Powell, Koput, & Smith-Doerr 1996

Inter-organizational Collaboration and the Locus of Innovation: Networks of Learning in Biotechnology

Abstract

Argument

When the knowledge base of an industry is both (1-1) complex and (1-2) expanding and the sources of expertise are (2) widely dispersed

-> the locus of innovation will be found in networks of learning, rather than in individual firms.

The fundamental and pervasive concern with access to knowledge -> The large-scale reliance on inter-organizational collaborations in the biotechnology industry

Scheme

A network approach to organizational learning 

Firm-level longitudinal hypotheses that link (1) research and development alliances, (2) experience with managing interfirm relationships, (3) network position, (4) rates of growth, and (4) portfolios of collaborative activities

Test

A sample of dedicated biotechnology firms in the years 1990-1994

Pooled, within-firm, time series analyses support a learning view 

Broad implications for future theoretical and empirical research on organizational networks and strategic alliances

* Introduction

Past

Corporate partnering and reliance on various forms of external collaboration (Hergert and Morris, 1988; Mowery, 1988; Hagedoorn, 1990, 1995; Badaracco, 1991; Hagedoorn and Schakenraad, 1992; Gulati, 1995).

Firms organized research and development (R&D) internally and relied on outside contract research only for relatively simple functions or products (Mowery, 1983; Nelson, 1990a).

Today

Companies in a wide range of industries are executing nearly every step in the production process, from discovery to distribution, through some form of external collaboration

Various types of interfirm alliances:

From R&D partnerships to equity joint ventures to collaborative manufacturing to complex co-marketing arrangements. 

The most common rationales: Combine complementrary assets

Some combination of risk sharing, obtaining access to new markets and technologies, speeding products to market, and pooling complementary skills (Kogut, 1989; Kleinknecht and Reijnen, 1992; Hagedoorn, 1993; Mowery and Teece, 1993; Eisenhardt and Schoonhoven, 1996)

+ Industries in which knowledge is developing rapidly

Correlation

R&D intensity of level of technological sophistication of industries

& The intensity and number of alliances in those sectors (C. Freeman 1991; Hagedoorn 1995)

Two forms of technological change

1) When advances build on existing know-how, established firms reap the bulk of the benefits. 

2) But when new discoveries create technological discontinuities or radical breaks from previously dominant methods, incumbents can be robbed of many of their advantages

+ Moreover, new kinds of organizational practices may emerge to exploit these novel developments (Schumpeter, 1934; Abernathy and Clark, 1985; Tushman and Anderson, 1986; Tushman and Rosenkopf, 1992)

Scope

Biotechnology represents a competence-destroying innovation because it builds on a scientific basis (immunology and molecular biology) that differs significantly from the knowledge base (organic chemistry) of the more established pharmaceutical industry. Consequently, biotech provides enhanced research productivity, with less risk and with more speed and potentially higher rewards (Weisbach and Moos, 1995).

To examine the organizational arrangements that have arisen in response to the technological ferment generated by biotechnology

Focus on forms of collaboration undertaken by dedicated biotechnology firms 

Assess the contribution of cooperative ventures to organizational learning

To map the network structure of this emerging industry and explain the purposes served by the extensive connections that typify the field.

COLLABORATION AND ORGANIZATIONAL LEARNING

Regime

A regime of rapid technological development

: research breakthroughs are so broadly distributed that no single firm has all the internal capabilities necessary for success.

-> The rewards go to the swifts

Logic

New technologies are both a stimulus to and the focus of a variety of cooperative efforts

Cooperative efforts that seek to reduce the inherent uncertainties associated with novel products or markets

Collaboration enhances organizational learning (Hamel, 1991; Dodgson, 1993)

1) Strategic (Teece, 1986; Williamson, 1991). 

The choice to pool resources with another organization depends on calculations involving risk versus return

Reliance on external partners involves hazards (Powell, 1990; Sabel, 1993)

Barriers to effective collaboration: Lack of trust between the parties, difficulties in relinquishing control, the complexity of a joint project, and differential ability to learn new skills

Who is an ally and who is not.

In those industries in which interfirm agreements are relatively frequent, there can be competitive confusion about who is an ally and who is not.

Depends on (1) each partner's size and position in the "value-chain," (2) the level of technological sophistication, (3) resource constraints, and (4) prior experiences with alliances.

The form of collaboration is purported to vary according to the specific types of skills and resources to be exchanged (Hennart, 1988; Pisano, 1989; Parkhe, 1993).

Posed this way, the decision to collaborate is a variant of the make-or-buy decision, framed largely in terms of transaction cost economics.

Turn to collaboration to acquire resources and skills they cannot produce internally, when the hazards of cooperation can be held to a tolerable level

2) Learning is a social construction process (Brown and Duguid, 1991).

What is learned is profoundly linked to the conditions under which it is learned

Sources of innovation do not reside exclusively inside firms; instead, they are commonly found in the interstices between firms, universities, research laboratories, suppliers, and customers (Powell, 1990)

The degree to which firms learn about new opportunities is a function of the extent of their participation in such activities (Levinthal and March, 1994)

# Then how do focal actors participate in the community

Brown and Duguid (1991: 48) summarized this view nicely by stating that learning is about becoming a practitioner, not learning about a practice

Von Hippel (1988) has shown that the trading of know-how often requires the establishment of long-term relationships in which exchange occurs within a learned and shared code.

Neat distinction

March (1991) captured these divergent views of learning in his discussion of the differences between exploration and exploitation in organizational learning

Intertwined

Exploitation and exploration, and calculation and community are intertwined.

Organizational learning is both a function of access to knowledge and the capabilities for utilizing and building on such knowledge

We follow Nelson (1990b) and Stinchcombe (1990) in arguing that organizational arrangements that provide access to knowledge quickly and reliably produce competitive advantage.

Rather than seeing such activity as calculative or strategic, we draw on a long line of research that stresses the centrality of building skills and exercising routines in organizations (Cyert and March, 1963; Nelson and Winter, 1982; Stinchcombe, 1990).

The complex reality of rapidly developing fields, in which knowledge is both sophisticated and widely dispersed, transcends the simple calculation of a make-or-buy decision.

More important than the number of authors is the diversity of sources of innovation and the wide range of different organizations involved in these breakthrough publications.

Learning through Networks

Argument: when knowledge is broadly distributed and brings a competitive advantage, the locus of innovation is found in a network of inter-organizational relationships (Powell and Brantley, 1992)

Companies must be expert at both in-house research and cooperative research with such external partners as university scientists, research hospitals, and skilled competitors

A network analog to Cohen and Levinthal's (1989, 1990) concept of "absorptive capacity."

A firm with a greater capacity to learn is adept at both internal and external R&D, thus enabling it to contribute more to collaboration as well as learn more extensively from such participation. 

Internal capability and external collaboration are not substitutes for one another, but complementary (Mowery and Rosenberg, 1989; Arora and Gambardella, 1994)

Networks of learning

1) Interorganizational collaborations are not simply a means to compensate for the lack of internal skills (Collaboration further develops and strengthens those internal competencies)

2) nor should they be viewed as a series of discrete transactions (deepen their ability to collaborate by instantiating and refining routines for synergistic partnering)

# These approaches are interesting

The development of cooperative routines goes beyond simply learning how to maintain a large number of ties. Firms must learn how to transfer knowledge across alliances and locate themselves in those network positions that enable them to keep pace with the most promising scientific or technological developments

Collaborations in high-tech industries typically reflect more than just a formal contractual exchange

# Formal contract exchange

"the tip of the iceberg-it excludes dozens of handshake deals and informal collaborations, as well as probably hundreds of collaborations by our company's scientists with colleagues elsewhere."

"In order for industrial research organizations to be in close contact with new advances in basic science, it is important . .. to be an active participant at the leading edge of world science. Effective technical interchange requires that the industrial organization have its own basic research results . . . to use as a currency of exchange"

# Informal = R&D collaboration (based on participation in technology development with internal R&D)

R&D collaboration is both an admission ticket to an information network and a vehicle for the rapid communication of news about opportunities and obstacles

When such projects lead to the possibility of a new medicine, relationships become formal and contractual.

# Evolution of the tie

R&D alliances serve as a platform for diverse network activity

When the sources of expertise are disparate, 

Collaborative R&D opens an organization's eyes to the need for accessing ideas and information from a variety of sources, to exploit the research findings in a commercial context

Both skill and experience are needed to accumulate the capability to benefit from the interdependencies across diverse collaborative behaviors. 

In addition, experience at collaborating is necessary to manage a diverse portfolio of ties. 

Hence, we argue that firms learn from exploration and experience how to recognize and structure synergies across different types of alliances

Hypothesis 1: The greater the (a) number of research and development alliances and (b) experience at managing R&D and other types of collaborations a firm has at a given time, the greater the number of non-R&D collaborations it subsequently pursues; and, in turn, the more diverse its future portfolio of ties will become, controlling for prior levels

# R&D tie & Collaboraiton experience -> More non R&D collaboration -> Diverse portfolio

An organization simultaneously learns which collaborations to pursue and how to function within a context of multiple cooperative ventures

Collaboration becomes emergent- stemming from ongoing relationships-informal, and nonpremeditated (Von Hippel, 1988; Hakansson, 1990).

We contend that firms with more experience have more ties and that the ties they have provided more central connectedness. Moreover, experience with diverse research-driven networks and the connections that experience brings determine how well situated a firm becomes

Hypothesis 2: The greater (a) the number of R&D alliances, (b) the diversity of ties, and (c) the experience at managing R&D collaborations or other ties that a firm has at a given time, the more centrally connected the firm subsequently becomes, controlling for the total number of ties and prior connectedness

# R&D tie, Diverse tie, Collaboration experience -> Centrally connected (control total tie and prior connectedness)

Central connectedness shapes a firm's reputation and generates visibility, producing access to resources via benefit-rich networks

Firms more centrally located should have more timely access to promising new ventures, while those with more collaborative experience should be better positioned to exploit them.

Hypothesis 3: The greater a firm's (a) centrality in a network of relationships and (b) experience at managing ties at a given time, the more rapid its subsequent growth, controlling for prior growth.

# Centrality, Collaboration experience -> Faster growth (controlling prior growth)

Finally, we expect that these returns from experience with diverse collaborative activity should elicit positive feedback.

Centrality in a network facilitates common understandings and shared principles of cooperation, thus enhancing further exchange

Hypothesis 4: The greater a firm's centrality in a network of relationships at a given time, the greater its number of subsequent R&D collaborations, controlling for prior collaborative R&D activity.

# Centrality -> More R&D (controlling prior R&D)

The Biotechnology Industry

In many respects, biotechnology is not an industry per se, but a set of technologies with the potential to transform various fields-pharmaceuticals, chemicals, agriculture, veterinary science, medicine, even waste disposal. 

Many researchers (e.g., Barley and Freeman, 1992; Amburgey, Shan, and Singh, 1994) treat the wide array of biotechnology companies as comparable. 

In contrast, we intentionally restrict our attention to only those for-profit firms engaged principally in human therapeutics and diagnostics, hereafter referred to as dedicated biotech firms, or DBFs. # DBF definition

The therapeutics sector is driven by a different research agenda and operates within a distinctive regulatory regime

We intentionally omit firms engaged only in agricultural, veterinary, or bioremediation activity and exclude peripheral companies that produce only equipment, materials, or test kits for the industry.

As the human biotechnology industry developed in the 1980s, it became clear that the full range of required skills (e.g., basic research, applied research, clinical testing procedures, manufacturing, marketing and distribution, and knowledge of and experience with the regulatory process) could not be easily assembled under one roof.

These practices help promote a common technological community between universities and DBFs.

Venture capitalists fueled most of the initial discoveries and guided many firms through their early years. But moving from basic research to product development required not only- lots of money, it also demanded expertise in conducting extensive clinical trials and securing federal regulatory approval.

Neither universities nor DBFs have been well equipped for these tasks, but large pharmaceutical firms most certainly have

So the various participants in biotech have turned to joint ventures, research agreements, minority equity investments, licensing, and various kinds of partnerships to make up for their lack of internal capabilities and resources (Pisano, 1989, 1991; Arora and Gambardella, 1990, 1994; Powell and Brantley, 1992).

Consequently, circumstances of mutual need develop. Small firms require large firms' financial support and regulatory savvy, while larger corporations desire access to the research prowess of smaller companies.

METHOD

Data

To explain the pattern of inter-organizational agreements that structure learning in the field of biotechnology

Agreements: Formal ties with expensive investments

Relational database that contains separate files for (1) DBFs in human therapeutics and diagnostics, (2) the formal contractual, interorganizational agreements involving DBFs, and (3) the partners to these agreements.

Data: founding date, employment levels, sources of financing, and collaborative agreements, which we treated as tie

Agreement description: Types of tie, Typical partners 

Began assembling the database in 1990, using Bioscan, an independent industry directory published six times a year that lists a great range of organizations (domestic and foreign, commercial, nonprofit, or government-owned, biotech, and diversified health care corporations). 

Bioscan lists information on a firm's ownership, its current products, and its research in progress.

For the five-year period 1990-1994

When information was missing from Bioscan, we consulted numerous other industry directories, such as various editions of Genetic Engineering and Biotechnology Related Firms Worldwide, Dun & Bradstreet's Who Owns Whom?, and listings in Moody's and Standard & Poor's. In addition, we consulted annual reports, Securities and Exchange Commission filings, and, when necessary, made phone calls to companies.

Sample: Research-driven DBFs in human therapeutics resulted in a sample of approximately 225 independently owned companies

Operationalizations and Measures 

Dependent variables. 

Hypotheses 1-4 predict the subsequent number and diversity of ties, network position in terms of central connectivity, and rates of growth for firms in our sample.

Number of R&D ties at time t + 1: The number of research and development ties a firm has captures the extent of its involvement in the core activities of the industry. 

# R&D tie

Number of ties of each type (other than R&D) at time t + 1: We used the disaggregated number of ties for each stage of the product life-cycle (see Table 1) to capture the non- R&D network activity in which firms are involved. These efforts as a way to exploit R&D discoveries and thus as a complement to R&D networks

# Non R&D tie

Network portfolio diversity at time t + 1: The range of ties that a firm is engaged in at any given time reflects a firm's portfolio of collaborative activities (Concentrated) 0~7/8 (Diverse)

# Diversity

Central connectivity at time t + 1. Before describing the measures of network centrality that we used, a few remarks are in order

Used only direct agreements between biotech firms and their partners.

+ Counted a connection between two DBFs when there was a direct tie (degree one) and when the DBFs were linked (at degree distance two) through a common partner.

# Direct tie, Strucutral equivalence confused

For similar reasons, in measuring centrality we did not discriminate among connections involving different function

Membership in the main component at time t + 1: A component is generally defined as a maximally connected subgraph, or a set of points that are connected to one another by paths of any distance (Scott, 1991: 104)

MainComp, that takes the value of 1 if a firm is in the main connected component and 0 if not.

Degree centrality at time t + 1: Centrality is a measure of how well connected, or active, a firm is in the overall network.

The number of other firms connected to that firm, ignoring how well those partners are connected

Closeness centrality at time t + 1: We also used a measure of centrality based on the concept of closeness (L. Freeman, 1979), which captures independence from the control of others. 

Closeness centrality was computed for each firm as the reciprocal of the sum of the degree distance to each other firm

Growth at time t + 1. We indexed growth in two ways. 

1) We used the reported number of employees at time t + 1 as a measure of size.

2) We also created a dummy variable, public, that takes on the value of 1 if the firm is publicly traded at time t + 1 and 0 otherwise.

Independent variables. We based our predictions on prior measures of collaborative research activity and experience, non-R&D network experience, diversity of alliances, and network 'centrality. 

We used the number of R&D ties, network portfolio diversity, and central connectivity as defined above, but at time t. 

We also introduced measures of network experience

Collaborative R&D experience at time t was measured as the time since inception of a firm's first R&D alliance. This was computed for each firm in each year as the current date minus the date at which the firm's initial R&D tie was formalized

Non-R&D network experience at time t, an additional measure of experience at managing ties, is the time since inception of a firm's first tie for any purpose other than R&D. This was computed for each firm at time t as the current date minus the date at which the firm's first nonresearch alliance was formalized.

Control variables

The prior total number of ties: Total ties at time t is the aggregate number of ties of all types and serves as a control in our predictions involving central connectivity

Firm age: firm's calendar age at time t

Size: the number of employees

Statistical Methods

Five years of cross-sectional records

Panel regression model

Fixed firm effect

To do so, we included fixed year effects

RESULTS

Hypothesis 1 predicted positive effects of the number of R&D ties and network experience at time t on the number of ties of each other type at time t + 1 and, in turn, on the diversity of ties at time t + 1.

Hypothesis 2 predicted positive effects of the number of R&D alliances, portfolio diversity, and network experience at time t on how centrally connected a firm becomes at time t + 1.

Hypothesis 3 predicted positive effects of the measures of central connectedness and network experience at time t on size and public at time t + 1.

First, age has no effect, while size is an outcome, not a predictor of network behavior. Growth is a process that requires time. Unlike biotic species, however, organizational growth is not programmed from age. Rather, it is the initiation of collaboration that sets the growth clock in motion, with centrality as a further stimulus. Second, network position (central connectedness) has reciprocal influences on R&D alliances, investment ties, and total collaboration. We have argued and shown that R&D ties, experience, and diversity produce central connectedness.

DISCUSSION

limitation

This is a young industry, with many of the founding firms not quite 20 years old. 

We lack data on firms that were founded and subsequently failed before 1990.

# Selection bias

Patterns that have emerged over a five-year period, albeit a crucial one in the industry's emergence from infancy to adolescence, do not provide the full story

First, numerous analysts have commented on the industry's very low mortality rate in its early years (Barley and Freeman, 1992; Burrill and Lee, 1993). 

Second, the rules of the industry have developed and become elaborated during precisely the period we are studying.

Finally, we think there are interesting and suggestive points of commonality between our measures of network experience and centrality and various, albeit less precise rankings of success in the industry.

Table 6: Sales in 1993 for the top-ten biotech products

A possible linkage between networks of learning and firm performance

Four firms are responsible for developing these medicines, although for six of the cases, a larger company is responsible for sales and marketing.

Using our network measures as rankings, we find three of these firms on the list of the most central DBFs, and all four are among those firms most steeped in experience.

Table 7: How biotech firms compare with firms in other industries, as well as with universities and research laboratories

Business Week publishes an annual R&D scorecard, examining the level of R&D investment by U.S. firms

The first six are all biotechs, and of this group, five of the companies appear on either our experience or centrality lists, and three companies (once again, Biogen, Chiron, and Genentech) appear on both lists.

Table 7: citation data to emphasize biotechnology's position in the research community

The data in Tables 6 and 7 combine with our statistical results to suggest that being centrally connected is necessary to achieve valued organizational outcomes.

On all these dimensions of production, our measures of learning are associated with those firms that have, thus far, been industry leaders.

# Being a market leader?

CONCLUSION

Several standard organizational characteristics, such as age and size, appear to be ancillary in accounting for patterns of collaboration. 

Neither growth nor age reduced the propensity to engage in external relationships. 

Instead, age, per se, proved unimportant in the context of network experience, and size was an outcome rather than a determinant of partnerships.

We found a path-dependent (Arthur, 1990) cycle of learning in which an early choice of exploration elicited positive feedback

In our view, the development of absorptive capacity (Cohen and Levinthal, 1989, 1990) and skill at managing collaborations, as well as the increased awareness of new projects and reputation as a valuable partner, are all serendipitous benefits of collaboration.

# What is the unit of analysis?

Equally important are changes at the network and industry levels.

A field is becoming more tightly connected not in spite of, but because of a marked increase in the number of partners involved in alliances with DBFs. Network density (based on connections) has increased 50 percent from .06 to .09, while the number of firms has dropped slightly (from 241 to 226).

First, firms are increasingly using ties to enhance the inflow of specific information, resources, and products. 

Second, firms are becoming much more adept at and reputed for the general practice of collaboration with diverse partners.

This, in turn, promotes a sense of community-level mutualism (Barnett, 1990)

As a result of this reciprocal learning, both firm-level and industry-level practices are evolving, with boundaries becoming ever more permeable

In contrast to the much-discussed liability of newness hypothesis (Stinchcombe, 1965; Hannan and Freeman, 1989), there appears to be a liability of unconnectedness (Baum and Oliver, 1992) at work in biotechnology, and other fields in which intellectual developments are expanding rapidly.

When the sources of knowledge are disparate and the pathways of technological development uncharted, we would expect the emergence of networks of learning.