in H3b, there is a negative

relationship between partners’ patenting capabilities, consistent with the idea that patenting

capabilities are substitutes. The fourth interaction term shows that firms and scientists whose

research is concentrated in fewer scientific domains create more value by teaming up with

more knowledge-diversified partners. An unexpected result is the negative relationship

between firms’ patenting capabilities and professors’ publishing capabilities, while H2

predicted a positive relationship. This finding echoes Gittelman and Kogut (2003), who found

that biotechnology firms’ high-impact patents did not build on important scientific papers, a

result the authors attributed to the fact that valuable inventions and valuable scientific

knowledge follow different and conflicting evolutionary selection ‘logics’. While Gittelman

and Kogut’s study was based on patents and publications by biotechnology firms, their

arguments offer a possible explanation for the negative relationship between firms with

higher stock of important patents and faculty with a higher stock of important publications.

However, I cannot rule out a simpler argument that the operationalization of variables did not

capture all relevant aspects for the division of innovation labor between firms and scientists.

The inclusion of various controls in the equation revealed interesting aspects of matching.

All other things being equal, faculty with a higher stock of important patents create higher

synergies by partnering with firms of smaller size. This result reinforces the arguments

behind the finding that scientists’ and firms’ patenting capabilities are substitutes. Contrarily,

professors with a better publication record create higher synergies with more resourceful

firms. This finding echoes a longstanding view in the literature that university−industry

collaboration is driven by scientists’ need to obtain resources from industry partners.

However, this result also shows that, all else equal, access to firms with more resources is

given to (and more valued by) prominent scientists.

We can also observe a negative relationship between lab size and firms’ publishing

capabilities, suggesting that firms with higher publication capabilities get more rewards from

collaborating with individual scientists or smaller teams. Likewise, it shows that firms’

publishing capabilities are more important for smaller labs than for larger ones.

The relationship between firm age and academic age is negative, an indication that more

experienced partners are substitutes for less experienced partners. Other interesting effects

can be observed when the role of firm experience (age) and resources (size) are examined in

interaction with faculty academic title.16 In particular, both larger firms and younger firms

create higher synergies by partnering with professors of high standing, such as chairmen and

chiefs of various divisions. Thus, both large and startup firms compete for higher status

faculty. These findings suggest that on the one hand, professors that can manage larger-scale

projects match with firms that can provide an appropriate level of resources; on the other

hand, faculty with higher standing in administration match with younger firms, which are

riskier but potentially, more rewarding partners.

The estimated sign of the baseline is positive, indicating complementarity between

partners’ resources. In terms of economic significance, the baseline represents the marginal

effect on the innovation output created by increasing the size of the firm and the lab by one

standard deviation. In Model (1), the baseline effect on the production function was

16 A model where I included the relationship between firm age, respectively, firm size with a variable measuring

the academic title on a scale from 1 (assistant professor) to 4 (chairman or chief) did not produce significant

coefficients, which is not surprising given that age/size and academic title interact in a non-monotonic way.

normalized to +1 and it represents the scale to which we can interpret the magnitude of other

coefficients. The results show that increasing the publication capabilities of both partners by

one unit (i.e., one standard deviation in the stock of quality-adjusted publications), has a nine-

times-higher effect on the innovation output than the baseline. Further, in absolute value, the

marginal impact of a one-standard-deviation change in partners’ patenting capabilities is 15

times more important than the baseline, and the marginal effect of a unit change in

knowledge-diversity is four times more important. 17

As well as being significant, the variables measuring partners’ capabilities explain a fair

amount of the variation in the data. Model (1) predicts 74% of inequalities.18 For comparison,

Model (2) shows the results of an equation that has only control variables; few coefficients

were significant and the model explained 59% inequalities. Model (3) contained all

relationships except those predicted by the main hypotheses and explained 61% inequalities.

Robustness checks

I performed several robustness checks to test the sensitivity of results to alternative model

specifications and variable transformations. Models (4) and (5) test whether the main

relationships predicted in H1 to H4 remain robust when firm age, respectively, industry

effects are taken into account. In relative terms, the impact of the main relationships on the

alliance output is smaller in these specifications, but previous results remain significant and

of the same sign. The analysis in Model (4) shows that firm age does not play an important

role when considered in combination with scientists’ publishing and patenting capabilities,

17 Because of the substitution relationship, the impact is positive in absolute terms when one unit increase in

patenting capabilities of one partner is associated with one unit decrease in patenting capabilities of the other

partner. The same reasoning applies for knowledge diversity: in absolute terms, one unit increase in knowledge

diversity on one side of the market, together with one unit decrease in knowledge diversity on the other side of

the market has a four times higher impact on the alliance output than the baseline.

18 In a market with k agents on each side, the maximum number of inequalities predicted by a random

configuration where k-2 pairs are identical to ones in the ‘true’ model is: (k-2) (k-3)/k(k-1). For the 10 markets

in the sample, we obtain (by multiplying the corresponding number of inequalities for each market) .386 or

38.6% inequalities. However, the expected value for such arrangement to occur is equal to (k-2)(k-3)/k(k-

1)*(1/(k!-1)), which is close to zero in the data.

but younger firms and more knowledge-diversified scientists create higher synergies together.

In Model (5), the results remain robust to the inclusion of a dummy variable for

biotechnology firms. Interestingly, scientists with higher patenting capabilities and those with

a more diversified knowledge base create higher synergies by collaborating with

biotechnology firms. However, the relationship between faculty publishing capabilities and

biotech dummy is not significant. Finally, Models (6) and (7) in Table 4 show that results of

the empirical test are robust to the log transformation of variables.19

Limitations and future research

The results of the matching analysis presented here are based on several assumptions. The

model discounts search costs (e.g., the cost of finding a partner), the potential initial

uncertainty about the match value (e.g., adverse selection), and other types of market

frictions, such as institutional policies of technology transfer that might impede the formation

of public-private partnerships. These factors should be taken into account for a more

comprehensive approach to match formation. For example, with high search costs,

geographical location might be an important consideration in partner choice. Because

collaboration contracts in this paper are from one university, all scientists in the sample are

equally located from the firms’ point of view. Data from more universities would allow us to

examine the role of geography in university−industry partnerships. Likewise, data from

universities with different technology-transfer policies would allow researchers to determine

if frictions created by technology transfer terms (e.g. rigid rules on intellectual property

rights) might preclude some matches from happening.

19 A small constant of 0.001 was added before logging to insure that values equal to zero remained in the

estimation. The constant shifts the distribution to the right and the log transformation minimizes the importance

of extreme values. The baseline now represents the marginal effect on the innovation output generated by

increasing the size of the firm and the lab size by 1%. Because of the log transformation, the interpretation of

coefficients is in terms of ‘percentage change’. For example, Model (7) in Table 4 indicates that relative to 1%

increase in firm size and lab size, a 1% increase in publishing capabilities of both partners (measured on the

new scale) yields a 4% increase in the alliance output. Additional analyses (available upon request) show that

results are not driven by the presence in the sample of firms and scientists without publications or patents.

This paper was based on the premise that the objective of a research alliance is to

generate new knowledge. Future work could examine if considerations other than research

productivity, such as perceptions of social status or affiliation with a particular social

network, provide additional explanatory power or change the understanding of matching

observed here.

The theoretical model presented in this paper addressed questions related to value

creation and abstracted away the costs of governance, value appropriation, and partnership

evolution. The connection between matching and these important aspects of alliances creates

an interesting opportunity for additional research.

Contribution to the literature of university-industry knowledge transfer

This study made a first step toward unpacking the dimensions of firm−scientist matching in

research collaboration. The empirical analysis showed that research partners complement

each other in publishing capabilities, but substitute each other in patenting capabilities.

Knowledge diversity also matters in partner choice, as shown by the complementary

relationship between knowledge-specialization and knowledge-diversification.

These findings are interesting both in the context of university−industry knowledge

transfer and, more generally, for the literature on alliance partner selection. The results

corroborate previous work that found a performance-enhancing effect for firms connected

with university professors with an outstanding research record. However, this study

contributes to a better understanding of the underlying causes of these effects. While previous

work has focused the value-added contribution of ‘stars’ in the innovation process, I show

that, through the ex-ante matching process, ‘star’ scientists team-up with firms that also have

outstanding publication records. Further, although publishing capabilities are important in the

market for research, I show that firm−scientist alliance formation is driven by a more

complex set of considerations, notably, by anti-assortative matching based on patenting

capabilities.

Contribution to the literature on alliance formation

Despite the importance of matching in partner choice, theoretical and empirical work in this

area is in an early stage. This paper argued that the drivers of alliance performance cannot be

fully understood without taking into account the matching process that precedes the

formation of research alliances, and it provided a rigorous approach for examining

partnership formation. The paper introduced a theoretical model that explicitly deals with the

endogeneity problem created by the self-selection of ties in specific partnerships. The model

demonstrated that anticipating synergistic gains and competing to ally with better partners

leads to sorting in the market, which, in turn, explains why certain firm−scientist alliances

create more value and enjoy higher innovation performance. The empirical analysis followed

the theoretical model closely and focused on identifying the underlying sorting patterns

created by complementarity and substitutability of partner attributes. Both the theoretical

framework and the methodology can be applied to settings other than knowledge creation.

ACKNOWLEDGEMENTS

This paper has significantly benefited from comments and suggestions by the members of my dissertation

committee, Rajshree Agarwal, David Audretsch, Janet Bercovitz, Glenn Hoetker and Joseph Mahoney. I am

also thankful to the editor Will Mitchell and three anonymous referees for their suggestions for improvement

during the reviewing process. Several people have provided useful comments at various stages of this project, in

particular, Gautam Ahuja, Thomas Astebro, Pierre Azoulay, Dan Elfenbein, Alfonso Gambardella, Bernard

Garrette, Shane Greenstein, Nicola Lacetera, Kristina McElheran, Deepak Somaya and Scott Stern. I am grateful

to Janet Bercovitz for generously sharing data for this project. The Ewing Marion Kauffman Foundation and the

Academy for Entrepreneurial Leadership at the University of Illinois at Urbana-Champaign have provided

invaluable financial support. Any remaining errors are my own.

REFERENCES

Agarwal R, Crosson R, Mahoney J. 2010.The role of incentives and communication in strategic alliances: an

experimental investigation. Strategic Management J. 31(4): 413–437.

Aghion P, Dewatripont M, Stein JC. 2008. Academic Freedom, Private-Sector Focus, and the Process of

Innovation. The RAND J. of Economics 39 (3): 617–635.

Ahuja, G. 2000. The duality of collaboration: inducements and opportunities in the formation of interfirm

linkages. Strategic Management J. 21(3): 317–343.

Anton JJ, Yao DA.2003. Patents, invalidity, and the strategic transmission of the enabling information. J. of

Economics and Management Strategy 12(2): 151–178.

Arora A, Gambardella A. 1994. Explaining Technological Information and Utilizing It: Scientific Knowledge,

Capacity, and External Linkages in Biotechnology. J. of Economic Behavior and Organization 24(1): 91–114.

Audretsch DB, Stephan P. 1996. Company-Scientist Locational Links: The case of Biotechnology. American

Economic Review 86(3): 641–652.

Azoulay P, Ding W, Stuart T. 2007. The determinants of faculty patenting behavior: Demographics or

opportunities? J. of Economic Behavior and Organization 63(4): 599–623.

Becker GS. 1973. A Theory of Marriage: Part I. J. of Political Economy 81(4): 813–846.

Belderbos R, Carree M, Lokshin B. 2004. Cooperative R&D and Firm Performance. Research Policy 33(10):

1477–92.

Bercovitz J, Feldman M. 2007. Fishing Upstream: Firm Strategic Research Alliances with Universities.

Research Policy 36(7): 930–948.

Bercovitz J, Feldman M. 2008. Academic Entrepreneurs: Organizational Change at the Individual Level

Organization Science 19(1): 69–89.

Brusoni S, Prencipe A, Pavitt K. 2001.Knowledge specialization, organizational coupling, and the boundaries of

the firm: why do firms know more than they make? Administrative Science Quarterly 46(4): 597–621.

Cassiman B, Veugelers R. 2006. In Search of Complementarity in Innovation Strategy: Internal R& D and

External Knowledge Acquisition. Management Science 52(1): 68–82.

Chatain, O.2010.Where Do Client-Specific Scope Economies Come from? A Revealed Preference Analysis.

Working paper, Wharton Business School.

Cockburn I, Henderson R.1998. Absorptive capacity, coauthoring behavior, and the organization of research in

drug discovery. J. of Industrial Economics 46(2): 157–182

Cohen WM, Nelson RR, Walsh JP. 2002. Links and impacts: The influence of public research on industrial

R&D. Management Science 48(1): 1–23.

Dahlin KB, Weingart LR, Hinds PJ. 2005.Team Diversity and Information Use. Academy of Management J.

48(6): 1107–1123.

David PA, Hall BH, Toole A. 2000. Is public R&D a complement or substitute for private R&D? A review of

the econometric evidence. Research Policy 29(4-5): 497–529.

Elfenbein D. 2007. Publications, patents, and the market for university inventions. J. of Economic Behavior and

Organization 63(4): 688–715.

Fabrizio KR. 2009. Absorptive capacity and the search for innovation. Research Policy 38(2): 255–267.

Fernando C, Gatchev V, Spindt PA. 2005. Wanna Dance? How Firms and Underwriters Choose Each Other. J.

of Finance 60 (5): 2437–2469.

Fleming L, Sorenson O. 2004. Science as a map in technological search. Strategic Management J. 25(8/9): 909–

928.

Fox J. 2009. Estimating Matching Games with Transfers. NBER working Paper 14382. National Bureau of

Economic Research, Cambridge, MA. Available at: http://www.nber.org/papers/w14382.

Fox J. 2010. Identification in matching games. Quantitative Economics 1: 203–254.

Gans JS, Stern S. 2010. Is there a market for ideas? Industrial and Corporate Change 19(3): 805–837.

Gittelman M, Kogut B. 2003. Does Good Science Lead to Valuable Knowledge? Biotechnology Firms and the

Evolutionary Logic of Citation Patterns. Management Science 49(4): 366–382.

Hall BH, Ziedonis RH. 2001. The patent paradox revisited: An empirical study of patenting in the U.S.

semiconductor industry, 1979–1995. RAND J. of Economics 32(1): 101–128.

Henderson R, Cockburn I. 1996. Scale, Scope, and Spillovers: The Determinants of Research Productivity in

Drug Discovery. Rand J. of Economics 27(1): 32–59.

Helfat CE, Finkelstein S, Mitchell W, Peteraf MA, Singh H, Teece, DJ, Winter SG. 2007. Dynamic

Capabilities: Understanding strategic change in organizations, Blackwell Publishing.

Hicks D.1995. Published papers, tacit competencies, and corporate management of the public/private character

of knowledge. Industrial and Corporate Change 4(2): 401–424.

Jensen R, Thursby M. 2001. Proofs and Prototypes for Sale: The Licensing of University Inventions. American

Economic Review 91(1): 240–259.

Jones B. 2009. The Burden of Knowledge and the Death of the Renaissance Man: Is Innovation Getting Harder?

The Review of Economic Studies 76(1): 283–317.

Kenney M.1986. Biotechnology: The University Industrial Complex. Yale University Press: New Haven, CT.

Lacetera N. 2009. Different missions and commitment power in R&D organization: theory and evidence on

industry- university relations. Organization Science 20(3): 565–582.

Lippman SA, Rumelt RP. 2003. A Bargaining Perspective on Resource Advantage. Strategic Management

Journal 24(11): 1069–86.

Lowe RA. 2001. The role and experience of start-ups in commercializing university inventions. G. Libecap, ed.

Entrepreneurial Inputs and Outcomes. JAI Press, Amsterdam, The Netherlands: 189–222.

Mansfield E., Lee JY (1996). The modern university: contributor to industrial innovation and recipient of

industrial R&D support. Research Policy 25(7): 1047–1058.

Milgrom P, Roberts J. 1995. Complementarities and fit: strategy, structure, and organizational change in

manufacturing. J. of Accounting and Economics 19(2): 179–208.

Mortensen DT. 1988. Matching: Finding a Partner for Life or Otherwise. American J. of Sociology 94,

Supplement: S215-S240.

Murray F. 2011. The oncomouse that roared: hybrid exchange strategies as a source of distinction at the

boundary of overlapping institutions. American J. of Sociology 116(2): 341–88.

Murray F, Stern S. 2007. Do formal intellectual property rights hinder the free flow of scientific knowledge? An

empirical test of the anti-commons hypothesis. J. of Economic Behavior and Organization 63(4): 648–687.

Nelson RR. 2004. The market economy, and the scientific commons. Research Policy 33(3): 455–471.

Orsenigo L.1989. The Emergence of Biotechnology. St Martin Press, New York.

Parmigiani A, Mitchell W. 2009. Complementarity, capabilities, and the boundaries of the firm: the impact of

within-firm and interfirm expertise on concurrent sourcing of complementary components. Strategic

Management J., 30(10): 1065–1091.

Rosenkopf L, Almeida P. 2003. Overcoming Local Search Through Alliances and Mobility. Management

Science 49(6): 751–766.

Roth AE, Sotomayor M. 1990. Two-Sided Matching: A Study in Game-Theoretic Modeling and Analysis.

Cambridge University Press, Cambridge, MA.

Santiago D, Fox JT. 2008. A Toolkit for Matching Maximum Score Estimation and Point and Set Identified

Sub-sampling Inference. Available at: http://www-personal.umich.edu/~jtfox/

Shaver JM. 1998. Accounting for Endogeneity When Assessing Strategy Performance: Does Entry Mode

Choice Affect FDI Survival? Management Science 44(4): 571–585.

Sorensen M. 2007. How smart is smart money? A two-sided matching model of venture capital, J. of Finance

62(6): 2725–2762.

Stephan P. 1996. The Economics of Science. J. of Economic Literature 34(3): 1199–1235.

Stephan P. 2010. Economics of Science. In Handbook of Economics and Innovation (vol. 1). Hall BH,

Rosenberg N (eds.). Elsevier: Amsterdam, North-Holland: 217–274.

Stern S.1995. Incentives and Focus in University and Industrial Research: The Case of Synthetic Insulin. In The

University−industry Interface and Medical Innovation. Gelijns A, Rosenberg N (eds.). Institute of Medicine,

National Academy Press: Washington, D.C.

Stuart TE, Ding W. 2006. When Do Scientists Become Entrepreneurs? The Social Structural Antecedents of

Commercial Activity in the Academic Life Sciences. American Sociological J. 112(1): 97–14.

Thursby JG, Thursby MC.2002.Who is Selling the Ivory Tower? Sources of Growth in University Licensing.

Management Science 48(1): 90–104.

Train KE.2004. Discrete Choice Methods with Simulation. Cambridge University Press: Cambridge, UK.

Uzzi B.1996.The Sources and Consequences of Embeddedness for the Economic Performance of Organizations.

American Sociological Review 61(4): 674–698.

Wang Q, von Tunzelmann N. 2000. Complexity and the Functions of the Firm: Breadth and Depth, Research

Policy 29(7/8): 805–818.

Yang Y, Shi M, Goldfarb A. 2009. Estimating the Value of Brand Alliances in Professional Team Sports.

Marketing Science 28(6): 1095–1111.

Zucker LG, Darby MR, Armstrong J. 1998. Geographically localized knowledge: Spillovers or markets?

Economic Inquiry 36(1): 65–86.

Zucker LG, Darby MR, Armstrong J. 2002. Commercializing Knowledge: University Science, Knowledge

Capture, and Firm Performance in Biotechnology. Management Science 48(1): 138–153.

Table 1.Variable Definition

Firms

Size

(FIRM SIZE)

Number of employees at contract year.

Source: Factiva; Corporate Affiliations; Thomson One

Publishing

Capabilities

(FIRM PUB CAP)

Ten-year stock of publications weighted by the number of citations (as of April 2010) and

a 10% yearly depreciation rate.

Source: ISI Web of Science *, combined with Corporate Affiliations to account for

organizational structure at the publication time

Note that ISI is the only collection of medical journals that contains information on

authors’ institutional affiliations during the sample timeframe. However, ISI provides this

information only for the first ten authors. Thus, this measure constitutes an accurate, albeit

conservative record of firms’ publications.

Patenting

Capabilities

(FIRM PAT CAP)

Six-year stock of patents weighted by citations (as of June 2010) and a 10% yearly

depreciation rate.

Source: Delphion combined with Corporate Affiliations to account for organizational

structure at the patent application time

Knowledge breadt

(FIRM BREADTH

)

Entropy measure of knowledge diversification based on inventive IPCR patent classes.

Higher values indicate higher degree of knowledge diversity. I used the full range of

inventive IPCR patent classes, not just the main patent class.

Source: Firms’ patent stock (see above)

Age

(FIRM AGE)

Number of years since founding.

Source: Corporate Affiliations; Factiva; Companies’ web pages

Biotechnology firm

(BIOTECH)

Coded 1 if the firm’s main activity was in the biotechnology industry

Source: Corporate Affiliations; Factiva; Companies’ web pages

Principal Investigators (PI)

Lab Size

(LAB SIZE)

Number of people associated with Principal Investigator’s lab.

I obtained accurate information on the lab size for 40% of the sample. For the remaining

scientists, I estimated the lab size from corroborating various sources (e.g. faculty web

sites, local news articles), and in particular, from carefully identifying the investigator’s

co-authors at the same university during a three-year window prior to the contract date.

Although not perfectly measured, this variable captures the variation in the human capital

resources that principal investigators could mobilize for a project.

Source: Elsevier’s Scopus; ISI Web of Science; Scientists’ web pages; Lexis-Nexis

Publishing

Capabilities

(PI PUB CAP)

Cumulative number of publications weighted by the number of citations (as of April

2010). A 10% yearly depreciation rate has been applied.

Source: ISI Science Citation Index Expanded

Patenting

Capabilities

(PI PAT CAP)

Number of patents and patent applications in which the scientist appears as inventor

cumulated over a six-year window prior to collaboration and weighted by patent citations

(self-cites excluded) as of June 2010. A 10% yearly depreciation rate has been applied.

Source: Delphion

Knowledge breadt

(PI BREADTH)

Entropy measure of diversification based on MESH descriptor/qualifier keyword

combination.

Source: Medline and Medical Subject Headings thesaurus of the National Library of

Medicine

Academic age

(PI AGE)

Number of years since the highest degree was obtained.

Source: Professors’ CV; University Registrar

Academic title Dummy variables for each title category: Assistant; Associate; Professor; Chairman/Chief

Source: Professors’ CV; University Registrar

Table 2. Variable measurement and empirical predictions

Hypothesis Firm Side Scientist Side Theoretical

Prediction

Relationship between Variable Measurement

and Expected Sign

Expected

ign

H 1 Publishing

Capabilities

Publishing

Capabilities

Complementarity High values on both variables indicate superior

publishing capabilities. A positive sign is consistent

with a complementary relationship.

H2 Patenting

Capabilities

Publishing

Capabilities

Complementarity High values indicate superior patenting, respectively,

publishing capabilities. A positive sign is consistent

with a complementary relationship.

H3a Patenting

Capabilities

Patenting

Capabilities

Complementarity High values on both variables indicate superior

patenting capabilities. A positive sign implies

complementarity.

H3b Patenting

Capabilities

Patenting

Capabilities

Substitution High values on both variables indicate superior

patenting capabilities. Hence, a negative sign (where

high levels on one side of the market are associated

with lower levels on the other side of the market)

implies substitution in patenting capabilities.

H4a Knowledge

Breadth

Knowledge

Specialization

Complementarity On both sides, the entropy-based ‘Knowledge

diversification’ measure increases with the diversity of

knowledge, such that high values imply ‘breadth’ or

‘diverse knowledge’ and lower values indicate

‘specialization’ or ‘focused expertise’. Theoretically,

complementarity is expected between breadth (i.e. high

values of the “Knowledge diversification” index) on

the firm side and specialization (i.e. low values of the

‘Knowledge diversification’ index) on the scientist

side. Hence, when the direction of measurement of the

‘Knowledge diversification’ variables is taken into

account, a negative sign indicates complementarity.

H4b Knowledge

Specialization

Knowledge

Breadth

Complementarity From a theoretical perspective, complementarity is

expected between specialization (i.e. low values of the

‘Knowledge diversification’ index) on the firm side

and breadth (i.e. high values of the ‘Knowledge

diversification’ index) on the scientist side. Thus, a

negative sign is consistent with the predicted

complementarity relationship.

Table 3. Descriptive Statistics and Correlations

Mean Std. Dev. Min Max 1 2 3 4 5 6 7 8 9 10 11 12 13 14

1.Size 23110.73 34486.16 1 156400 1

.Publishing Capabilities 37835.48 57591.10 0 221509 0.71 1

.Patenting Capabilities 5041.54 8551.80 0 65582 0.26 0.30 1

.Knowledge breadth 4.41 2.09 0 7.19 0.64 0.61 0.43 1

.Age 47 46.84 1 155 0.52 0.31 0.16 0.64 1

6. Biotechnology firm 0.26 0.43 0 1 -0.54 -0.29 -0.31 -0.52 -0.4 1

.Lab Size 17.40 13 1 41 -0.10 -0.09 -0.09 -0.10 -0.11 0.00 1

.Publishing Capabilities 1219.06 1383.27 1 12671.43 -0.06 0.04 -0.01 0.09 0.02 -0.04 0.55 1

.Patenting Capabilities 41.97 293.22 0 4382.99 -0.04 -0.09 -0.07 -0.05 -0.01 0.05 -0.06 0.31 1

10.Knowledge breadth 4.75 0.52 2.20 5.79 -0.09 -0.06 -0.12 -0.09 -0.14 0.09 0.56 0.37 0.26 1

11.Academic age 17.12 7.35 3 53 -0.09 -0.13 -0.13 -0.09 -0.10 0.07 0.40 0.12 0.01 0.50 1

12.Associate Professor 0.29 0.45 0 1 0.07 0.06 0.12 0.08 0.07 -0.06 -0.08 -0.05 -0.10 0.04 -0.06 1

13.Full Professor 0.35 0.48 0 1 -0.13 -0.10 -0.13 -0.06 -0.06 0.05 0.44 0.43 0.13 0.44 0.39 -0.36 1

14.Chair/Chief 0.12 0.32 0 1 0.01 0.03 -0.02 0.04 -0.04 0.05 0.26 0.40 0.23 0.31 0.34 -0.23 -0.20 1

Table 4. Point Estimates and 95% Confidence Intervals

Z-Score Transformation Log Transformation

Relationship Model 1

Model 2

Model 3

Model 4

Model 5

Model 6 Model 7

Baseline FIRM SIZE *

LAB SIZE 1 -1 -1 1 1 1 1

H1 FIRM PUB CAP *

PI PUB CAP

9.22

(5.33; 11.97) … …

3.56

(2.97; 7.10)

6.78

(3.00; 9.95)

3.35

(1.25; 4.72)

4.03

(1.66; 5.77)

H2 FIRM PAT CAP*

PI PUB CAP

-1.01

(-3.18; 2.17) … …

-1.02

(-2.96; 3.21)

-0.87

(-3.03; 3.88)

-1.04

(-5.60; 1.02)

-1.26

(-6.49; 2.95)

H3 FIRM PAT CAP*

PI PAT CAP

-15.91

(-18.78; -10.36) … …

-7.23

(-11.35; -5.32)

-4.61

(-9.71;-2.18)

-2.83

(-5.42; -1.07)

-2.66

(-5.49; -1.43)

H4 FIRM BREADTH*

PI BREADTH

-4.78

(-6.65; -1.24) … …

-1.93

(-4.91; -0.84)

-2.89

(-5.12; -0.71)

-4.24

(-10.98; -3.46)

-13.67

(-16.88; -7.79)

5 FIRM SIZE*

PI PUB CAP

5.26

(3.21; 8.15) … 9.72

(3.05; 12.30)

3.97

(1.64; 5.86)

6.61

(2.75; 9.36)

0.17

(-0.66; 3.24)

1.35

(0.92; 3.02)

6 FIRM SIZE*

PI PAT CAP

-21.54

(-23.85; -17.25) … -12.85

(-18.81; -4.23)

-9.37

(-11.47; -5.11)

-3.21

(-6.95; -1.43)

-1.02

(-3.19; 0.99)

-0.45

(-4.29; 1.53)

7 FIRM SIZE*

PI BREADTH

-0.31

(-4.40; 1.15) … -6.09

(-7.54; -2.18)

-0.35

(-3.28; 1.63)

-5.86

(-8.17; -2.27)

-19.74

(-24.40; -15.71)

-10.25

(-16.68; -7.83)

8 FIRM PUB CAP *

LAB SIZE

-3.66

(-6.45; -1.16) … -1.25

(-6.45; -0.37)

-3.94

(-7.86; -1.43)

-7.48

(-10.60; -3.54)

-2.23

(-5.88; -1.05)

-2.64

(-6.76; -1.15)

9 FIRM PAT CAP *

LAB SIZE

-0.78

(-2.59; 4.06) … 0.23

(-4.77; 2. 57)

-0.64

(-5.77; 2.41)

-0.32

(-3.15; 4.16)

1.46

(-4.48; 3.80)

1.01

(-3.65;5.16)

10 FIRM BREADTH*

LAB SIZE

-0.24

(-2.40; 5.73) … 0.10

(-3.30; 4.81)

0.68

(-3.60; 5.94)

0.85

(-2. 55; 5.48)

8.92

(2.55; 9.28)

6.61

(1.73; 8.33)

11 FIRM AGE *

PI AGE

-0.93

(-3.92; -0.08)

0.04

(-5.90; 2.87)

-0.60

(-4.63; 2.39)

-0.63

(-3.90; 1.11)

-0.89

(-4.20; -0.12)

-2.86

(-5.90; -0.53)

-5.89

(-9.33; -2.98)

12 FIRM SIZE *

PI AGE

-1.11

(-4.07; -0.52)

-0.24

(-4.78; 3.49)

-1.36

(-3.10; -1.06)

-0.98

(-4.01; -0.50)

-0.60

(-3.27; 0.29)

-5.06

(-10.77; -2.78)

-3.06

(-6.50; -0.75)

13 FIRM AGE *

ASSOCIATE

1.77

(-0.84; 4.15)

0.56

(-1.64; 4.22)

1.98

(-4.39; 3.25)

3.97

(-0.53; 5.58)

3.46

(-0.49; 5.83)

2.54

(-0.93; 5.36)

2.60

(-3.30; 5.27)

14 FIRM AGE *

PROFESSOR

5.84

(1.13; 6.42)

0.30

(-0.84; 4.87)

0.63

(-4.64; 2.81)

1.72

(0.74; 4.63)

2.63

(1.40; 6.14)

11.51

(9.65; 17.56)

3.90

(1.34; 8.56)

15 FIRM AGE *

CHIEF

-4.87

(-7.58; -2.21)

-0.65

(-6.70; 0.26)

-1.84

(-6.90; -1.39)

-4.15

(-8.42; -2.22)

-6.33

(-8.90; -3.93)

-9.00

(-16.26; -4.64)

-8.00

(-13.59; -5.23)

16 FIRM SIZE *

ASSOCIATE

-0.02

(-2.67; 3.50)

0.35

(-5.88; 3.11)

0.69

(-3.87; 3.83)

1.86

(-4.24; 3.13)

0.82

(-3.48; 2.40)

4.24

(-2.37; 6.41)

-0.47

(-4.81; 5.31)

17 FIRM SIZE *

PROFESSOR

-17.01

(-19.77; -11.34)

-2.12

(-6.96; -1.50)

-8.64

(-13.52; -4.96)

-3.73

(-7.13; -1.76)

-8.85

(-10.48, -3.55)

-5.45

(-8.89; -2.57)

-4.11

(-9.82; -2.63)

18 FIRM SIZE*

CHIEF

5.78

(3.30; 8.12)

2.01

(1.89; 7.28)

3.26

(0.92; 5.32)

7.53

(4.34; 9.85)

6.41

(2.80; 8.33)

4.92

(1.76; 7.08)

3.13

(1.17; 6.55)

19 FIRM AGE*

PI PUB CAP

0.60

(-2.30; 3.19) ……… 0.81

(-4.30; 3.23) ………

20 FIRM AGE*

PI PAT CAP

-0.42

(-4.20; 3.62) ……… -2.56

(-5.06; 2.47) ………

21 FIRM AGE*

PI BREADTH

-3.52

(-8.30, -1.92) ……… -23.72

(-27.31; -16.63) ………

22 BIOTECH*

PI PUB CAP

-1.76

(-4.26; 2.85) -6.36

(-8.19; 3.09)

23 BIOTECH *

PI PAT CAP

2.33

(1.26; 6.79) 11.99

(5.52; 14.13)

24 BIOTECH *

PI BREADTH

2.52

(1.49; 5.25) 8.80

(6.45; 12.30)

% Inequalities

predicted 74% 59% 61% 75% 76.5% 73.5% 74%

Note: Parameters in bold are significantly different than zero.

To estimate the model, I adapted Santiago and Fox (2008)’s template in Mathematica software for the particular production functions that are

tested here. For optimization, I used the standard differential evolution procedure in Mathematica 7.0.

Appendix

I illustrate how the method works by taking two random matches from the sample and

estimating a model based only on the main four hypotheses. Below, I denote the two matches

as Match 1= {Firm 1, Scientist 1} and Match 2= {Firm 2, Scientist 2}.

MATCH 1 MATCH 2

ATTRIBUTES Firm 1 Scientist 1 Firm 2 Scientist 2

Size

(Firm, respectively, Lab)

21000 18 140 4

PubCap

Publishing Capabilities

29000 2700 700 1200

PatCap

Patenting Capabilities

2500 0 200 10

KDiv

Knowledge Diversity

5 4 2.5 3

The match output (i.e. alliance production function) takes the following functional

form:

g (F,S |β)= β0 FirmSize * LabSize + β1 PubCapf * PubCaps + β2 PatCapf *

PubCaps + β3 PatCapf * PatCaps + β4 KDivf * Kdivs

Note that we don’t have information on the left hand side of the equation (we don’t

know how much innovation/knowledge output each alliance generates). On the right hand

side, the only relevant relationships for estimating complementarity and substitution are the

interaction terms between the capabilities of the partners.

To generate the point estimates, the method compares the output generated by any

two observed matches - as above, {Firm 1, Scientist 1} and {Firm 2, Scientist 2} - with the

counterfactual matches obtained by exchanging the alliance partners: {Firm 1, Scientist 2}

and {Firm 2, Scientist 1}.

The match production functions of the observed matches are:

{Firm 1, Scientist 1} generate together:

(β0 21000*18 + β1 29000 *2700 +β2 2500 *2700 + β3 2500 *0 + β4 5 *4) (1)

{Firm 2, Scientist 2} generate together:

(β0 140* 4+ β1 700 *1200+ β2 200 *1200 + β3 200 * 10 + β4 2.5 *3 ) (2)

The match production functions of the counterfactual matches are:

{Firm 1, Scientist 2} generate together:

(β0 21000 * 4 + β1 29000*1200 + β2 2500 * 1200+ β3 2500*10 + β4 5 *3) (3)

{Firm 2, Scientist 1} generate together:

(β0 140 *18+ β1700*2700+ β2 200 * 2700+ β3 200*0 + β4 2.5*4 ) (4)

The inequality that lies at the core of the estimation procedure is the following:

(1) + (2) > (3) + (4)

This relationship translates into:

(378000 β0 + 78300000 β1 +6750000 β2 + 20 β4) + (560 β0 + 840000 β1 + 240000 β2 + 7.5

β4) > (84000 β0 +34800000 β1 + 3000000 β2 + 25000 β3 + 15 β4) + (2520 β0 + 1890000 β1 +

540000 β2 + 10 β4)

Equivalently: (292040 β0 + 42450000 β1+ 3450000 β2 - 25000 β3 + 2.5 β4) > 0

Similar inequalities are written for all other observed pairs in a market and the

corresponding counterfactuals. As explained in the paper, each year constitutes a market in

this empirical context (up to a total of ten markets from 1995 to 2004). Note that partner

exchanges occur only among pairs belonging to the same market-year.

The method estimates both the sign and the magnitude of the coefficients β1 to β4 but

only the sign of β0. The coefficient β0 represents the ‘scale’ of the match production output.

The estimated values of β coefficients are those that maximize the number of

inequalities satisfied. The significance of the coefficients is given by the 95% confidence

intervals obtained from generating a large number of subsamples of matches.

How to Synergize Different Institutional Logics of Firms in Cross-border Acquisitions: A Matching Theory Perspective

Article

Full-text available

Feb 2023

When cross-border acquisitions take place, whose institutional logics should dominate the control in the daily operation of the acquired entity — the acquirer or the acquiree? To answer this question, this paper investigates how Geely, a Chinese automaker, successfully acquired Volvo from Sweden, a developed country, and managed the differences in the institutional logics of the two firms by transforming its dominant institutional logic. Through the lens of matching theory, we employ an inductive approach and conduct a longitudinal case study of Geely’s acquisition. We find that matching commercial and social logic is an inevitable requirement to cope with external institutional pressures in cross-border acquisitions. We further reveal that the structured interaction between compatibility and complementarity facilitates the matching process between commercial logic and social logic, resulting in a synergistic logic via the three different mechanisms of compatibility, complementarity, and co-evolution. Our findings challenge previous research that focuses on conflicts as the foremost drivers of the transformation of different institutional logics in organizations. Based on our findings, we develop a matching process model that offers insights for firms from developing economies to navigate dominant institutional logic transformation and thrive in the marketplace through their strategic cross-border acquisitions.

Managing open innovation with science-based vs. market-based partners: board of directors as a contingency

Article

Full-text available

Mar 2024

Drawing on the open innovation literature, we examine the relationship between alliances with science-based and market-based partners on the one hand, and impactful and lower-impact innovations, on the other hand. Specifically, we predict that alliances with science-based partners will boost impactful innovations while alliances with market-based partners will boost low-impact innovations. We also examine how the social capital of the Board of Directors moderates these relationships. We base our analyses on a large dataset of Chinese firms constructed from diverse sources and find strong support for our hypothesized relationships. We identify the theoretical and managerial implications of our study.

Institutional logics and founders' identity orientation: Why academic entrepreneurs aspire lower venture growth

Article

Jan 2023
RES POLICY

This study examines how the identity orientation of founders, i.e., the extent to which they define themselves in terms of their relationships to others and to social groups, is imprinted by their professional logic and influences their ambitions for venture growth. We draw on existing insights regarding the Darwinian, Communitarian, and Missionary orientation of entrepreneurs and on interviews with 29 founders to develop our hypotheses, which we then test in a sample of 58 academic and 113 non-academic founders that participated in a venture competition. We argue that, compared to non-academic institutional logics, academic logics are tied to a stronger Communitarian and Missionary orientation and a weaker Darwinian orientation in founders. A stronger Darwinian orientation values venture growth, whereas a stronger Communitarian orientation appraises the benefit of the technology for a restricted set of people at the expense of such growth ambitions. A stronger Missionary orientation values welfare maximization for society which may to some degree entail higher growth aspirations. We argue and empirically confirm that these identity orientations explain why academic founders hold lower growth aspirations for their start-up than non-academic founders. Our findings can at least partially clarify why academic start-ups do not grow according to expectancies. They theoretically advance our insights in academic entrepreneurship and founders' growth aspirations while also extending the literature on founders' identity orientation.

Critical Success Factors of University-Industry R&D Collaborations

Article

Full-text available

Mar 2023

University-industry R&D collaborations (UICs) play a vital role in stimulating open innovation that leads to new products, processes, and services that creates value for customers and broader societal impact. UICs, however, commonly fail to meet these stakeholders' benefits. This study identifies thirty-four critical success factors (CSFs) for improving UIC success. The study includes a systematic literature review and a longitudinal UIC case study between Bosch Car Multimedia in Portugal and University of Minho, a multi-million Euro R&D collaboration from 2013 to 2021. The importance of the CSFs is discussed in the context of the UIC lifecycle. A survey among researchers and industry practitioners involved in R&D collaborative projects was completed to confirm the analysis of the empirical results. This paper provides UIC managers with CSFs, which, when addressed competently, can provide a basis for successful UIC projects and sustainable university-industry collaborations.

How do university-firm interactions affect firm innovation speed? The case of Chinese science-intensive SMEs

Article

Full-text available

May 2024
RES POLICY

This study examines how university-firm (U-F) interactions affect innovation speed in science-intensive small and medium-sized firms (SISMEs). We distinguish between formal and informal U-F interactions and build on dynamic capability theory to argue that (1) U-F R&D alliances enhance innovation speed through firm-level entrepreneurial orientation (EO), and (2) frequent U-F informal contacts weaken the effects of U-F R&D alliances on innovation speed. Analyzing a sample of 268 SISMEs from 10 science parks in China, the results of the partial least squares structural equation modeling (PLS-SEM) support our hypotheses. Furthermore, fuzzy-set qualitative comparative analysis (fsQCA) identifies various configurations of U-F R&D alliances, U-F informal contacts and EO, along with other organizational, science park and environmental conditions, that lead to higher or lower innovation speed in SISMEs. Our findings offer valuable theoretical and practical insights, advancing our understanding of the complex relationship between U-F interactions and innovation speed in SISMEs.

Objectives of University Industry Collaboration in Nepal: Observation of Students, Professors and Industrial Leaders

Article

Dec 2023

The collaborative relationship between university and industry is becoming crucial research phenomena for professors, students, industrial leaders, government and the general public in the advanced developing as well as underdeveloped economies. It has become more imperative to the management schools and institutes and research centers. In the prevailing knowledge economy universities are called to produce innovative and creative manpower ready to job in the organizations. This research was conducted in Nepalese business management schools operating under the three leading universities, namely Tribhuvan University, Pokhara University and Kathmandu University. Apart from the students and professors, the perception of business community leaders also has been included in regard to attainability of the objectives of university industry collaboration in Nepal. The objective of the study was to find the perception of students, course facilitators and industrial leaders about the level of attainability of the stated objective of UIC. With some differences all respondentsfound the objectives of UIC are attainable in Nepal. Such kind of research is important to the universities, especially business management educators, the industries and the government to frame education policy and curriculum.

Recruiting Talent Through Entrepreneurs’ Social Vision Communication

Article

Full-text available

Apr 2023

For-profit social ventures are proliferating. They often communicate social visions, presenting an ideal future where the ventures resolve environmental or societal issues. We study whether social vision communication helps a startup to recruit talent—a fundamental problem for growth. We argue that jobseekers are less likely to apply to ventures communicating a social vision as they perceive reduced career advancement opportunities. We conducted two complementary studies to test our theory. Study 1 enlisted data from a job board for startups to show that ventures communicating a social vision receive 46.3% fewer job applications. Study 2 replicated this finding in a field experiment that further reveals the underlying mechanism: social vision communication limits jobseekers’ perceived career advancement opportunities. Both studies show that higher remuneration can compensate the negative effect of social vision communication. Our findings advance research on purpose-driven organizations, human resources, entrepreneurship, and vision communication to caution entrepreneurs against social vision communication as a recruitment strategy. Supplemental Material: The online appendix is available at https://doi.org/10.1287/orsc.2023.1671 .

Function Mechanism of Intellectual Property Capability on Relay Innovation Based on CWLBGSO-DAG-Bootstrap SEM: Mediating Effect of Knowledge Matching and Moderating Effect of Relationship Learning

Article

Full-text available

Jun 2022
Comput Intell Neurosci

Taking unicorn enterprises as the research objects, this study constructs the conceptual model of function mechanism of intellectual property capability on relay innovation performance and introduces knowledge matching as the mediating variable and relationship learning as the moderating variable. Hybrid methods composed of fusion of rough set and binary firefly algorithm based on community weak connection mechanism (CWLBGSO), directed acyclic graph (DAG), and Bootstrap structural equation modeling (Bootstrap SEM) are used to carry out empirical analysis of the conceptual model of function mechanism. The empirical analysis results show that strong intellectual property capability has a positive role in promoting the relay innovation performance of unicorn enterprises, knowledge matching plays the positive mediating role between intellectual property capability and relay innovation, relationship learning plays the moderating role of the relationships between intellectual property capability and relay innovation, relationship learning plays the moderating role of the relationships between intellectual property capability and knowledge matching, and relationship learning plays the moderating role of the relationships between knowledge matching and relay innovation.

The motivations of stakeholders when developing university industry collaborations in an Australian university: three case studies

Article

Mar 2023

The need for universities and industry to collaborate has been well established. In this paper, we present three cases studies, including 14 semi-structured interviews, across a range of disciplines. In the design and analysis of the study, we apply management and political science theories to better understand how stakeholder’s motivations influence the development of university industry collaboration within an Australian university environment. We present our findings across cases, and through five emerging themes: (a) stakeholder – trust and respect; (b) business alignment; (c) employability – mutual benefits; (d) commitment to cause – alignment of values; and (e) innovation/research focus. This study provides real-world insight into influencing the development of university industry collaborations enhancing the student experience, and potentially impacting employability after graduation.

This is what you came for? University–industry collaborations and follow‐on inventions by the firm

Article

Nov 2022

The positive impact of university‐industry collaborations on firms’ innovative performance is well recognized. However, to date, the existing heterogeneity within university‐industry collaboration processes and the sources of value creation underlying the resulting inventions are left underexplored. As a result, our understanding as to why some of the joint inventions resulting from such collaborations turn out to present a more fertile source of follow‐on developments and value for the collaborating firm than others is limited. The present paper sheds light on this question through the application of a knowledge recombination perspective. Hence, we open the black box of innovation and put a spotlight on the knowledge components that make up the joint inventions resulting from university‐industry collaborations. We evaluate how the nature—scientific vs. technological—and origins—internal vs. external to the collaborating partners—of these knowledge components relate to the inventive impact of the partners’ joint invention. Examining a sample of 9,102 USPTO co‐patents, joint inventions created through university‐industry collaborations are shown to be most fertile as a source of firm follow‐on inventions when they are the result of a recombination process that includes technological knowledge components stemming from both collaborating partners. This effect is most pronounced when the partners’ technological knowledge contributions are moderately similar. In contrast, when the joint technology development takes place in a technology domain that is novel to the firm, the resulting joint inventions are most fertile as a source of firm follow‐on inventions when the university contributes through the input of technological knowledge components situated in exactly this technology domain that is novel to the firm. Remarkably, no evidence for such direct effects is found regarding the partners’ scientific contributions. Together, these findings provide important insights for firms who intend to spur their internal technology development through collaboration with a university partner. This article is protected by copyright. All rights reserved.

The Role of Incentives and Communication in Strategic Alliances: An Experimental Investigation

Article

Full-text available

Jan 2010
STRATEGIC MANAGE J

This paper experimentally examines the determinants of the deviation between potential and realized value creation in strategic alliances. To better understand how decision making in alliances may influence success, we use an experimental design that juxtaposes two important factors that affect alliance members' decisions: economic incentives and communication. The evidence from our experiment sheds light on the relative impact of each, and more importantly, how both factors interact to explain successful outcomes.

The Economics of Science

Article

Full-text available

Sep 1996
J ECON LIT

Paula E. Stephan

This chapter examines the contributions that economists have made to the study of science and the types of contributions the profession is positioned to make in the future. Special emphasis is placed on the public nature of knowledge and characteristics of the reward structure that encourage the production and sharing of knowledge. The role that cognitive and noncognitive resources play in discovery is discussed as well as the costs of resources used in research. Different models for the funding of research are presented. The chapter also discusses scientific labor markets and the extreme difficulty encountered in forecasting the demand for and supply of scientists. The chapter closes with a discussion of the relationship of scientific research to economic growth and suggestions for future research.

The duality of collaboration: inducements and opportunities in the formation of interfirm linkages

Article

Mar 2000
STRATEGIC MANAGE J

Gautam Ahuja

I argue that the linkage‐formation propensity of firms is explained by simultaneously examining both inducement and opportunity factors. Drawing upon resource‐based and social network theory literatures I identify three forms of accumulated capital—technical, commercial , and social —that can affect a firm’s inducements and opportunities to form linkages. Firms possessing these capital stocks enjoy advantages in linkages formation. However, firms lacking these accumulated resources can still form linkages if they generate a radical technological breakthrough. Thus, I identify paths to linkage formation for leading as well as peripheral firms. I test these arguments with longitudinal data on technical collaborative linkages in the global chemicals industry. Copyright © 2000 John Wiley & Sons, Ltd.

A Theory of Marriage: Part I

Article

Jan 1973

G.S. Becker

Evaluating Technological Information and Utilizing it: Scientific Knowledge, Technological Capability and External Linkages in Biotechnology

Article

Jan 1994
J ECON BEHAV ORGAN

Identification in matching games

Article

Nov 2010

Jeremy T. Fox

I study a many-to-many, two-sided, transferable utility matching game. Consider data on matches or relationships between agents but not on the choice set of each agent. I investigate what economic parameters can be learned from data on equilibrium matches and agent characteristics. Features of a production function, which gives the surplus from a match, are nonparametrically identified. In particular, the ratios of complementarities from multiple pairs of inputs are identified. Also, the production function is identified up to a positive monotonic transformation.

Commercializing knowledge: University science, knowledge capture, and firm performance in biotechnology

Article

Jan 2002

Commercializing knowledge involves transfer from discovering scientists to those who will develop it commercially. New codes and formulae describing discoveries develop slowly - with little incentive if value is low and many competing opportunities if high. Hence new knowledge remains naturally excludable and appropriable. Team production allows more knowledge capture of tacit, complex discoveries by firm scientists. A robust indicator of a firm's tacit knowledge capture (and strong predictor of its success) is the number of research articles written jointly by firm scientists and discovering, "star" scientists, nearly all working at top universities. An operationally attractive generalization of our star measure - collaborative research articles between firm scientists and top research university scientists - replicates the impact on firm success. In panel analyses, publications by firm scientists with stars and/or top 112 university scientists increase the number and citation rate for firm patents. Further, star articles increase these rates significantly more than other top 112 university scientists' articles. Cross-sectional analyses of products and employment show a similar pattern of positive effects on firms' success of collaborations with stars or top university scientists, but estimates of differential effects are nonrobust due to multicollinearity. Venture capital funding has significant, usually positive effects on firm success.

The patent paradox revisited: An empirical study of patenting in the U.S. semiconductor industry, 1979-1995

Article

Mar 2001

We examine the patenting behavior of firms in an industry characterized by rapid technological change and cumulative innovation. Recent survey evidence suggests that semiconductor firms do not rely heavily on patents to appropriate returns to R&D. Yet the propensity of semiconductor firms to patent has risen dramatically since the mid-1980s. We explore this apparent paradox by conducting interviews with industry representatives and analyzing the patenting behavior of 95 U.S. semiconductor firms during 1979-1995. The results suggest that the 1980s strengthening of U.S. patent rights spawned "patent portfolio races" among capital-intensive firms, but it also facilitated entry by specialized design firms.

The Burden of Knowledge and the "Death of the Renaissance Man": Is Innovation Getting Harder?

Article

Jan 2009

Benjamin Jones

This paper investigates a possibly fundamental aspect of technological progress. If knowledge accumulates as technology advances, then successive generations of innovators may face an increasing educational burden. Innovators can compensate through lengthening educational phases and narrowing expertise, but these responses come at the cost of reducing individual innovative capacities, with implications for the organization of innovative activity - a greater reliance on teamwork - and negative implications for growth. Building on this "burden of knowledge" mechanism, this paper first presents six facts about innovator behaviour. I show that age at first invention, specialization, and teamwork increase over time in a large micro-data set of inventors. Furthermore, in cross-section, specialization and teamwork appear greater in deeper areas of knowledge, while, surprisingly, age at first invention shows little variation across fields. A model then demonstrates how these facts can emerge in tandem. The theory further develops explicit implications for economic growth, providing an explanation for why productivity growth rates did not accelerate through the 20th century despite an enormous expansion in collective research effort. Upward trends in academic collaboration and lengthening doctorates, which have been noted in other research, can also be explained in this framework. The knowledge burden mechanism suggests that the nature of innovation is changing, with negative implications for long-run economic growth.

Two Sided Matching: A Study in Game-Theoretic Modeling and Analysis.

Article

Nov 1992

Value creation in university-firm research collaborations: A matching approach

Abstract and Figures

Recommended publications

Corporate Effectuation

Impact of relationship strength and integration capability on the efficiency of knowledge transfer a...

COOPERATION, COLLABORATION AND PARTNERSHIPS: A DISCUSSION OF FAILURE FACTORS IN STRATEGIC ALLIANCES

The Start Uprising: Eighteen Months of the Startup America Partnership