Advancing Knowledge and the Knowledge Economy

Kahin_jkt.qxd

8/22/06

9:13 AM

Page 1

Advancing Knowledge and the Knowledge Economy edited by Brian Kahin and Dominique Foray The revolution in information technology transforms not only information and its uses but knowledge and the ways we generate and manage it. Knowledge is now seen as input, output, and capital, even if imperfectly accounted for or understood. Many businesses and public agencies are convinced that knowledge can be managed in sophisticated, rational ways and that networking and information technology are essential tools for doing so. In this collection, experts from North America and Europe look at the transformation of knowledge in the global economy in light of the rapid changes in information technology, the resulting explosion of data, the recognition of intangibles as sources of value and liability, and the increasingly blurred distinction between private and public knowledge. The appeal of the Internet as boundary-spanning knowledge infrastructure, bridging all sectors of the economy, is shadowed by another infrastructure of rights-based contracts, practices, and institutions. The contributors address the ways in which the processes for creating and organizing knowledge interact with information technology, business strategy, and changing social and economic conditions. They discuss the balkanization that results from the complexity of the knowledge economy, the variety of knowledge resources, the great diversity of institutional and market contexts, and competing models of control and cooperation — and of proprietary and nonproprietary knowledge. Brian Kahin is Visiting Professor at the University of Michigan. He is a coeditor of Transforming Enterprise (MIT Press, 2004) and many other books. Dominique Foray holds the Chair in Economics and Management of Innovation and is Director of the College of Management of Technology at École Polytechnique Fédérale de Lausanne. He is the author of The Economics of Knowledge (MIT Press, 2004).

Transforming Enterprise The Economic and Social Implications of Information Technology

edited by William H. Dutton, Brian Kahin, Ramon O’Callaghan, and Andrew W. Wyckoff Innovators across all sectors of society are using information and communication technology to reshape economic and social activity. Even after the boom — and despite the bust — the process of structural change continues across organizational boundaries. Transforming Enterprise considers the implications of this change from a balanced, post-bust perspective.

The MIT Press Massachusetts Institute of Technology Cambridge, Massachusetts 02142 http://mitpress.mit.edu

0-262-61214-3

978-0-262-61214-2

Kahin and Foray, editors

O F R E L AT E D I N T E R E S T

Advancing Knowledge and the Knowledge Economy

B U S I N E S S / C O M P U T I N G / T E L E C O M M U N I C AT I O N S

Advancing Knowledge and the Knowledge Economy EDITED BY

Brian Kahin and Dominique Foray

Advancing Knowledge and the Knowledge Economy

Advancing Knowledge and the Knowledge Economy

edited by Brian Kahin and Dominique Foray

The MIT Press Cambridge, Massachusetts London, England

( 2006 Massachusetts Institute of Technology All rights reserved. No part of this book may be reproduced in any form by any electronic or mechanical means (including photocopying, recording, or information storage and retrieval) without permission in writing from the publisher. MIT Press books may be purchased at special quantity discounts for business or sales promotional use. For information, please email [email protected] or write to Special Sales Department, The MIT Press, 55 Hayward Street, Cambridge, MA 02142. This book was set in Sabon on 3B2 by Asco Typesetters, Hong Kong, and was printed and bound in the United States of America. Library of Congress Cataloging-in-Publication Data Advancing knowledge and the knowledge economy / edited by Brian Kahin and Dominique Foray. p. cm. Conference papers. Includes bibliographical references and index. ISBN-13: 978-0-262-11300-7 (alk. paper)—ISBN 978-0-262-61214-2 (pbk. : alk. paper) ISBN-10: 0-262-11300-7 (alk. paper)—ISBN 0-262-61214-3 (pbk. : alk. paper) 1. Knowledge management—Congresses. 2. Information technology—Economic aspects— Congresses. I. Kahin, Brian. II. Foray, Dominique. HD30.2.A345 2006 658.40 038—dc22 2005058229 10

9

8 7 6

5 4

3 2

1

Contents

Preface

ix

1

Prospects for Knowledge Policy Brian Kahin

1

2

Optimizing the Use of Knowledge Dominique Foray

3

OECD Work on Knowledge and the Knowledge Economy Berglind A´sgeirsdo´ttir

I

Measuring Knowledge

4

Measuring Knowledge and Its Economic Effects: The Role of Official Statistics 27 Fred Gault

5

Assessing Innovation Capacity: Fitting Strategy, Indicators, and Policy to the Right Framework 43 Reinhilde Veugelers

II

Knowledge Communities

6

Interactive Learning, Social Capital, and Economic Performance Bengt-A˚ke Lundvall

7

Social Capital, Networks, and Communities of Knowledge Tom Schuller

8

Knowing Communities in Organizations Patrick Cohendet

9 17

25

61

91

75

63

vi

Contents

III The Changing Role of Institutions 9

111

Epistemic Infrastructure in the Rise of the Knowledge Economy Margaret Hedstrom and John Leslie King

10 Universities and the Knowledge Economy Robin Cowan

113

135

11 The Impact of ICT on Tertiary Education: Advances and Promises Kurt Larsen and Ste´phan Vincent-Lancrin

151

12 The Bayh–Dole Act of 1980 and University–Industry Technology Transfer: A Policy Model for Other Governments? 169 David C. Mowery and Bhaven Sampat IV Knowledge and Place

191

13 The Changing Dynamics of the Global Market for the Highly Skilled Andrew Wyckoff and Martin Schaaper 14 Knowledge in Space: What Hope for the Poor Parts of the Globe? Jan Fagerberg V

New Models of Innovation

193

217

235

15 Democratizing Innovation: The Evolving Phenomenon of User Innovation 237 Eric von Hippel 16 Innovation, Experimentation, and Technological Change Stefan Thomke 17 Knowledge, Platforms, and the Division of Labor W. Edward Steinmueller

257

277

18 Between ‘‘Knowledge’’ and ‘‘The Economy’’: Notes on the Scientific Study of Designs 299 Carliss Y. Baldwin and Kim B. Clark VI Models of Control and Cooperation

329

19 Patent Quantity and Quality: Trends and Policy Implications Dietmar Harhoff

331

Contents

vii

20 Blurred Boundaries: Tensions Between Open Scientific Resources and Commercial Exploitation of Knowledge in Biomedical Research 351 Iain M. Cockburn 21 The Economics of Technology Sharing: Open Source and Beyond Josh Lerner and Jean Tirole

369

22 ‘‘Open and Collaborative’’ Biomedical Research: Theory and Evidence Arti K. Rai 23 Critical Tensions in the Evolution of Open Source Software Brian Fitzgerald VII Emerging Infrastructure

415

429

24 Toward a Cyberinfrastructure for Enhanced Scientific Collaboration: Providing Its ‘‘Soft’’ Foundations May Be the Hardest Part 431 Paul A. David 25 Cyberinfrastruture-in-the-Making: Can We Get There from Here? C. Suzanne Iacono and Peter A. Freeman Contributors and Affiliations Index 481

479

455

391

Preface

At an elemental level, it is easy to see that information technology transforms information and the way we use it. The impact on knowledge is less apparent, in part because knowledge is a more complex and elusive matter than information. Nonetheless, we now see many businesses and public agencies taken with the idea that knowledge can be managed in rational ways and that networking and information technology are essential tools for doing so. We see knowledge assets embraced as a form of capital that is imperfectly understood and accounted for—even though it differentiates position and potential in an intensely competitive world, for nations as well as firms and institutions. Advancing Knowledge and the Knowledge Economy is the third of a series of projects that examine the economic and social implications of information technology. Beginning with Understanding the Digital Economy in 1999, followed by Transforming Enterprise in 2003, we have sought to showcase disinterested analysis in a way that helps a broader audience understand how information technology is changing the way we live and work. From a broad assessment at the peak of the Internet boom, we have moved toward a deeper, longer-term understanding of transformative processes. In this project, we look specifically at the transformation of knowledge as a resource underlying all fields of endeavor. To this end, we have sought collaborators in Europe. Not only is there a rich body of work on the economics of knowledge in Europe, but the European Commission has paid close attention to the social and economic underpinnings of research and innovation. We were pleased not only to work with Information Society Directorate General again but also, for the first time, with the Directorate General for Research. Special thanks are due to Andrew Sors and Nikolaos Kastrinos of the Strategy and Policy Unit of the Directorate for Social Sciences and the Humanities for their participation and support in bringing European researchers to the conference. We also thank Jesus Villasante of the Information Society Technologies Program for his continuing involvement in this effort.

x

Preface

The Organisation for Economic Cooperation and Development, more commonly known as OECD, played a major role in this project. Advancing Knowledge and the Knowledge Economy was inspired by a panel on the transformation of knowledge at the Transforming Enterprise conference. My coeditor Dominique Foray was then at the OECD Centre for Educational Research and Innovation, where he pursued this project on behalf of OECD, and he continued to contribute his time and vision from his new position at the E´cole Polytechnique Fe´de´rale de Lausanne. Kurt Larsen and Andrew Wyckoff at OECD also played important roles in guiding the project, and we are grateful to Deputy Secretary General Berglind Asgeirsdottir for her conference keynote and introductory essay. Principal funding for the Advancing Knowledge project was provided by the Digital Society and Technology Program and the Digital Government Program of the National Science Foundation.1 We would like to thank Bill Wulf, president of the National Academy of Engineering, for sponsoring the use of the National Academies auditorium and for opening the conference. We would also like to thank Comptroller General of United States David Walker for his inspiring keynote address, which is not included here but can be found online at: http://www.gao.gov/cghome/ ke01112005/. Finally, we would like to thank the many who contributed in other ways to making this project a success, including our panel moderators and the staff at the University of Michigan, the National Science Foundation, OECD, the National Academies, and the National Coordination Office for Networking and Information Technology Research and Development, Quantum Publishing Services, and The MIT Press. Brian Kahin University of Michigan Note 1. This material is based upon work supported by the National Science Foundation under Grant No. 0444492. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

1 Prospects for Knowledge Policy Brian Kahin

The maxim that the business of everybody is the business of nobody applies with special poignancy to knowledge. There are no ministries of knowledge. Knowledge lacks the cachet of technology and innovation. It lacks the specificity and urgency of information. Knowledge is diverse, complex, and context dependent—and rarely a topic of public discourse. The transformative effects of information technology and the global economic environment are changing the nature and uses of knowledge—and challenging the policy domains in which it plays a critical role. The generation and management of new knowledge is linked to innovation, wealth creation, and economic growth. Europe has conspicuously embraced the goal of becoming the world’s most competitive knowledge-based economy. However, linking knowledge to economic growth is difficult. The value of knowledge can lie in its ‘‘infinite expansibility’’—or in its novelty and enforced scarcity. The multifaceted and multivalent nature of knowledge makes it opaque—for academics and policymakers alike. There is too much to know about knowledge to be able to make intelligent decisions about it. Rapidly changing technological and economic conditions make the problem worse. Knowledge is more important, more multifaceted, more multivalent than ever. In relative terms, we know less and less about more and more. Yet specific changes show the growing significance and need for informed policy perspectives on knowledge. One can point to 

The rise of information technology and the Internet as knowledge infrastructure,



The Internet-enabled explosion of codified knowledge,



Emergence of innovation policy as an essential tool for economic growth,



The growing scope and significance of intellectual property,



Recognition of intangibles as sources of value and liability,

2

Brian Kahin

Disaggregation of the firm and the emergence of markets for knowledge and technology,





Emergence of knowledge management as a practical discipline, and



Increasing movement, sharing, and use of knowledge across boundaries.

While these developments challenge us to come to grips with the fabric of ‘‘knowledge policy,’’ they also show that our understanding of knowledge is changing in ways that are exceedingly difficult to keep in focus. Knowledge policies remain balkanized and isolated under different institutions and areas of expertise. Unlike the information revolution, the knowledge transformation has no analog-todigital shift and no discrete units like bits and bytes. Unlike other products of the information revolution, the transformation of knowledge cannot be readily priced and measured. The unspeakable complexity of the knowledge economy is reflected in the struggle over how to understand, represent, and account for intangible sources of value.1 But this specific and technical debate is only the most visible and persistent manifestation of the problem of generating usable knowledge about knowledge—a challenge whose infinite recursiveness seems to consign it to philosophy rather than social science. We do understand pieces of the transformation from personal experience. Our knowledge about people, firms, and institutions has come to be constructed and framed by websites. We know the extraordinary power of search engines for extracting approximations of contextual knowledge. The success of open source software development offers graphic evidence of the economic power of knowledge networks operating outside (but adjacent to) priced markets and incentives. Knowledge policy, such as it may be, remains widely dispersed in areas as diverse as intelligence and security, K–12 education, healthcare, patents, agency rulemaking, research funding, and the dissemination of agency information. It is hard to imagine a relationship between the established knowledge and basic skills taught in elementary school and the quest for new knowledge in science and technology. Yet K–12 education creates human capital that will serve the knowledge needs of the future. The balkanization of knowledge policy was less at issue in the past because knowledge moved more slowly. Patterns of dissemination were institutionalized and stable. Without digital technology, knowledge did not readily transcend geographic location or institutional boundaries. National boundaries were less permeable, private knowledge was confined within vertically integrated firms, and intellectual property controls were more limited. Higher education was successful as a bridge between fundamental knowledge and new knowledge, but students rarely

Prospects for Knowledge Policy

3

came in contact with proprietary knowledge. In universities and the public sector, knowledge was presumptively public unless it was classified. (This presumption was especially strong in the United States, which rejects the notion of government copyright and, since 1974, has had a strong, broadly applicable Freedom of Information Act.) The appeal of cyberinfrastructure lies in enabling users to overcome barriers of space, discipline, and institutional practice—and to liberate knowledge from original context. While it arises from the needs of scientific research, its greatest potential is as boundary-spanning general-purpose knowledge infrastructure available to school, work, and home. Like the Internet and the Web, it should bridge all sectors of the economy, public and private, facilitating the flow of knowledge within and across them. The debates over privatization and commercialization of the Internet 15 years ago were resolved by defaulting to openness and interconnection among heterogeneous networks and users. Today, there is concern that security must be addressed in future generations of information infrastructure. How does this concern translate beyond technical requirements and to higher-level knowledge infrastructure? Behind the enabling vision of cyberinfrastructure lies an expanding shadow infrastructure of rights-based contracts, practices, and institutions. Controls on intellectual property encourage disclosure and sharing of specific knowledge, at least within limited contexts such as business relationships or joint research projects. Yet in the aggregate these controls seem to become too trivial and commonplace, too hard to identify, and too easy to trip over. What was once a relatively clear-cut distinction between open/public and controlled/private knowledge has been blurred. Boundary-spanning economic activity flourishes: joint ventures, alliances, standards consortia, open source development. . . . Ownership of knowledge is crafted to varying degrees of centralization and different configurations of openness and control. Just as a variety of financial instruments have proliferated, the benefits and risks of knowledge can be allocated and modulated ad infinitum by creative contracting. The excessive variety of open source licenses is evidence.2 At the same time, markers of intangible value are increasingly articulated and costly to interpret. Patents have more claims, embrace a much greater range of subject matter, and are written with a wider variety of strategic interests in mind. Unlike real property, a patent is not prima facie a right to exploit, it is a right to exclude, and its value depends greatly on the scope and strength of its exclusionary power. It may be very costly to determine who owns what knowledge with confidence, taking into account interpretation, other patents, ambient interests, and the likelihood of finding prior art that invalidates the patent. What are the values of

4

Brian Kahin

complementary knowledge assets? What are the owner’s interests in asserting or sharing the property? Must these questions be addressed now or can/should they be deferred until later, when more may be known about technological potential, market demand, or competitive conditions? How should business relationships, industry norms, and the high cost of legal analysis and litigation be factored in? Like intellectual property, real property may be unique, but there are time-tested ways of determining its value with reasonable accuracy at a low cost. The problem is compounded by the fragmentation of the ‘‘owner.’’ As the Enron debacle shows, institutional and firm boundaries can be obscured or confused by the proliferation of partially controlled entities and privately contracted ownership interests. The fragmentation and blurring of ownership interests in elemental knowledge is disconcerting because it promotes the fragmentation of knowledge and its uses. A profusion of property interests cannot be managed with due attention, understanding, and deliberation on a cost-effective basis, especially when the interests are of low or indeterminate value. We see a number of market-based responses to this problem, such as: Patent pools—Rights to technologies needed to perform well-defined functions are assembled with a specific royalty and allocation of the revenue stream to patent holders.



Cross-licensing—Rights to use portfolios are traded, often with side payments that compensate for aggregate differences in value.



Nonassertion agreements—Mutual promises are made not to sue for patent infringement.



‘‘Mutually assured destruction’’—There is an implicit understanding not to sue because infringement is commonplace and likely to be mutual.



Patent pools are both knowledge-intensive and rights-intensive, but the fact that both enabling knowledge and knowledge about rights to control are involved compounds the problem. It took longer to negotiate the MPEG 4 patent pool than it did to develop the standard in the first place. Patent pools also raise antitrust problems when a choice is made among substitutable technologies. While patent pools may be quasi-public because of the competition policy implications, the other mechanisms operate privately with little accountability. Furthermore, they seem to undermine the exclusivity and disclosure that the patent system is supposed to provide. Opacity does not merely inspire efforts and mechanisms to reduce the costs and risks of navigating highly distributed, poorly defined rights. It leads to asymmetries of knowledge that can be exploited to perpetuate market advantage or dominance. Those who lack sophistication in knowledge management, who lack resources to as-

Prospects for Knowledge Policy

5

sert or defend themselves, or who have sunk investments based on incomplete information are the most at risk. The nascent ‘‘patent troll’’ is a knowledge arbitrageur who is able to take advantage of opacity, asymmetries, surprise, and the vulnerability of sunk investments. Judicious avoidance of knowledge is not necessarily a bad thing. Human attention and absorptive capacity are scarce. Opportunity costs may be high. We may do better to leave the details to trained professionals who have the epistemological and experiential frameworks needed to process knowledge. Users of knowledge and technology do not want to be overwhelmed by choices and the demands of decision-making that is peripheral to their core business. They want to trust their suppliers, and they want enduring relationships with both suppliers and customers. They want to reserve their attention for when and where their attention and action can have the most impact. In the words of Alfred North Whitehead: It is a profoundly erroneous truism, repeated by all copybooks and by eminent people when they are making speeches, that we should cultivate the habit of thinking of what we are doing. The precise opposite is the case. Civilisation advances by extending the number of important operations which we can perform without thinking about them.3

The same holds true for firms and institutions—all the more so when competition demands conservation and strategic allocation of business focus. Tools for performing without thinking are increasingly diverse and sophisticated. Just as writing obviates time spent memorizing, software drastically reduces the time spent on repetitive tasks of informing and learning. Firms buy off-the-shelf software because they do not want to expend the time and attention to develop their own word processors and spreadsheets. Policymakers know well the necessity of making decisions based on incomplete knowledge abstracted under severe practical constraints from staff and outside experts. They know, as do writers and editors, that impact is not based on volume of expression but on the ability to connect with an audience that may have little patience and many competing demands. Yet we have also become skeptical and distrustful of politicians, lawyers, the media, and other knowledge intermediaries. Businesses have become increasingly wary of being locked in to particular consultants, technologies, and sole-source solutions. But where to draw the line is continuous in any competitive or public operation, especially where fast-moving technology is constantly reframing and re-presenting the problem. Do we buy-in or preserve options? Buy or build? In what time frame? Do we invest in internal capacity and ownership—or somebody else’s tools and skills? Intellectual property or freedom of action? Not only tomorrow but 5, 15, and 30 years from now?

6

Brian Kahin

As global markets and the quest for sustainable advantage grow, these become questions of national and regional strategy—and therefore of public policy. Today, concern about software dependency and local capacity to develop and customize software is a factor in policy debate over government use of open source software, a long-term policy problem outside the conventional ‘‘total cost of ownership’’ calculus. At the same time, there is growing recognition of the economic complementarity of commonalities and property rights, especially in the ICT sector, where standards are an essential platform for innovation and market growth. The report of the National Innovation Initiative embraces a broad practical vision of intellectual property as both open and proprietary. A section entitled ‘‘Proprietary and Public Domain Intellectual Property’’ speaks of ‘‘intellectual property’’ as a knowledge asset that can be private or public, observing [T]he evolution of the innovation enterprise—the trend toward user co-creation, the need for interoperability in complex IT networks and revolutionary advances in understanding about human biological networks—is putting pressure on traditional IP models and strategies.

More explicitly: From an intellectual property perspective, open and proprietary IP models should not be seen as mutually exclusive; rather the IP framework must enable both approaches. Because collaborative innovation is relatively new, however, the structure and processes to accommodate ownership, openness and access are evolving. New creative models are emerging across sectors. A mature, balanced understanding of the purpose and practice of standards, including the important role of open standards and global harmonization, is essential to further interoperability, spur technological innovation and expand market applications.4

Instead of a bright-line dichotomy between exclusive and a pure public domain, we now have a growing variety of models and strategies, often shared by multiple entities, for mixing openness and control. This mix not only involves degrees of control, but it reflects the importance of complementarity—the need to examine the context that grounds new knowledge in shared understanding and common language. This is nothing new. Basic science has long provided a nonproprietary platform from which proprietary technology can be derived. However, the Internet and the Web have greatly expanded recognition of the importance of freely usable platforms into the realm of applied technology and services. Complementarity is an essential characteristic of systems technologies. Its nonproprietary/proprietary variant has become pervasive and important—in the ICT sector, in market-oriented innovation, and in free Web-based services such as Google. At the same time, institutional and economic forces have pushed the legal boundary between proprietary and nonproprietary in the opposite direction. Standards of

Prospects for Knowledge Policy

7

utility and inventiveness have been lowered, making patents easier to get. At the same time, patents have become easier to assert—and, at least in the United States, available in all fields of human activity. ICT-based systems of enablement and legally based systems of control do not pull against each other directly. Indeed, technology can be used to control, and law can be used to enable. Both are fed by the explosive growth of information and knowledge. In fact, these seemingly antithetical developments extend orthogonally, defining an exponentially expanded strategic space that offers vast new opportunities for combination, interaction, and complementarity. While not necessarily antagonistic, these two dimensions of knowledge management remain rooted in two fundamentally different perspectives on the value of knowledge. One says that value lies in scarcity, the other that value lies in ubiquity. Scarcity-based value can be linked to priced markets; ubiquity-based value does not show on the books but helps create and maintain markets. Businesses in the digital economy must be able to work with both with sustained focus in the design of competitive offerings. Policymaking by contrast is concerned with environments and how different enterprises and institutions may compete and evolve within future environments. The vision of cyberinfrastructure promises coherence and integration, but we know now that digitization brought far more differentiation than convergence, and that complex environments do not extrapolate well. Knowledge itself is increasingly protean, proprietary to infinitely varying degrees—promising access but demanding protection and inviting arbitrage. And policymakers, inured to acting on incomplete information, are already overwhelmed by the demands on their attention. Politicians recognize the ascendance of knowledge, but what can they do about it? The exploding scope, volume, and significance of knowledge in the global economy now exceeds the more slowly developing analytic frameworks and statistical bases on which informed public policy can be made. Inherited models of the physical world —assembly lines, pipelines, hierarchies, ledgers, warehouses—die hard and slowly. We know from software that knowledge can quietly encode and extend itself into infrastructure. We know from living that knowledge extends backward into its roots in the human psyche. We know that it spans the world outside and the world within. We may be slipping into the riddles and paradox. Perhaps we are revisiting an earlier era when men and women tried to make sense of a world in which they had been thrust—a world with plenty of signs but without coherent explanation, except whatever stories they could conjure up to string the fragments together. Today we test our stories against each other. We hope that these stories are compelling enough to carry beyond this book. We hope they speak to a world where

8

Brian Kahin

value is increasingly accorded to the creation, management, and distribution of what remains unknown. Notes 1. For example, Margaret M. Blair and Steven M. H. Wallman, Unseen Wealth: Report of the Brookings Task Force on Intangibles (Brookings Institution Press, 2001). 2. http://www.opensource.org/licenses/. 3. An Introduction to Mathematics (London: Williams and Norgate, 1911). 4. National Innovation Initiative, Innovate America, Council on Competitiveness, December 2004, pp. 15, 44.

2 Optimizing the Use of Knowledge Dominique Foray

Knowledge has always been at the heart of economic growth.1 From the 19th century (at least), the ability to invent and innovate, that is, to create new knowledge and new ideas that are then embodied in products, processes, and organizations, has served greatly to fuel development. And there have always been organizations and institutions capable of creating and disseminating information and knowledge in an efficient way. The knowledge economy, however, is a recent term that signifies a change from the economy of earlier periods. The knowledge economy is an economy in which much greater strategic importance is given to the allocation of resources in 

R&D and other formal modes of knowledge creation,



The formation of human capital through education and training,

The management of information, knowledge, and expertise through investments in codification and the building of social networks, and





The organization of markets of rights in knowledge.

This is also an economy in which a general-purpose technology (information technology) provides a powerful infrastructure that increases productivity and offers new opportunities to any knowledge-driven activity. The knowledge economy is, therefore, a useful framework for speaking of changes related to the production and distribution of knowledge in modern societies. These changes are numerous. However, there a few broad themes that merit comment. Efficient and Effective Deployment of Information and Communication Technologies (ICTs) as Knowledge Instruments ICTs are technologies geared to the production, processing, and dissemination of knowledge and information, and their effective deployment is central to the chal-

10

Dominique Foray

lenge of advancing the knowledge economy. One part of this book is dedicated to showing how ICTs are revolutionizing scientific and engineering research,2 education, and training as well as any activities involving collective actions, based on sharing rich messages and resources among many people. The unifying framework here is the concept of a general-purpose technology, stressing the potential of ICTs for broad applicability and therefore wide economic impact and also showing the crucial complementarities between the invention of ICTs and the development of applications in the general dynamics of the technology.3 The production of knowledge on the user side—that is, the development of new applications—is crucial to ensure the effective diffusion of the technology. This process is called ‘‘coinvention of applications’’ to show the creative aspect of it. The complementarities between the invention of ICTs and the coinvention of applications inject a dynamic feedback loop in which advances in ICTs lead to unpredictable inventions in applications, which in turn raise the return to improvement in ICTs. The positive externalities generated by such complementarities are extremely powerful in shaping a rapid and effective development path:3 One externality deals with the falling of coinvestment costs: Early users’ experience lowers later users’ costs of coinvention.



A second externality flows through the supply of improved technology, which depends of the number of total adopters and their likely adoption dates. The more extensive and earlier the adoption, the higher the incentive for suppliers to invest, race, and compete for the business created by the application.



This externality structure creates two lags: between invention and first coinvention and between early coinvention and the generalization of coinvented ideas. This is why the productivity impact comes late and is spread over time. It is often ‘‘too early to be disappointed’’4 when we look at the poor productivity impact of new ICT inventions. The coinvention of the related applications takes time! This is why the first episode of the ‘‘new economy’’ was characterized by productivity growth limited to the producer sector (the computer industry) prior to its impact on the many user sectors.5 The externality structure described above is particularly strong and powerful in activities like scientific and engineering research, where some specific circumstances create a very favorable environment for the coinvention of applications. In such circumstances (not limited to scientific research), the externalities between inventors of ICTs and coinventors of applications are considerable.

Optimizing the Use of Knowledge

11

Emergence, Transformation, and Path-Dependent Evolution of Institutions Devoted to the Creation and Transmission of Knowledge This is in a sense the essence of the economics of knowledge as a discipline: to explore socioeconomic institutions that can be relied upon to produce knowledge in an efficient manner. The unifying framework here is the character of knowledge as a semipublic good, with difficult-to-enforce property rights. Its diffusion is in principle good for social well-being but bad for private returns: No one wants to invest in the creation of new knowledge if the rents generated are not at least partly appropriable. Institutions that govern the creation and diffusion of knowledge are shaped by this trade-off: On the one hand they need to meet the objective of providing the ideal motivation to the private producer of knowledge while on the other they have to fulfill the social objective of ensuring efficient use of knowledge once it has been produced. This book discusses various institutional issues rather deeply: The necessary transformation of an old institution—the patent system—is the focus of several chapters. There are some doubts that in the age of the knowledge economy, the new equilibrium, involving intensive patenting activities, a large amount of cross-licensing, and aggressive patent enforcement strategies, is ‘‘better’’ than the preceding one that was characterized by a moderate level of patenting activities, with firms allowing diffusion of their own knowledge in return for lowcost absorption of others’ knowledge. The latter seems to be a system with lower transaction costs, while the former does not seem distinctly superior in terms of knowledge creation.



Some contributions deal with the increasing importance of resource allocation mechanisms that replicate some of the structures of open science, supporting a more original approach to rewarding inventors and managing spillovers than that suggested by the use of intellectual property rights to exclude potential users. It is interesting to see that this particular class of social organizations, in which high rates of innovation are correlated with rich voluntary spillovers, is of broader applicability than is suggested by the predominant and sometimes only reference, open source software.



New trust-building mechanisms are critical to the effective development of virtual relations in the knowledge economy. These involve a very broad range of issues, such as the construction of ‘‘emotional trust’’ in situations where it is unlikely to emerge spontaneously or the construction of cognitive trust, such as the certification of knowledge circulating on the Internet, teachers’ competences, and curriculum



12

Dominique Foray

supplied by long-distance organizations. The concept of social capital6 is considered in our book as a useful way to capture the importance of this type of knowledge that is more a relational asset than a personal attribute possessed by individuals. It can be seen as a form of glue that holds the constituent members of society together and permits them to function more productively in the economic sphere as well as in other interpersonal transactions where trust is beneficial. Addressed also in this book is the issue of the new division of labor between the old and the new institutions designed to collect, organize, preserve, and provide access to knowledge-bearing objects. The old LAM (libraries, archives, and museums) and the new Internet are strongly complementary and must be deployed in a coherent way to ensure the effective development of the epistemic infrastructure. The advent of the Web has reignited naı¨ve notions of a single universal collection of all knowledge while also amplifying preexisting sources for information access in ways heretofore impossible. The LAM and the Web in fact prove complementary in four key areas: access, information quality insurance, social memory, and information property.



Universities—a key institution for the knowledge economy—are under pressure to redefine their places and roles in this new division of labor. History shows that this is not the first time Western universities have been pressed to reinvent themselves. After the Bolonia model (universities as a place for education) came the Humboldt model (a place linking education and research) and then the MIT model (a place linking education, research, and innovation). What is the next step? The academic capitalism model strongly emphasizes the commercialization of some of the basic functions of universities. The post Bayh–Dole university involves the mission of commercializing research through an increasing patenting and licensing activity. In the chapter devoted to this topic, the author pleads for a sort of retro-evolution to push universities back to produce the kind of public goods they traditionally generated and that serve to help the innovation system, but only in an undirected way: The question ‘‘What is the market value of this knowledge’’ should be replaced in many circumstances by ‘‘Is it true? Let’s stop and think.’’



The Coevolution of Technologies and Institutions ICTs and institutions are mutually interdependent processes. This is strongly emphasized in a couple of chapters: Creating the right and proper institutional design to fully realize the potential of ICTs has proved to be at least as difficult as advancing the technology itself. Systemic discrepancies between the development of ICTs and the development of institutional equipment are a strong source of turbulence. Pro-

Optimizing the Use of Knowledge

13

viding the soft (institutional) foundations of cyberinfrastructure may be the hardest part.7 Knowledge Division and Dispersion The division of labor and increasing specialization in knowledge production cause the knowledge base to fragment and disperse. Division (due to the natural development of science and technology), dispersion (associated with ‘‘democratization’’ of innovation), and the increasing ability of firms and agents to realize ‘‘experiments’’ make broad integrated understanding difficult. This is a source of inefficiency and ignorance at the micro level that can have disastrous consequences at the level of global policy making. The structures of the knowledge constantly need to be rebuilt, which requires knowing how to organize fragmented knowledge. This is why the creation of integrative knowledge and coordinating mechanisms is important—that is, norms, standards, infratechnologies, and architectures, which lead in turn to new coordination mechanisms (such as modularity) that are economically efficient and effective. Communities are coordinating and integrating mechanisms, for creating and developing common languages and codes and increasing the number of interpersonal transactions within fields of knowledge or practice. Community-developed codes, such as language, and techniques help integrate dispersed and divided knowledge. However, they have an ambivalent role. Codes also raise entry costs and keep outsiders at bay. There is thus a need for meta-standards to ensure that highly integrated communities will not isolate themselves and their knowledge from the rest of the world. Proximity—geographic agglomeration and clustering—is also an important factor for integrating knowledge. The need for knowledge integration becomes an important factor in locational problems and may counterbalance the mitigation of geographical constraints due to the fact that the costs of moving knowledge is dramatically decreasing (at least relative to the cost of moving people). The Importance of Public Knowledge In a period of wild passion for private property in the realm of knowledge creation, it is important to reassert the importance of public knowledge. In any process of knowledge creation, integration, and exploitation, the existence of a freely accessible stock of knowledge—a knowledge commons—is crucial. By public knowledge, I do not necessarily mean public-sector, but rather the sharing and pooling of knowledge/information as a result of freely revealing it, involuntary spillovers,

14

Dominique Foray

academic norms, etc. Deeper appreciation of the role of public knowledge as an efficiency driver is needed to balance a preoccupation in knowledge management literature with how private actors can capture information and how businesses can control strategic knowledge assets. Open, Distributed Systems Mixing a number of these ingredients together—ICTs, freely revealing behavior, integration, trust—points to an emerging paradigm of open, distributed systems of innovation and learning. The various organizational forms and functions of open, distributed systems are in fact assuming wider relevance and prominence in the knowledge economy. Historically speaking, scientific communities have been pioneers in the development of such open, distributed systems, but in the knowledge economy this particular decentralized mode of knowledge creation and distribution expands far beyond scientific activities toward software programmers, product and service development activities, and the many professional and expert communities that engage in knowledge sharing, information pooling, and learning by interacting. Toward Evidence-Based Knowledge Policy As the several chapters dedicated to the measurement of (activities related to) knowledge show, major progress in the economics of knowledge has been achieved on the empirical side. During the past several decades a number of efforts have overcome the nonobservability problem to make knowledge a more measurable phenomenon. As a result, knowledge is a bit less slippery for economists who want to study it, and policy makers have access to accumulating evidence on many aspects of the knowledge economy. Policy development should no longer rest only on a casual understanding and vague perception of problems and issues. However, increasing the efficiency of institutions for generating and using knowledge requires a clearer picture of the ways knowledge, information, and know-how are generated and used. In the chapters focusing on measurement issues, the authors demonstrate the kind of ‘‘evidence-based policy research’’ that is useful for knowledge policy making. However, the economic foundations of knowledge policy are still in their infancy, and we are still stuck with a collection of sectoral policies dealing with R&D, education and training, intellectual property rights, etc. Given the characteristics of knowledge as an economic good, it is possible to identify resource allocation mechanisms, socioeconomic institutions, that can in princi-

Optimizing the Use of Knowledge

15

ple produce and allocate knowledge in an efficient manner. However, institutional dynamics of institutions exhibit inherent market failures. Better understanding of these institutions and the processes governing their evolution is, therefore, needed to inform policy design. Knowledge policies are needed as tools to improve the working of institutions but also to inform larger social choices about what kind of institutions and mechanisms will lead to outcomes that are ‘‘optimal’’ at national, regional, and global levels. Notes 1. I am grateful to Brian Kahin for scholarly advice and invaluable editorial help. 2. In the opening speech of the Advancing Knowledge and the Knowledge Economy Conference, Bill Wulf talked of computing as the ‘‘fourth modality of scientific research’’ (after observation, theory, and experiment). 3. T. Bresnahan, The Mechanisms of IT’s Contribution to Economic Growth (Paris: St Gobain Centre for Economic Research, 2000). 4. M. Abramovitz and P. A. David, ‘‘Two Centuries of American Economic Growth: From Exploitation of Resources Abundance to Knowledge-Driven Development.’’ Stanford Institute for Economic Policy Research, Discussion Paper No. 01-05, 2001. 5. In 1999, experts wrote ‘‘to date, computer technology has proved unbelievably effective at reproducing itself; beyond that, its apparent influence on productivity has so far been somewhere between imperceptible and adverse’’ (‘‘The New Economy,’’ The Economist, July 24, 1999), while in 2004 they wrote ‘‘The core of the US success story, and the source of its difference from Europe, appears to have been in ICT-using industries’’ (R. Gordon, ‘‘Why Was Europe Left at the Station When America’s Productivity Locomotive Departed?’’ CEPR, London, Discussion Paper No. 4416, 2004). The two statements are true. There is simply a lag between these two channels through which ICT has impacted productivity. 6. However, the use of the metaphor of ‘‘capital’’ and the term social capital may not be appropriate here (see Arrow, ‘‘Observation on Social Capital,’’ in Social Capital: A Multifaceted Perspective, ed. Dasgupta and Serageldin [Washington, D.C.: The World Bank, 1999). 7. To borrow from the title of the chapter by Paul A. David in this volume.

3 OECD Work on Knowledge and the Knowledge Economy Berglind A´sgeirsdo´ttir

While knowledge always has been at the heart of economic development, there is substantial evidence that the capacity to produce and use knowledge has much more explanatory value in determining levels of economic welfare and growth than in the past. The OECD has now for two decades worked on knowledge and the knowledge economy. The TEP (Technology Economy Programme) in the late 1980s was one of our first attempts to clarify the relations between technology developments and economic performance. Since then there have been many other projects involving many other people. A few of the more recent OECD studies with an analytical focus on knowledge and the knowledge economy are 1. The Growth Project, aiming at identifying the factors determining growth in OECD countries, 2. The economic impacts of ICT investments, 3. Intellectual assets and innovation, 4. Knowledge management—analyzing the production, dissemination, and application of knowledge in different sectors, 5. Human and social capital investments and returns, and last but not least 6. Statistical and other indicator development to measure the knowledge economy, such as the Science, Technology, and Industry Scoreboard, knowledge management practices in the private sector, intangible investments in enterprises, etc. Space does not allow me to summarize all this work. I will therefore concentrate on four key messages on the important issues that need to be taken into account when countries and their institutions want to promote their knowledge economy.

18

Berglind A´sgeirsdo´ttir

Figure 1 Important factors shaping the knowledge economy.

My first message is as follows: Good ‘‘economic fundamentals’’ are important for stimulating the knowledge economy. By good ‘‘economic fundamentals’’ I understand Stable macroeconomic policies that allow long-term planning, including wellfunctioning product and capital markets,





Well-functioning labor, product, and capital markets;

Efficient training policies that help ensure that the lesser educated are equipped with the right skills, thus avoiding ‘‘knowledge divide,’’





Competition policies, which drive down the costs of technologies,



Liberalization of telecommunication policies, and



Openness to trade and foreign direct investments to let in ‘‘new ideas.’’

I have chosen to illustrate the important factors shaping the knowledge economy as a ‘‘Greek temple’’ with four pillars (Figure 1). The ‘‘economic fundamentals’’ are the base on which the four pillars stand. The four pillars also illustrate that for the knowledge economy to develop and grow, it is not enough to focus on a single policy or institutional arrangement. A whole range of policies and coordinated actions to create the right conditions is necessary. The ‘‘policy mix’’ must be based on a comprehensive strategy suited to each country or circumstance and will include the four pillars ‘‘innovation,’’ ‘‘new technologies,’’ ‘‘human capital,’’ and ‘‘enterprise dynamics.’’ At the top of the Greek temple, I have put ‘‘globalization,’’ which is a driver that influences all four pillars and all four key factors, which are becoming increasingly

OECD Work on Knowledge and the Knowledge Economy

19

Figure 2 First pillar, innovation: R&D growth by industry structure (percentage point in business R&D industry as a share of GDP).

mobile and global under the globalization process: ‘‘research and development,’’ ‘‘Internet,’’ ‘‘highly skilled,’’ and ‘‘multinational companies.’’ My second message is therefore as follows: The development of the knowledge economy is dependent on four main ‘‘pillars’’: innovation, new technologies, human capital, and enterprise dynamics. I will briefly go through the main developments in the four pillars that have occurred during the past decade in OECD countries. The first pillar is innovation. R&D expenditure, patents, etc. all grew in the second half of the 1990s in most OECD countries. Innovation has now become more widespread, with increasing activities especially in services, ICTs, and pharmaceuticals, as can be seen in Figure 2. The importance of innovation as a key competitive factor has forced a faster cycle time and meant that firms had to experiment with new ways to acquire innovations either through links to universities, alliances with each other, or mergers and acquisitions. There is little doubt that ICTs are the technologies of our era. The comparisons with bio- or nanotechnologies are weak. The United States and small E.U. countries have had a relatively large impact of ICT investment in terms of its contribution to

20

Berglind A´sgeirsdo´ttir

Figure 3 Second pillar, information technology: ICT capital to GDP growth (in percentage points).

Figure 4 Third pillar, human capital: Population that has attained at least upper secondary education.

GDP growth; the effect has been much smaller in France, Germany, and Italy (Figure 3). One of the key issues is to what extent the rapid technological progress in ICT goods and services contributes to productivity growth. The Federal Reserve Bank of New York has recently published a study saying that there is now a consensus that a large portion of the productivity growth in the United States can be traced to the sectors of the economy that produce information technology or use ICT equipment and software most intensively. The third pillar is human capital—the knowledge, skills, and competences instilled in workers (Figure 4). Human capital is very important for developing

OECD Work on Knowledge and the Knowledge Economy

21

Figure 5 R&D share of foreign affiliates/total business and R&D.

a knowledge economy in several respects. First, we know that there is a wellestablished relationship between human capital and labor productivity, and human capital is therefore a significant determinant of growth. Second, the two previously mentioned pillars of the knowledge economy—innovation and new technologies— are not effective without a stock of trained and qualified workers to realize their benefits. OECD countries have increased the percentage of the population that have attained at least a secondary education in order to meet the increased demand for ‘‘knowledge-intensive’’ employment. The fourth pillar is enterprise dynamics. Newly created firms have spurred innovation in many areas. They have been responsible for an increasing share of the growth in the private R&D and patent activity in the United States and a number of other OECD countries (Figure 5). The dynamics in firm turnover (exit and entry) reflect the ability of countries to expand the boundaries of economic activity, shift resources, and adjust the structure of production to meet consumers’ changing needs. My third message is as follows: Globalization is a pervasive factor that affects all four pillars of the knowledge economy. Globalization is not new, but it has been strengthened by the international mobility of highly skilled workers, ICTs, faster and cheaper transportation, trade liberalization, global capital markets, etc. The R&D share of foreign affiliates compared to the total business R&D is substantial in some countries, for example, Canada,

22

Berglind A´sgeirsdo´ttir

Ireland, Spain, and Sweden, and less important in France, Japan, and the United States, and in all countries exempt Ireland it increased during the period 1995–2001. My fourth message is as follows: New organizational innovations and knowledge management practices have to be developed to deepen the benefits of the knowledge economy. The ‘‘softer’’ social and organizational changes are in many cases very important for the development of the knowledge economy. Investments in ICTs or R&D without the management and organizational structures in enterprises that enable productive use of knowledge workers are less productive. These include teamwork, flatter management structures, and stronger employee involvement, and they often entail a greater degree of responsibility of individual workers regarding the content of their work. The adoption of work practices and the presence of labor–management institutions tend to facilitate take-up of new technology. Organizations are increasingly paying attention to their systems of knowledge management to ensure that they are capturing, sharing, and using productive knowledge within their organization to enhance their learning performance. Joint work by Statistics Canada and OECD on knowledge management indicates that knowledge management practices in companies seem to have a far from negligible effect on innovation and other aspects of corporate performance. A survey on knowledge management practices in French companies has shown that whatever the company’s size, industry, or R&D effort, firms innovate more extensively and file more patents if they set up knowledge management policies. Social capital in the form of networking and trust can help realize innovative environments such as Silicon Valley. Trust-based relations facilitate cooperation and are essential for good economic performance and innovation. Trust reduces transaction costs and improves the flow of information and thus has direct economic effects as well as indirect and wider outcomes. It aids innovation by improving communication flows and the diffusion of knowledge, within and between organizations. The knowledge economy cannot simply be characterized by higher ‘‘knowledge intensity,’’ as for example more highly skilled people in the labor force. Increasingly, countries will have to think about how education promotes effective participation in communities of knowledge, and this will include social and moral competences as well as technical ones. Conclusion This synopsis of what the OECD knows about knowledge and the knowledge economy makes us aware that there is much more that we need to know. There is a huge

OECD Work on Knowledge and the Knowledge Economy

23

agenda for researchers, policy-makers, and others to develop better understandings and policies for the knowledge economy. Acknowledgments It was a great pleasure to be involved in this project on Advancing Knowledge and the Knowledge Economy and to know that it is the result of a partnership of OECD, the National Science Foundation, the European Commission, the U.S. Interagency Working Group on IT Research and Development, and the University of Michigan. I especially want to thank the National Academies for hosting the event—it was a marvelous place to have a debate about the role and nature of knowledge in our economies and societies.

I Measuring Knowledge

4 Measuring Knowledge and Its Economic Effects: The Role of Official Statistics Fred Gault

Introduction This chapter is about measuring knowledge and its economic effects, and the role played in this by official statistics. Knowledge is pervasive, and more and more consumption involves products of the mind, which have economic and social consequences. The chapter focuses on the generation, transmission, and use of knowledge, the indicators used to describe these activities, and the economic effects that they produce. With the information and communication technology (ICT) infrastructure in place, electronic products can be traded globally and material goods can be purchased electronically from anywhere. These products embody knowledge, some more than others, and their production and trade can be monitored, using the existing statistical infrastructure, with some additional assumptions. These statistics can contribute to discussions of the knowledge-based economy (KBE). The chapter distinguishes between the KBE and ‘‘knowledge’’ and goes on to discuss measures of knowledge activities, linkages, and outcomes before placing them in an economic and social context. Some time is spent on the activity of learning since without learning through education, training, and doing, knowledge cannot flow, and without flow it cannot have an economic effect. The chapter ends with a discussion of the knowledge system, which consists of institutional actors engaged in the generation, transmission, and use of knowledge, and identifies gaps in the statistics that describe the system and the implication for evidence-based policy. The KBE and Knowledge Before looking at the measurement of knowledge and its economic effects, the economic analysis of the KBE is examined briefly and compared with other special frameworks.

28

Fred Gault

To describe the KBE, some industries can be identified as knowledge intensive, and once that is done, all of the statistics available in the System of National Accounts (SNA) can be made available as a special aggregation. Special aggregations have been done for some years for the energy and tourism sectors, and their definitions are found in annexes to the International Standard Industrial Classification of All Economic Activities, revision 3 (ISIC.rev.3) (UN 1990). More recently, a definition of the Information and Communication Technology (ICT) sector has been added to ISIC.rev.3.1 (UN 2002), and these aggregations of four-digit ISIC classes allow analysts to track the behavior of these sectors and to compare a sector with other sectors, or four-digit classes. As a concrete example, Statistics Canada publishes monthly GDP figures for the ICT sector (Statistics Canada 2004) using this aggregation. The classification of industries as knowledge intensive, or as high, medium, and low tech, based on R&D intensity (Hatzichronoglou 1997) is not without its problems since agreement has to be reached on the assumptions needed to do this. Once industries are classified, firms in knowledge-intensive industries may produce products that are not knowledge intensive, with the converse being true for firms in non– knowledge intensive industries. If the concept of the KBE is to be taken further, there must be a classification of knowledge-based products (goods or services). If there is such a classification, trade figures can be produced and studies done of the economic and social impacts of the acquisition and diffusion of knowledge-based products. This has been done for the ICT products by the Working Party on Indicators for the Information Society (WPIIS) (OECD 2002c), and Hatzichronoglou (1997) has done it for high, medium, and low technology. However, there is no equivalent to these internationally agreed definitions for knowledge-based industries or products nor agreement on the economic and social surveys required to collect information on their production and use. Measuring knowledge itself is more challenging, if not impossible (Foray and Gault 2003, p. 18). There is no unit of knowledge that corresponds to a currency unit in the SNA, and there is nothing comparable to the concepts of current and constant currency units that supports comparisons of the economic system over time. There is also nothing comparable to purchasing power parities (PPP) that supports comparisons across space or price indices for comparison over time. Knowledge is different from conventional economic products. It can be sold or given away, but it is still retained by its original owner, unlike a tangible good (a brick) or an intangible good (music, text, or video on a medium). Knowledge can, in some cases, be disseminated instantaneously (an encryption key), unlike goods,

Measuring Knowledge and Its Economic Effects

29

which have to be transported, or services, which have to be rendered (haircuts, computer system designs, or sessions of financial advice). In other cases, the dissemination is far from instantaneous. It takes months to become proficient in a foreign language or to able to solve a differential equation. It may take an eternity to achieve knowledge of the presence of a supreme being. Knowledge is a complex subject and growing in complexity (Hodgson 2000), as is the measurement of knowledge, assuming that measurement is possible. This chapter addresses knowledge that can result in economic or social outcomes and then focuses on economic effects (Stehr 1996). Three kinds of knowledge activity are considered: generation, transmission, and use (Foray and David 1995). In each of these activities, knowledge can be codified and stored in documents, or embodied in a product, or tacit and present in a person, a team, an organization, a region, or a country. Moving knowledge requires the capacity to transmit it, to publish, to demonstrate, to mentor, or to teach. It also requires the capacity to absorb, or to learn, through education and training, and by doing. Knowledge activities are more encompassing than just those associated with natural sciences and engineering; they include social science and humanities activities. This gives rise to a different set of policy issues that go beyond promoting the creation of new knowledge through curiosity driven research. For example, the involvement of living things in research introduces ethical issues, and some activities, such as human cloning or stem cell research, raise moral issues. National security can also circumscribe the creation, transmission, and use of knowledge about, for example, encryption, the artificial creation of viruses, or applications of nano-devices. Conveying a capacity for action is broader than research activities and their associated social and economic boundaries. Using a taxonomy developed by Lundvall and Johnson (1994), knowledge is conveyed in schools, where students learn to ‘‘know what’’ (the density of lead), in universities, where they learn to ‘‘know why’’ (the laws of quantum physics), and in the workplace, where they learn to ‘‘know how’’ (on-the-job training), and as they become part of networks, they learn to ‘‘know who’’ (Lundvall 2000). Since this chapter is about measurement of knowledge and its economic effects, examples of knowledge activities, linkages, and outcomes and their related indicators are first presented. These lead to a discussion of how the indicators could be used in an integrated manner to describe the generation, transmission, and use of knowledge, and of where there are gaps in such an approach. It follows the observation of Foray (2004, p. 18) that ‘‘The aim of the economics of knowledge is thus to analyze and discuss institutions, technologies, and social regulations that can facilitate the efficient production and use of knowledge.’’

30

Fred Gault

Knowledge Activities The description of activities starts with formal knowledge generation as a result of research and development (R&D), continues with invention, innovation, adoption, and diffusion of practices and technologies, and then moves to the human resources related to all of these activities. Once people are brought into the analysis, the measurement problems and the policy issues multiply, and that leads to the section on knowledge linkages that follows. The order of presentation is not meant to suggest a linear model view of knowledge activities. Each activity can, and does, occur independently, as well as in conjunction with other knowledge activities. In innovation, this is described by the ‘‘chain-link’’ model of Kline and Rosenberg (1986). Research and Development This is the creation of new knowledge, which may appear as a seminar, a publication, or a patent or as knowledgeable graduate students and researchers. The indicators for R&D are well established (OECD 2002a) and include expenditure on the funding and performance of R&D and the human resources allocated to the doing of R&D (OECD 2004a). With the increasing global nature of R&D, attention is turning to the international payments and receipts for R&D services (OECD 1990) and the role of networks, alliances, and partnerships in the doing of R&D (Hagadoorn 2001). Invention Inventions can be an outcome of R&D, a formal process of knowledge generation, they can result from learning by doing in a production or delivery environment, and they can be influenced by suppliers and clients. Inventions may be protected by intellectual property instruments, such as a patent, a trademark, or a copyright, or by trade secrecy. Patent indicators are produced by the OECD (2004a). Guidelines for developing and using patent indicators are also provided by the OECD (1994). Innovation For the past 20 years technological innovation has been regarded as the bringing of new or significantly improved products (goods or services) to market or the introduction of new or significantly improved processes for the production or delivery of products. This can be regarded as the use of knowledge, both internal and external to the firm, to create value. Guidelines on how to measure the activity of innovation were codified in 1992 as the Oslo Manual (OECD 1992) and revised in 1997 (OECD/Eurostat 1997). The indicators, resulting from three rounds of the European Community Innovation Survey (CIS), are published by the European Union (Com-

Measuring Knowledge and Its Economic Effects

31

mission of the European Communities 2004a) and by individual governments, as are indicators derived from surveys in non–E.U. countries (Gault 2003). They include the propensity to innovate as well as a number of indicators related to sources of information, outcomes, use of intellectual property, and barriers to innovation. What has become clear during the last decade of formal measurement of innovation in many countries is the importance of organizational structures and management practices as well as the development of new and existing markets to the innovation process. This would have been no surprise to Schumpeter (1947), but it has taken some time for the empirical world to move from technological innovation rooted in natural science and engineering to a broader concept that includes knowledge drawn from the social sciences and humanities. The most recent revision of the OECD/Eurostat Oslo Manual incorporates these changes and was released in 2005. This will lead to new indicators of innovation that will show how knowledge from different sources combines to add value to the firm and to have impact on businesses and people. Use of Technologies and Practices Innovation is classified according to its novelty. It can be a world first, a market first, or a firm first. This means that firms that adopt new technologies and practices can lay some claim to a firm-first innovation. Indicators of the adoption and diffusion of technologies and practices are guides to how the economy and society are changing and to the impacts resulting from such change. There are many indicators of the activity of adopting technologies and practices and of related activities, much as is the case for the activity of innovation. The OECD publishes indicators of the use of biotechnologies, information and communications technologies, and nano technologies (OECD 2002b, 2003) and of knowledge management practices (Foray and Gault 2003). These indicators illustrate the use of knowledge by the firm to adopt new processes in order to do better what it does. Human Resource Development A key to all of these activities is the human element. It is the person or, more likely, the team (Prusak 2001), that changes behavior and creates additional value. The question is how the knowledge held by the people involved is measured. One way is to identify the percentage of the labor force with particular educational attainments on the assumption that the higher the attainment, the more likely the people are to be sources of knowledge and to be able to capture knowledge for value creation. This approach is used in the European Innovation Scoreboard (EIS) (Commission of the European Communities 2004b), which records the percentage of the

32

Fred Gault

labor force in the Eurostat Labour Force Survey in the age range 25–64 with tertiary education. An attempt at codifying measures of the characteristics of people in science and technology led to the Canberra Manual (OECD/Eurostat 1995), which focused on level of education and on occupation. This has proved difficult to apply since, outside of a population census, there are few surveys that collect data from a large enough sample to support analysis of the characteristics of people by industry, level of education, and occupation. An alternative approach, used in the EIS, is to record the percentage of science and technology graduates in the population with an age range of 20 to 29. While the characteristics of the highly qualified personnel in the population and the labor force are important indicators of the stock, the policy preoccupation has been shifting to mobility. Mobility of people is a flow of embodied knowledge and of absorptive capacity. This is addressed in the next section, on linkages. The attraction and retention of the highly qualified are becoming more important as retirements increase because of the age distribution of the labor force, and there is concern that the production of newly qualified graduates by the higher education sector is not able to deal with the demand and that immigration cannot fill the gap as it has in the past in some countries. The market for the highly qualified has moved from overlapping local markets to a global market, and this is especially true for academics. The question is how to capture this in official statistics, and that is discussed in the next section. Knowledge Linkages Many of the knowledge activities in the previous section are covered by official statistics, although there are still gaps. The same cannot be said for knowledge linkages, which tie together knowledge activities. These ties can be bilateral one- or twoway flows, or they can be multilateral and multidirectional. Understanding them and providing robust indicators of them is another challenge that must be met if there is to be a beginning of an understanding of how this dynamic system works. This section is divided into two subsections, one on sources of knowledge and one on networks of knowledge. Sources of Knowledge The R&D department in a firm, a government laboratory, or a university department can use and generate knowledge, but for it to enable the creation of value, the knowledge must be transferred to the market, directly or indirectly. There are several ways of doing this.

Measuring Knowledge and Its Economic Effects

33

Intellectual property can be transferred within the firm that has generated it or licensed to an affiliated or unaffiliated firm or to a new ‘‘spinoff’’ firm established to develop the knowledge in the marketplace. In all of these cases, the knowledge flow can be one- or two-way. If it is two way, the market experience can be fed back to the knowledge creators, and the knowledge can develop in response. The role of the client as a source of ideas has been developed by von Hippel (1988, 1998), but they are not the only source. Suppliers of materials, equipment, and people may be sources of ideas for knowledge activities in an organization, and ideas for doing new things do not just come from the R&D department. They may come from the production unit, if management is aware enough to harvest them, or from the providers of finance, who wish to grow their investment. In the development of official statistics on innovation and the use of technologies and practices, questions are added to surveys to identify the sources of ideas for the activity, and the answers illuminate policy discussions about the purpose of government laboratories or the role of research in universities. Little happens without money, and large firms are able to fund new activities from retained earnings or through debt. A small firm in the service sector will have limited revenue and no assets beyond the minds of its people. For it to bring new products to market or to introduce a new process for production or delivery, it needs help, and that is a role for the angel investor or the venture capitalist. Financial organizations, especially if they have been working in the industry, bring to the firm knowledge about pitfalls and opportunities, and the transfer of that knowledge to the firm is key to its success. Knowledge can be licensed, acquired by listening to clients, suppliers, and the product producers, but it is people who do the licensing and the listening and without them, nothing happens. The key issue of absorptive capacity is dealt with later, but the question here is where do the highly skilled and highly qualified people come from? People, who embody knowledge, come from postsecondary institutions, other organizations in the same business, and other organizations that do business with the organization. These sources may cross national boundaries, and they raise the question of mobility of the highly qualified and the ease with which immigrants can be assimilated into the economy and society (OECD 2004b). There is a joint initiative of the OECD and the UNESCO Institute of Statistics (UIS), supported by the National Science Foundation, to look at the career paths of doctorate holders. This involves the development of surveys of earned doctorates not dissimilar to the NSF survey of that name, and the objective is to initiate pilot

34

Fred Gault

surveys in a number of OECD countries. This will result in indicators of the work intentions of new doctorate holders, including plans to work outside the country where the doctorate was awarded. This raises the problem of how to share such information among the statistical agencies involved in the collection of the data so that the holders of doctorates who cross borders can be included in surveys that track these highly qualified people throughout their careers. The indicators from a network of surveys of earned doctorates will provide measures of knowledge flow across economic sectors and national boundaries, and they will support discussion of education and of immigration policies. There are also indirect links of knowledge to the market. An example is the generation of knowledge to support regulation of drugs, foods, and health care products, quality of education, and health and social services. Such knowledge tends to be generated and applied in governments and in other public institutions. Networks of Knowledge The indicators in the first part of this section address individual sources of knowledge transfer. In this part, indicators of knowledge networks are discussed. With the development of bibliographic databases, a whole field of bibliometrics has developed that includes measures of knowledge flow. An example is copublication analysis, which captures the implicit link in academic papers, or patents, written by authors from different institutions. The extreme example is the experimental high-energy particle physics paper with more than a hundred authors and from many institutions all over the world, but collaboration is not just limited to highenergy particle physicists. In a global village, collaboration is increasing, and so is the knowledge flow associated with it. A paper by Godin et al. (1998) mapped copublication in Canada and produced indicators showing the links between government, higher education, and the private sector, and it demonstrated the significant role of the universities in Canadian knowledge flow. While the paper was published by the statistical office as part of a program of indicator development, the indicators never became ‘‘official statistics.’’ Because the database from which they were derived was commercially available, any institution could produce such indicators, and in Canada, they have evolved outside the statistical office. Another indicator of knowledge flow is a contract. One institution agrees to pay money to another institution for the delivery of a product, which may be new knowledge. Contracts for R&D can be tracked in R&D surveys, and they provide evidence of knowledge flow in the R&D community. With globalization, the flow of money across borders for R&D is becoming an area of interest. Who is buying and who is selling knowledge becomes important when much domestic knowledge is rooted in an institutional infrastructure that is supported by public funds.

Measuring Knowledge and Its Economic Effects

35

A contract is, by its nature, restrictive, while joint ventures are less so. They will have a common goal to which the partners subscribe, but once established, there is scope for adjusting the projects within the undertaking. An example is an industry– university collaboration around the needs of an industry, with practical and well funded projects being available to academics and their graduate students. As the knowledge is developed collaboratively, the intellectual property can be managed differently from that of academics wishing to commercialize the results of their research through the university technology transfer office. Both of the examples just offered give rise to indicators of knowledge flow that is at least two way and that can be many way. Knowledge Outcomes Indicators of activities support public policy debate on the funding for the activity, its industry, and its location. Indicators of linkages show how different parts of the economy connect, and that adds a dimension to the policy debates on the activities. However, as corporate and public accountability become more widespread, interest groups want to know what happened as a result of resources committed to knowledge activities. This raises the question of what indicators are needed to help answer these questions. An outcome of the activity of knowledge creation can be more knowledge, which feeds back into knowledge creation. This can be measured by academic publication, and its relevance by academic citation. The activity can give rise to commercially valuable intellectual property protected by patents, trademarks, or copyrights, or it can result in new products or processes, new forms of organization, greater market share, or the opening of entirely new markets. Indicators of outcomes can be directly linked to a project, such as the number of papers, patents, graduate students, academic promotions, and Nobel prizes resulting from an NSF funded university research project, or they can be lagged or indirectly linked. The link between photonics in 2005 and Einstein’s paper on the photoelectric effect in 1905 is an example of the latter. Somewhere in between are indicators of employment levels and skill levels in firms resulting from the introduction of new processes and products that result from the flow of knowledge from sources inside and outside the firm. The need to be able to say what the government, the business, or the academic institution got in return for committing scarce resources to an activity will continue to drive the development of indicators based on outcomes. Demand for official statistics produced at arms length from the policy process and interest groups will also grow.

36

Fred Gault

Related Characteristics of Institutions and Regions While indicators of knowledge activities, linkages, and outcomes reflect the state of the system at a given time, additional indicators are needed to provide context. Some of these are the size of the institution, the industry it is in, the quality of its labor force, the characteristics of its location, and the legal conditions that govern its operation. There are also the framework conditions that govern the economy and society. Size matters in two ways. The propensity to engage in R&D, to patent, and to adopt and adapt new technologies and practices increases with the size of the firm. However, this variation with size is not continuous for all activities (Earl and Gault 2003). Work on knowledge management practices suggests that some are appropriate for small firms and others are more important for large firms. The propensity to use such practices rises with size to an upper bound, and then they are overtaken by other practices that can be supported by the infrastructure of a larger organization. These findings have implications for policy interventions to promote innovation and knowledge flow. Industry matters since not all industries engage in knowledge activities to the same degree. For example, some make extensive use of patents, some do not. Others prefer trade secrecy. The quality of the human resources in the firm is related to the propensity to engage in knowledge activities. Geography is an issue even in a global economy since firms relate to their competitors, to local institutions of education, and to affiliate firms, which may be geographically close if the firm is a spinoff. While there is a considerable literature on clusters and other geographical characteristics (OECD 2001; Florida 2002), stimulated by the work of Porter (1990), statistical organizations and academics are still working on indictors that could provide reproducible evidence of the existence and performance of a cluster. While Florida has a number of indicators that can be derived from census of population data and that can distinguish regions more likely to attract the highly qualified labor force needed for knowledge activities in firms, there are also indicators of educational attainment and of the quality of the educational institutions. The OECD conducts the Program for International Student Assessment (PISA), and the resulting indicators are able to differentiate OECD countries and regions within them according to the performance of 15-year-old students in mathematics, reading, sciences, and problem solving. In addition to the performance in the four areas, PISA provides information on family characteristics (OECD 2004c). Throughout this chapter the assumption has been made that knowledge indicators are being developed within a functioning market economy with well established

Measuring Knowledge and Its Economic Effects

37

institutions that deal with legal issues, education, finance, government programs, and trade and link to the international institutions. In many countries of the world this is not the case, and the first step in dealing with knowledge and its economic effects requires the strengthening of institutions and their links with international organizations (Dahlman and Aubert 2001). There is also the need to bridge the ‘‘knowledge divide’’ that separates the industrialized from the nonindustrialized (Chataway et al. 2003). Learning and Doing Learning is essential to any of the activities of knowledge generation, transmission, and use, and it is not just the learning of an individual, but of the team, the institution, the region, and the nation. Official statistics of education provide indicators of learning achievement, and they have been well developed over the years, but they deal with the individual, not the group. However, it is the group that is instrumental in transmitting the economic effects of knowledge (Dierkes 2001). The R&D establishment of a firm can be used as proxy for the ability of the firm to absorb knowledge from outside and from inside, if the R&D team is connected to the production and the sales departments in a way that supports the flow of useful knowledge. However, not all firms engage in the activity of formal R&D, devoting at least one person-year to it. In Canada, less than 4% of manufacturing firms do R&D, but around one in 10 produces world-first innovations. Where does absorptive capacity come from for the innovating firms that do not do R&D? This raises the need for indicators of the characteristics of the labor force, as a well educated labor force, with good management and communication skills that can capture ideas, technologies, and practices from a variety of sources and integrate them into the value chain. Some proxy indicators are the percentage of the labor force with tertiary education and measures of the use of practices by the firm to attract, train, and retain workers of high quality. Firms and other organizations can engage in alliances, networks, and partnerships to enhance their opportunity to learn and to capture and absorb knowledge. Indicators of networks have been developed over the years (Hagadoorn 2001; de la Mothe and Link 2002) but have not yet become part of official statistics. The System In order to measure knowledge and to determine its economic effects, there have to be indicators that describe knowledge activities, linkages, and outcomes. The

38

Fred Gault

indicators discussed in this chapter have been a mixture of those based on official statistics and those developed by academic institutes, and in their present state they are insufficient to provide a clear understanding of ‘‘knowledge and its economic effects.’’ The question is whether it is possible to develop such a set of indicators, and then, whether it is feasible. If the System of National Accounts is taken as a model, it is difficult to establish a correlation between investment in fixed capital in a firm and economic outcomes, such as changes in quality of life, in the skill requirements of the labor force, or in employment levels. There are many studies on the correlation between investment in ICTs or R&D and economic growth, but they rely on a wealth of economic data made possible by many years of developing and using the standards of the System of National Accounts or of R&D measurement. In the wider context of knowledge measurement, the history is not yet there. There are organizations that work with member countries to set standards for definitions and the interpretation of data collected using the definitions. The SNA manual brings together the work of the Commission for the European Communities, the International Monetary Fund, the OECD, the United Nations, and the World Bank (1993). This community is now discussing whether to treat research and development as a capital investment rather than as an operating expense, a decision that could have considerable impact on official statistics and on how R&D is treated and analyzed in national accounts. This leads to a discussion of the economic effects of knowledge stocks, as well as flows. The OECD brings together 30 member countries to set the standards used to measure and interpret technological and related organizational change. The standards for R&D were established in the 1960s (OECD 1963), and those for innovation were first codified in the 1992 Oslo Manual (OECD 1992). In 2005, the OECD released a revision of the current Oslo Manual (OECD/Eurostat 1997) that will incorporate nontechnological innovation such as the development of markets or the finding of new markets or changes in organization or in the use of management practices. It will also incorporate a chapter on linkages to stress the importance of tying together sources of knowledge from outside the firm, the activity of innovation, and the outcomes of the activity. This is a significant step in making the point that indicators of activities, presented and used in isolation, are not the most effective way of describing the system that surrounds the activity. The need to describe the (knowledge) system has implications for how the statistical work is done and for policy development that uses the indicators. Greater collaboration in the development of indicators is needed, bringing together official statistics on education, training, lifelong learning, and related topics with those on technological and related organizational change.

Measuring Knowledge and Its Economic Effects

39

Understanding the system is one thing; influencing it is another. Official statistics on the financial system combined with government statements on the need to control inflation reduce expectations of inflation, and this has implications for wage settlements and the domestic or international mobility of the highly qualified. If the objective is to promote innovation or entrepreneurship, official statistics play a different role. They describe the present state of the system and allow it to be compared with, or benchmarked against, other systems. The official statistics then support the monitoring of progress toward the targets of government and can be used to identify best practices in other countries that could be implemented domestically. An example of this approach is the EU Innovation Scoreboard. Conclusions Measuring ‘‘knowledge’’ is a complex, if not impossible, undertaking, and relating knowledge to economic effects is more complex. If understanding of the knowledge system is to be advanced in a cost effective way, the policy community must involve itself in the development of new indicators and standards related to their development. This will ensure that the new indicators are used. The shift from producing indicators of activity toward indicators of linkages, or flows, and on to indicators of outcomes should continue if there is to be any hope of understanding the knowledge system, its actors, activities, and linkages, and its outcomes. This will make survey questionnaires longer and more complicated than they would be if only the activity were measured, but that increase in burden should be justified by more effective evidence-based policy. The breadth and complexity of the knowledge system must be acknowledged. The knowledge system does not involve just the firm and its activities, but schools, colleges, and universities, municipal, state, and federal governments, and organizations providing cultural and other leisure activities. The firm that is producing the value added is connected locally, since that is where its people live, but also globally to other firms involved with the same kind of knowledge. While statistics and policies relating to all of these areas could be brought together more easily under a central government, such integration presents a real challenge to federal governments. It does, however, underline the need for collaborative development of both policy and measurement across jurisdictions and disciplines. Acknowledgments The author is grateful for comments from colleagues from Statistics Canada and from the international community involved in measuring knowledge, the information society, and technological and organizational change.

40

Fred Gault

References Chataway, Joanne, Paul Quintas, David Wield, and Fred Gault (2003). ‘‘From Digital Divide to Knowledge Divide—A Primer.’’ In Monitoring the Digital Divide and Beyond, George Sciadas, editor. Montreal: Orbicom. Commission for the European Communities, the International Monetary Fund, the Organisation for Economic Co-operation and Development, the United Nations, and the World Bank (1993). System of National Accounts 1993. Brussels/Luxembourg, New York, Paris, Washington, DC. Commission of the European Communities (2004a). ‘‘Innovation in Europe: Results for the EU, Iceland and Norway.’’ Luxembourg: Office for Official Publications of the European Communities. ——— (2004b). ‘‘European Innovation Scoreboard 2004: Comparative Analysis of Innovation Performance.’’ Commission Staff Working Paper SEC (2004) 1475. Brussels: Author. Dahlman, Carl J., and Jean-Eric Aubert (2001). ‘‘China and the Knowledge Economy: Seizing the 21st Century.’’ WBI Development Series, Washington DC: World Bank. de la Mothe, J., and A. N. Link, editors (2002). Networks, Alliances, and Partnerships. Boston: Kluwer Academic Publishers. Dierkes, Meinolf (2001). ‘‘Practice and Knowledge Management.’’ In Knowledge Management in the Innovation Process, pp. 9–42, John de la Mothe and Dominique Foray, editors. Boston: Kluwer Academic Publishers. Earl, Louise, and Fred Gault (2003). ‘‘Knowledge Management: Size Matters.’’ In Measuring Knowledge Management in the Business Sector: First Steps, Dominique Foray and Fred Gault, editors, pp. 169–188. Paris: OECD. Florida, Richard (2002). The Rise of the Creative Class, and How It’s Transforming Work, Leisure, Community and Everyday Life. Philadelphia: Basic Books. Foray, Dominique (2004). The Economics of Knowledge. Cambridge MA: MIT Press. ———, and Paul David (1995). ‘‘Accessing and Expanding the Science and Technology Knowledge Base.’’ STI Science, Technology, and Industry Review 16: 13–68. ———, and Fred Gault (2003). Measuring Knowledge Management in the Business Sector: First Steps. Paris: OECD. Gault, Fred, editor (2003). Understanding Innovation in Canadian Industry. Montreal and Kingston: McGill-Queen’s University Press. Godin, B., Y. Gingras, and L. Davignon (1998). ‘‘Knowledge Flows in Canada as Measured by Bibliometrics.’’ Catalogue No. 88F0006XIE19980010, Ottawa: Statistics Canada. Hagadoorn, J. (2001). ‘‘Inter-Firm Partnerships—An Overview of Major Trends and Patterns Since 1960.’’ in Strategic Partnerships: Proceedings from an NSF Workshop, J. E. Jankowski, A. N. Link and N. S. Vonortas, editors, pp. 63–92. Washington DC: National Science Foundation. Hatzichronoglou, T. (1997). ‘‘Revision of the High-Technology Sector and Product Classification.’’ OECD STI Working Paper 1997/2, Paris: OECD.

Measuring Knowledge and Its Economic Effects

41

Hodgson, G. M. (2000). ‘‘Socio-Economic Consequences of the Advance of Complexity and Knowledge.’’ In The Creative Society of the 21st Century, pp. 89–112. Paris: OECD. Kline, S. J., and N. Rosenberg (1986). ‘‘An Overview of Innovation.’’ In The Positive Sum Strategy: Harnessing Technology for Economic Growth, R. Landau and N. Rosenberg, editors, p. 289. Washington, DC: National Academy Press. Lundvall, B. (2000). ‘‘The Learning Economy: Some Implications for the Knowledge Base of Health and Education Systems.’’ In Knowledge Management in the Learning Society, pp. 125–141. Paris: OECD. ———, and B. Johnson (1994). ‘‘This Learning Economy.’’ Journal of Industry Studies 1, No. 2 (December): 23–42. OECD (1963). Proposed Standard Practice for Surveys of Research and Development: The Measurement of Scientific and Technical Activities. Directorate for Scientific Affairs, DAS/ PD/62.47, Paris: OECD. ——— (1990). Proposed Standard Method for Compiling and Interpreting Technology Balance of Payments Data: TBP Manual 1990. Measurement of Scientific and Technological Activities Series, Paris: OECD. ——— (1992). Proposed Guidelines for Collecting and Interpreting Technological Innovation Data: Oslo Manual. OECD/GD(92)26, Paris: OECD. ——— (1994). Using Patent Data as Science and Technology Indicators—Patent Manual 1994. The Measurement of Scientific and Technological Activities Series, Paris: OECD/ GD(94)/114, 1994, Paris: OECD. ——— (1995). The Measurement of Human Resources Devoted to Science and Technology—Canberra Manual. Paris and Luxembourg: OECD and Eurostat. ——— (2001). Innovative Clusters: Drivers of National Innovation Systems. Paris: OECD. ——— (2002a). Frascati Manual: Proposed Standard Practice for Surveys on Research and Experimental Development. Paris: OECD. ——— (2002b). OECD Science, Technology and Industry Outlook. Paris: OECD. ——— (2002c). Measuring the Information Economy. Paris: OECD. ———(2003). OECD Science, Technology and Industry Scoreboard. Paris: OECD. ——— (2004a). Main Science and Technology Indicators, Vol. 2004. Paris: OECD. ——— (2004b). ‘‘Meeting of the OECD Committee for Scientific and Technological Policy at Ministerial Level: Science, Technology and Innovation for the 21st Century.’’ OECD news release January 29–30, 2004, Final Communique´, Paris: OECD. ——— (2004c). Learning from Tomorrow’s World: First Results from PISA 2003, Paris: OECD. OECD/Eurostat (1995). The Measurement of Human Resources Devoted to Science and Technology—Canberra Manual. Measurement of Scientific and Technological Activities Series. Paris and Luxembourg: OECD and Eurostat. ——— (1997). Proposed Guidelines for Collecting and Interpreting Technological Innovation Data: Oslo Manual. Paris and Luxembourg: OECD and Eurostat. Porter, M. (1990). The Competitive Advantage of Nations. New York: Free Press.

42

Fred Gault

Prusak, Larry (2001). ‘‘Practice and Knowledge Management.’’ In Knowledge Management in the Innovation Process, John de la Mothe and Dominique Foray, editors, pp. 153–157. Boston: Kluwer Academic Publishers. Schumpeter, J. (1947). Capitalism, Socialism and Democracy, 2nd ed. New York: Harper and Row. Statistics Canada (2004). The Daily (December 23), Catalogue No. 11-001E, Ottawa: Author. Stehr, Nico (1996). ‘‘Knowledge as a Capacity 88F0017MIE1996002, Ottawa: Statistics Canada.

for

Action.’’

Catalogue

No.

UN (1990). International Standard Industrial Classification of All Economic Activities, Revision 3, Series M, No. 4/Rev. 3. New York: Author. ——— (2002). International Standard Industrial Classification of All Economic Activities, Revision 3.1, Series M, No. 4/Rev. 3.1. New York: Author. von Hippel, E. (1988). The Sources of Innovation. Cambridge, MA: Cambridge University Press. ——— (1998). ‘‘Economics of Product Development by Users: The Impact of ‘Sticky’ Information.’’ Management Science 44: 5.

5 Assessing Innovation Capacity: Fitting Strategy, Indicators, and Policy to the Right Framework Reinhilde Veugelers

Introduction Europe’s growth performance has been the subject of increasing scrutiny during recent years, most notably in the context of the Lisbon process to make the European Union the most knowledge-based economy in the world and its efforts to encourage governments to introduce employment- and productivity-enhancing reforms. This reform agenda is all the more pressing given that the European Union’s underlying growth rate has been trending downward since the second half of the 1990s. The much debated analysis of the contribution to overall productivity growth from ICT production and use, described below, indicates the European Union’s difficulty in reorienting its economy toward the newer, higher productivity growth sectors such as ICT. At the same time, it raises the broader issue, discussed later, of whether the European Union is insufficiently capable of creating and exploiting new technologies in general. Tackling the deficient E.U. innovative capacity requires a broad systemic policy framework that goes well beyond targeting R&D budgets but unfortunately includes many factors difficult to document with statistical indicators. We evaluate the actual policy strategy developed to tackle the European Union’s growth challenge, namely the Lisbon strategy. More particularly we examine the choice of policy priorities and structural reforms for tackling the deficiencies in the innovative capacity. In addition, we analyze the set of indicators chosen to evaluate progress. We then draw the policy conclusions. Are the right indicators chosen for informing improvement in innovative capacity? What are the interactions and complementarities between the various reforms and indicators? We also consider the need to monitor and evaluate the indicators—and whether this should be done at aggregate or sectoral levels, at the E.U., national, or regional level.

44

Reinhilde Veugelers

Assessing the Problem: The European Union’s Relative Productivity Performance Enhancing productivity growth is fundamental to realizing the Lisbon ambition. We first focus on the nature and source of the deterioration in the European Union’s productivity growth performance relative to that in the United States since the mid1990s, which will serve the subsequent discussion of the policy approach to be adopted in order to remedy this situation. Where Are the European Union’s Productivity Growth Problems Coming From? The Importance of Knowledge Production and Diffusion One of the most popular explanations for the diverging productivity fortunes of the European Union and the United States has been the relative exposure of both areas to ICT. A primary source of U.S. productivity acceleration in the 1990s has been the increasing share of ICT production in the United States, combined with extraordinary gains in productivity. However, given the general purpose technology characteristics of ICT, one should also see productivity gains from using that technology, further sustaining the ICT effect on aggregate productivity (e.g., Jovanovic and Rousseau forthcoming; Bresnahan and Trajtenberg 1996). In fact, using a growth accounting framework, both the ICT producing manufacturing and intensive ICTusing private services categories are causing the 1996–2000 divergences in European Union–United States productivity growth rates. It is precisely in these two areas of the economy where the European Union fares most poorly relative to the United States, either in terms of the size of the respective industries (i.e., small shares of overall E.U. output) and/or in having relatively low productivity growth rates (European Commission 2004b, Annual Review).1 An important question to examine is the extent to which the example of ICT is an isolated case or is likely to be replicated in other high-growth, high-tech industries. If this is a credible risk, then the key question is whether the European Union has specific problems in relation to its innovation infrastructure and whether the United States has a better innovation capacity in general than the European Union. In overall terms, when one assesses the evidence in relation to the manufacturing sector, it is fair to conclude that the overall R&D infrastructure of the United States seems to dominate that of the European Union. Not only does the United States display a higher R&D intensity overall, it also has a larger weight of its production concentrated in high-tech sectors, and it realizes a better growth performance in high-tech sectors. Hence, differences in innovative capacity are a prime candidate to explain the European Union–United States differences in productivity growth performances, particularly in high-tech manufacturing industries (see Table 1).

Assessing Innovation Capacity

45

Table 1 Comparison of European Union–United States differences in R&D spending and productivity growth in high-tech manufacturing sectors

High-tech manufacturing ICT Non–ICT

E.U.–U.S. gap in R&D spending

E.U.–U.S. gap in VA (specialization)

E.U.–U.S. gap in productivity growth

1991– 1995

1996– 1999

1991– 1995

1996– 2000

1991– 1995

1996– 2000

0.686 0.552 0.783

0.621 0.411 0.813

0.825 0.45 0.98

0.826 0.42 1.01

0.48 0.23 1.15

0.41 0.27 2.81

Note: U.S. ¼ 1. Source: European Commission 2004b, Annual Review.

While Table 1 illustrates the United States’ higher specialization, larger R&D intensity, and higher productivity growth in ICT sectors compared to the European Union, for non–ICT high-tech sectors, the picture is less devastating for the European Union, particularly in the second part of the 1990s. There is no difference in specialization in these industries, nor a productivity disadvantage. The gap in total expenditures on R&D is also smaller than in total. Unfortunately, these sectors, often being only medium- to high-tech, have far less scope for productivity growth than the ICT industries. The main conclusion is that while there are examples of good performance, in particular sectors and particular member states, overall the E.U. innovation environment remains weak in a number of key ‘‘input’’ indicators. But in addition to investments in R&D, there are also other characteristics of the E.U. innovation system that need to be looked at to explain its problems, to focus on the high-productivity growth areas, and to gain a higher rate of return from its knowledge investments. What Determines a Nation’s Innovation Capacity? What are these factors that determine an economy’s ‘‘national innovation capacity,’’ defined as the ability of a nation not only to produce new ideas but also to commercialize a flow of innovative technologies over the longer term? Using the insights from macro- and microeconomics and national innovation systems models (e.g., Aghion and Howitt 1992; Romer 1990; Lundvall 1992; Freeman 1987; Nelson 1993) applied economic theorists (e.g., Furman et al. 2002) have synthesized what determines an economy’s national innovation capacity. From this perspective a range of factors are deemed to be important for effective innovation effort. A

46

Reinhilde Veugelers

sufficiently developed ‘‘supply’’ side of R&D (as reflected in the amount of R&D carried out or the number of skilled researchers) is a necessary but insufficient condition for successful innovation. Broader framework conditions are important as well, including a sufficient ‘‘demand’’ for innovation to reward successful innovators. This requires sophisticated lead users willing to pay for innovations, effective intellectual property rights (IPR) schemes, a favorable macroeconomic environment, and effective competition in output markets, and especially market entry and exit processes. But perhaps the most critical element in the framework is the interconnectedness of the agents in the system, linking the common innovation infrastructure to specific technology clusters. This requires good industry–science links and well-functioning capital and labor markets. In the national innovation capacity perspective, country differences with respect to innovation and growth might reflect not just different endowments in terms of labor, capital, and the stock of knowledge but also the varying degrees of the ‘‘knowledge distribution power’’ or the efficiency of the innovation system. Overall, this perspective warns against looking at statistical indicators individually to assess the performance of a national innovation capacity. Rather, a systemic approach should be taken to understand the relationships between science, technology, and innovation and socioeconomic development. The problem with this approach, however, is to approximate empirically the institutional framework and the ‘‘knowledge distribution power’’ of nations. What is available at present are only pieces of statistical evidence showing the importance of interactions, such as the availability of venture-backed financing, cooperation in R&D among firms and between science and industry, (international) copatenting, the number of researchers employed by business, etc. (see, for example, Furman et al. 2002, Economic Policy Committee 2004). These framework features, although more difficult to document with statistical indicators, need however to be taken into account when we want to understand the relative overall effectiveness of a country’s innovative systems, if not quantitatively, then at least qualitatively. The Economic Policy Committee (2004), from combining fieldwork evidence and an analysis of statistical indicators, concludes that ‘‘market pull conditions’’ and knowledge networks are the key areas of E.U. weakness. The European Union generates a great deal of knowledge in its universities and research institutes and produces large numbers of skilled personnel. But often it does not exploit this knowledge and expertise for social and economic needs (the ‘‘European paradox’’). This compares to the United States, which has a better connectedness of science and industry with an openly competitive system of private and public universities and government subsidies to universities through peerreviewed research grants, which result in a higher quality of the research base. Other

Assessing Innovation Capacity

47

important framework conditions present in the United States are the advantage of a large, unified market unencumbered by differences in language, customs, and standards, a clearer and stronger U.S. intellectual property rights system, more flexible financial markets, making available venture capital finance to innovating firms, and more flexible labor markets, affecting both internal migration and the international immigration of highly skilled people (Gordon 2004b). The Policy Reaction: The Need for a ‘‘Systemic Approach’’ The overarching policy implication from all this is a need for a ‘‘systemic’’ policy approach to improve the European Union’s innovative potential. At the European Council of March 2000 in Lisbon, the European Union launched a comprehensive set of integrated structural reforms geared toward the general objective of becoming ‘‘the most dynamic and competitive knowledge-based economy in the world, capable of sustainable economic growth with more and better jobs and greater social cohesion’’ as well as ‘‘an increasing respect for the environment.’’ With the adoption of the strategy of Lisbon, as it became known, the European leaders acknowledged the need for profound reforms in the European Union. The scope of the Lisbon strategy has been wide from the outset, not only in terms of objectives but also in terms of the policy tools to be used. Table 2 classifies the Table 2 Lisbon reforms Investments in the knowledge-based economy Invest in R&D and innovation Invest in education and training Encourage production and use of ICT Product and capital market reforms Improve the business environment Improve the functioning of the internal markets for goods and services Promote E.U. financial integration

Labor market reforms Improve incentives to participate and remain in the labor market Improve matching between human resources and vacancies Increase labor market flexibility

Social policy reforms Modernize social protection systems Improve working conditions and skill levels

Environmental policy reforms Improve understanding of environmental problems Increase use of cost–benefit analysis Increase use of market based instruments

Source: Adapted from Economic Commission 2005.

48

Reinhilde Veugelers

Lisbon reforms into five categories: investments in the knowledge-based economy, product and capital market reforms, labor market reforms, social policy reforms, and environmental policy reforms. As the previous sections have documented, tackling the deficiencies in the innovative capacity requires a systemic policy perspective. The Lisbon strategy embodies the idea that to yield maximum synergies from structural reforms, they are best implemented in a comprehensive and coordinated way. Therefore, beyond stimulating the research inputs from the public and private sectors, it is important that other structural reforms are part of the Lisbon agenda as well, to improve the European Union’s innovative capacity. Investment and innovation benefit from a more competitive and entrepreneurial environment, fostered by structural reforms on product, capital, and labor markets that improve the transfer of resources from low-productivity to higher productivity use. With well-functioning product markets, firms will have incentives to innovate and new firms embodying new ideas can flow into the market. Furthermore, new business opportunities can be taken advantage of only if appropriately educated and skilled workers can be hired under the right conditions. This requires flexible labor markets to provide innovators access to researchers and skilled human capital. Similarly, well-functioning risk capital markets ensure innovators access to financial capital to finance their risky projects. High-tech start-ups, often an important source of breakthrough innovations, in particular need open product markets with low entry barriers and access to capital, especially early stage financing of high-risk ventures. Implementing the Policy Strategy: Need for Indicators The wide scope of the Lisbon strategy made it necessary to focus and identify a restricted, well-defined set of targets and policy measures necessary to achieve the objectives and at the same time a corresponding restricted set of indicators to monitor progress on the targets. Defining the Targets and Policy Measures With respect to research and innovation, a key element of the Lisbon strategy has been to speed up the transition toward a knowledge-driven economy under the umbrella of a European Knowledge Area (EKA). Action has been shaped around a range of initiatives ranging from e-Europe and the creation of a European Research Area to promoting innovation and establishing common objectives at the E.U. level for national education policies. For innovation and research, the following list summarizes the targets and policy measures with respect to the EKA:

Assessing Innovation Capacity

49

Research Network national and joint research programs on a voluntary basis around freely chosen objectives.




Improve the environment for private research investment, R&D partnerships, and high-technology start-ups.







Develop an open method of coordination for national research policies.




Roll out a world-class research communications infrastructure.

Remove obstacles to the mobility of researchers; attract and retain high-quality research talent in Europe.







Introduce a cost-effective community patent.

Harness new and frontier technologies, notably biotechnology and environmental technologies (Stockholm).







Fully implement the e-Europe Action Plan by 2005.

Information Society


Ensure that all teachers have training in digital skills by 2003.

Ensure access to a widespread, world-class communications infrastructure and ensure significant reduction in the cost of using the Internet (local loop unbundling).







Create conditions for e-commerce to flourish.




Prevent info exclusion.




Stimulate e-government.

Support takeup of 3G mobile communications and introduction of Internet Protocol version 6.




Education


Achieve a substantial increase in per capita spending on human resources.




Promote lifelong learning.




Adapt skills base better to needs of knowledge society.




Better recognize qualifications.

Promote learning of E.U. languages and introduce a European dimension to education.







Promote school twinning via the Internet.

This has been translated into specific and measurable targets for the European Knowledge Area:

50

Reinhilde Veugelers




Increase R&D spending with the aim of approaching 3% of GDP by 2010.




The proportion financed by business should rise to two-thirds of that total by 2010.




100% of schools should be connected to the Internet by 2002.




100% of teachers should have training in digital skills by 2003.




Internet penetration in households should reach 30% by 2002.




Basic governmental services should be 100% online by 2002.

Indicators Structural Indicators To monitor the progress on the main targets of the Lisbon strategy, the European Commission and Council agreed on a list of 14 main structural indicators. Member states’ performance on these indicators is continuously being assessed. For the European Knowledge Area, R&D expenditures as a percentage of GDP, with a target of 3%, is the main structural indicator.2 But beyond this main indicator, there are other (secondary) structural indicators of the EKA, which are continuously being monitored:


Spending on human resources (public expenditure on education),




GERD (gross domestic expenditure on R&D) by source of funds (private–public),




Level of Internet access—households/enterprises,




Science and technology graduates—total/females/males,




Patents—EPO/USPTO,




Venture capital investments—early stage/expansion and replacement, and




ICT expenditure—IT/telecommunications expenditure.

Innovation Indicators Beyond the structural indicators that cover all Lisbon areas, the Lisbon European Council also requested, for the area of innovation and R&D, the development of the European Innovation Scoreboard (EIS) by DG Enterprise.3 The 2003 EIS contains 19 main indicators, selected to summarize the main drivers and outputs of innovation. These indicators are divided into four groups: human resources for innovation (five indicators); the creation of new knowledge (four indicators); the transmission and application of knowledge (three indicators); and innovation finance, output, and markets (seven indicators). The EIS mainly uses Eurostat data, covering 32 countries. Six of the 19 indicators are drawn from the E.U. Structural Indicators.

Assessing Innovation Capacity

51

1. Human resources 1.1 Science and engineering graduates (percentage of 20–29 years age class) 1.2 Population with tertiary education (percentage of 25–64 years age class) 1.3 Participation in life-long learning (percentage of 25–64 years age class) 1.4 Employment in medium-high and high-tech manufacturing (percentage of total workforce) 1.5 Employment in high-tech services (percentage of total workforce) 2. Knowledge creation 2.1 Public R&D expenditures (GERD–BERD) (percentage of GDP) 2.2 Business expenditures on R&D (BERD) (percentage of GDP) 2.3.1 EPO high-tech patent applications (per million population) 2.3.2 USPTO high-tech patent applications (per million population) 2.4.1 EPO patent applications (per million population) 2.4.2 USPTO patents granted (per million population) 3. Transmission and application of knowledge 3.1 SMEs innovating in-house (percentage of manufacturing SMEs and percentage of service SMEs) 3.2 SMEs involved in innovation cooperation (percentage of manufacturing and service SMEs) 3.3 Innovation expenditures (percentage of all turnover in manufacturing and services) 4. Innovation finance, output, and markets 4.1 Share of high-tech venture capital investment 4.2 Share of early stage venture capital in GDP 4.3.1 SME sales of ‘‘new to market’’ products (percentage of all turnover in manufacturing and service SMEs) 4.3.2 SME sales of ‘‘new to the firm but not new to the market’’ products (percentage of all turnover in manufacturing and service SMEs) 4.4 Internet access/use 4.5 ICT expenditures (percentage of GDP) 4.6 Share of manufacturing value-added in high-tech sectors 4.7 Volatility rates of SMEs (percentage of manufacturing and service SMEs)

52

Reinhilde Veugelers

Research Indicators In support of the European Research Area (ERA) initiative, DG Research was entrusted with a mission to produce a set of indicators and a methodology for benchmarking research policies in the member states. A set of 20 indicators was proposed to help monitor and report on progress toward the ERA. Most of the indicators are already used in other commission publications. Eight indicators are also used in the European Innovation Scoreboard (DG Research 2003). Evaluating Performance on the Selected Indicators In 2004 at midterm, the Lisbon strategy was evaluated by a group of reviewers chaired by Mr. Kok, known as the ‘‘Kok group.’’ The objectives initially set out were far from being achieved. During the last four years, the overall growth performance of the European economy has been disappointing. In fact, after having peaked in the mid-1990s at around 97% of U.S. levels, E.U. labor productivity per hour is projected to deteriorate to around 88% in 2005, which is close to its relative level in the early 1980s. This post-1995 deterioration in relative productivity levels reflects a sharp decline in E.U. productivity growth rates relative to those of the United States during the period in question (European Commission 2005). One of the most disappointing aspects of the Lisbon strategy to date is the performance on R&D. On the R&D target, only two countries (Finland and Sweden) currently have R&D spending exceeding 3% of GDP; in these same two countries business is achieving the goal of spending the equivalent of 2% of GDP on R&D. The rest are behind on both scores. Progress in providing every teacher with digital training is very disappointing. On a positive note, member states have progressed in the spread of ICT and Internet use in schools, universities, administration, and trade. Household Internet penetration, for example, has risen rapidly, with 12 member states meeting the targets. The little progress that has been made on R&D is all the more remarkable taking into account that the Lisbon European Council rightly recognized that Europe’s future economic development would depend crucially on its ability to create and grow high-value, innovative, and research-based sectors. However, the knowledge society is a larger concept than just an increased commitment to R&D expenditures. Further zooming in on innovation and R&D, Table 3 provides a look at a selected combination of indicators from the innovation indicators discussed previously for the EU15 relative to the United States and Japan. As Table 3 indicates, Europe continues to lag behind the United States and Japan on several indicators. Although science and technology graduates show no gap, both Japan and the United States have significantly more working population with

Assessing Innovation Capacity

53

Table 3 Selection of main EIS indicators—A triad comparison EU15 1. Science and engineering graduates 2. Working population with tertiary education 3. Total R&D expenditures (GERD as percentage of GDP) 4. Public R&D expenditures (GERD–BERD) (percentage of GDP) 5. Business expenditures on R&D (percentage of GDP) 6. Early stage venture capital in GDP 7. EPO patent applications (per million population) 8. USPTO patent applications (per million population) 9. EPO high-tech patent applications (per million population) 10. Share of high-tech sectors in manufacturing value added 11. ICT expenditures (percentage of GDP)

U.S.

Japan

11.3 21.5 1.99

10.2 37.2 2.80

33.8 3.01

0.69

0.76

0.81

1.30

2.04

2.28

0.037 161.1

0.218 169.8

174.7

80.1

322.5

265.2

31.6

57.0

44.9

14.1

23.0

18.7

7.0

8.2

9.0

Source: European Commission (2004a).

tertiary education. Both government and, in particular, business expenditures on R&D are considerably lower within Europe. Moreover, growth rate differentials reveal a similar trend, suggesting a further widening of this gap in the near future. Also, with respect to invested venture capital, within Europe the amounts of resources available (divided by GDP) are clearly lower. This difference in the level of science and technology ‘‘inputs’’ between the European Union and the United States is accompanied by lower levels of technological output. Both the United States and Japan outperform Europe in terms of technological performance as measured by the number of patents per million inhabitants, with this ‘‘gap’’ being even more pronounced for high-tech patents. Similarly, in terms of the share of added value within manufacturing industries, the difference from the United States is especially striking. Broadly speaking, a picture emerges from the indicators in which the European Union continues to lag behind on the level of technological performance and related—technology intensive—economic activity. All of this confirms the analysis of the persistent deficiency of the European Union’s innovative and growth capacity

54

Reinhilde Veugelers

underlying the rationale for the Lisbon strategy, as discussed previously. Hence, the conclusion seems inescapable that Lisbon as yet has failed to deliver with respect to improving the European Union’s innovative capacity. Conclusions for the E.U. Policy Process At midterm it is clear that the Lisbon strategy has not delivered what was expected, particularly with respect to the knowledge economy, calling for an evaluation of the policy strategy. Does the European Union Have a Systemic Policy Approach? A ‘‘systemic’’ approach builds on complementarities among reforms: In order to reach maximum effectiveness, measures in one reform domain need to be accompanied by flanking measures in another domain. Measures that increase the level of competition in product markets, for example, often lead to economic restructuring, implying job losses in some sectors and employment creation in others. Wellfunctioning labor markets would tend to facilitate such a transition. Hence, in view of the complementarities, Lisbon should be broad, i.e., should cover multiple policy areas. The problem is, however, that the Lisbon strategy is not sufficiently connected to be understood as a truly ‘‘systemic’’ endeavor. Lisbon is a collection of policy initiatives rather than a truly integrated view. This is why the Kok group successfully called for more focus.4 For Europe to increase its living standards, it needs to focus on employment and productivity growth. But in line with the systemic approach, this needs to be done through a wide range of reform policies as well as a wider macroeconomic framework as supportive as possible of growth. No single action will deliver higher growth. Rather, a series of interconnected initiatives and structural changes are needed. In line with a ‘‘systemic perspective,’’ the renewed Lisbon strategy, although it clearly carries as a priority progress on the knowledge society, calls for action in other areas of policy as well, such as the completion of the internal market and improving the business climate. Evaluating the Choice of Indicators for Improving the Innovative Capacity Are the Sets of Indicators Chosen to Evaluate Progress the Right Indicators for Informing Improvement Toward the Lisbon Objectives? The set of indicators—both the structural indicators and those specific for research and innovation—although being restricted by data availability, clearly look like be-

Assessing Innovation Capacity

55

ing inspired by the specific weaknesses of the E.U. innovative capacity and the ‘‘systems’’ approach toward improving this capacity. Although R&D spending is a central structural indicator, it fits into a set of other structural indicators, allowing integration with labor, capital, and product market reforms. The targets in other areas are also important to improving the innovative capacity. For instance, assessing the ease of entry of new firms and their survival is important to have new ideas coming to market. For this, targets like the Risk Capital Action Plan, lowering the cost of doing business, further opening up markets, are important to assess as well. But also targets like lifelong learning and the reduction of barriers to labor mobility between member states will improve the human capital resources necessary for implementing innovation strategies. Even sustainable development targets, when directed towards green technology development, could be developed into an E.U. innovative strength. Are We Measuring the Right Indicators for Informing Improvements in Innovative Capacity? A key message from a systemic approach is that the effectiveness of innovation systems depends on the balanced combination of creative capacity, diffusion capacity, and absorption capacity. These dimensions are all somehow reflected in the selected indicators. The targets selected for the European Knowledge Area, beyond R&D expenditures, reflect the importance of a highly educated labor force as central in the European Union’s creative, distributive, and absorptive capacity. It also reflects the specific importance of ICT in the European Union’s growth agenda as a general purpose technology and recognizes the importance of financing for innovation. The area of indicators that is least represented is scientific performance and industry–science links. Especially the lack of industry–science link indicators is disturbing since this is one of the particular deficiencies of the E.U. innovative capacity (cf. European Paradox). What we are missing in the set of main indicators are those on industry–science links such as, for instance, cooperation between firms and research institutes, copatenting and copublishing, researcher mobility between industry and science, private research funding of basic research, patenting by universities and public research institutes, and spinoffs. This is partly due to a lack of systematic data on this, but clearly more could be done here (Gault, this volume). Are We Measuring the Indicators at the Right Level of Aggregation? Underlying any aggregate innovation indicator is the structural makeup of the economy, which differs greatly between E.U. countries. Such structural differences can have an important role in explaining some of the differences in innovation performance. The main reason is that there is a great deal of diversity among industrial

56

Reinhilde Veugelers

sectors in terms of innovation process, innovation inputs, and outputs. First, technological opportunities differ across sectors, with ICT as a prime example of the high-growth sector, with huge opportunities for technological advance. Other major differences across sectors are the sizes of the innovating units and the objectives of innovation, with product versus process innovations. There is also a great deal of diversity among the sources of innovation: in-house R&D laboratories, suppliers, users, public research institutes. This implies that there will be major differences across sectors in many of the indicators used in the EIS, for example those based on R&D, patenting, SMEs, and innovation expenditures. Since the systemic approach operates at the specific technology or sectoral level, this implies that indicators should be traced at technology/sectoral level. The single most important constraint is the lack of data at the sector level for some key variables. Nevertheless, the main conclusion from an exploratory analysis of the EIS indicators at sectoral level is that there is a great deal to be gained by analyzing innovation performance across sectors. Another area of overaggregation is the geographical dimension. Regional-level data are of value for two reasons. Innovation policies are often developed and implemented at the regional level, in addition to national- and E.U.-level policies. Regional indicators can help inform these policies. Beyond the selection of indicators and the level at which they should be evaluated, there is also the issue of the systemic approach to evaluating the indicators. Since multiple dimensions need to be measured for innovation capacity, multiple indicators need to be developed and assessed simultaneously. The recognition of the need to measure multiple dimensions has led to an emerging call for composite indicators. Also, the European Union had advocated the use of composite indicators. The EIS 2003, for instance, contains two such composite indicators as well as the 3% Action Plan. However, the current use and implementation of composite indicators reflects that the main motivation is the need to summarize different indicators, rather than reflecting the need for a systemic evaluation of indicators. This is clear since the weighing of indicators is mostly statistical rather than guided by a conceptual model. Implications for STI Policies We conclude by drawing some implications for a systemic policy process. Enhancing Horizontal Policy Coordination Increasing the efficiency of STI policies implies improving the policy arena in terms of coordination among various policy makers. STI policies should not be designed in isolation from each other (research policies, innovation and education policies)

Assessing Innovation Capacity

57

in close interaction with other policy areas (financial markets, labor markets, product markets, macroeconomic stability, environmental policies). Close cooperation among decision-making instances or even integration should be explored to guide prioritization processes and to better exploit synergies. Enhancing Vertical Policy Coordination The natural tendency for R&D resources to concentrate geographically should be reflected in a decentralized policy design, but this should be accompanied by coordination of policies among regional, national, and international policy makers. The Lisbon strategy and the ERA should not be thought of as a harmonization process: Innovative and productive structures differ across countries and regions. A decentralized policy approach implies more possibilities of adaptation to local specific needs in order to better align the various complementary local actors. Flexibility of policy measures is needed at the various administrative levels, especially between national and regional levels. Nevertheless, coordination among the various policy levels is important. The progressive opening of national programs, cross-fertilization measures, and international mobilization of human resources need to be promoted. The idea is to facilitate cooperation and to boost diffusion and uptake of knowledge by increasing the efficiency of the resources used. Improving the Management of the Policy Framework Well-developed skills and competencies are needed within the policy world itself. Inventiveness and creativity in policy building will be enhanced if policy makers can access experiences of other countries, provided these are presented in their context and evaluated properly. Benchmarking exercises involving policy makers should be conceived as ‘‘learning-by-interacting’’ exercises rather than ‘‘diffusion of best practices.’’ In addition, STI policies need to be supported by monitoring and evaluation (scientific, external) practices, which then feed back into the policy process. Finally, involving stakeholders in policy making is necessary. This emphasizes the importance of an appropriate governance system for policy. Acknowledgments The author acknowledges support from the European Commission Key Action ‘‘Improving the Socio-Economic Knowledge Base’’ through contract No. HPSE-CT2002-00146, the Belgian Federal Government DWTC (IUAP P5/11/33), KULeuven (OT/04/07A), and the Flemish Government (Steunpunt O&Ostatistieken). The chapter reflects only the views of the author and does not commit the European Commission.

58

Reinhilde Veugelers

Notes 1. Regarding manufacturing, two sectors dominate the overall productivity patterns, namely semiconductors and office machinery. These are the two industries where the United States is clearly ahead, with semiconductors contributing five times more to U.S. productivity growth compared to the equivalent gains for the European Union and with office machinery contributing more than twice as much. Of the service industries that individually contributed significantly to overall productivity growth, the United States is dominant in the financial services area and wholesale and retail trade. 2. For a full list of performances on all structural indicators, see European Commission (2005). 3. In addition, in the framework of enterprise and industrial policy, there is the complementary Enterprise Policy Scoreboard. Several indicators in both scoreboards are identical. 4. At the same time, the midterm reviewers have called for an improvement in the governance of the Lisbon process. A major deficiency of the Lisbon strategy is the governance of the policy process, with a lack of peer pressure at the level of the member states and poor communication about the benefits to all actors involved. An ambitious and broad reform agenda needs a clear narrative in order to be able to communicate effectively about the need for it.

References Aghion, P., and P. Howitt (1992). ‘‘A Model of Growth Through Creative Destruction.’’ Econometrica 60: 323–351. Bresnahan, R., and M. Trajtenberg (1996). ‘‘General Purpose Technologies: Engines of Growth?’’ Journal of Econometrics 65: 83–108. David, P., and D. Foray (1995). ‘‘Accessing and Expanding the Science and Technology Knowledge Base.’’ STI Review 16. DG Research (2003). ‘‘Investing in Research: An Action Plan for Europe.’’ Economic Policy Committee (2004). ‘‘Mid-Term Review of the Lisbon Strategy: Progress Report by the Economic Policy Committee,’’ Annex B. European Commission (2003a). ‘‘3th Science & Technology Indicator Report.’’ DG Research. ——— (2003b). ‘‘European Innovation Scoreboard.’’ Technical Paper No. 1-4. ——— (2004a). ‘‘2003 Innovation Scoreboard.’’ Commission Staff Working Paper. ——— (2004b). ‘‘The EU Economy: 2004 Review.’’ European Economy. ——— (2005). ‘‘The Economic Costs of Non–Lisbon.’’ Occasional Papers No. 16, European Economy. Freeman, C. (1987). Technology and Economic Performance: Lessons from Japan. London: Pinter. Furman, J., M. Porter, and S. Stern (2002). ‘‘The Determinants of National Innovation Capacity.’’ Research Policy: 899–934.

Assessing Innovation Capacity

59

Gordon, R. (2004a). ‘‘Five Puzzles in the Behaviour of Productivity, Investment and Innovation.’’ CEPR Discussion Paper No. 4414. ——— (2004b). ‘‘Two Centuries of Economic Growth: Europe Chasing the American Frontier?’’ CEPR Discussion Paper No. 4415. ——— (2004c). ‘‘Why Was Europe Left at the Station When America’s Productivity Locomotive Departed?’’ CEPR Discussion Paper No. 4416. Jones, C. I. (2002). ‘‘Sources of US Economic Growth in a World of Ideas.’’ American Economic Review 92: 220–239. Jovanovic, B., and P. Rousseau (forthcoming). ‘‘General Purpose Technologies.’’ In Handbook of Economic Growth, P. Aghion, editor. Lundvall, B.-A., editor (1992). National Systems of Innovation: Towards a Theory of Innovation and Interactive Learning. London: Pinter. Nelson, R., editor (1993). National Innovation Systems: A Comparative Analysis. Oxford, UK: Oxford University Press. Romer, P. (1990). ‘‘Endogenous Technological Change.’’ Journal of Political Economy 98: S71–S102.

II Knowledge Communities

6 Interactive Learning, Social Capital, and Economic Performance Bengt-A˚ke Lundvall

Introduction In general, with the exception of new growth theory and the literature on national systems of innovation, there is a tendency for scholars working on innovation and knowledge not to confront the macroeconomic issues. Seen from the other side, it has been quite acceptable among macroeconomists to assume that what happens at the macroeconomic level can be well understood without bothering too much about technology and institutions related to innovation and learning. One temporary exception, when macroeconomists actually did get involved in the debate, was the ‘‘new economy’’ episode, when even mainstream OECD economists—stimulated by Alan Greenspan’s speeches on the new economy—for a short period referred to technology, and especially information technology, as a factor that might affect the workings of the aggregate economy. It was a pity that the basic hypotheses behind this concept were too crude. Their assumptions that it would be rather simple for firms and for the economy as a whole to absorb the new technology and to transform it into economic growth reflected some lack of understanding of institutions and innovation processes (Lundvall 2003). This episode, and what appeared as the burst of the IT bubble, might actually have reaffirmed to macroeconomists that it is best to stay away from difficult themes having to do with knowledge and innovation. This essay presents some preliminary arguments for why this might not be a good idea. Specifically, it will be argued that the kind of interactive learning that interconnects users and producers in processes aiming at new products may have a major impact on economic performance of the economy. To understand the prerequisites for such learning to take place should therefore be of major concern not only for management but also for policy-makers at the national level.1 In this chapter we link ‘‘learning by interacting’’ to macroeconomic dynamics. The argument is that this form for learning is fundamental since it transforms the

64

Bengt-A˚ke Lundvall

outcomes of learning by doing and learning by using from being local to becoming nonlocal. Embodying knowledge in new services and products may be seen as an alternative to codification as a mechanism of generalizing local knowledge. The argument is predominantly conceptual and will be built up through references to contributions from a handful of outstanding economists. The chapter is a follow-up of ideas first developed in the booklet Product Innovation and User– Producer Interaction (Lundvall 1985). Learning as the Major Source of Economic Growth—Pasinetti on Growth and Structural Change Pasinetti (1981) is one of the major postwar contributors to economic theory. Pasinetti builds a theoretical model with vertically integrated sectors each ending up with final commodities for private consumption. The dynamics driving structural change come from producer learning, resulting in productivity growth in these sectors, while the uneven growth in demand reflects consumer learning in connection with consumers’ adoption of new and old consumption goods. It is characteristic for Pasinetti’s argument that product innovation—in the form of new consumption goods—is seen as a prerequisite for avoiding stagnation and thus for sustainable growth. This is in contrast to most production theory that simply ignores product innovation. And it is also characteristic that the diffusion of new consumption goods is seen as requiring consumer learning. Consumers’ preferences are shaped in such learning processes and not given once and for all. These assumptions bring us closer to the realities of the modern economy than standard production and consumption theory. Empirical studies show that new products, together with change in process technology and in work organization, are key elements in economic growth (Christensen and Lundvall 2004). But the major aim of Pasinetti’s model is to build a theory explaining the logic of value creation in a dynamic context without specifying the institutional context, and the author, while recognizing the critical importance of the phenomena, abstains from explaining how and why learning takes place in respectively consumption and production. To get closer to understanding how learning takes place, it is actually necessary to open the black boxes of the vertically organized production chains and to be explicit regarding institutional framework and its impact on interactive learning. We will argue, first, that the organization of these chains in terms of markets, hierarchies, and networks is of crucial importance for the rate of productivity growth within each chain and, second, that the framework conditions at the level of the national system have a major impact on the organization and thereby on the actual rate of growth of the whole economy.

Interactive Learning, Social Capital, and Economic Performance

65

The Division of Labor and the Static Scale Effects What follows is very much in the spirit of Adam Smith (1776 [1904]). It may be seen as an attempt to bring his analytical framework up to date by taking into account the speedup of change and the increasing importance of knowledge creation and learning (David and Foray 2002, p. 4). The starting point is his fundamental hypothesis, that economic development may be defined as a process where the degree of specialization and the division of labor becomes more developed and complex. For our purpose it is fundamental to distinguish between two different dimensions of specialization, one where the specialization takes place within an organization and one where activities become separated from each other by organizational borders. Specialization within the firm we call internal specialization and that between organizations we call external specialization. The more developed internal specialization makes it possible to reap static economies of scale within the organization. These economies emanate from reducing the frequency of shifts in job tasks for each worker or team as well as from workers and teams learning to pursue their tasks more skillfully through longer cycles of repetition. We call this the internal static effect.2 A similar static effect emanates from increasing the vertical division of labor among organizations. A single firm specializing in producer goods—components, equipment, and systems—may—instead of serving a user within the same organization—serve many external clients and thereby obtain scale economies in production. Organizations become more efficient by reaping scale effects, and by focusing their use of specialized resources on tasks that are well suited for these resources, they also learn to pursue the tasks involved more proficiently. We call this the external static effect.3 Adam Smith also took into account dynamic effects affecting the rate of innovation, but these will be left aside now and returned to later. Entering Transaction Costs—Williamson (1975) Oliver Williamson’s seminal book from 1975 has had a major impact on the understanding of industrial organization and especially on the analysis of vertical integration and disintegration. The argument is that scale economies (‘‘technology’’ in Williamson’s terminology, internal and external static effects in ours) are of secondary importance compared to transaction costs when it comes to explaining vertical integration. Williamson’s arguments are well known. Uncertainty coupled with opportunism —agents pursuing their aims with guile—will, in contexts of ‘‘small numbers,’’

66

Bengt-A˚ke Lundvall

lead to high transaction costs and thereby give incentives for firms to integrate vertically, i.e., substituting a hierarchy for the market. Today his analytical model has become more complex, emphasizing asset specificity, taking hostages, and holdup situations. But the main line of argument remains unchanged. Starting from Williamson’s analysis it is a paradox that product innovation (innovations addressing the market) is a frequent phenomenon—at least as frequent as process innovation (Lundvall 1985). It can also be argued that ‘‘perfect’’ competition—with anonymous relationships between seller and buyer and with price/quantity as the only information accessible for agents—forms a very poor climate for product innovations. One way to solve the paradox is to introduce the idea of ‘‘organized markets’’ and different mechanisms that might limit opportunism in such markets. Seen in this light you need to take into account not only transaction costs but also benefits emanating from interactive learning between users and producers across these organized markets (Lundvall 1992). The Organization of Industry—Richardson’s Contribution When it comes to explaining vertical integration, most scholars associate with transaction cost theory and the contributions by Williamson. But, actually, G. B. Richardson had presented a different approach to vertical integration already in 1972 in his article ‘‘The Organisation of Industry’’ (Richardson 1972). Richardson argued simply that what firms would choose to do inside the organization would be ‘‘similar’’ activities—in a resource-based context it meant that they should draw upon similar capabilities. Ahead of his time, he went further and explained network formation as a governance form in between market and hierarchy, arguing that activities that were ‘‘complementary’’ but not ‘‘similar’’ would tend to be traded in the network form.4 In a more recent paper he revisits his analysis from 1972 (Richardson 2002). Here he argues that an economy with vertically integrated firms would be quite vulnerable to shifts in demand. If demand is reduced for an end product, a proportional reduction would be imposed at all the earlier stages of the production chain. If you assume limited interorganizational mobility of resources, this would in an economy with demand shifts contribute to low rates of capacity utilization.5 Richardson argues that for the economy as a whole this rigidity can be overcome by vertical disintegration. In the case of parallel vertically integrated production chains a fall in final demand for specific end products would affect the productivity of the vertically organized production chains significantly downward. In the vertically disintegrated production system the specialized producer can address other users and even reorient capabilities to other purposes more easily. We call this the external flexibility effect.6

Interactive Learning, Social Capital, and Economic Performance

67

Learning by Doing and Learning by Using There is growing consensus that one new tendency in the economy is a speedup of change and a simultaneous reduction in the life expectancy of knowledge (Foray and Lundvall 1996). One way to characterize the combination of the growing importance of knowledge in the economy and the more rapid depreciation of knowledge capital is to refer to a ‘‘learning economy’’ (Lundvall and Johnson 1994; Lundvall 1996). The most fundamental characteristic of this new stage is a rapid rate of change driven by technology, not least information technology, globalization, and market deregulation, forcing a big proportion of firms and of the workforce to engage in building new competencies in order to survive or avoid exclusion. This change in context makes the following considerations especially urgent. The concept ‘‘learning by doing’’ was made known by Kenneth Arrow (1962). He used empirical observations of learning curves and productivity growth patterns from the production of airplane bodies as inspiration for the idea. Later Rosenberg (1982) developed the idea of ‘‘learning by using’’ to explain the rapid reduction in the cost of using complex systems as users become more familiar with them. His empirical reference was to airway companies and their use of new generations of airplanes. As a follow-up Von Hippel and Tyre (1995) gave interesting illustrations of how the introduction of new process technology normally involves a phase of solving unforeseen problems—a phase during which both operators and technology developers learn by doing/using. These kinds of learning take place to different degrees in all parts of the economy. The more innovations in terms of new products and systems, the more learning will be imposed upon developers, producers, and users. But you might argue that their impact in relation to the whole economy is limited since the learning is ‘‘local’’ and ‘‘specific’’ to one specific user or producer, or perhaps it even remains embodied in individuals (this is argued in Foray 2000).7 This brings us to the core argument that ‘‘learning by interacting’’ is fundamental for the generalization of local learning. (With the side conclusion that ‘‘generalization of local knowledge’’ does not always take the form of codification!) Learning by Interacting One major argument against Williamson’s assumption that calculating transaction costs would be sufficient to analyze and explain vertical integration is that the separation of users from producers into two different organizations actually might enhance interactive learning.8 The idea is simple. If a producer integrates with a user, or the other way around, the integrated couple tends to become less attractive as partners for interaction, information exchange, and learning seen from the viewpoint

68

Bengt-A˚ke Lundvall

of the remaining independent users or producers. The independent units have very good reasons to be wary about the self-interest of the integrated units and be reluctant to share information about what they have learned from doing and using. We do not need to introduce ‘‘opportunism’’—it is simply a question of clear and legitimate self-interest. The reduction in transaction costs for the integrated couple might be substantial, but the long term loss from being locked into learning with only one user (producer) may be much bigger, especially if we are in a sector with turbulent technology and rapid change in user needs. In the learning economy it is important to be able to draw upon a diversity of experiences among users (producers). And, what is more important for the argument in this chapter, from the viewpoint of the whole economy the learning by interacting has the effect of transforming local learning into general knowledge embodied in, for instance, new machinery, new components, new software systems, or even new business solutions. This argument is parallel to Richardson’s argument on ‘‘flexibility,’’ but here his argument is taken one step further; vertical disintegration is seen as fundamental for stimulating learning-based growth in the economy as a whole. We might call this the external learning effect. Adam Smith and Two Modes of Innovation—DUI and STI Adam Smith’s arguments for the development of the division of labor went farther than static internal and external economies of scale. He links the development of the division of labor to innovation in two different ways, and he actually indicates two different modes of innovation. One is experience based and corresponds to DUI learning, referring to learning by doing, using, and interacting. The other mode is science based and corresponds to STI learning, referring to science, technology, and innovation (see Jensen et al. 2004). In the beginning of Volume I of Wealth of Nations he gives the example of innovation based upon learning by doing: The boy who develops an easier way to handle a process in order to get more time to play with his friends (see Box 1). But immediately after that he refers to ‘‘men of speculation’’—the scientists—who are ‘‘often capable of combining together the powers of the most distant and dissimilar objects.’’ Both of these examples are relevant for our reasoning about vertical disintegration, diversity, and interactive learning. For instance, the producer of process equipment may be involved in an interaction with users where he draws upon the experiences of operators in user firms when developing new models and systems. But he might also be involved in an interaction with knowledge institutions, as suppliers, in order to be updated on technological opportunities or even to buy R&D results.

Interactive Learning, Social Capital, and Economic Performance

69

Box 1 Adam Smith (1776: p. 8) on the DUI mode of learning A great part of the machines made use of in those manufactures in which labour is most subdivided, were originally the inventions of common workmen, who, being each of them employed in some very simple operation, naturally turned their thoughts towards finding out easier and readier methods of performing it. Whoever has been much accustomed to visit such manufactures, must frequently have been shown very pretty machines, which were the inventions of such workmen, in order to facilitate and quicken their own particular part of the work. In the first fire-engines, a boy was constantly employed to open and shut alternately the communication between the boiler and the cylinder, according as the piston either ascended or descended. One of those boys, who loved to play with his companions, observed that, by tying a string from the handle of the valve which opened this communication, to another part of the machine, the valve would open and shut without his assistance, and leave him at liberty to divert himself with his play-fellows. One of the greatest improvements that has been made upon this machine, since it was first invented, was in this manner the discovery of a boy who wanted to save his own labour. All the improvements in machinery, however, have by no means been the inventions of those who had occasion to use the machines. Many improvements have been made by the ingenuity of the makers of the machines, when to make them became the business of a peculiar trade; and some by that of those who are called philosophers or men of speculation, whose trade it is not to do any thing, but to observe every thing; and who, upon that account, are often capable of combining together the powers of the most distant and dissimilar objects. In the progress of society, philosophy or speculation becomes, like every other employment, the principal or sole trade and occupation of a particular class of citizens. Like every other employment too, it is subdivided into a great number of different branches, each of which affords occupation to a peculiar tribe or class of philosophers; and this subdivision of employment in philosophy, as well as in every other business, improves dexterity, and saves time. Each individual becomes more expert in his own peculiar branch, more work is done upon the whole, and the quantity of science is considerably increased by it.

In both cases the separation line (some kind of market) between the producer and user may benefit interactive learning at the level of the involved parties as well as knowledge diffusion at the level of the economy as a whole. In the case of STI learning a certain amount of in-house R&D may be needed to absorb knowledge from the outside sources. But the diversity argument remains relevant also here. The research laboratory or the software firm that addresses many users with different needs and experiences will learn more by doing so than the in-house lab or software department getting feedback only from in-house users. But even if similar mechanisms are at work, it might still be useful to make a distinction between the two modes because the prerequisites for interactive learning to

70

Bengt-A˚ke Lundvall

take place are different. In the DUI mode the generalization of local learning will typically be embodied in new machinery and components, while in the STI mode innovations may reach the user in the form of disembodied codified knowledge. The first type of interaction may be based on social interaction and trust in a broad sense, while the second may be more demanding in terms of overlapping scientific competences, and it might therefore thrive only on the basis of a common professional background. For instance, firms that are users of knowledge produced by academia may need to have, in house, employees with an academic background. Douglass North on Institutions, Organizations, and Market Competition In a brief essay developed on the basis of his Adam Smith lecture in 1994, North introduces a discussion parallel to the one entered here (North 1996). His essay contrasts the gains from increasing division of labor with the costs to make the system work. He identifies a ‘‘transaction sector’’ and refers to data showing that already by 1970 as much as 45% of the GNP in the United States could be defined as ‘‘transaction costs.’’9 North does not make a clear distinction between internal and external transaction costs, and he tends to see all government expenditure as part of transaction costs. Controlling shirking within the organization is referred to in parallel with costs of stipulating and enforcing contracts. In the beginning of his essay he refers to ‘‘human learning, the most important source of long run economic change,’’ but in the rest of the paper institutions are discussed not in terms of how they affect learning but instead in regard to their impact upon transaction costs. But his observations that it is more or less difficult to establish ‘‘efficient markets’’ in different national economies and that the concept ‘‘social capital’’ may be seen as opening up a new way of tackling this issue are useful for the purpose of this chapter. But in this chapter the adjective ‘‘efficient’’ has a different meaning. North says that ‘‘the key to efficient markets are institutions that result in low costs of transacting.’’ Here we add that (in the learning economy) dynamic efficiency needs to be taken into account and dynamic efficiency has to do with how far institutions support learning within and between organizations.10 There might be overlap between institutional forms that bring down transaction costs and forms that stimulate learning—the presence of trust and the absence of opportunism are obvious examples. But they are certainly not identical. For instance, a restrictive intellectual property right regime may reduce transaction costs while at the same time reducing benefits from learning by interacting. But the most obvious case where the two deviate is when producers and professional users are engaged in a process of interactive learning in connection with prod-

Interactive Learning, Social Capital, and Economic Performance

71

uct innovation. Here you might reduce transaction costs by vertical integration, but by doing so you would reduce the contribution from interactive learning to technical progress both on the user and the producer side. Conclusions User–producer interaction in connection with product innovation is certainly of interest for management scholars. Getting the interaction to work well is a key to successful innovation. In this chapter it has been argued that creating institutional frameworks at the national (and international) level that promote this kind of interaction in ‘‘organized markets’’ is of major interest also for enhancing macroeconomic dynamic performance. North makes the point that the vertical division of labor will be affected by ‘‘social capital.’’ It is a somewhat amorphous concept but it certainly points in the right direction. Here it has been implicitly defined as ‘‘the willingness and capability of citizens to make commitments to each other, collaborate with each other, and trust each other in processes of exchange and interactive learning.’’ A crucial issue is who is regarded as the ‘‘significant others’’—is it the members of the family, tribe, or nation or does it include everybody on the globe? Defined in this way, it is true that a society rich in terms of ‘‘social capital’’ will operate with a more developed vertical division of labor—including more organizational borders cutting through the production chains. Such a society would be more successful in terms of net wealth creation because its interactive learning would be based on more diversity and local learning would be more widely generalized and diffused more widely in the economy. It would also be more flexible in its response to shifting demand—capacity utilization would be higher, ceteris paribus (cf. Richardson argument above). As pointed out by North, in a society poor in terms of ‘‘social capital’’ with a high GNP per capita and a highly developed division of labor much of the wealth would be absorbed by social costs—costs to control workers inside organizations and costs to control market transactions and access to intellectual property outside/between organizations. This raises the next question—what determines the willingness and capability of citizens to make commitments to, collaborate with, and trust others in processes of exchange and interactive learning? Here we propose that economists need help from other disciplines with more insight into what humanity and human societies are about. George Herbert Mead (1934) and the social interactionists have important things to say that could help us analyze these issues.

72

Bengt-A˚ke Lundvall

A predictive and transparent legal system may be helpful both in its own right and as bolstering trust when corruption in the private and public sectors is not supportive to social capital. Social inequality and unequal opportunities undermine social capital, especially when seen as an injustice by the have-nots. Therefore, following North, and seeing all public expenditure including social welfare and regional policies as ‘‘transaction costs,’’ may be a mistake. It might alternatively be seen as investment in social capital. Taking it for granted that all citizens are free riders, always ready to tell a lie and engage in ‘‘holdup’’ whenever they see an advantage in doing so, might be a serious mistake, making our economy much less efficient than it could be. If this pessimistic view were correct, the learning economy would soon dwindle. Nothing is more demanding in terms of trust than interactive learning. And, as Arrow (1971) says, ‘‘you cannot buy trust—and if you could buy it . . . it would have no value whatsoever.’’ This is why the learning economy must anchor its efficiency outside the economy, i.e., in society and good citizenship—greed is not enough to make the system work. And perhaps there are good reasons to be grateful for that. Notes 1. The argument in this chapter relates to the contribution to this volume by Eric von Hippel, who demonstrates that, in certain areas, users—including consumers—become increasingly active in developing new products for their own use and that this phenomenon may have a positive impact on economic welfare. Superficially, there is a contradiction with the view in this chapter that ‘‘process innovation’’ is less useful at the systems level since it remains local. If we introduce feedback mechanisms from users to producers, entrepreneurial initiative among users, and ‘‘communities of sharing’’ in the von Hippel framework, the two perspectives may be seen as complementary since user experimentation contributes to diversity in learning. Without specifying such mechanisms it might, actually, be difficult to discuss the welfare impact of user innovation. 2. It is observed in the so-called Horndahl effect, where plant productivity grows year by year with little change in organization, equipment, and products (Lundberg 1961). 3. This perspective is presented in Stigler (1951) to explain the vertical division of labor. 4. A recent interesting contribution applying the resource-based view of the firm to vertical integration and confronting it with Williamson’s transaction cost approach is from Jacobides and Winter (2004). 5. In order to reap the static internal benefits, the division of labor within the organization needs to be characterized by a certain constancy. If workers were floating freely in and out of the organization, their learning would be limited. A different argument in the Williamson spirit would be that we must take into account that ‘‘transaction costs’’ are higher in the labor market than in most other markets since what is bought is ‘‘labor power’’ rather than concrete ‘‘labor.’’

Interactive Learning, Social Capital, and Economic Performance

73

6. Vertically organized firms might try to overcome the problem by introducing management techniques that promote functional flexibility. One aim of functionally flexible organizational forms would typically be to reduce the negative effects on internal flexibility of a highly developed division of internal specialization. Learning organizations may be seen as a solution to rigidities that alternatively would require vertical disintegration. 7. There are different managerial ways to try to compensate for the limited learning capability of hierarchies. The establishment of a ‘‘learning organization’’ where horizontal communication and interdivisional groups are combined with external networking may be seen as such an attempt. But learning organizations are as important for innovative capability in firms that have focused their attention on a few steps in the production chain and that operate in technologically dynamic sectors (Christensen and Lundvall 2004). 8. The idea of benefits from interactive learning was originally inspired by a case where a Swedish dairy technology producer (Alfa Laval) kept an affiliate in Denmark in spite of its having losses year after year. Asked why they did not close it down, the management of Alfa Laval responded that they were willing to pay a price for being close to and learning from the most advanced dairy technology users in the world. 9. Anne P. Carter (1994) has introduced an alternative perspective where several of the posts that North count as ‘‘transaction costs’’ are defined as ‘‘costs of change’’—that is, as resources that have been used either to incite change or to cope with the consequences of change. Actually, this might be a more relevant perspective in the rapidly changing learning economy. 10. At the DRUID conference on Bornholm in 1998, Oliver Williamson was asked if he would define interactive learning as just another form of transaction. He responded modestly that his analytical model has its limitations, especially when it comes to capturing dynamic processes such as learning and innovation.

References Arrow, K. J. (1962). ‘‘The Economic Implications of Learning by Doing.’’ Review of Economic Studies 29, no. 80. ——— (1971). ‘‘Political and Economic Evaluation of Social Effects and Externalities.’’ In Frontiers of Quantitative Economics, M. Intrilligator, ed. Amsterdam: North Holland. Carter, A. P. (1994). ‘‘Production Workers, Metainvestment and the Pace of Change.’’ Paper prepared for the meetings of the International J. A. Schumpeter Society, Munster, August 1994. Christensen, J. L., and B.-A˚. Lundvall, eds. (2004). Product Innovation, Interactive Learning and Economic Performance. Amsterdam: Elsevier. David, P. A., and D. Foray (2002). ‘‘Economic Fundamentals of the Knowledge Society.’’ SIEPR Discussion Paper No. 01-14, Stanford University. Foray, D. (2000). The Economics of Knowledge. Cambridge, MA: MIT Press. ———, and B.-A˚. Lundvall (1996). ‘‘From the Economics of Knowledge to the Learning Economy.’’ In Employment and Growth in the Knowledge-Based Economy, D. Foray and B.-A˚. Lundvall, eds. Paris: OECD.

74

Bengt-A˚ke Lundvall

Jacobides, M. G., and S. G. Winter (2004). ‘‘The Co-Evolution of Capabilities and Transaction Costs.’’ The Leverthulme Trust Digital Transformations Programme Working Paper (August). Jensen, M. B., B. Johnson, E. Lorenz, and B.-A˚. Lundvall (2004). ‘‘Absorptive Capacity, Forms of Knowledge and Economic Development.’’ Paper presented at the Second Globelics Conference in Beijing, October 16–20. Lundberg, E. (1961). Produktivitet och Raentabilite. Stockholm: Norstedt and Soener. Lundvall, B.-A˚. (1985). Product Innovation and User-Producer Interaction. Aalborg: Aalborg University Press. ——— (1992). ‘‘Explaining Inter-Firm Cooperation and Innovation: Limits of the Transaction Cost Approach.’’ In The Embedded Firm: On the Socioeconomics of Industrial Networks, G. Grabher, ed. London: Routledge. ——— (1996). ‘‘The Social Dimension of The Learning Economy.’’ Aalborg University, DRUID Working Paper 1996-1. ——— (2003). ‘‘Why the New Economy Is a Learning Economy.’’ Economia e Politica Industriale: Rassegna trimestrale diretta da Sergio Vacca` 117: 173–185. ———, and B. Johnson (1994). ‘‘The Learning Economy.’’ Journal of Industry Studies 1, no. 2 (December 1994): 23–42. Mead, G. H. (1934). Mind, Self, and Society. Chicago: University of Chicago Press. North, D. (1996). ‘‘Organisations, Institutions and Market Competition.’’ Working Paper, Washington University, St. Louis, MO. Pasinetti, L. (1981). Structural Change and Economic Growth. Cambridge: Cambridge University Press. Richardson, G. B. (1972). ‘‘The Organisation of Industry.’’ Economic Journal 82: 883–896. ——— (2002). ‘‘The Organisation of Industry Revisited.’’ DRUID Working Paper No. 02-15. Rosenberg, N. (1982). Inside the Black Box: Technology and Economics. Cambridge: Cambridge University Press. Smith, Adam (1776). An Inquiry into the Nature and Causes of the Wealth of Nations, 5th ed. 1904, Edwin Cannan, ed. London: Methuen and Co. Stigler, G. J. (1951). ‘‘The Division of Labor Is Limited by the Extent of the Market.’’ Journal of Political Economy 59: 185–193. von Hippel, E., and M. Tyre (1995). ‘‘How Learning by Doing Is Done: Problem Identification and Novel Process Equipment.’’ Research Policy 24, no. 5. Williamson, O. E. (1975). Markets and Hierarchies: Analysis and Antitrust Implications. New York: Macmillan.

7 Social Capital, Networks, and Communities of Knowledge Tom Schuller

Introduction This chapter considers the role of networks in knowledge economies. Unlike most of the chapters in this volume, it does not focus heavily on new technologies in advancing knowledge but concentrates on the nature of relationships within knowledge communities and deals with the topic primarily from a conceptual viewpoint. I use terminology developed within a social capital perspective but accept that similar insights could be gained using different labels. A major advantage of using social capital is the emphasis it encourages on values and norms as a defining characteristic of effective networks and communities (Schuller et al. 2000). These express themselves behaviorally as well as attitudinally and as such are open to fruitful empirical investigation. The ways in which social capital complements the more traditional concept of human capital to improve economic prosperity and enhance social well-being have been explored in OECD’s 2001 publication, The Well-Being of Nations (WN). The key thrust of the argument in WN is that conventional usage of human capital alone is inadequate as a tool to understanding the development and functioning of knowledge and skills. This is the case even when this is interpreted in terms of economic objectives only but still more so when wider objectives are factored in. I take up some of the issues raised in WN and suggest ways of getting some purchase on the varied roles of networks in the generation, dissemination, and validation of knowledge. The underlying premise is that we need to elaborate ways of differentiating networks and communities and to incorporate into our analyses a sense of how such networks change over time. In addition to the broadening of scope that the application of social capital entails, there is a significant shift in many OECD member countries to a concern with outcomes rather than inputs and with the mechanisms that translate inputs into outcomes. In relation to education, this means that instead of focusing predominantly

76

Tom Schuller

on the numbers or proportions of the population who participate in human capital formation (education or training) or the amounts invested in those activities, there is growing interest in what happens as a result of the investment: public as well as private returns, and in domains such as health and social inclusion as well as for employment and productivity.1 As for knowledge advancement, the OECD’s Science, Technology and Industry Scoreboard (2003, p. 16) defines investment in knowledge in terms of expenditure on R&D, expenditure for higher education (public and private), and investment in software. The limited range of inputs is related to productivity and GDP, but there is clearly a need to explore both a wider range of intermediate and final outcomes resulting from such investment and the mechanisms that turn the investment into results, in whatever degree of efficiency. This is not so much a criticism of the scoreboard as a confirmation that we have major gaps in our conceptual apparatus for understanding the process of knowledge creation, as well as in the data needed. WN defined social capital as ‘‘networks together with shared norms, values and understandings that facilitate cooperation within or among groups’’ (OECD 2001, p. 41). I follow this definition, although I shall have some observations to make on the extent to which such sharing takes place or is expected to take place. Trust is not explicitly included in this top-level definition but figures prominently in the discussion, as both a source and an outcome of social capital. In the latter part of the chapter I deal briefly with the role of trust in promoting the validation and critique of knowledge creation. Interactions Between Bonding and Bridging Social Capital The literature commonly identifies three basic forms of social capital, following Woolcock (1998): Bonding SC which refers to relationships within or between relatively homogeneous groups. In the context of knowledge advancement, this may refer particularly to intradisciplinary or intraprofessional affiliations.



Bridging SC referring to relationships within or between relatively heterogeneous groups. In this context, it refers particularly to interdisciplinary or interprofessional connections.



Linking SC referring to relationships between people or groups at different hierarchical levels.



The conventional distinction between bridging and linking is that the former refers to horizontal connections and the latter to vertical ones within power structures. For

Social Capital, Networks, and Communities of Knowledge

77

the purposes of this chapter the distinction is not strongly relevant, so I shall not deal with linking SC as a separate component. Although these forms are now very familiar in the literature, it is striking how little consideration is given to how one might analyze or even conceptualize the interaction between them, to how far they complement or conflict with each other in different contexts, and especially to how these interactions may play out over time. A first general proposition of this paper is that the relevance of SC, and network thinking, to the knowledge economy will be greatly enhanced if we can further develop our understanding of these interactions. Pointing to high levels of social capital or strong networks tout court will bring only limited insights. The effect of social capital can be better understood if there is a clearer focus on the different forms it takes and on how these forms reinforce or impede each other.2 One can have bonding SC without bridging, but not vice versa. Almost any form of social life involves bonding, whether the basis for the bonds is predominantly normative or functional. Bonding may be tight or loose, but tightness or looseness does not itself tell us that much about the effects. Bonding has its limitations, for those who are within the group as well as those excluded. This is demonstrated in Granovetter’s seminal work on weak ties (1973), which shows that employment prospects are helped more by knowing people with whom one has loose links than by those one associates with closely. In relation to knowledge generation and innovation, excessive bonding encourages groupthink, discourages new perspectives, and narrows the potential range of skills and expertise available to the group. However, defining at what point bonding becomes dysfunctional is difficult. Bonding entails exclusion, as its mirror image. A social group that is infinitely porous and has no conception of who does and does not belong to it cannot have much reality. ‘‘Exclusion’’ in this sense is socially neutral. It may be gentle and weak, so that it is easy for outsiders to join, but there must be some notion of differentiation between those inside and outside the group. By contrast, of course, the exclusion may be harsh rather than gentle, so that aspiring entrants are rudely rejected (for example, at a macro level the forcible exclusion of unskilled economic migrants from a country or region, or at a micro level racist blackballing from a social club), and it may be strong though unarticulated (or perhaps because unarticulated; cf. the existence of shadowy but powerful old boy networks where the rules of membership are never explicit, making it doubly difficult for outsiders to break in). Even where the bonding is strong, this does not mean it is exclusionary in any pejorative sense. Thus, a research group may exercise very strong disciplinary bonding that is highly exclusionary, because it is difficult for most people to acquire the necessary disciplinary expertise to become part of it, but socially and economically functional

78

Tom Schuller

because it enables the group to advance its knowledge base rapidly and efficiently and to put this to wider use. The distinction between bonding and bridging is almost entirely context dependent. Very few groups are entirely homogeneous. For our purposes it may be convenient to think of academics—one category of knowledge producers—as at one level a single social group with its own norms and values (see Tony Becher’s catchily titled anthropological work Academic Tribes and Territories [1989]), but very obviously any investigation of how they generate and disseminate knowledge will quickly break the group down, most likely into different disciplines and then into subdisciplines and yet further into the proliferating branches of each evolving field of knowledge. Each of these branches sustains some form of communality of understanding with its fellow branches, but the links spanning the community as a whole grow weaker the closer one looks. The same process necessarily applies to other groups—in other words, the more fine-grained the examination the greater the apparent heterogeneity. Academics may share some universal common values, such as a commitment to seeking after truth or a belief that universities should be adequately funded, that unite them at certain times. But these do not have much power to tie them together in very effective or meaningful ways (remember Clark Kerr’s acid definition of a university as a crowd of disaffected academics united only by common grievances over car parking). The role of ICT in shifting the characteristic profile of academic communities away from its physical institutional base to more virtual forms is well known. The value of bridging, i.e., of making connections with people or groups that are dissimilar in their approach to knowledge generation, is that it can extend the range of ideas, expertise, and contacts to which one has access and the range of opportunities for applying knowledge, commercially or not. It is therefore likely to be particularly important for those concerned with extending the frontiers of knowledge, at least for parts of the process (see discussion of trajectories below). This thesis is well suited for empirical application (see Burt 1982 on bridging within organizations). But my point here is that we can understand it only if we look at both sides of the coin. So I offer a second general proposition, that a successful and dynamic interaction between bonding and bridging forms of SC is likely to be a key component of sustained knowledge generation. The relationship between bonding and bridging forms of SC may be complementary or conflicting, in most cases lying somewhere along the continuum between the two. In some contexts, bonding is achieved only by enforcing total commitment to the group in question, forbidding cross-membership of contiguous or related groups, minimizing external contacts, and rigorously filtering new memberships. Cult religious groups are an extreme case but are hardly an appropriate one for study in

Social Capital, Networks, and Communities of Knowledge

79

the context of knowledge advancement. More relevant are lesser degrees of extremity, where adherence to a particular paradigm of knowledge accumulation is nevertheless regarded as radically incompatible with sympathy for any other approach. Closer to home is the question of the place of randomized control trials in social research, following the norm of medical research: At one extreme is the group that sees RCTs as generating the only form of truly robust evidence, to the exclusion of all others, while at the other end are those who see any use of such an approach as wholly inappropriate to social or educational purposes and not even to be discussed as an option (Oakley et al. 2003). Even where paradigm wars as such have not broken out, there will often be ongoing fluctuating tensions in the extent to which groups see benefit in accepting the norms or values of others’ approaches to knowledge generation. It is hard also to define what might constitute ideally complementary forms of bonding and bridging.3 But arguments over interdisciplinarity are a useful example of how such forms need to be analyzed. Lip service for a long time has been paid to the idea of interdisciplinarity (cf. OECD 1972) as a classic form of bridging, but it generates inherent tensions. It must entail, up to a point, the reduction of disciplinary boundaries in order to increase communication and cross-fertilization, but just how far should this process go? It is sometimes implied that the aim should broadly be for dissolution of the distinctions between disciplines and that the farther this process goes, the better. An alternative view, to which I myself subscribe, is that effective interdisciplinarity depends on most (not all) participants continuing to have access to a secure disciplinary base, from which they can make effective connections, returning iteratively to their discipline to move it forward and change its shape as a result of the interaction. The same applies, mutatis mutandis, to professions and related knowledge practitioners. The function of a bridge is to link two discrete terrains, not to obliterate the differences between them. Moreover, there may be inherent limits to the range of disciplines that can be coherently involved in any single exchange. But as the process of interdisciplinary collaboration develops, one can expect participants to reach out to more partners, without fearing that they are abandoning their own original reference group. It is easy to picture a simple matrix as a way of considering the set of relationships between networks, or other social units, in these terms (Figure 1). A group of, say, researchers or other knowledge builders that has high levels of both bonding and bridging would presumably be secure in its own identity and internally cohesive but manage this without closing itself off to new ideas and expertise. This looks like the ideal combination for advancing knowledge; however, the equilibrium is not necessarily one that can be easily sustained, because the energies successfully

80

Tom Schuller

Figure 1

generated produce their own destabilizing effects. The diametric opposite—low on both counts—is likely to be unstable from the start, but in a nonproductive sense. It is low on trust and xenophobic—at least in the intellectual if not the social sense. It is therefore unlikely to generate much from its own resources nor to derive benefit from external contacts. High bonding and low bridging has the strengths and weaknesses alluded to above. It is a profile that may be functional and comfortable for a (long) time but at a lower energy level because it lacks infusion of foreign ideas. This implies a risky degree of closure—risky because in a context of rapid knowledge obsolescence the community will find itself isolated and left behind. The fourth category, with high bridging but low bonding, looks like a group that is ready to redistribute itself through lack of central cohesiveness or whose members do not have this group as their primary reference point but use it only instrumentally. It could be a disciplinary association or research group whose subject area is running out of steam or, on the other hand, that is so dynamic it is unlikely to sustain a common identity for much longer. These are all ideal types, ripe for empirical illustration. It would be interesting to compare, for example, different disciplines within an academic institution or different epistemic communities linked by similar research concerns to see how the dynamics evolve. The sense of dynamic takes me to a third proposition, which concerns the significance of typical or atypical trajectories over time. Forms of social capital, and the relative importance of bonding and bridging for any given social or knowledgegenerating unit, will change over time, so that there is rarely a stable equilibrium (and where this exists, it will be time limited). Early in a network’s formation, bonding will be particularly important to establish trust and ensure that there is an adequate sharing of norms and values, with the benefits this brings in terms of sharing of ideas, data, and so on. None of this needs to be explicit, but it will be hard for collective knowledge generation to occur if a basic level of bonding SC is not built up.

Social Capital, Networks, and Communities of Knowledge

81

Figure 2

The interesting empirical question is at what point(s) in the trajectory of the network does bonding SC begin to need to be complemented by bridging SC and then whether the growth of bridging SC leads to a transformation in the identity of the network. Schematically, one can envisage various forms of trajectory on this kind of measure (see Figure 2). A simple linear one (A) denotes the continued steady growth of a knowledge-generating group (not necessarily growth in numerical terms; it could be in output or in influence or reputation) that maintains internal strength without recourse to bridging. Alternatively, without such recourse the group may lose impetus, lacking external stimulus, and wither away (B). A third trajectory (C) shows initial growth in bonding followed by some fallback but then a reinvigoration when links are made with external groups, bringing in new research ideas or fresh approaches to the implementation of research results. Many other more complex trajectories are obviously possible. So the temporal dimension is crucial, but we need to add in one further aspect. Thus far I have spoken largely in disciplinary terms, with research groups as the model of knowledge generators. Under this line of thinking the bridging is between disciplines and researchers. But innovative knowledge advancement depends in part on the interaction between knowledge generators and knowledge users. This is the case increasingly in many sectors, e.g., software development (Foray 2004; von

82

Tom Schuller

Hippel this volume). So effective advancement will depend at some point on bridges being built with the users of knowledge and its outputs. When the product is a marketable good, the users are the consumers or the intermediate producers. For some services this is also the case; i.e., the consumers purchase the service, and the market is the mechanism through which their preferences are transmitted. But public services in particular are a huge area where markets do not operate, or operate only partially, and the primary user is the professional service provider, such as a teacher or health worker. If there is no communication or trust between the knowledge generators and those responsible for applying the knowledge, it is unlikely to succeed (see OECD 2003). The implications of this fairly banal observation are considerable. The ‘‘coproduction’’ of knowledge in these cases does not mean the elimination of differences in role and competence, just as interdisciplinarity does not mean the elimination of disciplines. But it does mean a level of mutual understanding and respect. Here I need to return for a moment to the WN definition of social capital as the sharing of norms, values, and understandings. This could be taken to imply wholesale identity in respect of each of these three terms. In other words, for social capital to exist in significant measure, whether in bonding or bridging form, the members of the network need to be wholly congruent in their norms, values, and understandings. This would be to misconstrue the nature of such networks. It is unrealistic to expect or aspire to such congruence. It is not unrealistic to expect a considerable degree of overlap. But the important thing, arguably, is that there is a recognition and acknowledgment of each other’s norms and values—even if these are not completely shared in the sense that each party subscribes to them equally. This is especially true in respect of bridging social capital. In fact, it may be worth considering a revision of the conceptual structure so that bonding SC is defined in terms of sharing norms and values with other network members, whereas bridging SC is defined in terms of acknowledging the validity of others’ norms and values without necessarily having to share them.4 For example, researchers in health or education may have as their primary goal the advancement of particular forms of knowledge, of pushing forward the boundaries of their field in intellectually exciting ways. These are not the values that principally characterize practitioners’ work, which is why researchers as a bonded group will find it difficult to include practitioners who cannot see relevance in the researchers’ enthusiasm (and are unlikely to comprehend the conversation if it is cutting edge). But if practitioners can recognize these values as crucial components in the work of researchers while on the other side the researchers acknowledge the practitioners’ valuing of accessible and applicable results, there should in principle be better communication, less frustration, and higher levels of mutually satisfactory

Social Capital, Networks, and Communities of Knowledge

83

Figure 3

output. At least the odds of success will be higher (though there is no guarantee that the obstacles of communication and of different timeframes will be overcome). Successful partnerships are built on this kind of recognition of mutual advantage and not on an unrealistic assumption of total identity of interest. It is this line of argument, incidentally, that best defends social capital against the charge that it smuggles in some quasi-coercive form of communitarianism, which imposes a single world-view on all its members. To sum up this very schematic discussion: 1. Social capital, with its notions of bonding and bridging, offers a useful if not entirely new way into thinking about the role of networks in knowledge generation. 2. Bonding and bridging have to be seen in relation to each other and the pattern of their interaction traced over time. 3. Empirical work can trace out these different trajectories as they appear in different knowledge fields and social contexts. 4. There are significant measurement issues to be resolved in analyzing the extent of both bonding and bridging. I turn now to a different application of social capital, to the theme of knowledge economies. Mapping SC in Knowledge Economies A knowledge-based economy is characterized by 

The growing importance of economic transactions dealing with knowledge itself,



Rapid qualitative changes in goods and services, and

84

Tom Schuller

The incorporation of the creation and implementation of knowledge into the mission of the personnel involved.



Figure 3 derives from a number of simple propositions: 

Knowledge is not just a matter of accumulating facts, or even information.

The distribution of knowledge and of the competences required to generate it will impact on the efficiency with which knowledge can be generated.



These competences are not the property of individuals alone but of collectives/ groups.



Axis A refers to the volume of knowledge accumulated. As we grow more ‘‘knowledge-rich,’’ we shift upward on this axis. There is no need to go here into the question of how this is to be measured, but traditional nested hierarchies such as facts/information/knowledge/wisdom are relevant concepts. The accumulation is not of course a simple linear process. In any case, it is clear that the mere accumulation, even of wisdom (or whatever is conceived of as at the top of the hierarchy), is not enough to guarantee progress and satisfaction. Trust accelerates the accumulation of knowledge, but crucial questions remain of who owns the knowledge, who has access to it, and who is able to take advantage of such access. Therefore, axis B refers to the extent to which the knowledge mapped on axis A is accessible, formally or actually. As the distribution of knowledge grows broader, we shift outward on this axis. The nature of accessibility reflects the complexity of knowledge. It might refer to formal accessibility (for example, in respect of patents and IPR or of the spread of ICT-driven access) or to the skills required to take advantage of opportunities (cf. Sen 1999 on capabilities). Immediately the traditional efficiency/equity issue presents itself: To what extent is there a trade-off between speed of knowledge accumulation on the one hand and on the other a desire to see maximum participation in this process and the private or public benefits it generates? The trade-off may not exist at all, and equity may complement, not conflict with, efficiency, but this is not my concern here. Instead I am simply suggesting that efficiency considerations on their own may require at least a basic distribution, such that the process of knowledge advancement is partially conditioned by the relationship between the vanguard and others. Trust, locally or at higher levels, represents a collective investment that enhances accessibility and broad distribution. This is therefore an issue for public policy. Axis C refers to the cultural and organizational processes that govern the validation of the knowledge. Shifting outward on this axis implies that these processes are growing stronger, enabling us (the participants in whatever the social unit is) to have greater confidence in the quality of the knowledge.5 This occurs through

Social Capital, Networks, and Communities of Knowledge

85

professional, public, and private critique and exchange. As increasing amounts of information—from cutting-edge research findings to lay advice—wash across the world, conventional mechanisms for sifting and testing it are stretched beyond their limits. Advancing knowledge generates major challenges for our validating standards and mechanisms. Trust is an essential component in remedying the weaknesses of formal knowledge-validating systems. Traditional means of developing, establishing, and implementing professional knowledge are put under strain by a number of factors: the sheer growth in the volume of output, already mentioned; the speed of circulation of information, powered primarily by new technologies, which defies the capacity of conventional modes of scientific examination such as peer review to keep up with output and to sift out the good from the dross; the erosion of established forms of professional authority, for example, the status of academics, which means less certainty even when established professionals do offer judgment; and the shift toward Mode 2 knowledge, which is less bounded by disciplinary divisions (Gibbons et al. 1994). In some measure those involved in the generation and the use of knowledge will rely more on trusted contacts than on external sources to enable them to discriminate between reliable/less reliable and valuable/less valuable knowledge. This will especially be the case in fast-moving fields where recourse to traditional methods of knowledge validation will simply take too long to satisfy the needs of the individual researcher, the research community, or the user. This in turn raises interesting issues about how far informal trust relationships can be expected to assume the burden of knowledge validation and what the implications are for public accountability, especially, though not only, where there is no market mechanism as one test of quality. Historically, informal relationships of this kind have been used to sustain existing structures of power and to exclude nonconforming knowledge challenges, as well as for more positive functions. In one sense, the more such informal relationships are utilized, the more opaque becomes the process of accumulating and developing innovation. Like social capital generally, trust relationships can be put to uses that are biased or regressive in social terms. But overall there is good reason to suppose that higher trust levels will sustain the overall quality of information. Social capital, with its emphasis on trust, thus reenters the discussion via axis C, but linking to the model as a whole. Faced with proliferating rivers of knowledge— potentially valuable but also potentially overwhelming even where valid and potentially misleading—we need processes for rendering them accessible and manageable so that they do not become floods of muddy and polluted water. Gaining access to valuable new knowledge is only one side of the picture. It is equally important to be able to distinguish and discard irrelevant, dubious, or downright false information.

86

Tom Schuller

To do this, we need to be able to rely on others when our own levels of understanding are insufficient. Sometimes this can be done contractually, for example, by paying people to gather and sift information, whether as part of a research team or in a service capacity. It may also be done by relying on formal or informal networks of colleagues or friends whom we trust, personally and/or professionally. In other words, for the purposes of knowledge validation there is arguably a shift from trust in recognized external sources of authority to more personalized forms of trust that depend on one’s judgment of the individual or body concerned—their technical expertise, of course, but also their trustworthiness and their commitment to similar values. How are such contacts formed and used? The interaction between bonding and bridging SC is paralleled by one between different forms of trust: on the one hand wholesale reliance on a known person or body as authority source (akin to bonding) and on the other recourse to a less familiar but potentially enriching partner. It is worth noting the role that trust has in validating skills as well as knowledge, especially in areas where traditional certification procedures are inadequate. Trust reduces transaction costs and improves the flow of information and thus has direct economic effects as well as indirect and wider outcomes. It aids innovation by improving communication flows and the diffusion of knowledge, within and between organizations (Maskell 2000; PIU 2002). But it is also a key part of building knowledge-generating capacity. Recruiting people into leading-edge knowledge occupations often means using contact networks where trust complements or even replaces more formal procedures. These processes of accumulation, access, and validation will play out in different ways in different contexts. The health sector provides a brief illustration. It involves massive amounts of research (on products and services) and presents particular challenges in respect to how the knowledge generated is utilized. Much health service research involves mixes of Mode 1 and Mode 2 knowledge generation (Ferlie and Wood 2003). It combines traditionally determined, discipline-bound inquiry based on academic standards of expertise with work that is more socially distributed, boundary-crossing, and iterative between different stakeholders. The language of Mode 2 may be less academically esoteric, but communication is more complex because of the range of interested parties. Second, there is a shift toward new relationships between knowledge producers and users. This requires new bridges to be built, bringing together into a direct partnership groups that previously had an arms-length relationship. One aspect of this has been labeled ‘‘cocreation,’’ where the experience of health service users is tapped and developed to change and improve delivery (Cottam and Leadbeater 2004). To

Social Capital, Networks, and Communities of Knowledge

87

some extent the producer–user divide is blurred, because the users not only supply inputs from their experience but also become more able to supply their own care, notably in respect to conditions such as diabetes or arthritis. Third, the effectiveness of service delivery—the application of knowledge— depends in part on the quality of communication between provider and consumer. Education can strongly influence the pattern of communication, enabling consumers to articulate their needs, which in turn enables the provider to identify the appropriate type of service (Hammond 2003; Schuller et al. 2004). The growth of internetbased self-doctoring has a different set of implications for consumer education and for professional–patient interaction. All of these relationships involve different forms of trust and mutual confidence, which shape the speed and quality of knowledge advancement. Conclusion This chapter has presented nothing in the way of empirical substance. While it may be obvious that functioning networks and trust levels generally enhance the creation and dissemination of knowledge, the ways in which this works at the frontiers of knowledge still remain to be understood. All I have aimed to do is open up some lines of thinking and analysis, applying a social capital perspective to knowledge advancement. I conclude with three rather diverse policy questions that relate to the field of education: 1. What factors, social as well as technological, have general significance in shaping the ways networks function? To take just one example: how will the change in relative educational achievement by men and women, and the consequent shift in the gender profile of knowledge occupations, affect these functions? 2. The knowledge economy cannot be characterized simply as one that employs high proportions of educated people, as measured by standard levels of individual qualification. Increasingly countries will have to think about how education promotes effective participation in communities of knowledge, and this will include social and moral competences as well as technical ones. Assessing these, and their outcomes, is a major challenge. 3. How far does the physical architecture of our educational institutions reflect and complement the demands of a knowledge-based economy? Equipping schools, colleges, and research institutions with up-to-date technologies for the accumulation of knowledge is one thing, but constructing facilities that enhance effective access and validation demands other kinds of creative approaches.

88

Tom Schuller

Notes 1. CERI’s activity on Measuring the Social Outcomes of Learning (http://www.oecd.org/edu/ socialoutcomes) addresses these issues directly. 2. The same general point is arguably true of human capital, but social capital is more strongly identified with relationships than with attributes, so the interactions are correspondingly more central. 3. Steinmuller’s (2003, p. 63) discussion of relationships between communities of practice addresses this issue and introduces a number of concepts—boundary objects, translation processes—that could well form the basis for a systematic analysis of the nature of internetwork interactions. As he says, however, this is as yet a very young field of study. 4. A similar distinction, far more elaborated, is made in respect of innovation by Duguid (2003), between communities of practice and networks of practice: CoP members not only have a practice in common but coordinate that practice with each other; NoP members share some form of practice but do not systematically coordinate it. 5. We are, I acknowledge, caught in a basic epistemological circularity: given that the issue is one of knowledge validation, who is to determine whether the processes are actually growing stronger, and by what criteria? However, I do not think this detracts from the applicability of the model.

References Becher, T. (1989). Academic Tribes and Territories. Buckingham: Open University Press/ SRHE. Burt, R. (1982). Towards a Structural Theory of Action. New York: Academic Press. Cottam, Hilary, and Charles Leadbeater (2004). Health: Co-Creating Services. Red Paper 01. London: Design Council. Duguid, Paul (2003). ‘‘Incentivizing Practice.’’ Institute for Prospective Technological Studies (IPTS) Workshop on ICTs and Social Capital in the Knowledge Society, Seville, November. Ferlie, E., and M. Wood (2003). ‘‘Novel Mode of Knowledge Production? Producers and Consumers in Health Services Research.’’ Journal of Health Services Research and Policy 8, Suppl 2: 58–61. Foray, Dominique (2004). The Economics of Knowledge. Boston: MIT Press. Gibbons, M., C. Limoges, H. Nowotny, S. Schwartzman, P. Scott, and M. Trow (1994). The New Production of Knowledge: Science and Research in Contemporary Societies. London: Sage. Granovetter, M. (1973). ‘‘The Strength of Weak Ties.’’ American Sociological Review 78(6): 1360–1380. Hammond, C. (2003). ‘‘How Education Makes Us Healthy.’’ London Review of Education 1(1): 61–78. Maskell, Peter (2000). ‘‘Social Capital, Innovation and Competitiveness.’’ In Social Capital: Critical Perspectives, S. Baron, J. Field, and T. Schuller, eds. Oxford: OUP.

Social Capital, Networks, and Communities of Knowledge

89

Oakley, A., V. Strange, T. Toroyan, M. Wiggins, I. Roberts, and J. Stephenson (2003). ‘‘Using Random Allocation to Evaluate Social Intervention: Three Recent U.K. Examples.’’ Annals AAPSS 589 (September): 170–189. OECD (1972). Interdisciplinarity. OECD, Paris. OECD (2001). The Well-Being of Nations. OECD, Paris. OECD (2003). The Science, Technology and Industry Scoreboard. OECD, Paris. PIU (Performance and Innovation Unit) (2002). ‘‘Social Capital: A Discussion Paper.’’ Cabinet Office, UK Government, London. Schuller, Tom, Steve Baron, and John Field (2000). ‘‘Social Capital: A Review and Critique.’’ In Social Capital: Critical Perspectives, S. Baron, J. Field, and T. Schuller, eds., pp. 1–42. Oxford: OUP. Schuller, T., J. Preston, C. Hammond, A. Brassett-Grundy, and J. Bynner (2004). The Benefits of Learning: The Impact of Education on Health, Family Life and Social Capital. London: Routledge Falmer. Sen, A. (1999). Development as Freedom. Oxford: OUP. Steinmuller, Edward W. (2003). ‘‘Communities of Practice and Their Effects on Performance and Functioning of Organizations.’’ Institute for Prospective Technological Studies (IPTS) Workshop on ICTs and Social Capital in the Knowledge Society, Seville, November. Woolcock, M. (1998). ‘‘Social Capital and Economic Development: Towards a Theoretical Synthesis and Policy Framework.’’ Theory and Society 27: 151–208.

8 Knowing Communities in Organizations Patrick Cohendet

Introduction A knowing community can be defined as a gathering of individuals who accept to exchange voluntarily and on a regular basis about a common interest or objective in a given field of knowledge.1 Through this regular exchange, common cognitive platforms and common social norms are built. These ensure the cohesion of the community and guide the behavior of newcomers. The critical role of knowing communities in the building of useful knowledge for society has been popularized recently by successful examples such as the Linux open source community or the role of the community of ‘‘reps’’ at Xerox.2 These examples underline the role of communities as modes of economic coordination that ‘‘economize’’ on hierarchy to produce useful knowledge. As the knowledge-based economy expands, it is said that knowing communities play an increasing role, because they can take charge of significant parts of the ‘‘sunk costs’’ associated with the process of generation or accumulation of specialized parcels of knowledge.3 These costs correspond, for instance, to the progressive construction of languages and models of action and interpretation that are required for the implementation of new knowledge and that cannot be covered through the classical signals of hierarchies (or markets). This setting is likely to compensate for some organizational limitations (learning failures) that firms face when confronted with the need to continuously innovate and produce new knowledge. Knowing communities might be found in traditional work divisions and departments, but they also cut across functional divisions, spill over into after-work or project-based teams, and straddle networks of cross-corporate and professional ties. ‘‘For example, within firms, classical communities include functional groups of employees who share a particular specialization corresponding to the classical division of labor (e.g., marketing or accounting). They also include teams of employees with heterogeneous skills and qualifications, often coordinated by team leaders and

92

Patrick Cohendet

put together to achieve a particular goal in a given period of time’’ (Amin and Cohendet 2004). Knowing communities can be of a different nature in the way they deal with knowledge: Some of them may focus on the accumulation and exploitation on a given field of knowledge (communities of practice), some of them on the exploration of a new field of knowledge (epistemic communities). The role of knowing communities seems to be particularly important for organizations, in particular for firms. As Brown and Duguid (1991) suggest, the firm can be viewed as a collective of communities, not simply of individuals, in which enacting experiments are legitimate, separate community perspectives can be amplified by interchanges among communities. Out of this friction of competing ideas can come the sort of improvisational sparks necessary for igniting organizational innovation. Thus large organizations, reflectively structured, are perhaps well positioned to be highly innovative and to deal with discontinuities. If their internal communities have a reasonable degree of autonomy and independence from the dominant worldview, large organizations might actually accelerate innovation.

Brown and Duguid’s statement opens the field for a renewed vision of the economic approach of the firm, at least in three directions: First, behind the scene of the ‘‘visible’’ infrastructure of knowledge defined by the hierarchy, firms can be considered as a set of knowing communities that are repositories of useful knowledge, which is embedded in their daily practices and habits. As Wenger (1998) noted, a community drawing on interaction and participation to act, interpret, and innovate acts ‘‘as a locally negotiated regime of competence.’’ Therefore, communities are suppliers of sense and collective beliefs for agents and play a central role of coordination in the firm.



Second, an important part of the innovative potential of firms may reside in the interaction between different knowing communities of the organization. The quality of this innovative potential depends on the nature of the interaction between communities. Thus, a critical role of the hierarchy of the firm is to contribute to organize efficient platforms of knowledge within firms in order to facilitate the interaction between knowing communities. Also, firms should design internally the mechanisms that ensure a satisfactory circulation of the knowledge held by knowing communities to the formal structures of the firm.



Third, large firms may find in the functioning and activation of their communities considerable potential of innovation that could contribute to counterbalance the traditional bureaucratic handicap of large firms toward innovation.



This contribution aims at assessing and explaining the role of knowing communities in organizations along the three perspectives opened by Brown and Duguid. In the first part, we discuss the nature of knowing communities, their characteristics, and

Knowing Communities in Organizations

93

the reasons for the growing interest in this mode of generation of knowledge in society. In the second part, we focus on the interaction between heterogeneous communities as a strategic locus of generation of innovation in a given organization. This issue also raises the problem of the delicate matching between the ‘‘visible’’ hierarchical structure of the firm and the ‘‘invisible’’ process of active knowledge formation that occurs within the myriad of communities in a given firm. The last part is devoted to a discussion of the role of communities as a critical potential for large firms to increase their degree of innovativeness. Knowing Communities: Characteristics, Properties, and Limits In this contribution, we will consider the notion of knowing communities as a general one, without specific distinction. However, it is important to recall that the literature has identified different types of knowing communities,4 among which the two main types are the following: 1. Communities of practice The concept was emphasized by Lave and Wenger (1991), who identified the existence of groups of persons engaged in the same practice, communicating regularly with one another about their activities. Members of a community of practice essentially seek to develop their competencies in the practice considered. Communities of practice can then be seen as a means to enhance individual competencies, and this goal is reached through the construction, the exchange, and the sharing of a common repertoire of resources (Wenger 1998). Self-organization is an essential characteristic of communities of practice. 2. Epistemic communities are small groups of ‘‘knowledge-creating agents who are engaged on a mutually recognized subset of questions, and who (at the very least) accept some commonly understood procedural authority as essential to the success of their collective activities’’ (Cowan et al. 2000, p. 234). Because of agents’ heterogeneity, for the sake of knowledge creation the first task of epistemic communities is to create a codebook. And because agents lack deeply shared values, knowledge creation mode is much like a form of externalization (conversion of tacit into explicit knowledge, in the sense of Nonaka and Takeuchi 1995). As informal groups, knowing communities exhibit specific characteristics that distinguish them from the traditional organized entities usually analyzed in economics or business science: Communities have no clear boundaries, and there is no visible or explicit hierarchy at the top of them that can control the quality of work or the respect of any standard procedure.



94

Patrick Cohendet

It has been repeatedly argued that what holds the community together is the passion and commitment of each of its members to a common goal, objective, or practice in a given domain of knowledge. Thus, the notion of contract is meaningless within the members of the community, and in particular there is a priori no motive to think of any financial or contractual incentive devices to align the behavior of the members of the community.



The interactions between members of the community are governed by a type of trust grounded in respect for the common social norms of the community. Trust within the community can be measured when one can observe that the behaviors of the participants, after being exposed to an unexpected event, are not guided by any form of contractual scheme but by the respect of the social norm of the group.



Recent literature has emphasized that some of the specific motives that guide the behaviors of members of the community could have an economic interpretation. Frequency of interactions within the community considerably reduces opportunistic behaviors. With repeated interactions, holdups and moral hazard problems will be attenuated through the creation of norms of cooperation and routines5 as well as the intensification of reputation mechanisms. Therefore, a large part of agency problems will be resolved spontaneously in the knowledge-based economy.



The validation of the knowledge takes place in first analysis within a given community. In the same way, the interpretation of the knowledge provided by the outside (in particular by the hierarchy) is examined, criticized, and reprocessed (to lead sometimes to creative adaptations) within communities.



All the above features that characterize learning communities contribute to making clear the differences between communities and the other forms of coordination units that can be found within the firm: Knowing communities differ from functional groups. Contrary to communities, these units are under the responsibility of a hierarchy at the top of them, with clear boundaries separating those belonging to the unit and those apart. Of course, such functional units can contribute to the process of accumulation of knowledge. However, when compared to communities (where cognitive links are continuously activated and enhanced ‘‘naturally’’ between members), these units require considerable efforts to be made in order to activate the conservation of routines, the power of replication of the routines, and the continuous improvement of the routines between members. Moreover, while communities are loci of active and deliberate learning processes between members in order to create, exchange, and accumulate knowledge, functional units are mainly characterized by passive modes of learning, such as, for instance, ‘‘learning by doing,’’ which has been heavily described in the literature.



Knowing Communities in Organizations

95

Knowing communities differ from project teams, or task forces. These teams of employees with heterogeneous skills and qualifications are often coordinated by team leaders and put together to achieve a particular goal in a given period of time. Communities may share some common traits with teams: For instance, the group’s interests generally coincide with the interests of the members. However, there is no visible hierarchy in communities, nor any constraint of time in the process of knowledge generation and accumulation.



Knowing communities differ from coalitions in the sense that the strategic calculation of agents does not generally determine their adherence to a given community. Moreover, a coalition has by definition a clear boundary, contrary to communities.



Knowing communities differ from cliques in network theories. If cliques share with communities the characteristic of having no clear boundaries, they differ in the sense that the relationships between agents within a clique do not express a generally cognitive dimension. They do not address a clear objective of creation and accumulation of knowledge.



Knowing communities are thus repositories of useful knowledge, which is embedded in their daily practices and habits. The local daily interactions constitute an infrastructure that supports an organizationally instituted learning process that drives the generation and accumulation of knowledge by the community. Most of the time, the accumulation of knowledge by a given community is shaped by a dominant mode of learning (such as ‘‘by circulation of best practices’’) adopted by the community and can circulate through the existence of a local language understandable by the members only. As Wenger (1998) noted, a community drawing on interaction and participation to act, interpret, and innovate acts ‘‘as a locally negotiated regime of competence.’’ The communal setting provides the context in which the collective beliefs and the representations structuring the individual choice are built. Communities allow the strengthening of individual commitments in an uncertain universe. Individuals remain attentive to the specific contexts and can therefore update the shapes of their cooperative engagements. However, as an organizational mechanism, communities have limits. In particular, since a given community focuses on a given domain of knowledge, the issue of the lack of diversity resulting from the interplay of communities is raised. Diversity in organization can be obtained by the interaction between different communities, but nothing guarantees a priori the systematic concordance of interests and objectives of the different communities on the spot. Constituent communities of the organization are not necessarily all homogeneous or convergent toward a common objective. Risks of intercommunal conflicts, autism, or parochial partitioning are latent. The issue is the extent to which the knowledge held by one community can

96

Patrick Cohendet

be transmitted to the rest of the organization: to the ‘‘visible’’ hierarchical structure and to the myriad of other communities in the firm. The coherence of the firm and the existence of common referential systems become therefore crucial questions. The coexistence of hierarchies and communities within the firm poses the problem of the spontaneous or deliberate emergence of referential systems structuring individual and collective beliefs constructed in the process of decision. Such critical issues call for an in-depth discussion of the nature of the interaction between communities. This discussion is proposed in the following part. The Nature of Interaction Between Knowing Communities in the Firm, and the Role of the Hierarchy How, within a given organization, do communities interact? Intuitively, we can assume that as for individuals, too distant communities within a firm will not lead to innovative solutions. But if the cognitive distance between communities is too small, the innovative potential of the firm will fade away.6 However, the interactions between communities reveal specific properties. More than interaction between individuals in an organization, the interaction between communities can be conceived as a game of language and translation, depending on the one side on the habits and social norms of the respective communities and on the other side on the number and behavior of ‘‘browsers’’ (individuals who navigate between different communities). Last but not least, this game is arbitrated by the hierarchy. The hierarchy cannot in principle influence the internal functioning of a given community, but it can influence the nature of interactions between communities.7 Hierarchy can facilitate the building of a common platform of knowledge between communities and find ways to let the knowledge accumulated by communities flow and bring value to the firm. This raises the issue of the delicate matching between the ‘‘visible’’ hierarchical structure of the firm and the ‘‘invisible’’ process of active knowledge formation that occurs within the myriad of communities in a given firm. A too strict control of hierarchy forcing members to follow the rules decreed by the ‘‘visible structures’’ would prevent the firm from benefiting and deriving value from the knowledge accumulated by the ‘‘invisible communities.’’ Such decisions certainly would not have eliminated the functioning of the community (people would continue to talk and exchange about their practices), but the flow to the rest of the organization of useful knowledge accumulated at the level of the respective communities would presumably have been hampered. On the other hand, leaving to the communities all the process of knowledge creation and formation will expose the firm to risks of incoherence, inconsistency, and anarchy. To a large extent, hierarchy is in charge of the fine tuning of the cognitive distance between communities.

Knowing Communities in Organizations

97

The above developments claim for a precise examination of the nature of the interactions between communities. To do so, considering that the unit of analysis is the community, let us focus on how to describe the nature of interactions between communities. In other words, what is the meaning of a ‘‘strong’’ or a ‘‘weak’’ interaction between communities? We propose to explore this issue through two main hypotheses: Our first hypothesis is that the structure of interactions between communities can be defined by two main factors: the repetitiveness of interactions between communities and the quality of communication between communities, which could be assimilated to the cognitive distance between communities. Indeed, the two phenomena have common features, but distinguishing them is important to clarify the different contexts of interactions between communities. The repetitiveness of interactions between communities expresses the ‘‘quantitative’’ dimension of the relationships between communities.8 Some communities may meet frequently (e.g., workers and managers using the same canteen), and this can generate some benefits for the firm (e.g., formation of a certain common knowledge, circulation of news that ‘‘something isn’t going well’’), even though the intensity of communication between them is low (e.g., minimal common language or grammar to improve the circulation of knowledge between the communities). A high degree of repetition of interactions between knowing communities contributes to stimulate the processes of learning, create favorable conditions for the resolution of conflicts, and encourage the realization of economies of scale. Organizational devices, such as group projects or frequent meetings encouraging the socialization of experiences, are regularly introduced by the management to compensate for the lack of spontaneous interaction between heterogeneous communities. This enables us to better understand the importance given to the construction of privileged learning platforms by firms (‘‘ba,’’ in the sense of Nonaka and Konno 1998). Frequent quantitative interactions between communities contribute to lower the cognitive distance between communities but do not guarantee in the long term the existence of a common grammar and codes between heterogeneous units.



The quality of communication between communities expresses the ‘‘qualitative’’ dimension of the relationships between communities. Some communities can be joined together through a rich texture of communication, even if the quantitative ‘‘degree of repetition’’ of interaction is low. Mintzberg (1979), for example, quotes the well-known example of operations in hospitals, where the members of the different communities involved (surgeons, anesthetists, nurses) meet infrequently, but when they do so, they know exactly what to do and how to work together (thanks to the possibility of communication provided during their respective training).



98

Patrick Cohendet

Table 1 Different types of organizational contexts of interactions between communities within the firm Low repetitiveness of interactions between communities

High repetitiveness of interactions between communities

Low quality of communication between communities

Weak interactions (weak ties, strong cognitive distance)

Moderate interactions I (strong ties, strong cognitive distance)

High quality of communication between communities

Moderate interactions II (weak ties, weak cognitive distance)

Strong interactions (strong ties, weak cognitive distance)

Circulation of knowledge in an innovating firm is based essentially on the sharing of codes and languages allowing various communities to interact. Thus, it is a question of relational or cognitive proximity (Nooteboom 2000a) between distributed units, requiring attention to syntactic, semantic, and pragmatic communication, shared tacit knowledge, flow and interpretation of information, and trust or other conventions of collaboration. As a result, the specific combinations of the two dimensions (repetitiveness of interactions and quality of communication) lead to four different organizational contexts corresponding to different structures of interactions between communities (see Table 1). From the above, it follows that the coherence of the firm requires that hierarchical structures (including the style of management, the system of decision-making, etc.) should vary with the organizational contexts of interactions between communities. In order to extract the potential benefits from intercommunity interactions, the role, the nature, and the design of hierarchies should strongly differ according to the different organizational contexts. In a knowledge-based context, where an increasing part of useful knowledge is held by communities, the delicate matching between the functioning of autonomous communities and the hierarchical system heavily depends on the context of interactions between communities. We thus consider that the degree of repetition of interactions combined with the likeness of representations due to rich communication modes between communities are the essential elements to understand not only the convergence of actor anticipations, and adaptation to common norms, but also the coherence of the firm as a whole. Obviously, there is no ‘‘ideal’’ organizational configuration for knowledge creation (a one best way). Each configuration might be fitting to a specific environment.

Knowing Communities in Organizations

99

But we argue that as the knowledge-based economy develops, decentralized configurations as proposed above (cases 3 and 4 in Table 1) will be increasingly salient. Cowen and Parker (1997, p. 28) explain with respect to internal organization that ‘‘market changes are moving manufacturing farther and farther away from steadystate, low variety, long-batch production runs, relevant to Taylorist methods, to high variety and small runs . . . Organizations are adopting new forms of decentralization to cope with the instability, uncertainty, and pace of change of the marketplace . . . In cluster or network working, employees of undifferentiated rank may operate temporarily on a certain task or tasks in teams. The clusters are largely autonomous and engage in decentralized decision-making and planning . . . They are conducive to individual initiative (intrapreneurship) and faster decision-taking. They facilitate organizational flexibility.’’ The implications of the above hypotheses lead to the following typology (Table 1) of interactions between communities. 1. The first category (low repetitiveness of interactions between communities, low quality of communication between communities), corresponds to a situation where communities do not interact in the organization and have no means to do so. Thus, the coherence of the firm and the innovative impulse of the organization have to come from an external entity. In such cases, the essence of coordination relies on intensive managerial coordination that establishes ex-ante top-down rules and procedures to be followed by the entire organization and centralizes the global vision of the product creation process in what can be called the traditional sequential process mode of management (as in a typically Taylorist organization). The strong division of work relies on specialized units that do not interact on a frequent basis and do not develop rich modes of communication. Classic incentive and coordination mechanisms such as Taylorist time and motion management principles drive decisionmaking. Management by design clearly dominates management by communities (Amin and Cohendet 2004), although local mechanisms of learning in communities (e.g., at shop floor level) can transmit learning-by-doing effects at the global level of the organization. In this particular environment, the cognitive distance between communities does not change through time. Communities remain distant, and there is no force in the organization that favors a reduction of the cognitive distance between communities. 2. The second category (high repetitiveness of interactions, low quality of communication between communities) corresponds to the overlapping problem-solving mode (as in matrix types of organizations) that aims at bridging and cross-fertilizing through repeated informational exchanges between specialized groups in the organization. When some groups gathering members of different communities are formed,

100

Patrick Cohendet

these groups (such as team projects) are temporary, and each individual keeps the jargon and codes of his community of origin. The absence of a rich architecture of communication between the groups leads to an expensive search for a cognitive consensus between communities and calls for active managerial involvement, mostly expost, to solve disputes and conflicts between communities, but also designated to implement common knowledge and to coordinate beliefs while producing sense. The low intensity of communication between communities, especially noticeable in emergent relations built around many communities, can lead to an expensive search for cognitive alignment between communities. Coordination by leadership (necessarily conscious and intentional) appears to be the ideal solution in instances where the costs of communication or compatibility are onerous or where the resolution of coordination problems is urgent. Therefore, a script of leadership emerges, charged with coordinating intricate actions or beliefs while producing sense. Foss (1999) has shown that, in some circumstances, leadership can offer less expensive solutions than complex mental processes or formation of conventions. This type of situation also requires a specific coupling between management by design and management by communities. Part of the solution might reside in the hands of ‘‘middle management’’ that plays, for authors such as Nonaka and Takeuchi (1995), a decisive role in the innovative quality of the business. The middle managers can be seen as mediators who know the norms and habits of the communities sufficiently well to translate messages of the hierarchy into a jargon intelligible to different communities, and in turn, to translate the messages coming from communities for the hierarchy. In such contexts, the high frequency of interactions between communities may reduce the cognitive distance between them. However, the reduction of the cognitive distance is an uncertain long-term result that may depend on other characteristics of the relationship between communities such as the duration of the interactions (Bogenrieder and Nooteboom 2004). For instance, in short-term projects it is unlikely that the interactions between agents during the project have significant impacts on the cognitive distance between communities. 3. The third category (low repetitiveness of interactions and high intensity of communication between communities) corresponds to the existence of cognitive structures of interaction between communities. The modular organization, based on cognitive platforms that allow rather independent heterogeneous communities to interact efficiently, is the emblematic example of this situation. In such contexts, learning at the component level is insulated from disruptions by unexpected changes in product architecture during development projects. The role of hierarchy is to define ex ante the nature of the platform, and ex post to redefine the platform if radical innovations are unavoidable. The existence of

Knowing Communities in Organizations

101

such an infrastructure of knowledge (common grammar, common codes, common languages) may be due to very different historical factors (a type of education that has anticipated the cognitive forms of relationships between heterogeneous communities, shared experience that has lasted long enough to permit a common grammar to be built, a decision taken by the hierarchy to build a modular platform of knowledge, etc.). But whatever the reason, the common infrastructure of knowledge has taken time and sunk costs to be built. It not only defines what the communities have in common, but it also implicitly defines what they do not have in common. Standardized interfaces between each community and the common platform of knowledge allow each community to work independently of others. This implies specific advantages, in particular the fact that provided that the platform holds, the need for coordination by hierarchy is significantly reduced. The functioning of the modular system implies that the cognitive distance between heterogeneous communities is maintained constant. The innovativeness of the global system results from the freedom offered to each community to explore in-depth the variety of options in its specialized field of knowledge, provided that the standard interfaces are respected. In this case, management by communities temporarily dominates management by design. However, if the constraint of the interfaces cannot be respected, then the efficacy of the common platform becomes severely questioned. This could happen, for example, when emergent innovations in one community imply the reformulation of the whole cognitive platform. In such a context, sense-making interventions by the hierarchy may be needed to decide if the novelty produced requires reformulating the common platform. If so, a new cognitive process of definition of a common grammar, codes, and language has to be initiated. In summary, the role of the hierarchy is to intervene at critical moments when the need to reformulate a common platform of knowledge between communities is perceived as essential. Category 3 is thus a case where management by design and management by communities sequentially alternate as dominant modes of coordination. 4. In this fourth category (high repetitiveness of interactions and high intensity of communication between communities), we can envisage governance by community alone, with hierarchy needed only to ‘‘authorize’’ or ‘‘enact’’ the organizational forms produced by the interactive autonomous communities. The organization can extensively operate in a self-organized manner as one is located in either a consolidated or an emergent context. It is probable that in such a situation, the unceasing efflorescence of communities allows the organization to innovate constantly. This mode could be called management by enactment, echoing the work done by Ciborra (1996), who described the knowledge platform at Olivetti in such terms.

102

Patrick Cohendet

The organization can largely operate in a self-organized manner (including the determination of its core interests, which can occur without excessive market or hierarchical intervention) in either a consolidated or an emergent context. It is probable that in such a situation, the unceasing bubbling of communities allows the organization to innovate constantly since it does not disrupt corporate integrity (this dimension can be related to the creative spiral as conceived by Nonaka and Takeuchi [1995]). In such a context, where management by communities clearly dominates management by design, the main role of the hierarchy is to enact the innovative outcomes produced by the constant interactions of communities. Knowing Communities and the Innovative Potential of Large Firms Having analyzed the nature of the interactions between communities, we can now turn to the issue of the innovativeness of large firms. As suggested by Brown and Duguid, large firms might be highly innovative if they succeed in leveraging the innovative potential of the myriad of their small knowing communities. ‘‘One of the central benefits of these small, self-constituting communities is that they evade the ossifying tendencies of large organizations. Canonical accounts of work are not only hard to apply and hard to learn. They are also hard to change. Yet, the actual behaviors of communities of practice are constantly changing, both as newcomers replace old timers and as the demands of practice force the community to revise its relationship to its environment. Communities of practice like the reps’ continue to develop a rich, fluid, non canonical world view to bridge the gap between their organization’s static canonical view and the challenge of changing practice. This process is highly innovative’’ (Brown and Duguid 1991, p. 50). The literature on innovation (see, for instance, Symeonidis 1996) generally mentions some specific advantages of large firms as compared to small firms in their ability to innovate. These advantages include economies of scale and scope, ability to benefit from externalities between different projects, a larger market power, a larger financial potential, the possibility to concentrate R&D, etc. On the contrary, the small firms benefit from a larger flexibility and a larger potential to exploit decentralized initiatives from the agents belonging to the organization. In particular, SME offer a priori the advantages of avoiding bureaucratic costs associated with the management of large organizations. The rise of coordination costs as the size of the firm grows seems to be unavoidable and appears as the principal factor that hampers large firms from innovating. This increase of bureaucratic burden contributes to muffle the decentralized initiatives to innovate within large companies. What Brown and Duguid suggest is thus a strong potential counterargument to the intense debate found in the literature. Large firms, provided that they can exploit

Knowing Communities in Organizations

103

the innovative potential of the communities that compose the firm, could escape from the burden of increased bureaucracy: They can ‘‘economize on bureaucracy.’’ Knowing communities can bring forward innovative potential to a given organization for the following reasons: Knowing communities do not bear the risk of being ‘‘ossified’’ by the codified rule of a large corporation. The continuous circulation of best practice among the members of the community is a guarantee that within a knowing community the rules in place are continuously questioned.



Knowing communities are continuously enriched by the entrance of new members that bring new ideas and new human capital.



The functioning of knowing communities is not constrained by the boundaries of the organization. Knowing communities are open to the world, and members of a given community continuously exchange knowledge with individuals who do not belong to the organization (and sometimes even work for competing firms). They are global in scope, connecting practitioners worldwide. Members of a given community who belong to the same organization are thus permanently exposed to new ideas and knowledge coming from the outside world. Thus, it often happens that ‘‘communities know more than the organization does.’’ However, this feature has a negative counterpart, the risk that some of the strategic knowledge of a given organization is diffused to the outside world through the current functioning of a given community. Such leakages may be extremely difficult to control.9



The example of the reps at Xerox illustrates the innovative potential of a knowing community in an organization. The dynamic process of innovation activated by this community could be analyzed as a process of dynamic ‘‘cognitive contagion’’ between communities. The process starts at the interaction between the reps and the users.10 Then the reps alerted other communities within the organization (engineers, designers, etc.) to convince them of the need to develop creative ideas at Xerox in small photocopier lines of products. By a progressive contagion of communities within the organization, a kind of ‘‘percolation’’ threshold is achieved, from which the organization as a whole contributes to the development of the innovation. The dynamics of communication between communities can be approached through the principle of ‘‘translation/enrolment,’’ elaborated in particular by Callon and Latour (1991). According to these authors, the innovative diffusion of ideas (for example, from the lab to the market) can be interpreted as a process of progressive contagion of communities, where each community makes efforts to ‘‘command the attention’’ of other communities to convince them of the relevant interest of the knowledge it has elaborated. Of course, this dynamic perspective of communication between communities depends on the nature of the static organizational contexts of

104

Patrick Cohendet

interaction, which were discussed in detail previously. It is thus probable that Category 4 (‘‘Strong interactions, with strong ties and weak cognitive distance’’) is the context in which the innovative process has the greatest chance of being achieved. The group of agents who succeed in expressing and formalizing an innovative idea is confronted by one main difficulty: not the risk of being copied (at no cost), but rather the risk of being misunderstood by others (including agents belonging to the same institution), which then leads to their procedures and experience not being reproduced by others. Communities with creative ideas will thus undertake considerable efforts to alert other communities in order to convince them of the usefulness and potentials of their discovery, by forcing engagement, adherence, a common language, common beliefs, and alignment of interests in general (Callon 1999). The procedural construction of a cognitive weft between communities makes it possible to code the experiences and the history of the organization and therefore to give the sense, ex post facto, of the construction and stabilization of a common vision or culture, ensuring the global consistency of the distributed venture. Conclusion This contribution attempted to recognize and understand the increasing role of knowing communities in organizations. Companies are beginning to recognize that knowing communities can be supported, nurtured, and leveraged to benefit the members of the community and the organization as a whole. The critical role of communities as building blocks for the formation of knowledge and competences in organizations has been highlighted. However, the growing role of knowing communities raises many questions for corporate management. In particular, a key question is how can firms benefit from the useful knowledge held by its different distributed communities without compromising the organizational hierarchical structure geared toward the firm’s efficiency. Facing this dilemma, it is probable that as long as communities emit ‘‘weak signals’’ in the organization, hierarchy will not intervene. However, as a knowing community becomes increasingly influential, the firm will certainly try to exercise more control over the community and limit its growing influence by trying to convert this influential community into an organizational unit. This tendency is illustrated by Gongla and Rizzuto (2001, p. 273) in the ways they describe successful knowing communities at IBM Global Services experience: At this stage, the community innovates and generates, creating significant new business objects—new solutions, new offerings, new methods, new processes, and new groups. The community identifies, influences, and even creates trends in its area of expertise. The community’s innovation affects not just its members and the immediate domain within which it

Knowing Communities in Organizations

105

operates but other parts of the organization and external agencies as well. We hypothesize, however, that few communities will ever reach or sustain themselves as a community at the adaptive stage. The work being done by the community becomes too important to the organization for it to allow the community to continue as a self-governing body. There is a distinct likelihood that the organization will want more control and essentially convert the community into an organizational unit.

Notes 1. This chapter is issued from a series of works in process carried out at BETA Strasbourg, with M. Diani, O. Dupouet, E. Schenk, and F. Cre´plet. It also benefited from a discussion at the conference in honor of Bart Nooteboom, Nov. 25 and 26, 2004, University Erasmus of Rotterdam. 2. One of the most widely cited studies of knowing communities was carried out by Orr (1990) concerning activities of a team of photocopier reps at Xerox. According to Orr, the job of a rep is best described as a continuous improvisation taking place in a network of relationships between clients, machines, and other reps. Reps work autonomously. They intervene at Xerox clients’ places where they have to repair a machine, by themselves most of the time. Together, they form a community of practice, which allows mutual help and the collective problem solving of unusual breakdowns. When they talk about machines, technicians actually build up a common identity while exchanging what each of them gained from various experiences. They create a stock of operational competencies in sharp contrast with handbooks and user guides that are promoted by the hierarchy. This ever-actualized repertoire is transmitted through oral culture. It allows reps to cope with managerial evolutions that downplay their role by increasingly relying on work methods disconnected from the reality of workers and machines. The highly technical work of reps appears as a resource socially distributed, stored, and diffused above all by informal conversations. Moreover, through their ability to maintain constant interactions with other communities (engineers, designers, etc.) in the organization, reps are at the origin of many creative ideas at Xerox. 3. See, for instance, Cowan and Foray (1997). 4. In organizations, the knowing communities have been called by different names at various times, such as ‘‘learning communities’’ (Hewlett Packard), ‘‘family groups’’ (Xerox), ‘‘knowledge networks’’ (IBM Global Services), etc. In the academic literature as well, many types of knowing communities have been recently analyzed. Besides the main categories (communities of practice and epistemic communities detailed in the text), one can quote, for instance: Communities of creation were studied by Sawnhey and Prandelli (2000). These authors demonstrate that traditional organizational models associated to R&D departments do not allow firms to benefit from the full creativity and diversity that their partners could bring in. They help in setting interorganizational cooperation agreements on innovations.  Strategic communities have been identified by Storck (2000), who draws this concept from empirical studies carried out at Xerox. These communities might belong to the company or to operational units. They are formed by groups of collaborators specialized in a specific information technology. The variety of experiences and knowledge gained from belonging to 

106

Patrick Cohendet

these two dimensions provide them with sources of learning and organizational performance enhancement.  Lynn et al. (1996) refer to the concept of communities of innovation to account for modalities used by organizations to bring a technology to the market. These are a set of interdependent and integrated organizations involved in this process. Members are identified by examining sources and flux of assets and information circulating during the process of commercialization. Communities of creation, given their properties, are quite similar to epistemic communities as they are described in this paper. The two other types of communities are, on the contrary, different in nature from both epistemic communities and communities of practice (Storck, 2000, draws the differences existing between strategic communities and communities of practice). 5. The ‘‘truce’’ hypothesis in Nelson and Winter’s routine analysis (1982). 6. As discussed in Nooteboom, in order to achieve a specific joint goal, the categories of thought of the people involved must be coordinated to some extent. Different people have a greater or lesser ‘‘cognitive distance’’ between them (Nooteboom 1999). A large cognitive distance has the merit of novelty but the problem of incomprehensibility. In view of this, organizations need to reduce cognitive distance, i.e., achieve a sufficient alignment of mental categories to understand each other, utilize complementary capabilities, and achieve a common goal. ‘‘Cognitive distance yields both a problem and an opportunity. The opportunity is that we learn from others only when they see and know things differently. In the absence of claims of objective knowledge, interaction with others is the only path we have to correct our errors. The problem is that people may not understand each other and have to invest in understanding’’ (Bogenrieder and Nooteboom 2004). 7. Learning communities tend to maintain a cognitive distance vis-a`-vis the exterior, in particular vis-a`-vis the hierarchy of the organization. This can be interpreted as a ‘‘creative insubordination,’’ which is in particular expressed in the classical example of the ‘‘rep’’ from Xerox, rather than reinterpret in their own jargon, rules, and norms of the community the codified orders that are coming from the hierarchy. This natural tendency could be interpreted in terms of cognitive dissonance. Cognitive dissonance is a psychological phenomenon that refers to the discomfort felt at a discrepancy between what you already know or believe and new information or interpretation. It therefore occurs when there is a need to accommodate new ideas, and it may be necessary for it to develop so that we become ‘‘open’’ to them. Neighbour (1992) makes the generation of appropriate dissonance into a major feature of tutorial (and other) teaching: He shows how to drive this kind of intellectual wedge between learners’ current beliefs and ‘‘reality.’’ Cognitive dissonance has two major effects on learning: If someone is called upon to learn something that contradicts what they already think they know—particularly if they are committed to that prior knowledge—they are likely to resist the new learning. Accommodation is more difficult than assimilation, in Piaget’s terms.  If learning something has been difficult, uncomfortable, or even humiliating enough, people are not likely to admit that the content of what has been learned is not valuable. To do so would be to admit that one has been ‘‘had,’’ or ‘‘conned.’’ It is not, however, the qualities of the course that are significant so much as the amount of effort that participants have to put into it, so the same qualification may well be valued more by the student who had to struggle for it than the student who sailed through. 

Knowing Communities in Organizations

107

8. We have simplified the analysis of the quantitative dimension of the interaction between communities by focusing only on one characteristic (the frequency of interaction). Bogenrieder and Nooteboom (2004) suggested a richer definition based on four characteristics (one of them being the frequency of interaction): ‘‘In our view, the ‘strength of ties’ has four aspects. One aspect is intensity, which refers to the effort and commitment of resources involved and to the scope of activities taken up in the tie (share of total activities). The resources that are committed are not necessarily only resources of money, time, or effort, and may also include psychological resources (commitment, loyalty, fairness, and empathy). A second aspect is frequency of interaction, a third is openness of communication, and a fourth is duration of ties. Strong ties yield shared experience, which reduces cognitive distance. Durable ties enable the development of empathy and identification . . . as a basis for trust.’’ 9. Because virtual devices will expand, this issue is going to be a critical one for organizations. A solution to avoid excessive diffusion of strategic knowledge to the outside world from a given knowing community in the organization is to propose to the organizational members to sign a ‘‘charter’’ of responsive behavior toward the organization. 10. In the perspective of Eric von Hippel’s works.

References Amin, A., and P. Cohendet (2004). Architectures of Knowledge: Firms, Capabilities, and Communities. Oxford, UK: Oxford University Press. Bogenrieder, I., and B. Nooteboom (2004). ‘‘Learning Groups: What Types Are There? A Theoretical Analysis and an Empirical Study in a Consultant Firm.’’ Organization Studies 25: 287–313. Bourdieu, P. (1977). Outline of a Theory of Practice. Cambridge, UK: Cambridge University Press. Brown, J. S., and P. Duguid (1991). ‘‘Organizational Learning and Communities of Practice: Toward a Unified View of Working, Learning and Innovation.’’ Organization Science 2: 40– 57. ———, and P. Duguid (1998). ‘‘Organizing Knowledge.’’ California Management Review 40: 90–111. Callon, M. (1999). ‘‘Le re´seau comme forme e´mergente et comme modalite´ de coordination.’’ In Re´seau et Coordination, M. Callon et al., eds. Paris: Economica. ———, and B. Latour (1991). La science telle qu’elle se fait: Anthologie de la sociologie des sciences de langue anglaise. Paris: La De´couverte. Ciborra, C. U. (1996). ‘‘The Platform Organization: Recombining Strategies, Structures, and Surprises.’’ Organization Science 7: 103–118. Cohendet, P., F. Creplet, M. Diani, O. Dupouet, and E. Schenk (2004). ‘‘The Delicate Matching between Hierarchies and Communities.’’ Journal of Management and Governance 8: 27– 48. Cowan, R., and D. Foray (1997). ‘‘The Economics of Knowledge Codification and Diffusion.’’ Industrial and Corporate Change 6(3): 595–622.

108

Patrick Cohendet

———, P. A. David, and D. Foray (2000). ‘‘The Explicit Economics of Knowledge Codification and Tacitness.’’ Industrial and Corporate Change 9: 211–253. Cowen, T., and D. Parker (1997). Markets in the Firm: A Market-Process Approach to Management. London: Institute of Economic Affairs. Foss, N. (1999). ‘‘Understanding Leadership: a Coordination Theory.’’ Working DRUID. Gongla, P., and C. R. Rizzuto (2001). ‘‘Evolving Communities of Practice: IBM Global Service Experience.’’ IBM Systems Journal 40(4): 842–862. Grandori, A. (1997). ‘‘Governance Structures, Coordination Mechanisms and Cognitive Models.’’ The Journal of Management and Governance 1: 29–47. Granovetter, M. (1985). ‘‘Economic Action and Social Structure: The Problem of Embeddedness.’’ American Journal of Sociology 91: 481–510. Langlois, R., and N. Foss (1996). ‘‘Capabilities and Governance: The Rebirth of Production in the Theory of Economic Organization.’’ Kyklos 52: 201–218. Lave, J., and E. Wenger (1991). Situated Learning: Legitimate Peripheral Participation. Cambridge, UK: Cambridge University Press. Leonard-Barton, D. (1995). Wellsprings of Knowledge: Building and Sustaining the Source of Innovation. Boston: Harvard Business School Press. Lynn, L., N. Mohan Reddy, and J. D. Aram (1996). ‘‘Linking Technology and Institutions: The Innovation Community Framework.’’ Research Policy 25: 91–106. Mintzberg, H. (1979). The Structuring of Organizations. Englewood Cliffs, NJ: Prentice Hall. Neighbour, R. (1992). The Inner Apprentice. Plymouth, UK: Petroc Press. Nelson, R. R., and S. Winter (1982). An evolutionary Theory of Economic Change. Cambridge, MA: Harvard University Press. Nonaka, I., and N. Konno (1998). ‘‘The Concept of ‘Ba’: Building a Foundation for Knowledge Creation.’’ California Management Review 40: 40–54. ———, and H. Takeuchi (1995). The Knowledge-Creating Company: How the Japanese Companies Create the Dynamic of Innovation. Oxford: Oxford University Press. Nooteboom, Bart (1999). Inter-Firm Alliances: Analysis and Design. London: Routledge. ——— (2000a). ‘‘Learning by Interaction: Absorptive Capacity, Cognitive Distance and Governance.’’ Journal of Management and Governance 4: 69–92. ——— (2000b). Learning and Innovation in Organizations and Economies. Oxford: Oxford University Press. Orr, J. (1990). Talking about Machines: An Ethnography of a Modern Job. Ithaca, NY: Cornell University. Polanyi, M. (1967). The Tacit Dimension. New York: Doubleday. Sawnhey, M., and E. Prandelli (2000). ‘‘Communities of Creation: Managing Distributed Innovation in Turbulent Markets.’’ California Management Review 4: 24–54. Scholer, D. (1996). New Community Networks: Wired for Change. Reading, MA: AddisonWesley.

Knowing Communities in Organizations

109

Storck, J. (2000). ‘‘Knowledge Diffusion through ‘Strategic Communities.’ ’’ Sloan Management Review 41(2): 63–74. Symeonidis G. (1996). ‘‘Innovation, Firm Size and Market Structure: Schumpeterian Hypotheses and Some New Themes.’’ OECD Economic Studies 27: 11. von Hippel, E. (1988). The Sources of Innovation. New York: Oxford University Press. Wenger, E. (1998). ‘‘Communities of Practice: Learning as a Social System.’’ Systems Thinker 9: 2–3. ———, R. McDermott, and W. M. Snyder (2002). Cultivating Communities of Practice. Boston: Harvard Business School Press.

III The Changing Role of Institutions

9 Epistemic Infrastructure in the Rise of the Knowledge Economy Margaret Hedstrom and John Leslie King

Introduction The epistemic infrastructure of the knowledge economy1 arose from libraries, archives, museums, galleries, zoos, aquaria, and other systematic collections that enable individuals and societies to know what they know and to do what they do.2 The assembling, safe keeping, organizing, representing, and displaying of archival documents, plants and animals, common and rare objects, works of art, and so on is the heart of knowledge generation, learning, sense making, and commerce. This is an ancient art: As long ago as 5,000 years, collections of clay tablets, papyri, and inscriptions on stone kept track of laws and decrees, administrative and financial transactions, and ownership and control over property and reminded citizens and subjects of their duties and obligations.3 These collections were stored and maintained in centers such as the great library of Alexandria, as well as in similar libraries at Ephesus, Pergamum, Athens, and Ur. Ancient libraries can be found as well in sites from Macedonia to the Persian Gulf and from China and India to southeastern Asia. Even though elements of epistemic infrastructure have been with us for five millennia, there is no inherent reason for any particular collection or collecting institution to persist. In fact, we have to contend with the fact that the Alexandrian library was destroyed and that no library of Greek or Roman antiquity survived intact due to a combination of natural disasters, intentional destruction, and a decline in the relevance of their collections for new political regimes, ideologies, and epistemic cultures. The decline and rebuilding of collections and amassing of new ones both signals and serves as a trigger for significant transformations in the creation, dissemination, and exploitation of knowledge. The Dark Ages of Europe followed closely upon the purposeful destruction and dissipation of the ancient libraries of Greek and Roman antiquity in the 4th century. Had the learning captured in those collections not been preserved by Islamic scholars

114

Margaret Hedstrom and John Leslie King

of the Middle East and North Africa, it might have been lost forever. That learning was restored to the West beginning with the Islamic conquest of Iberia in the 8th century. The Christian retaking of Iberia in the late 11th century captured some of what had been preserved in these precious Islamic libraries, and portions of the collections were subsequently translated into Latin and disseminated. Christian Europe, awakening from centuries of intellectual impoverishment, began to evolve institutional mechanisms of knowledge creation and sharing that contributed directly to the Renaissance, the Scientific Revolution, and the Enlightenment. We attribute the persistence of collections in part to the adaptability of collecting practices and collecting institutions. As collections grew in size and scope, their custodians developed methods for organizing and managing larger collections and new types of material. In time, ancient collections and their new counterparts were transformed from cloistered secrets or private treasures to public goods that played vital service for science, education, and entertainment. This brief essay cannot cover completely the story of epistemic infrastructure. Rather, it pulls into the foreground attributes whereby epistemic infrastructure forms an essential foundation for the knowledge economy, and suggests how and why epistemic infrastructure is as critical today as it has been in the past. The Rise and Role of Epistemic Infrastructure Modern epistemic infrastructure has been built on a vast array of collections, as well as the systems of practice that have shaped their contents, organization, accessibility, and use. To illustrate the rise and role of epistemic infrastructure, we focus on collections of objects and texts in two classes of institutions, museums and archives/ libraries. The modern versions of these institutions arose from a revolution in thinking in early 17th century Europe that replaced the traditions of scholasticism with principles of rationality and empirical investigation. Slowly, the haphazard gathering of oddities evolved into a deliberate practice of collecting similar and dissimilar things and imposing upon them a systematic order that reflected a new way of thinking that became the foundation of the modern world.4 We begin with the Ashmolean Museum of Oxford University, which is among the oldest surviving museums in the world.5 The Ashmolean was never conceived of as a museum in the modern sense. It arose from rather informal collecting habits by two gardeners, a father and son team both named John Tradescant between 1610 and 1662. Tradescant Sr. began collecting plants, but expanded his collection to include shells, stuffed animals, works of art, and artifacts from afar.6 He eventually established a residence in South Lambeth, across the Thames from London at Westminster, and put his collection on display under the name The Ark. Tradescant Jr.

Epistemic Infrastructure in the Rise of the Knowledge Economy

115

became the keeper of the collection and expanded it after his father’s death. In 1650 he met Elias Ashmole, a successful lawyer who had studied at Oxford and who recognized the importance of the collection. Ashmole and a colleague, Thomas Wharton, produced a catalog of the collection titled Musaeum Tradescantianum in 1656. Following Tradescant Jr.’s death in 1662, Ashmole took control of the collection and donated it to Oxford University in 1683. This was not a unique occurrence. From the middle of the 16th century onward, many such collections were created.7 By the 18th century, Paris alone had at least 723 of these Cabinets of Curiosities, or Wunderkammern as they were commonly known.8 Such collections eventually found their way into institutionalized arrangements in universities and scholarly societies and as self-standing museums that afforded greater stability and allowed for scholarly interrogation. This paralleled the rise of the distributed ‘‘invisible college’’ of scholars, collaborating through meetings and correspondence under the auspices of organizations such as the Royal Society of Great Britain.9 The Royal Society set out in 1669 to construct a universal taxonomy of natural objects, and although the effort failed, it established a goal of systematic collecting in the natural sciences that continues to this day. Wunderkammern, and their later instantiation as museums, played a critical role in the rise of modern science and scholarship. Collecting became a form of inquiry: a means of creating a didactic resource that initially made sense only to the collector but with organization and codification was transformed into a resource that could be shared among collectors and with inquisitive people to create a common knowledge.10 This institutionalized clarification and correction of earlier evidence constituted a vital step in the rise of modern scholarly and scientific inquiry. When the skeleton of the dinosaur Iguanadon, discovered in 1822 in southern England, went on public display, visitors asked where such a thing came from. The public debate over the fossil record and the implicit arguments by visual metaphors that specimen collections presented paved the way for both easy acceptance of and widespread alarm at the publication of Darwin’s Origin of Species in 1859. Science came into the political sphere and remains there still. Wunderkammern also aided the rise of systematic method in the sciences by facilitating careful, repeated observation. Large collections permitted careful comparison of morphology among the specimens and sparked early efforts at taxonomy and classification that became the foundation of the life sciences. Documentation, in the form of written and printed catalogs and paintings of Wunderkammern, allowed representations of the collections to circulate.11 This made it possible to export knowledge from centers to the periphery and to import knowledge into the centers from far-flung places in an emerging global network of trade and commerce. Scholars on the periphery could compare local collections with

116

Margaret Hedstrom and John Leslie King

representations of other collections, raising questions that led to further discussion and examination via the invisible college.12 At the same time, new specimens and objects were imported from China, India, Africa, and the Americas into European commercial centers such as Venice, Seville, Lisbon, and Amsterdam, enabling what Burke calls ‘‘discovery in a global context.’’13 This emerging epistemic infrastructure laid the groundwork for Carl von Linne´ (Linnaeus) to construct his Systema Naturae in 1738, upon which all subsequent biological taxonomy was built. Wunderkammern sparked scholarly and scientific collaboration that formed the foundation of the scientific and industrial revolutions. In a short time, major institutionalized collections were formed. The nucleus of the French National Museum of Natural History was formed in 1749, and that of the British Museum in 1753. Their success led to a golden age of museum expansion and stabilization in the 19th century. Philadelphia’s Academy of Natural Sciences opened in 1812. The Peale Museum opened in Baltimore in 1814. The National Museum of Denmark was established in 1816. The ethnographic museum of the Academy of Sciences in Petrograd opened in 1836. The Smithsonian Institution got going in 1846. The British Great Exhibition of 1851 had a major influence on the relationship between display of collections and emerging industrial enterprise. Harvard University established museums for botany, zoology, and anthropology between 1858 and 1866. Museums became both forces for development and common fixtures in developed countries. Modern libraries and archives also emerged during the Scientific Revolution and the Enlightenment. The invention of printing with movable type around 1450, based upon earlier woodblock printing and papermaking technologies from China and the Islamic world, interjected a powerful new technology into the epistemic infrastructure.14 Historians of writing and literacy find many connections between the introduction of printing, the Scientific Revolution, and the spread of Enlightenment thought. For one, the technology of printing made it possible to reproduce manuscripts as books and greatly reduced the labor required to create multiple copies. The vast majority of early printed books were published editions of classical and religious texts that had been preserved in Islamic societies or salvaged thorough repeated copying and translation in the monasteries of Catholic Western Europe.15 Yet by the 16th century, printers were publishing a wide variety of new compilations of facts, drawings, and other data. These compendia of laws, astronomical charts, botanical drawings, maps, and the like created the initial constructs for analyzing abstract representations of the world, much like the early collectors of museum objects created a common language for analyzing objects of nature.16 Multiple copies of stable printed works created an opportunity for scholars, intellectuals, and the clergy to compare different texts and to discover similarities and

Epistemic Infrastructure in the Rise of the Knowledge Economy

117

anomalies among different representations of similar things and among different interpretations of similar phenomena. The circulation of printed works fundamentally altered the transfer of knowledge between religious and secular cultures, from expert to novice, from academy to academy, from scientist to artisan, and across generations. The power to reproduce texts and circulate them on a larger scale than ever before possible allowed scientists, philosophers, and theologians to mobilize and convince others to see the world the same way they did.17 The circulation of published works also created new discourse about secular society, religion, science, philosophy, and geography that eventually resulted in distinctions between science and magic, evidence and faith, and near and distant, which characterize the Scientific Revolution and Enlightenment thought. Printing texts in vernacular languages, with accompanying dictionaries, grammars, and spelling books, helped to consolidate local dialects into national languages, giving European languages their modern forms by the 17th century. This process hastened the differentiation of people from people and state from state on the basis of a shared linguistic and cultural identity.18 Like museums, libraries and archives grew hand in hand with private publishing and the book trade, the emergence of the modern university, the maturation of scholarly societies, and the rationalization of the administrative apparatus of modern states. These forces worked together to further accelerate the production of books, administrative documents, maps, journals, reports, and the like and to foster the collection of books, maps, drawings, and documents that flowed into private and public collections. Between the 16th and 18th centuries, many collections of texts were gradually appropriated from the church and the nobility and transformed into sources of information that the nascent disciplines of philosophy and science used to create new epistemic cultures. Private collectors gradually opened their collections for viewing by privileged elites and for research by qualified scientists, or they donated them to libraries and archives in universities, major municipalities, and provincial towns. Progressive monarchs donated their private libraries to the citizens, forming the early instances of national libraries. This effort was aided by the advent of mandatory legal deposit. France established in 1537 that every printer must deposit a copy of each title printed in the King’s castle, and similar laws were soon enacted throughout Europe. This greatly facilitated the creation of major libraries such as Oxford’s Bodelian Library, which grew out of an arrangement between Sir Thomas Bodley and the Stationer’s Company to place into the Oxford University library copies of everything the company published. The first copyright act, Britain’s Statute of Anne in 1709, provided a fixed term of protection for published works and required deposit of nine copies in libraries through the country. During the French Revolution, the Bibliothe`que du Roi, at that time the

118

Margaret Hedstrom and John Leslie King

largest collection of books in Europe, was seized and transformed into the Bibliothe`que Nationale de France. By 1800, national libraries existed in 20 countries, including the United States, which passed legislation that year establishing the Library of Congress. Archives also grew in number and scope during this period to meet the demands of administrators in the church, the state, and the commercial sector for organized records to keep track of land, facilities, production, extraction, and subjects in their growing domains of domination. A major breakthrough in the organization of archives occurred shortly after the French Revolution, when scholars of paleography and diplomatics at the Ecole des Chartes established the principle of provenance. When faced with the task of organizing massive collections of books, archives, and manuscripts from prerevolutionary institutions that had been seized by the new state, archivists introduced the concept of respect des fonds. Under this principle, archivists maintain the original arrangement of collections that was established by the person or entity that produced the records. This helped overcome the confusion of earlier classification schemes for archives that were built on arbitrary criteria such as time period, subject matter, strict chronology, author, location, size, and even shape or color, in favor of classifying similar documents on the basis of their common origins or provenance.19 The use of archives to write the history of the French nation, based on documentary evidence rather than determinism or romanticism, started a transition that made archives valuable not only as instruments of administration but also as resources for learning from the past.20 The evolution of archives and libraries into instruments of knowledge acquisition and organization was not accidental. As the collections in libraries and archives grew, haphazard methods of storing and organizing collections became inadequate for locating material on particular topics. Efforts by collectors, scholars, and librarians to impose order on collections through organizational and classification schemes began in the 17th century. Early proposals for arranging books and catalogs, such as Naude’s Advice on Building up a Library (1627), de Ara´oz’s How to Arrange a Library (1631), or Leibnitz’s Plan for Arranging a Library (1679), reflected competing views of how best to divide knowledge into useful categories.21 Use of the term bibliothe`que (library) was not limited to the physical places where books were organized and housed. Bibliothe`que also denoted catalogs and inventories of all known books by a particular author, on a particular subject, or in a particular language, regardless of their physical location.22 Managing the proliferation of books required increasing specialization of libraries by subject, language, origin, and types of works. This created a need not only for universal classification schemes to organize the ideal universal library but also for increasingly specialized vocabularies and cataloging methods aligned with particular topics or specialties, much like

Epistemic Infrastructure in the Rise of the Knowledge Economy

119

the emergence of systematic classification in museums and the scientific disciplines that they supported. We opened this discussion with the rise of the Ashmolean Museum from a Cabinet of Curiosity, created for the pleasure of its collector and used as a form of public entertainment, to an organized museum at the end of the 17th century. We conclude this section with the Library of Congress as an example of how far the epistemic infrastructure had evolved by the beginning of the 19th century. The copyright provision in the U.S. Constitution already reflected the progressive goals of the new republic. Unlike earlier copyright laws, which were intended either as a means to censor unorthodox ideas or to protect the intellectual property of authors, the copyright provisions in the U.S. Constitution had the intention of promoting ‘‘the progress of science and the useful arts.’’ In 1802, President John Adams approved an appropriation of $5,000 to purchase ‘‘such books as may be necessary for the use of Congress.’’ The first books arrived from London in 1801 and were stored in the U.S. Capitol. After the Capitol burned during the War of 1812, Thomas Jefferson reestablished the Library by selling his own personal collection of 6,487 volumes to Congress in 1815. When Jefferson sold his multifaceted, multilingual collection to Congress, he felt the need to defend its diversity by stating that there was ‘‘no subject to which a Member of Congress might not have occasion to refer.’’23 The establishment and rebuilding of the Library of Congress illustrates a number of aspects of the development of epistemic infrastructure. By the early 19th century, Enlightenment thinkers, founding a new nation on the periphery, considered a national library essential for informed governance and for the ‘‘progress of science and the useful arts.’’ Jefferson’s private library, which reflected his own broad and cosmopolitan interests, became the core of a national resource. Science and the useful arts would serve as the engine for economic development, long before the concept of a ‘‘knowledge economy’’ came into circulation. Epistemic Infrastructure in the Industrial Era: Adaptability of Purpose Francis Bacon foretold the rise of epistemic infrastructure in 1594. In his Gesta Grayorum, he said that knowledge was acquired through libraries, botanical gardens, zoos, aquaria, museums, and laboratories.24 Two centuries later, the world was reaping the rewards of that infrastructure. Daniel Webster captured this in a speech given on June 17, 1825, at the groundbreaking for the Bunker Hill monument in Massachusetts, in which he said that ‘‘a vast commerce of ideas’’ had emerged wherein knowledge had ‘‘triumphed over distance, over differences of language, over diversity of habits, over prejudice, and over bigotry.’’25 Webster saw

120

Margaret Hedstrom and John Leslie King

that the pursuit of knowledge had grown from a pastime for curious individuals into a great dynamo of social and economic advancement. By the time of Webster’s speech, the steam engine had been applied to ship and rail transport, mechanistic agriculture had begun with the cotton gin, vaccination against smallpox was under way, and manufactured methane was being used to light cities. Within fifty years of the speech, Henry Bessemer revolutionized the making of steel, the railroad industry invented operations management, the telegraph was in regular use, Pasteur pioneered profound changes in human health, and warfare reached new levels of carnage through new weaponry such as the machine gun. The changes went far beyond economic progress. By 1835 the British monarchy had weakened in favor of Parliament, the American and French revolutions had transpired, and modern democratic institutions had been established. By 1875, most industrial nations had abolished slavery, instituted compulsory primary and secondary education, and were beginning major expansions in higher education. The 19th century saw the creation of nearly all of the world’s great museums, the rise of the great state libraries, the establishment of many academic libraries, and the beginning of public library systems. Industrialization, urbanization, and mass literacy created further pressures to transform private collections of information and artifacts into public goods. During the latter half of the 19th century, private benefactors founded and provided resources for countless museums, galleries, historical societies, and libraries so that a broader public could appreciate and learn from collections of treasures that had accumulated over the centuries.26 These institutions aided the social mission of universal education in consort with mandatory education and the establishment of countless public and private colleges and universities. Public libraries provided citizens with information and tools to make informed decisions and taught immigrants and new urban residents how to make intelligent use of their leisure time.27 Museums and archives contributed to nation building by assembling documents and artifacts that provided a basis for a shared sense of the past and a common national and cultural identity. During this expansion, librarians, bibliographers, archivists, and curators developed highly refined practices of organization and classification for texts, objects, living plants and creatures, and cultural artifacts to serve the specific needs of increasingly specialized collections tuned to particular audiences. One vital innovation in modern epistemic infrastructure was the creation of systematic methods of cataloging that could scale to the industrial production of books, magazines, newspapers and other mass-produced texts. This required cataloging and classification schemes that were built on explicit principles and rules so that efforts to create bibliographic information for one work could be shared and reused by other libraries that also owned a copy of the same work. Sir Anthony Panizzi, a librarian at the

Epistemic Infrastructure in the Rise of the Knowledge Economy

121

British Museum, first developed 91 rules for cataloging in 1832. Melvil Dewey introduced the more precise Dewey Decimal System in 1876, which classified knowledge into 10 categories, each of which could be subdivided decimally in subcategories, sub-subcategories, etc. A key breakthrough came near the end of the 19th century, when Charles Cutter built on Dewey’s concepts to create a highly logical and scholarly system. Inspired by Dewey’s work, Cutter began work on his own classification system in the 1880s while serving as Librarian of the Boston Athenaeum. Although Cutter’s classification scheme was not adopted as widely as the Dewey system, he introduced basic principles for library cataloging that inform most bibliographic classification systems today. Cutter believed that if a patron knew the author, title, or subject of a desired work, he or she ought to be able to locate that item in the library catalog. Cutter also proposed an expansive system of classification, with seven levels ranging in specificity from one level for the smallest libraries to all seven levels for the largest libraries.28 Dewey was also concerned with the standardization and efficiency of library operations. He established the Library Bureau in 1876 ‘‘for the definite purpose of furnishing libraries with equipment and supplies of unvarying correctness and reliability.’’29 When he founded the first American ‘‘library school’’ at Columbia University, it was named the School of Library Economy. In the early 20th century, when the Library of Congress started organizing its one million volumes for public use, librarians adapted Cutter’s scheme, known today as the LC classification scheme. This created a platform for the development of uniform standards for cataloging, classification, and eventually interlibrary lending of materials. In 1902, the Library initiated a card distribution service that made it possible for other libraries to purchase preprinted library cards for their catalogs, rather than cataloging their collection according to idiosyncratic and institution-specific practices. The extension of the library’s classification and cataloging schemes to the rest of the nation led to a uniformity of cataloging across libraries subscribing to the service, but more significantly the standard classification system provided a common ‘‘user interface’’ to print publications through the card catalog. This epistemic infrastructure became further refined during the mid-20th century when the Library of Congress and national libraries elsewhere developed standard classification schemes and cataloging rules for subjects, names, titles, and an expanding variety of new media types. The selection and organization in museums, libraries, and archives of specific types of collections for particular audiences conveyed to the public where to go to seek types of information and objects of interest. Careful selection of the best exemplars or most appropriate materials put marks of authenticity, legitimacy, and authority on collections. The naturalist seeking physical specimens, the historian

122

Margaret Hedstrom and John Leslie King

searching for documents, or the private citizen looking for uplifting fiction could trust the epistemic infrastructure to provide the best evidence, the highest quality sources, or the most useful reading material. This infrastructure was a source of continuity, with a focus on the perpetual care of knowledge even when knowledgebearing objects had become obsolete, outdated, or irrelevant. The preservation function reinforced a conservative ethos in the institutions of the infrastructure that forced them to balance continuity with change. Institutions often became reluctant to change their practices for selection, access, and exhibition and tended to resist pressure to discard or limit access to information deemed subversive, dangerous, or politically unpopular. This pull toward conservatism in collection development and practice intersected with a variety of social, cultural, and economic pressures in the late 20th century to put epistemic infrastructure under stress. Established institutions found audiences dwindling for exhibits that were viewed as elitist, and user communities began shrinking in relationship to collections viewed as conservative, nationalistic, and bourgeois. In response, starting in the late 1960s, many museums redefined their role and image from that of a temple containing clearly interpreted objects and toward an interactive forum for learning and open-ended interpretation.30 Public libraries built new facilities in rapidly growing suburban areas, actively targeted young readers, expanded services such as with outreach programs, and organized events that helped to integrate libraries into neighborhoods and communities.31 In response to a growing interest in social history, local history, and genealogy, many archives reassessed their collecting policies and aggressively sought materials on women, ethnic and linguistic minorities, popular culture, and social movements. In recent years, core elements of the epistemic infrastructure have faced both declining public subsidies and rising costs. Some institutions have responded primarily with internal structural changes that rationalize collecting, make internal processes more efficient, and take advantage of networks for sharing little-used materials. Research libraries, for example, under the pressure of rising acquisition costs for books and scholarly journals, have reduced acquisitions and begun to develop networks for sharing expensive but little-used items.32 At the same time that networks for cooperation and sharing have evolved, cultural institutions have been forced to compete with each other for limited funding from private foundations and public agencies in an economy that is increasingly hostile to the concept of public goods. Museums, galleries, zoos, aquaria, and other institutions of display have adopted market-oriented strategies such as charging admission, introducing museums shops and cafeterias as profit centers, and seeking corporate sponsorship for capital campaigns and high-profile exhibits. Ironically, museums that emerged from the Wunderkammern tradition of entertainment and

Epistemic Infrastructure in the Rise of the Knowledge Economy

123

display have once again begun to place great emphasis on exhibition and outreach. This has increased tension over the professional authority and autonomy of curators and librarians to control the content and presentation of their collections. Responding to the demand-based and pay-as-you-go economic model raises the specter of curators forced to produce exhibitions or librarians required to build collections that are popular and trendy rather than critical and thought provoking. These shifts have spurred intense debate over what constitutes transparency in the epistemic infrastructure and who has the right to question the legitimacy of the contents, organization, display, and interpretation of knowledge and cultural heritage. The very act of selecting what will be displayed, and how those displays will be presented, has become increasingly politically charged. This is seen with clarity in recent efforts by fundamentalist religious groups to prohibit science museums from displaying materials that contradict their notions of creation and by parents who wish to ban from school libraries books they do not agree with, but the issue is broader. Should holocaust museums focus only on the Nazi extermination of Jews, even though the Nazis also targeted the Romany, the mentally ill, homosexuals, socialists, and communists for extermination? Should local taxpayers insist that public libraries install filters to block pornography from young people when those same filters also erroneously block legitimate medical information?33 Sometimes everyone agrees on the appropriateness of having a given exhibition but disagrees over the way the exhibition is framed. In the mid-1990s, the Smithsonian Institution’s National Air and Space Museum planned an exhibit of the newly restored Enola Gay, the B-29 airplane that dropped the atomic bomb on Hiroshima, along with an exploration of the larger context of using nuclear weapons at the end of World War II. Curators were soon caught in a crossfire between those who wished to focus on the decision to use nuclear weapons and those who wished to focus on the heroism of the bomber crew. Tom Crouch, curator of the exhibit, remarked, ‘‘Do you want to do an exhibit intended to make veterans feel good, or do you want an exhibition that will lead our visitors to think about the consequences of the atomic bombing of Japan? Frankly, I don’t think we can do both.’’ Crouch was proved right: The museum could not do both, and the exhibit was constructed to make veterans feel good.34 Constructing narratives remains a perplexing challenge for epistemic infrastructure. A narrative is the means by which materials are used to create a story for the visitor that cannot emerge simply from the presence of the materials themselves. Going back to the skeleton of Iguanadon, that narrative was nothing more than the display itself in light of growing tension between strictly biblical accounts of creation and new interpretations of the geological record in early 19th century science. The presentation of the Iguanadon skeleton was itself a challenge: ‘‘Explain this.’’ If

124

Margaret Hedstrom and John Leslie King

Iguanadon was destroyed in Noah’s flood, what else failed to make it onto the Ark? The concreteness of the fossils threw the accepted narrative of the time into confusion and laid the groundwork for the controversy that surrounded Darwin’s theory of evolution in 1859. Museums still face that controversy 150 years later. The Enola Gay exhibition tried and failed to create a master narrative for multiple audiences, enlightening all and offending none. No one disagreed about the facts of the Enola Gay, the war, the bombing, and the aftermath. The argument was about what it all means. Epistemic infrastructure is increasingly pushed past the problem of knowledge and into the realm of political discourse, from the question of how things are to the question of how things ought to be.35 Epistemic Infrastructure in the Knowledge Economy Epistemic infrastructure is facing a period of deep uncertainty at the dawn of 21st century. Economic pressures, competition from alternative service providers, changing expectations from consumers and users, and a fundamental shift in their material base from tangible objects to digital representations are part of this uncertainty. There is evidence to suggest that significant shifts in the organization and economics of epistemic infrastructure are under way. One instance is the development of commercial alternatives to the services that traditional components of the epistemic infrastructure provide. The online bookseller Amazon.com has an extensive on-line database, not unlike a library catalog, listing several million book titles, 250,000 CD titles, and data about 250,000 motion pictures and entertainment programs from 1891 to the present. Amazon’s ‘‘collection’’ probably compares well with major research libraries in the number of titles available, and it is likely that more people search the Amazon databases on any given day than search the catalogs of any single research library, and possibly all research libraries combined. Moreover, Amazon competes with other on-line booksellers by adding services that mimic some of the selection and legitimating functions that libraries traditionally perform, such as alerts and recommendations based on user profiles and past purchases and opportunities for consumers to post and read book reviews. Another instance is illustrated by new enterprises that take advantage of potential complementarities between investments in public goods and commercial interests. Google Print is a case in point. The search engine company is working with several major libraries to digitize their print collections and make them available on-line to the extent possible under the current intellectual property regime.36 At no cost to the libraries, Google will invest millions of dollars in exchange for a digital version of their print content, which the company will use to draw more consumers to its site.

Epistemic Infrastructure in the Rise of the Knowledge Economy

125

Google clearly has a commercial stake in the project but also claims that its aim is to ‘‘help maintain the preeminence of books and libraries in our increasingly Internetcentric culture by making these information resources an integral part of the on-line experience. We hope to guide more users to their local libraries; to digital archives of some of the world’s greatest research institutions; and to out-of-print books.’’37 Google Print will point users to libraries in their vicinity that own the titles they are seeking and direct them to booksellers of new, used, and out-of-print books.38 Echoing the 18th century rise of national libraries that reinforced notions of national, cultural, and linguistic identity, the announcement of Google Print set off a firestorm of controversy in France and several other European countries about diversity and bias in the Google Print collection.39 There also are moves in the direction of making formerly private goods into public goods. The Open Source software movement has created an alternative to proprietary commercial products for developing and licensing software. Librarians have been working since the 1980s to develop and implement new models for scholarly publication and access to scholarly communications. By 2004 there were more than 700 open-access journals available in a wide variety of disciplines.40 Some government funding agencies are starting to require the authors of government-funded research to publish in open access journals so that the public at large can benefit from the results of public investments.41 ‘‘Knowledge conservancies,’’ such as the Creative Commons, are being established for owners of intellectual property to place their works voluntarily in the public domain in order to contribute to the larger social good. Finally, advocates for the next generation of research and learning environments are urging massive public investments in ‘‘cyber-infrastructure’’ that includes not only building bigger networks and faster computers but also developing the missing elements of the epistemic infrastructure in the form of digital repositories, powerful ontologies, and skilled persons to ensure persistence and coherence of the new digital collections.42 For most of the past two centuries, traditional epistemic infrastructure has operated as a public good in which a patronage structure of universities, governments, and nonprofit philanthropies provided funds and the institutions of libraries, archives, museums, galleries, botanical gardens, and so on carried out their missions as best they could. This arrangement never required robust means for judging the economic value of the infrastructure. Crude input/output models considered materials (papers, books, objects) inputs and the number of patrons served or the number of reference questions answered outputs. There was no cost accounting to show how invisible functions such as selection, organization, curation, and so on contributed to the services delivered or, more importantly, to the effect of those services in the economy.43 In a period of both skepticism about the need for public goods and a

126

Margaret Hedstrom and John Leslie King

reorganization of the provision of epistemic infrastructure, these models put traditional institutions at a disadvantage because they provide no way to determine the value of the actual work being done. The contributions of both individuals and institutions that have built and that maintain epistemic infrastructure are both invisible and taken for granted. They are not included in the balance sheet, and it is impossible to determine how any particular input leads to any changes in a user’s welfare. As long as the infrastructure was seen as an important public good to be funded at the appropriate level, this was fine, but new models of knowledge generation, dissemination, and exploitation call into question the value of the traditional infrastructure that has no useful measures of costs and benefits to fall back on. We are not arguing that the epistemic infrastructure of the industrial age could or should be mapped to the present environment. In fact, we contend that the established epistemic infrastructure and the institutions in which it is embedded have survived for millennia by finding the right balance between conservatism and innovation and by adapting to fundamental shifts in the production of knowledge. Those who care for the traditional infrastructure must decide whether and how to shape the next transformation while recognizing and convincing others that the market is not likely to provide all of the services required. Indeed, the new commercial enterprises, such as Amazon and Google Print, are deeply dependent on the products of traditional infrastructure. Amazon does not catalog the books, magazines, and CDs it sells. It uses the cataloging-in-publication infrastructure, which originated at the Library of Congress, in which quality control over cataloging and classification is managed by national libraries or by consortia such as OCLC, the Online Computer Library Center. Nor does Amazon maintain a warehouse of out-of-print or obsolete materials in anticipation of some potential demand long in the future; those services are provided by national and research libraries. The International Standard Book Number (ISBN) convention, arising from the traditional infrastructure, provides all booksellers with a powerful inventory control mechanism. Likewise, Google Print does not acquire simple bags of bits on which to unleash its powerful search engines when it scans the collections of research libraries. It also inherits the accumulated wisdom of generations of selectors who were knowledgeable about the needs of scholars and students in specific disciplines and well versed in the criteria for judging the quality of works. Web-enabled market forces and the traditional epistemic infrastructure are not inherently in competition with one another. In fact, they are complementary. The challenge is to find ways to exploit that complementarity that will also strengthen the knowledge economy. Four areas of complementary are worth noting: access, information quality and integration, social memory, and information property.

Epistemic Infrastructure in the Rise of the Knowledge Economy

127

Access The traditional epistemic infrastructure, together with the Web, can dramatically improve effective access to information for all strata of the population. Improvements in Information and Communications Technologies (ICTs) make it increasingly possible for people to participate in on-line discussions; seek information on healthcare, employment opportunities, or government benefits; use e-mail to communicate with friends and family; or post their own content on the Web. However, the technology alone is only part of a broader challenge that includes development of skills required to exploit the technology.44 The traditional epistemic infrastructure has long provided free access to materials that individuals cannot afford to purchase or that individual scholars could not collect with their own resources. Another complementarity is the collocation of physical collections and the information necessary to learn from them. The Web can make images of and textual information about objects in a museum’s collections available to remote users, and it can allow users of library, archival, and other collections to explore countless possibilities before traveling to consult particular resources. Preliminary reviews of on-line collections and commentary can dramatically narrow the search space and provide structure to the study when the individual can go to a culture institution in person. Provision of on-line information at varying levels of depth helps users gain knowledge tailored to their needs. The entire holdings of information on each object can be made available to the user at will, providing unparalleled access to collections for users on-site or anywhere. Information Quality Assurance The Web can permit easy access to information, but it is inadequate for teaching and research where definitive and high-quality information resources are instrumental for critical analysis, innovation, and new knowledge generation. Search engines index mainly the ‘‘surface’’ Web of unrestricted and static web pages. The ‘‘deep’’ Web remains unseen by most users, even though it contains as much as 500 times the information of the surface Web and is growing faster.45 Deep web resources, such as the curated collections in traditional institutions, also tend to be selected, indexed, and controlled for quality and authoritativeness by subject experts or editors. It is possible to put traditional epistemic infrastructure services on the Web, tied to large traditional collections and managed by experts who know the collections intimately, but doing so is not cost-free. One example is the Internet Public Library (http://www.ipl.org), which began in 1996 as a student experiment to create an online service providing features of a public library and has grown dramatically. The IPL is used heavily by students in elementary and secondary school for assistance

128

Margaret Hedstrom and John Leslie King

with their schoolwork, not only by providing on-line access to content selected for this audience but also through a global network of librarians and student volunteers who respond meaningfully to the kinds of reference questions that commercial sites such as Ask Jeeves could not begin to answer. One key to services that integrate traditional epistemic infrastructure and the Web is information quality: content authenticated by experts who know both the subject matter and the patterns of user demand. Social Memory A critical aspect of social memory is long-term preservation of vital knowledge, which requires both the mechanisms for preservation and the sophistication to know what to delete over time. Libraries, archives, and museums maintain collections over centuries and are the most important form of institutionalized long-term social memory. Even in the print world, the idea of ‘‘comprehensive collection’’ has largely disappeared: No library on Earth collects everything that is published, and it is virtually impossible to ascertain the fraction of total global titles held in all the libraries taken together. Large amounts of material disappear forever each year, even as librarians, archivists, and curators presume that other institutions are preserving those items. Contemporary social memory is created through the integration of widely distributed objects and collections using skills in organization and classification as well as understanding of the epistemic regimes that those systems of knowledge organization impose on the worlds they describe. Even in the world of physical evidence, there is no way to tell whether the material being lost is of longterm value. The situation is far worse on the Web, where loss of information is massive, routine, and undetectable. There are no established institutions to collect and preserve digital objects that are generated without consideration beyond first-order uses. Digital objects cannot be collected and organized the way physical books, documents, and objects have been for the past centuries, and it is not clear whether anyone is going to keep them at all. Libraries are reluctant to rely on publishers for archiving services because publishers have never been in the archiving business. Besides, they come and go. The University of Michigan Library, one of the participants in the Google Print project, will retain the physical copies of everything that Google digitizes and acquire a complete copy of the digital corpus. This decision acknowledges that Google has no commercial interest in long-term preservation. It further recognizes that the University of Michigan Library has been in existence for almost two centuries, while Google has been around less than a decade. With few tools to capture and preserve Web documents that are critical sources of information and important cultural artifacts, long-term preservation and social memory are endan-

Epistemic Infrastructure in the Rise of the Knowledge Economy

129

gered. This is one more area where the services of traditional epistemic infrastructure must not be taken for granted. Information Property Information property management is one of the most challenging but important areas of complementarity between the traditional epistemic infrastructure and the Web. Copyright, designed originally to provide authors with an inducement for creativity and generation of new knowledge, has expanded in scope and duration and limited the public’s alternatives to purchasing or leasing access to the increasing amount of content protect by copyright.46 The digital realm challenges this evolving tradition of copyright by making possible wholesale downloading, digital copying, and manipulation of digital objects to make exact replicas and create derivative works. Content providers are fighting back with encryption and digital rights management regimes that diminish the public domain and undermine the traditional ‘‘fair use’’ exceptions and practices upon which much of the epistemic infrastructure has relied. The neutral platform encouraging free flow of content that once characterized the Internet seems to be eroding, with potentially disastrous effects on highquality scholarship, academic freedom, collaboration, and creativity. Some argue that restrictions on digital content are becoming counterproductive to innovation and knowledge generation that copyright was supposed to encourage.47 Others argue that the ‘‘propertization’’ of knowledge is transforming key knowledge communities, such as universities, from ‘‘gift’’ economies to ‘‘market’’ economies, upsetting the basic model of academic operation of the past several hundred years.48 While the propertization of knowledge might produce new knowledge communities that rival universities in content and quality, this is by no means certain, and the potential for loss if the bargain goes bad is enormous. Alternative strategies for producing, evaluating, and distributing intellectual property on the Web are evolving, such as open source software, on-line preprint services, open access journals, and community knowledge projects such as Wikipedia (http://www.wikipedia.org). It remains to be seen whether these efforts will merely change the current parameters of information property or instead create fundamentally new means for valuing and distributing knowledge, but they are important for developing the complementarity between the Web-based market strategies and the strengths of the traditional epistemic infrastructure. Conclusion Two examples of contemporary problems readily illustrate society’s dependence on epistemic infrastructure. One is the prospect of global climate change. The empirical

130

Margaret Hedstrom and John Leslie King

basis for distinguishing variations in climate is deeply tied into long-standing epistemic infrastructure. No one deliberately collected data and specimens over the past centuries in anticipation of global climate change, but such data were nonetheless collected and conserved in libraries, archives, museums, zoos, and botanical gardens simply because they might be useful at some point.49 Bones, shells, fossils, and ice cores are now being calibrated with maps, ships’ logs, weather station reports, crop data, observations of bird migrations, and even personal diaries to develop a more complete picture of climate conditions over time. Without these seemingly mundane sources of information, there would be no factual basis to determine whether climate change was even occurring, much less to decide what to do about it. The other example involves the mapping of the human genome, which could lead to victory over myriad diseases. To use the genome data, it is necessary to distinguish strictly inherited diseases from those that have a genetic base but that must be triggered through environmental causes or diet, exercise, and other habits. The mapping of the genome created a whole new regime of intellectual property, including the astonishing case of a private firm patenting the entire genome of Iceland.50 The Icelandic genome is particularly valuable because of the isolation and relative stability of the population and because Iceland has deep epistemic infrastructure that makes the genome information useful. Icelanders have long kept family genealogies that can be used along with detailed medical records from the public health system to analyze the role of genetics in a wide range of diseases. The technical and scientific aspects of human genome mapping are justifiably seen as great accomplishments, and they are themselves part of the evolving epistemic infrastructure. Nevertheless, they are of relatively little practical value without the complementary utility of many other aspects of epistemic infrastructure that are typically overlooked. Epistemic infrastructure grew up around selection processes evolved by curators, librarians, and archivists to filter knowledge according to professional norms and standards, subject and domain knowledge, and attentiveness to the needs of user communities. This kind of systematic collecting builds trust in knowledge resources. A knowledge economy built on digital information will likewise depend on clear indicators of quality, authoritativeness, and authenticity. The lessons from the building of epistemic infrastructure in the 19th and early 20th centuries are powerful guides in this evolution. The knowledge economy will undoubtedly need new tools as it grows, but it already has a great deal of capacity and capability in the traditions of museums, archives, and libraries. Handled carefully, this traditional epistemic infrastructure will simultaneously build the value of knowledge in the society and decrease disparities between information ‘‘haves’’ and ‘‘have-nots’’ with respect to ability to acquire, evaluate, manipulate, and generate information. This infrastructure is modern society’s most vibrant and effective resource for dealing with ex-

Epistemic Infrastructure in the Rise of the Knowledge Economy

131

traordinarily challenging and conflicting demands. Those working at the forefront of the knowledge economy should recognize and strengthen it. Notes 1. The Center for Educational Research and Innovation (CERI) of the Organisation for Economic Cooperation and Development (OECD) sponsored this work as part of the project ‘‘Innovation in the Knowledge Economy: Implications for Education and Learning’’ (OECD 2004). An earlier version was published online by OECD (http://www.oecd.org/edu/km/ mappinginnovation). The authors gratefully acknowledge the assistance of Dominque Foray, Paul David, Michael Cohen, Paul Edwards, Bob Frost, and Brian Kahin. 2. Karin Knorr Cetina, Epistemic Cultures: How the Sciences Make Knowledge (Cambridge, MA: Harvard University Press, 1999). Our notion of epistemic infrastructure is related to the concept of epistemic cultures, but we use epistemic infrastructure on a macro level to discuss similarities and patterns across numerous environments and vast time spans. 3. Ernst Posner, Archives of the Ancient World (Cambridge, MA: Harvard University Press, 1972); Rosalind Thomas, Literacy and Orality in Ancient Greece (New York: Cambridge University Press, 1992); James P. Sickenger, Public Records and Archives in Ancient Greece (Chapel Hill: UNC Press, 1999); Lionel Casson, Libraries in the Ancient World (New Haven: Yale University Press, 2001); James J. O’Donnell, Avatars of the Word (Cambridge, MA: Harvard University Press, 1999); Ancient Archives and Archival Traditions: Concepts of Record-Keeping in the Ancient World (Cambridge: Oxford University Press, 2003). 4. See Peter Burke, A Social History of Knowledge (Cambridge: Polity Press, 2000), pp. 103– 106. 5. See Arthur MacGregor, Ark to Ashmolean: The Story of the Tradescants, Ashmole, and the Ashmolean Museum (Oxford: The Ashmolean Museum and the Tradescant Trust, 1997); Arthur MacGregor, The Ashmolean Museum: A Brief History of the Museum and Its Collections (London: Jonathan Horne Publications, 2001); C. H. Josten, Elias Ashmole, FRS (Oxford: Ashmolean Museum, 2000). Of special value is Prudence Leith-Ross, The John Tradescants: Gardeners to the Rose and Lily Queen (London: Peter Owen, 1984). 6. Patrick Mauries, Cabinets of Curiosities (London: Thames & Hudson, 2002), pp. 141– 145. 7. See essays contained in Oliver Impey and Arthur MacGregor, eds., The Origins of Museums: The Cabinet of Curiousities in Sixteenth- and Seventeenth-Century Europe (Oxford: Clarendon Press, 1985). Other sources of value include Krzysztof Pomian, Collectors and Curiousities: Paris and Venice, 1500–1800 (London: Polity Press, 1990); S. Greenblat, Marvelous Possessions: The Wonder of the New World (Chicago: Chicago University Press, 1991); A. Blair, The Theater of Nature: Jean Bodin and Renaissance Science (Princeton: Princeton University Press, 1997); Edward Miller, That Noble Cabinet: A History of the British Museum (London: Andrew Deutsch, 1973); Paula Findlen, Possessing Nature: Museums, Collecting, and Scientific Culture in Early Modern Italy (Berkeley: University of California Press, 1994). 8. Burke, p. 106.

132

Margaret Hedstrom and John Leslie King

9. Burke, Chap. 3, pp. 32–52. 10. Adalgisa Lugli, ‘‘Inquiry as Collection: The Athanasius Kircher Museum in Rome,’’ RES 12 (Autumn 1986): 109–124. This idea finds further expression in Lorraine Daston, ‘‘Curiosity in Modern Science,’’ Word and Image 11, no. 4 (1995); Lorraine Daston, ‘‘Marvelous Facts and Miraculous Evidence in Early Modern Europe,’’ Critical Inquiry 19 (1991): 93– 124; Lorraine Daston and K. Park, Wonders and the Order of Nature (New York: Zone Books, 1998); Lawrence Weschler, Mr. Wilson’s Cabinet of Wonder: Pronged Ants, Horned Humans, Mice on Toast, and Other Marvels of Jurassic Technology (New York: Pantheon, 1995). 11. No Cabinets of Curiosities have survived intact. What we know about their contents, organization, and aesthetics must be extracted from paintings and woodcuts that often served as frontispieces to their catalogs. See Mauries. 12. See Burke, Chap. 4, pp. 53–80. 13. Burke, pp. 79–80. 14. Elizabeth Eisenstein, The Printing Revolution in Early Modern Europe (Cambridge: Cambridge University Press, 1983); Jonathan M. Bloom, Paper before Print: The History and Impact of Paper in the Islamic World (New Haven, CT: Yale University Press, 2001). 15. David Diringer, The Hand-Produced Book (New York: Philosophical Library, 1953), pp. 275–335; Lucien Febvre and Henri-Jean Martin, The Coming of the Book, English ed. (London: Verso, 1976 [1997]); Mary Carruthers, The Book of Memory (Cambridge: Cambridge University Press, 1990); Ronald J. Diebert, Parchment, Printing and Hypermedia (New York: Columbia University Press, 1997). 16. Eisenstein, pp. 187–254. 17. Bruno Latour, ‘‘Visualization and Cognition: Thinking with Eyes and Hands,’’ Knowledge and Society: Studies in the Sociology of Culture Past and Present 6 (1986): 1–40. 18. Benedict Anderson, Imagined Communities (London: Verso, 1983); Diebert, pp. 104– 110. 19. Lara Moore, ‘‘Restoring Order: The Ecole des Chartes and the Organization of Archives and Libraries in France, 1820–1870.’’ PhD dissertation, Stanford University, UMI 3002025, March 2001. 20. Ann Laura Stoler, ‘‘Colonial Archives and the Arts of Governance,’’ Archival Science 2 (2002): 87–109; Burke, pp. 138–141. 21. Roger Chartier, The Order of Books (Stanford: Stanford University Press, 1994), pp. 62– 66; Burke, pp. 103–106. 22. Chartier, pp. 62–88. 23. Thomas Jefferson to Samuel H. Smith, September 21, 1814, Jefferson Papers, Library of Congress, as described in Cole, John Y. Jefferson’s Legacy: A Brief History of the Library of Congress (Washington, DC: Library of Congress, 1993). Available on-line at http://lcweb .loc.gov/loc/legacy/. 24. H. Helmes and D. S. Bland, eds., Gesta Grayorum: Or, the History of the High and Mighty Prince Henry, Prince of Purpoole, Anno Domini 1594 (Liverpool: Liverpool University Press, 1968). Available at http://fly.hiwaay.net/~paul/bacon/devices/gestaintro.html.

Epistemic Infrastructure in the Rise of the Knowledge Economy

133

25. Daniel Webster, Speech at the dedication of the Bunker Hill Monument, June 17, 1825, in Daniel Webster: ‘‘The Completest Man,’’ Shewmaker, Kenneth E., ed., (Hanover, NH: University Press of New England, 1990), pp. 99–104. Available at http://www.dartmouth .edu/~dwebster/speeches/bunker-hill.html. 26. Kevin M. Guthrie, The New-York Historical Society: Lessons from One Nonprofit’s Struggle for Survival (San Francisco: Jossey-Bass, 1996). 27. Abigail Van Slyck, Free to All: Carnegie Libraries and American Culture, 1890–1920 (Chicago: University of Chicago Press, 1995). 28. Maurice F. Tauber and Edith Wise, ‘‘Classification Systems,’’ in The State of the Library Art, ed. Ralph R. Shaw (New Brunswick, NJ: Rutgers University Graduate School of Library Service, 1961). 29. Kristen Patschke, ‘‘Melvil Dewey: The Father of Librarianship.’’ http://www .booktalking.net/books/dewey/. 30. Museums and Communities (Washington, DC: Smithsonian Institution, 1992). For an example of the consequence of a failure to maintain relevance to local communities, see Guthrie, The New-York Historical Society: Lessons from One Nonprofit’s Struggle for Survival. 31. Redmond Kathleen Molz and Phyllis Dain, Civic Space and Cyberspace (Cambridge: MIT Press, 1999). 32. Anthony M. Cummings, Marcia L. Witte, William G. Bowen, Laura O. Lazarus, and Richard Ekman, University Libraries and Scholarly Communications, A Study Prepared for the Andrew W. Mellon Foundation (New York: ARL and the Andrew W. Mellon Foundation, 1992). 33. Caroline Richardson, Paul Resnick, Derek Hansen, Holly Derry, and Vicky Rideout, ‘‘Does Pornography-Blocking Software Block Access to Health Information on the Internet?’’ Journal of the American Medical Association 288 (2002): 2887–2894. 34. Martin Harwitt, An Exhibit Denied: Lobbying the History of Enola Gay (New York: Copernicus, 1996). 35. For a discussion of ways to counter this trend, see David Carr, The Promise of Cultural Institutions (Walnut Creek, CA: AltaMira Press, 2003). 36. Brian Kladko, ‘‘Google Joins Effort to Put Millions of Books Online,’’ KRTBN KnightRidder Tribune Business News, The Record (Hackensack, NJ, January 18, 2005). 37. Google Library Project. http://print.google.com/googleprint/library.html (accessed June 29, 2005). 38. For an argument that Google is an entertainment company, see Dan Mitchell, ‘‘What Is Google, Anyway?’’ New York Times, Personal Business, What’s Online, Late Edition-Final (June 18, 2005): C5, Col. 1. 39. Elaine Ganley, ‘‘Europeans to Counter Google Print Project,’’ Associated Press Worldstream, International News (May 5, 2005). 40. Free Expression Policy Project, ‘‘The Information Commons: A Public Policy Report, Brennan Center for Justice, NYU School of Law, 2004. Available online at http://www .fepproject.org.

134

Margaret Hedstrom and John Leslie King

41. Business Wire, San Francisco, ‘‘The Public Library of Science Urges Researchers to Comply with the New National Institutes of Health’s Public Access Policy’’ (February 3, 2005). 42. National Science Foundation, ‘‘Report of the National Science Foundation Blue-Ribbon Advisory Panel on Cyberinfrastructure’’ (January 2003) http://www.nsf.gov/cise/sci/reports/ toc.jsp. 43. Denise A. Troll, ‘‘How and Why Are Libraries Changing?’’ p. 10. Draft White Paper for the Digital Library Federation, http://www.diglib.org/use/whitepaper.htm. 44. William A. Dutton, Society on the Line: Information Politics in the Digital Age (New York: Oxford University Press, 1999). 45. S. Lawrence and L. Giles, ‘‘Search Engines Fall Short,’’ Science 285 (1999): 295. 46. National Research Council, The Digital Dilemma (Washington, DC: National Academy Press, 2000). 47. Lawrence Lessig, The Future of Ideas (New York: Random House, 2001). 48. Corynne McSherry, Who Owns Academic Work? Battling for Control of Intellectual Property, pp. 74–76 (Cambridge: Harvard University Press, 2001). 49. Paul N. Edwards, ‘‘A Vast Machine: Standards as Social Technology,’’ Science 304 (May 7, 2004): 827–828. 50. Jocelyn Kaiser, ‘‘Population Databases Boom, from Iceland to the U.S.,’’ Science 298 (November 8, 2002): 1158–1161.

10 Universities and the Knowledge Economy Robin Cowan

Introduction The simplest rationale for the existence of a publicly funded university is that it provides some form of public good. If all the outputs of a university were privately owned, and privately appropriable, there would be no need for public funding. Either firms would fund the research and training that they could internalize, or students could fund the teaching through higher future earnings. Consequently, one way to pose the issue of the future role of universities is to ask what public goods they can provide that cannot be provided in other ways. There are many possible types of answers to this question, and different answers receive emphasis at different moments in history. During most of the modern period, though, we can observe one function of the universities that has been dominant and ongoing, lasting until some time in the 20th century. In The University in Ruins, Bill Readings coins the phrase ‘‘the university of culture.’’ Readings argues that the modern university system builds directly on the writings of Humbolt and the German idealists, who were very explicit about the social role of the university. The argument is that universities are uniquely placed to provide a sense of national culture. By studying and teaching the social and cultural history of a nation, this culture is extended through time. Where this is valuable, and how we can see this as a public good, is in the way it conditions the citizens of a nation. The teaching function of this university creates graduates who are all steeped in the same cultural tradition, at least within one nation state. They have a common view regarding their social and cultural roots; they share a world view; if the university is working well, the graduates fit well into the existing society and can further its aims in the future. All of this has clear advantages socially, administratively, and economically. In this tradition, of course, the humanities take a central place in the university structure, since they are the repository of social culture that is so important in this sort of cultural extension.

136

Robin Cowan

Historically, this has been extremely important, and is very apparent in the German case. To create a single nation out of a collection of provinces, it is necessary to create a national identity to which all citizens can relate. Without this common view, or social bond, there is no reason that citizens of one province will feel any obligation or allegiance to citizens of any other. Without this, a nation cannot exist. To this end, a ‘‘tradition’’ must be created, or found, and expounded, and to the extent that all students are taught or exposed to the same tradition, an identity will emerge. If this identity is strong, graduates will be ‘‘good citizens,’’ able to function well, achieve their ends, and still contribute effectively to the greater good. In the European context, within a nation (with some obvious exceptions) by and large this tradition has been a tradition of a literature, written in a particular language. Literature is social commentary, description, or analysis, and the literature of one language provides the basis of description of a single culture.1 One can see a parallel here, as Readings points out, between the university of culture and a national airline. Until recently, Air France has probably been the best remaining example. Why does Air France still exist? Definitely not because it has produced great financial profits for its owners.2 Air France still exists (as is the case with so many French institutions) as an extension of French culture. Externally, it shows French technological prowess, French styles of service, French cuisine (perhaps). It is a means of projecting the French presence, and here this must mean cultural presence, since as an economic endeavor it cannot be said to be an unqualified success. Within France, it plays a closely related role. ‘‘Projecting’’ is not quite the right word in the internal context, but perhaps ‘‘advertising’’ is. Air France serves to remind the French citizen of French technological prowess and so on, as it does the outside world. But in addition, it serves as a vital means by which all French citizens are linked to all others. It is a tool of cohesion. Flying unprofitable internal routes does not pay for Air France the company, but it does pay for Air France the promoter of French culture. And here is the public good that has been considered so valuable. The value this provides is to bind the French together (in a very oldfashioned way), which provides benefits throughout the society and polity and, it is crucial to point out, to other French industries. On the latter point, consider the ‘‘irrational’’ loyalty the French have to their wine industry. (Though to be fair, one does occasionally hear a French citizen admit that there might be one or two Italian wines worth drinking.) And because this public good is considered so valuable, Air France has been protected for the economic sphere. Just as, historically, the universities have been. The university of culture did provide economic benefits, through creating a cadre of administrators who shared a coherent view of the world and so were able to create policy with at least an overarching coherence, and it created a cadre of managers

Universities and the Knowledge Economy

137

whose view fit well with that of the policy makers. The most explicit exemplar here are the French (again) Grandes Ecoles, which have provided generation after generation of administrators and managers who all had a common goal (the furtherance of French prestige, roughly speaking) and common views both about the value of that goal and about how to achieve it in the broadest senses. But this is all clearly a public good. It is a benefit that is not appropriable by any single agent or even group of agents. Consequently, the university has been publicly funded and protected from the economic sphere. It is very important to notice one other thing in this respect, which is that while the state was funding the university, it was the university that was defining the nation or state, not vice versa. The university itself studied, and thereby defined the culture that was being promulgated, but it also determined how that promulgation took place. Those funding the university had a minimal role in directing it. This independence between the funding and the details of the activities has been extremely important for the system as a whole. This is a description of the universities until sometime in the middle of the 20th century. But since then, the university of culture has been disintegrating. The role of the university is no longer to provide the next generation of good citizens. Part of the reason is that this role builds and is built upon a relatively strong presence of nation states. But as internationalization proceeds apace, the place of the nation state recedes. We see this economically in the rise of the transnational corporations, which have become of the size of nations, and in international currency flows which are of the size of the GDP of nations. We see this socially through large international migrations. And we see this in the universities, particularly obviously in the ‘‘culture wars,’’ in which violent arguments, particularly in the English-speaking world, erupted about the definition of the culture we are studying. Fights over tradition and canon were essentially about the definition of ‘‘the culture,’’ and the existence of these fights indicates that the existence of a culture that defines a nation can no longer be taken for granted. The university of culture is no more. There is an irony worth pointing out. One of the effects of the university of culture was to facilitate and even drive a growth in national cohesion, which was particularly visible in Germany. Different regions within a nation, by exposure to the common cultural tradition that was developed and promulgated through the national university system, slowly converged, at least in their views of the cultural traditions. This opened the possibility for, and indeed helped to create a form of, national ‘‘social cohesion.’’ That phrase, ‘‘social cohesion,’’ is used deliberately, and anyone familiar with European Commission Framework Programme documents, and even some of their tender documents, will recognize it. One of the big issues within the Commission is creating cohesion across the different regions in Europe. As part of the published evaluation criteria for Framework proposals, the

138

Robin Cowan

contribution of the proposed project to social cohesion looms large. Just when potentially a new role, in Europe at least, for the university of culture emerges, the university of culture disappears. So if the old role for the university has gone, and old university structures are crumbling, what exists now, or what will rise to take its place? At this point I depart from Bill Readings and pursue a different tack. Readings’s view, which seems quite perceptive to me, is that we are now living in the University of Excellence, and he has insightful things to say about what that means.3 But I am going to turn to innovation and innovation systems. The Linear Model Abdux In the bad old days innovation was viewed as a linear process and was described using ‘‘the linear model.’’4 The general idea was that basic R&D provides the foundational knowledge for applied R&D, which provides the foundational knowledge for innovation, which then becomes a good to be diffused to users. The process is linear, with one stage feeding the next, and it is unidirectional. In this model, the role of universities was clear. It was to do the basic R&D, thereby providing the foundational knowledge, information, data, instrumentation, and so on, on which the entire rest of the innovation edifice is built. In terms of public goods, the role is also clear. Knowledge is a public good (Arrow 1962; Nelson 1959), so basic R&D, being the most widely applicable of all types of knowledge, will be severely undersupplied by the market. Problems of extensibility and appropriation will deter firms from producing this type of knowledge. Given its importance to the entire system, it had better be provided publicly. This is a very nice, simple model, in which it is relatively easy to see the roles played by different actors and institutions, and it has clear implications for research funding. As we proceed from basic R&D to diffusion, the outputs become more and more appropriable, so there is less and less need for public subsidy. While the model has these desirable properties, it was also seen as having one or two problematic aspects. First, it was difficult to find examples in which ‘‘pure’’ basic research led directly to products (see for example Rosenberg 1982, Chapter 7), though nylon and the laser are two favorites. Second, it was easy to find examples where most of the innovations in an industry were made by the users, involving very little, if any, science at all (see von Hippel 1976, 1977, 1986). The first observation makes us wonder what basic R&D is really doing in the context of an innovation system. The second observation implies that in our model we need a feedback, a way for users to be putting information into the process, rather than just using the outputs of it. So to our simple structure we should add a link, or more than one, in

Universities and the Knowledge Economy

139

which users are connected to other players, providing them with information or knowledge inputs for their activities. But once you start adding feedbacks to a model, any model, the temptation to add ‘‘just one more’’ is hard to resist. And soon emerged the system model of innovation. The system model of innovation, sometimes known as the National Innovation Systems model, emphasizes that there are many different types of actors and institutions that contribute information or knowledge to the innovation process, and that information flows in many directions between many different types of agents. In this model, many agents matter, and they matter in many ways. But if we say, with just a little bit of exaggeration, that when thinking about innovation, ‘‘everything matters,’’ it is quite natural to ask, ‘‘Does anything matter more than anything else?’’ or ‘‘Does everything, as input, matter for everything, as output?’’ One could argue that the system model of innovation has seriously overcorrected the faults of the linear model. Be that as it may, we can still sensibly ask, if universities are part of the innovation system, what is their role within it? There is an obvious and uncontroversial answer, namely that the universities provide highly trained graduates. (If this is all they do, then probably they should be transformed into vocational schools.) But if universities are linked in many ways to many different entities in the system, surely they contribute in other ways as well. The natural answer in its general form is clearly that universities are a source of knowledge and information (after all, those are the main products of research) on which other entities can build market-valued goods and services. And in this regard the systems model is very similar to the linear model. Where it differs in a very important way is that in the system model, the university can take in knowledge and information from many different sources. In the original linear model, by contrast, it appeared as if knowledge was generated in the universities without any inputs from the outside. In this regard, the systems model was a much better representation of university research (of which more below). But pushing hard the line that universities are a source of knowledge and information for other actors, the justification of the university system becomes the applicable knowledge it feeds to innovators. In recent years, there has been considerable interest in what are called the sciencebased industries. Biotechnology is the most prominent example. The claim is that innovation and output in these industries depends heavily on very recent advances in (university) basic research, and thus they provide the paradigm example of how universities contribute to the innovation system. We see concern in policy-making circles, particularly in Europe, to extend this vision to other industries as well.5 While this vision may apply to some industries,6 it is important not to let the biotech tail wag the innovation system dog.

140

Robin Cowan

Asking this question, and asserting as part of the answer that universities have a crucial role, invites difficulties for the justification of public support for the university. The reason is that by making a claim that universities are important players in a system of innovation, we are implicitly making the claim that the university has an economic role, and thus its support can be justified by economic arguments. By emphasizing that universities play this role, we move away from the cultural or social justification of the university, arguing that it makes a strong contribution to wealth generation, and over time, the contribution seems to be, in the arguments at least, more and more direct. We are inviting the removal of the protection of universities from the economic sphere. And indeed, we are getting what we asked for.7 The Bayh-Dole Act is a perfect manifestation of the view that universities can contribute very directly to wealth generation. The idea that there are lots of patentable, and therefore marketable and economically valuable, ideas lying around on lab benches makes the university contribution to wealth creation almost as direct as it is possible to be. As central players in the innovation system universities provide inventions (which they should patent, as a means of improving technology transfer from university to industry), they provide instrumentation, which is necessary for doing further, and probably more applied, research, and they provide skilled labor, which industry can employ in its own innovation activities. All of these are relatively direct contributions to the economy in general and to innovation in particular. The justification of the university here is as a creator of wealth, if indirectly. But where is the public good? Technology transfer via the patent system is by definition not a public good. The public good aspect remains the conventional one: the outputs of teaching and basic research. The Linear Model Redux I want to propose that perhaps there is a simpler change to the linear model which, while retaining its simplicity, addresses the two problems mentioned above and will lead to a slightly different view of the university’s role.8 It would remain a very stark model, missing many features of an innovation system, but in its starkness and simplicity, it does permit a certain focus, which can be difficult to obtain with more complex models. What I want to propose is simply moving basic R&D to the end, rather than the beginning, of the process and then adding a feedback, from the end to the beginning. The ivory tower is not a windowless edifice. Scientists do look out their windows from time to time, and what they see affects what they consider to be interesting problems. Peter Galison (2003) discusses the development of Einstein’s theories of

Universities and the Knowledge Economy

141

relativity and his concern with time and how we could know that events located at two distant points in space happen at the same time. Galison argues that it is no coincidence that Einstein was working on this at the moment at which railroads and cities (in Switzerland, and in Bern, near the patent office, in particular) were developing technologies to coordinate their clocks. It is also no coincidence that Poincare´ was working on virtually the same problem, independently of Einstein, just a few years earlier. He was, after all, in the Paris Bureau of Longitude, for which one of the major challenges of the time was to find a way for sailors to determine Paris time from anywhere in the world so that they could effectively measure the longitude at their own location. Poincare´ did not get to relativity, but his central concern was the same as Einstein’s initial concern, namely, what ‘‘simultaneity’’ means. Here we see an example of some extremely theoretical basic research that was closely connected, in inspiration at the least, to developments in technology. This position more generally, that science is not isolated from the ‘‘real world,’’ in a windowless ivory tower, is central in the recent sociology of science (see Barnes et al. 1996 for an example for a recent text). While it is true that the fundamental quest for science is the truth, there are very many possible true statements or facts that could be pursued.9 Why this one rather than that one? How do scientists decide what to investigate? In part at least, the answer is driven by the things they observe in the world, whether they be technological developments, new industrial processes, or economic phenomena. Rosenberg (1982, especially Chapter 7) argues that a typical pattern is that basic research follows technological or applied developments. The relationship between science and technology is complex (certainly more so than my simple revision of the linear model would allow), but it is common that basic research serves to provide the theoretical explanations of things that are already being used in practice. Phenomena observed in practice, whether puzzling or just interesting, prompt scientists to look for the underlying physical processes that explain them.10 Sometimes the phenomena are existing technological successes which can be understood at a more general level; sometimes they are existing technological problems in the solving of which basic results are produced. In either case, though, the direction of scientific or basic research is strongly affected by existing technological products. A big part of basic research picks up phenomena in the outside world, sometimes natural, often man-made, and tries to explain them by asking about their fundamental underlying principles. This seems like a rather pessimistic view of university research. University scientists here lag behind the ‘‘real world,’’ instead of leading it, and even so, they continue to engage in navel gazing. Now their connection to the outside world is purely that it serves as a source of navel lint.

142

Robin Cowan

But this interpretation is overly pessimistic and misses one very important aspect. A senior manager in charge of university relations at a major computer firm stated in conversation that his main goal in having professors visit his firm was to try to raise their intellectual curiosity about the types of developments the firm was involved in. He was explicitly not trying to get them to solve his technical problems; he was trying to steer their (basic) research interests. The value of the involvement of academics in a firm is not in immediate product development; rather, it lies in the future. When a scientist does basic research, he or she is observing some phenomenon and making a model of it. With a model, it is possible to extract the important, central features and causal connections driving the phenomenon and to understand ‘‘how it works.’’ When the phenomenon is technological, as in several of Rosenberg’s examples, it can involve the tacit knowledge of the practitioner. The process of basic research can be seen, by extension, as tantamount to a process of codification.11 A phenomenon or process is described in a compact, reproducible way in which its fundamental aspects are captured and secondary or unimportant features are set aside. In a completely new field codification involves developing a vocabulary, models, and a language in which the phenomena can be described. Depending on whether the phenomenon fits into an existing field or is entirely new, different activities receive more or less attention in this process, but all of them are central to the ability to teach the principles that are discovered. What is being codified, typically, in the research process is a causal structure that explains some phenomenon. A phenomenon is typically broken into several pieces, and then the pieces are linked. The research activity must decide first what to consider as ‘‘a piece.’’ There are a huge number of ways in which any thing can be partitioned, and the partitioning into constituent components is utterly central to the enterprise. Then, of course, the issue is how those pieces fit together, causally, to explain the observed effect.12 The reason I emphasize this has to do with causation. This is a tricky idea and many attempts, beginning with Aristotle, have been made to provide a theory of it. One such attempt seeks to connect the idea of causation with the idea of ‘‘control’’ or manipulability.13 That is, understanding the causes of a phenomenon implies understanding in principle how to control it. While this remains a view that has some difficulties to be addressed, the central idea indicates how basic research can be a powerful input into innovation. Innovators are centrally interested in how to control processes. It is newfound control over phenomena that permits them to produce new products and processes. So causal models are indeed inputs to the innovation process but, as Rosenberg and von Hippel would argue, not as direct as this paragraph seems to suggest.

Universities and the Knowledge Economy

143

The route I will suggest comes through teaching. Once a phenomenon is understood, once a causal model that explains it exists and has been codified, it can be taught. The role of basic research is as an input to teaching. It permits universities to create graduates who understand, at a relatively deep level, technologies, phenomena, processes that have already been observed or created. That is, today’s graduates understand yesterday’s innovations. And they understand them not only in the purely pragmatic way that interests the entrepreneur (if I do X, there is an effect Y, which can be translated into higher productivity or a better product), but they also understand why X produces Y. They are not taught, and this is important to acknowledge, how to turn Y into higher profits. This is an extremely important difference between an entrepreneur and a scientist. Scientists are attracted or driven by ‘‘interesting phenomena’’ and not by ‘‘interesting market opportunities.’’14 This observation of the difference between academics and entrepreneurs constitutes a strong reason for skepticism regarding the view that universities can be a source of directly applicable knowledge for innovators. The fact that today’s graduates understand yesterday’s innovations is not the problem it sounds, since today’s graduates, or at least those of them who escape the university, will be tomorrow’s innovators. Their innovations will build on their knowledge and their understanding of how things work. This is the reason it may be more apposite to speak of highly qualified rather than highly trained personnel, since the latter implies a strong degree of vocational skills while the former connotes a heavier emphasis on general skills. The ability to be a successful innovator depends on levels of general understanding, that is, the understanding that comes from the outputs of basic research. But again it is worth emphasizing that the results students are being taught are not of the ‘‘with a little refinement this theorem will become a product’’ variety. It is far more general than that. The conclusion here is disappointingly conventional. The role of universities is to couple research and teaching. The output is highly qualified personnel. The human capital embodied in graduates is not highly specific and is in fact general enough that it constitutes a public good, or at least, without labor contracts that are tantamount to indentured servitude, a nonappropriable good. Innovation Ignored Is there something beyond this disappointingly conventional conclusion about the place of the university? Does the university do more than simply analyze today’s phenomena and teach the results of that analysis to tomorrow’s innovators? I believe the answer is yes, or at least has been yes, but that the recent emphasis given to the university as a player in the innovation system threatens that activity.

144

Robin Cowan

The activity I refer to is ‘‘reflection.’’ The university is the only place in modern society in which nonteleological reflection is institutionalized. Part of the role of the university has been to provide a location in which members of the society could reflect on what the society is doing, to discuss any issue thought pressing, without reference to any outside constraints or goals. It provided a place for thought that was insulated both from the political process and from the market. It was a place in which the phrase ‘‘let’s stop and think’’ was a sort of trump card. Any issue can be worthy of careful, deliberate, unrushed discussion and analysis, and the university was a place in which this activity was revered. This idea, that somewhere someone should be able to resist any pressures to treat an issue either quickly or with a particular frame of reference, or even with a particular outcome in mind, was considered a fundamental part of a well-functioning society and was built into the university ethos. What goes hand in hand with this is a concern for truth. In the university setting, in principle, the truth trumps everything. When faced with a choice between action motivated by truth, either seeking it or acting upon it, and action driven by some other motive, the former is lexicographically preferred. Obviously, people can be mistaken about the truth, believing things that are not in fact true. But within the university setting, this is not an issue. Pursuit of, and statements about beliefs of, truth are meant to be the dominant currency. There is no other institution in society in which this is the case.15 And it seems patently obvious to me, at least, that this is a valuable function. But the insertion of the university into the innovation system, and thereby into the market sphere, is seriously threatening this role. It introduces a second, and growingly important, criterion on desideratum for action, namely the economic effects or value of it. If a university is passing things to the ‘‘outside world,’’ it must be aware of what its ‘‘clients’’ want and at some level deliver it to them. This has always been true to some extent, even in Readings’s University of Culture, since there the universities were delivering a social cohesion which the state (funding the universities) wanted. But notice that this is a demand specified in very general terms, and in fact the universities determined the details of how and what was delivered. When the demands becomes more specific, clients have a louder voice in the what and how. The pressure of the market means that even that most ‘‘ethereal’’ of departments, namely philosophy, is now offering courses to appeal to the market, for example in medical ethics. One could make the case that this is not necessarily bad and that many people outside academe now need to be able to think carefully and effectively about such issues, so this pressure on philosophy departments could constitute a very important contribution to society. But there are other cases that cannot be

Universities and the Knowledge Economy

145

explained away like this. They are emblematic, it seems, of the thought that the market has no interest whatsoever, except possibly instrumentally, in the truth. Two recent high publicity cases at the University of Toronto make the point. In the year 2000, David Healy was appointed as the Professor of Psychiatry and Head of the Mood and Anxiety Disorders Program within CAMH (Center for Addiction and Mental Health). The offer was formally made and accepted. In August, he gave a lecture in Toronto in which he discussed, among other things, the idea that serotonin reuptake inhibitors could contribute to suicidal tendencies. This was not a new idea of his; he had published papers and lectured on it many times in the past. Within days, his appointment was rescinded. Of course, no one knows exactly why this was done, so it may just be a coincidence that half16 of the funding of the mood disorders program at CAMH came from pharmaceutical companies.17 The second case in Toronto involved Dr. Nancy Olivieri, former head of the Sick Children’s Hospital’s hemoglobinopathy program. Her research on the drug deferiprone showed up unexpected side effects, so she decided to break a confidentiality agreement with Apotex, the Toronto pharmaceutical company that was sponsoring her research, tell her patients, and publish the results in the New England Journal of Medicine (Olivieri et al. 1998). She was subsequently threatened with legal action by Apotex and, even more astonishing, removed from her hospital post by the university. Naturally, this caused quite a stir, but the report on the affair observed, among other things, that this lack of independence of researchers was no longer uncommon. In other (quite ugly) words, industry sponsors of research appear now to be attempting to buy the results they want, or at least suppress the results they don’t want.18 And in doing so they seem to be able to dictate university policy.19 The ability to put truth at the head of the list of motivations and the ability to pause and reflect are both seriously threatened by the universities’ having been pushed toward the market. What it points to is a very difficult issue. In order to protect these aspects of university culture, it must be the case that research (and indeed other university activity) is to a very great extent independent of funding sources. Of course this makes justification of university activity extremely difficult, since funders naturally ask, ‘‘What are we getting for our money?’’ and demand, ‘‘Show that our money is not being wasted.’’ If funders feel they should have a loud voice in determining university activity, it is naturally quite easy for them to see waste in activities that do not directly further their aims. If funding agencies, whether public or private, are assured that universities will make great contributions to an innovation system, and funding is based on this argument, then activities that do not respond directly to that goal are waste. Activities that have the nature I have been describing in this section will almost automatically fall in this category.

146

Robin Cowan

Conclusion I stated at the outset that universities have always provided some public good, and this was how they justified public support. On the model described by Humbolt, which survived through most of the modern period, universities provided a steady stream of ‘‘good citizens.’’ Their value showed in their ability to administer the nation as a nation, their ability to integrate with existing enterprise to promote the national economy, and their ability, and indeed affinity, to support and further its cultural identity. Social cohesion was an important public effect. In the early 20th century, universities provided a large cadre of highly educated labor. On the back of this labor tremendous economic growth and social change was built. Again it was an output that due to its nature was nonappropriable. What is the public good of the future? I have argued above that again it is a body of highly qualified personnel. Again, as graduates they are not ‘‘trained’’ as they would be in a vocational school but rather educated in the broad sense. Vocational training can in principle be made a club good, but education remains a public good due to labor mobility. Adopting this model or view, especially in the European context, we open the door to the redistribution of human capital. If the labor mobility is across borders, so too is the redistribution. This will further break down the national identities that were such an important part of the German university model. On the other hand, it provides a route to create the Holy Grail of the EU Framework programs, and possibly of the European project itself, namely ‘‘social cohesion.’’20 But notice that this does not follow the model of the University of Culture, in that the goal here is not to find some European literary tradition, or on the American model a literary canon, through which to create and then pass on a cultural tradition. One suspects that the cultures are too diverse. This suggests a shift from the humanities to the social sciences as the intellectual center of the university. Here it may be possible to find more, broader-based cohesion. In terms of ‘‘channeling knowledge’’ my argument seems to come to the following conclusions. The systems model was right: Knowledge does flow into the university research environment. It does so by scientists observing interesting phenomena and asking how they work. Industry can provide a lot of this observation, and in doing so interest the university scientist in phenomena that industry perhaps uses but does not understand. But this is definitely not a new way of saying that universities can solve problems that industry is having. Rather, universities are codifying things that industry either does not understand or only knows tacitly. Here is the public good, which does serve to help the innovation system, though in a very general way: The university produces basic, public knowledge, and a stream of graduates who understand it.

Universities and the Knowledge Economy

147

Where does this leave reflection and truth? On this model of the university’s role, they are restored to, or perhaps supported in, their rightful place. Because industry, or the external innovation system in general, is not a ‘‘client’’ of the university, the latter does not need to produce something that industry wants to buy. The question ‘‘What is the market value of this activity?’’ again recedes in the university sphere and can be replaced by ‘‘Is it true? Let’s stop and think.’’ Acknowledgments I wish to acknowledge many useful conversations with William Cowan during the preparation of this paper. The contributions of the Blueland Library should also be acknowledged. Notes 1. Philosophy is the other discipline that looms large in this university, as it is a discipline that explicitly examines ‘‘how to live,’’ which is clearly a central question in the ‘‘soul’’ of a culture. 2. The situation of Air France has changed since 1993, when it received a 2 billion pound subsidy from the French government: It no longer enjoys this sort of protection. 3. One of the amusing things that Readings points out is that if you look at the public web pages of universities you discover that they all describe themselves as unique. Interestingly, they are all unique in exactly the same way: They are excellent. What it means to be excellent exactly is anyone’s guess and is pretty much up for grabs. 4. The history that follows here, while not a fiction, is meant to be rhetorical, rather than a detailed account of the history of thought in the economics of innovation. For that, see for example Rothwell (1992). 5. For evidence of just how much this is troubling EU policy makers, see the proceedings of the conference The Europe of Knowledge 2020: A Vision for University-Based Research and Innovation, Liege, April 2004, in which an ever-present theme is how the university system will support innovation in various industries and sectors in the coming decades. http://europa .eu.int/comm/research/conferences/2004/univ/index_en.html 6. Though one is tempted to ask why biotech is not able to do the research in-house, merely by hiring professors at large multiples of their academic salaries. Could it be that what biotech is buying is not knowledge or research but rather ‘‘independence’’ with which to impress regulators? But perhaps such cynicism is best left undeveloped. 7. This rhetoric is certainly unfair, in that it reads as if the purveyors of the systems model of innovation must take the credit or blame for the ongoing push toward the market. It is quite possible that this is a case rather of seeing which way the wind is blowing and bending with it.

148

Robin Cowan

8. I don’t mean, here, to suggest that the system model is wrong. There are typically many ‘‘right’’ models of any given phenomenon. Models are made, not found, which means that given that several right models can be constructed, the interesting issue becomes which one is most useful in the circumstances. 9. One can remain agnostic on the debate between realism and instrumentalism while making this claim, simply observing that even if ‘‘truth’’ is defined differently under the two views of the scientific enterprise (and even if some instrumentalists avoid the word ‘‘truth’’ altogether), searching for it remains an objective regardless of the outcome of this debate. On these issues see Putnam (1990) for example. 10. Some of Rosenberg’s examples: short-wave radio transmission and properties of the ionosphere; short circuits and whisker crystals; the properties of Bessemer and postBessemer–produced metals and metallurgy; aluminum alloys and ‘‘age hardening’’; petroleum cracking and the general effects of heat on hydrocarbons; semiconductor use and semiconductor theory. 11. See Cowan and Foray (1997) or Cowan et al. (2000) on the economics and process of codification. 12. Obviously, the partitioning and causal connection building are not independent activities. 13. See for example Collingwood (1940), Gasking (1955), or Menzies and Price (1993), or for a version less dependent on human intervention, Pearl (2000). 14. This can be interpreted as one of the reasons why the technology-push model was replaced by the demand-pull model, to use Rothwell’s (1992) terminology. 15. One might make a claim that the monasteries perform or did perform this function. As such, they bear similarities to universities: One could argue that monastaries were the preserve of those who glorify the god Yahweh, whereas universities are the preserve of those who glorify their own god, namely, consistency. 16. As reported by the Canadian Broadcasting Corporation. 17. In a similar, earlier case, Healy had written an article published by the Hastings Institute, in which he warned of possible problems with SSRIs and produced evidence that an increasing proportion of the therapeutics literature was ghost-written (Healy 2000). It turned out that Lilly was one of the biggest private funders of the Hastings Center, and following this article they withdrew their support. For more on these cases see http://www.healyprozac .com/AcademicFreedom/default.htm. 18. Olivieri was eventually reinstated to her post and otherwise vindicated (Spurgeon 1999). 19. It may only be coincidence that both of these cases involve medicine, and in particular pharmaceutical research. But if not, this may be another reason for fearing the biotech tail of the innovation dog. 20. The Commission does seem to be aware of this. It is actively trying to foment social cohesion through universities and research institutes. This is most obvious at the level of the (senior) researcher, through Framework Programme evaluation procedures, but it exists as well at the level of graduate students and postdocs through various Marie Curie activities. It is very much weaker at the undergraduate level, which is probably the most important level of the three. But perhaps the Bologna Accord is a step in that direction.

Universities and the Knowledge Economy

149

References Arrow, K. J. (1962). ‘‘Economic Welfare and the Allocation of Resources for Innovation.’’ In The Rate and Direction of Technical Change, R. Nelson, editor. New York: National Bureau of Economic Research. Barnes, B., D. Bloor, and J. Henry (1996). Scientific Knowledge: A Sociological Analysis. Chicago: Chicago University Press. Collingwood, R. (1940). An Essay on Metaphysics. Oxford: Clarendon Press. Cowan, R., and D. Foray (1997). ‘‘The Economics of Codification and the Diffusion of Knowledge.’’ Industrial and Corporate Change, vol. 6: 595–622. ———, P. David, and D. Foray (2000). ‘‘The Explicit Economics of Knowledge Codification and Tacitness.’’ Industrial and Corporate Change 9(2): 211–253. Galison, P. (2003). Einstein’s Clocks, Poincare´’s Maps: Empires of Time. New York: Norton. Gasking, D. (1955). ‘‘Causation and Recipes.’’ Mind 64: 479–487. Healy, D. (2000). ‘‘Good Science or Good Business?’’ Hastings Center Report 30: 19–22. Menzies, P., and H. Price (1993). ‘‘Causation as a Secondary Quality.’’ British Journal for the Philosophy of Science 44: 187–203. Nelson, R. (1959). ‘‘The Simple Economics of Basic Research.’’ Journal of Political Economy 67: 297–306. Olivieri, N. F., G. M. Brittenham, C. E. McLaren, D. M. Templeton, R. G. Cameron, R. A. McClelland, A. D. Burt, and K. A. Fleming (1998). ‘‘Long-Term Safety and Effectiveness of Iron-Chelation Therapy with Deferiprone for Thalassemia Major.’’ New England Journal of Medicine 339(7): 417–423. Pearl, J. (2000). Causality. New York: Cambridge University Press. Putnam, H. (1990). Realism with a Human Face. Cambridge, MA: Harvard University Press. Readings, B. (1996). The University in Ruins. Cambridge, MA: Harvard University Press. Rosenberg, N. (1982). Inside the Black Box: Technology and Economics. Cambridge, MA: Cambridge University Press. Rothwell, R. (1992). ‘‘Successful Industrial Innovation: Critical Factors for the 1990s.’’ R&D Management 22(3). Spurgeon, David (1999). ‘‘Toronto Research Funding Dispute.’’ British Medical Journal 318: 351. von Hippel, E. (1976). ‘‘The Dominant Role of Users in the Scientific Instrument Innovation Process.’’ Research Policy 5(3): 212–239. ——— (1977). ‘‘The Dominant Role of the User in Semiconductor and Electronic Subassembly Process Innovation.’’ IEEE Transactions on Engineering Management EM-24(2): 60–71. ——— (1986). ‘‘Lead Users: A Source of Novel Product Concepts.’’ Management Science 32: 791–806.

11 The Impact of ICT on Tertiary Education: Advances and Promises Kurt Larsen and Ste´phan Vincent-Lancrin

Introduction Knowledge and innovation as well as information and communication technology (ICT) have had strong repercussions on many economic sectors, such as health care, finance, and transportation (Foray 2004; Boyer 2002). What about education? The emphasis placed on knowledge as a crucial driver of economic development has set a new scene and new challenges for the education sector. First, education is a prerequisite of the knowledge-based economy: The production and the use of new knowledge both require a more (lifelong) educated population and workforce. Second, ICT is a very powerful tool for diffusing knowledge and information, a fundamental aspect of the education process. In this sense, it can play a pedagogic role that could in principle complement (or even compete with) the traditional practices of the education sector. The challenge for the sector is to continue to expand, with the help of (or under the pressure of) new forms of learning. Third, ICT sometimes induces innovative practices: For example, navigation does not involve the same cognitive processes since the Global Positioning System (GPS) was invented (e.g., Hutchins 1995); scientific research in many fields has also been revolutionized by the new possibilities offered by ICT, including digitization of information and new recording, simulation, and data processing possibilities (Atkins et al. 2003). Could ICT similarly revolutionize education, especially since education deals directly with the codification and transmission of knowledge and information—two activities whose power has been decoupled by the ICT revolution? The education sector has so far been characterized by rather slow progress in terms of pedagogic innovation. Educational research and development does not play a strong role in producing systematic knowledge that translates into ‘‘programs that work’’ in the classroom or lecture hall (OECD 2003). As a matter of fact, education is not a field that lends itself easily to experimentation, partly because experimental approaches in education generally cannot be described precisely enough to

152

Kurt Larsen and Ste´phan Vincent-Lancrin

be sure that they are really being replicated (Nelson 2000). There is little codified pedagogic knowledge within the sector, and the knowledge management mechanisms are underdeveloped. Moreover, learning typically depends on other learning inputs than those received in the classroom or formal education process. The success of learning depends on a range of multifaceted socioeconomic factors beyond the control of educators. Focusing on tertiary education, this chapter examines the promises of ICT in the education sector, especially as a way to stem pedagogic innovation. Leaving aside the impact of ICT on the research or e-science performed by tertiary education institutions (see Atkins et al. 2003; David 2004), we concentrate on e-learning, broadly understood as the use of ICT to enhance or support learning and teaching in (tertiary) education. E-learning is thus used as a generic term referring to different uses and intensities of uses of ICT, from wholly online education through to campusbased education and other forms of education supplemented with ICT in some way. The supplementary model would encompass activities ranging from the most basic use of ICT (e.g., use of PCs for word processing of assignments) through to more advanced adoption (e.g., specialist disciplinary software, handheld devices, learning management systems, etc.). However, we keep a presiding interest in more advanced applications including some use of online facilities. The remainder of the chapter is organized as follows. Drawing on the scarce existing evidence, including a recent survey on e-learning in postsecondary institutions carried out by the OECD Centre for Educational Research and Innovation (OECD/ CERI), we show that e-learning has not yet lived up to its promises, which were overstated in the hype of the new economy. ICT has nonetheless had a real impact on the education sector, inducing a quiet rather than a radical revolution. We then interpret the evidence by comparing it to a model for e-learning adoption and conclude that, while e-learning is still at an early stage of adoption, some experiments point toward further and more radical developments, notably learning objects and open educational resource (OER) initiatives. Finally, we highlight some of the challenges for further development of e-learning. Living Up to the Promises: A Quiet Rather Than Radical Revolution During the dot-com boom, e-learning embodied many promises: enhanced quality of teaching/learning, increased and widened access for students, and decreased costs for students and governments, as well as new business and organizational models for tertiary education institutions. The possibilities of cross-border delivery through e-learning were also seen as opportunities (and challenges) that would reshape national tertiary education systems and offer emerging economies and developing

The Impact of ICT on Tertiary Education

153

countries a quick way to build their human resource capacity. Many observers and institutions speculated on the emergence of a huge market for e-learning and created (or merely announced the future creation of) new dedicated ventures. Fully online learning and the shift from physical to virtual campuses was even sometimes seen as a probable future for tertiary education in the medium term. After the burst of the new economy bubble in 2000, irrational beliefs about the market value of e-learning and overinvestment were mocked, although the dot-com boom generated more announcements than actual delivery (Boyer 2002). Skepticism replaced overenthusiasm. While it is still growing at a rapid pace, but from a very low starting point, has e-learning lived up to the promises it once embodied? To some extent. The reality of e-learning has never matched its most radical promises (Zemsky and Massy 2004). Like other activities, e-learning has not yet proved its ability to generate high profits or to replace the old economy of learning. However, interpreting this as a failure of e-learning would oversimplify the reality. The temptation to throw the baby out with the bath water should be resisted. Although, perhaps unsurprisingly, e-learning has not led to the radical revolution in tertiary education that was sometimes prophesied, some of its forms are already pervasive in tertiary education and have already led to a quiet revolution. Its modesty should not encourage one to overlook its benefits. This section gives an overview of the limited evidence available on the adoption of e-learning in tertiary education. E-learning Adoption During the dot-com boom, advocates of e-learning sometimes claimed that fully online learning would progressively supersede traditional face-to-face learning and represent a competitive threat for traditional tertiary educational institutions. To some extent, this belief has been a reason for the creation of new ventures and for established institutions to enter this developing market: Early adopters expected to gain a brand name and a serious competitive advantage in the new market. The reality, however, is that while there are examples of successful operations, fully online learning has remained a marginal form of e-learning and has not come close to the ultimate goal or rationale for e-learning adoption. However, this does not mean that e-learning in other forms has not gained ground within tertiary education during the past decade. There is indeed some evidence to suggest a noticeable growth of e-learning adoption on both the demand and the supply side. One must bear in mind that e-learning encompasses a wide range of activities. Following the terminology used in the OECD/CERI survey (OECD 2005), different levels of online learning adoption can be distinguished as follows, from the least to the most intensive form of e-learning:

154



Kurt Larsen and Ste´phan Vincent-Lancrin

None or trivial online presence.

Web supplemented The Web is used but not for key ‘‘active’’ elements of the program (e.g., course outline and lecture notes online, use of email, links to external online resources) without any reduction in classroom time.



Web dependent Students are required to use the Internet for key ‘‘active’’ elements of the program—e.g., online discussions, assessment, and project/collaborative work—but without significant reduction in classroom time.



Mixed mode Students are required to participate in online activities, e.g., online discussions, assessment, and project/collaborative work, as part of course work, which replace part of the face-to-face teaching/learning. Significant campus attendance remains.



Fully online The vast bulk of the program is delivered online with typically no (or insignificant) campus attendance or through ‘‘learning objects.’’



What do we know about the major trends in the adoption of e-learning by institutions and students? First, e-learning has grown steadily in the past decade, at a relatively rapid pace, but from a very low starting point—and for some activities from scratch. The lack of comprehensive data renders these trends difficult to document, but existing surveys all commonly point toward an increasing activity/supply. A significant share of tertiary education institutions have developed some e-learning activities and strategies and believe in the critical importance of e-learning for their long term strategy. The 2003 Sloan Survey of Online Learning based on a sample of 1,000 U.S. institutions shows that only 19% of the institutions have no advanced e-learning activities—that is, web-dependent, mixed mode, or fully online courses (Allen and Seaman 2003). The remaining 81% offer at least one course based on those advanced e-learning activities. Second, this growth of e-learning under all its forms should continue in the near future. There is converging evidence that tertiary education institutions consider e-learning to be part of their future development strategy. The 2003 Sloan survey reported that fewer than 20% of the U.S. tertiary education institutions considered online education to not be critical to their long term strategy. The 2004 survey of online learning carried out by the Observatory on Borderless Higher Education (OBHE) revealed that 79% of the 122 responding Commonwealth universities had an institution-wide online learning strategy as such or integrated into other strategies (46%) or under development (33%). Only 9% of these institutions had no e-learning strategy in place or under development in 2004 (Garrett and Jokivirta 2004; Garrett and Verbik 2004). While these figures may reflect some self-selection in the respondents, they unambiguously show a significant adoption or willingness

The Impact of ICT on Tertiary Education

155

to adopt some form of e-learning in the coming future. Although reflecting different levels of adoption of e-learning, all tertiary education institutions participating in the qualitative OECD/CERI survey on e-learning matched these findings and reported plans to increase their level of online delivery or to maintain their already high levels (OECD 2005). Third, virtual universities are not likely to become the paradigm of tertiary education institutions. While growth is likely to continue, especially in distance institutions (see below), no evidence points toward a predominance of this form of e-learning in the near future in tertiary education. The mixed mode of learning blending online and on-campus courses now clearly appears as a better candidate. Institutions head toward the simultaneous offer of a variety of learning models. For understandable reasons, only a few campus-based institutions (that is, the bulk of postsecondary institutions) appear to aim at delivering a large share of their courses fully online or to becoming completely virtual. While some institutions participating in the OECD/CERI survey are at the avant garde of e-learning, no campus-based institution predicted to deliver more than 10% of its total programs fully online within three years (OECD 2005). In the United States, rather than offering only fully online courses (16%) or only mixed mode courses (10%), most institutions offer both fully online and blended courses; moreover, the majority (67%) of academic leaders believe that mixed mode and web-dependent courses hold more promise than fully online, against only 14% having the opposite view (Allen and Seaman 2003). This clearly reflects what we know about the main rationales for undertaking e-learning. The OBHE surveys show that on-campus enhancement of teaching and learning (first) and improved flexibility of delivery for on-campus students (second) are the two key rationales in institutional strategies of e-learning. Only 10% of the institutions considered the enhancement of distance learning as more important than on-campus enhancement. Interestingly, the level of importance granted to distance or fully online learning decreased between 2002 and 2004 among returning respondents. Distance or fully online learning remains the fifth most important rationale though (Garrett and Jokivirta 2004). Finally, while an application of the fully online model is probably not for tertiary education overall, at least in the medium term, this does not mean that fully online activities are not growing rapidly and have not gained ground at distance education institutions (Bates 1995). Fully online learning is clearly very important for distance institutions. In the OECD/CERI survey, the institutions willing to embrace fully online learning to the greatest extent were all virtual/distance learning–only institutions (or branches) (OECD 2005). To our knowledge, no data on fully online enrollments are available for other countries than the United States. According to the 2004 Sloan survey, more than 1.9 million students (i.e., about 11% of all U.S.

156

Kurt Larsen and Ste´phan Vincent-Lancrin

tertiary-level students) took at least one fully online course during the fall of 2003. According to the previous 2003 survey, about one-third of them, that is, 578,000 students, took all their courses online. The University of Phoenix, the largest university in the United States in terms of enrollments, has, for example, 60,000 of its 140,000 students online. The enrollments of fully online students in the United States were increased by about 20% between 2002 and 2003, to 1.9 million students (Allen and Seaman 2003, 2004). This growth rate, which is estimated at 25% for 2005, is much higher than the growth rate of total tertiary enrollments in the United States. From a low starting point, fully online learning is thus growing at a rapid pace, even if it is merely as a complement to face-to-face or mixed mode learning. In conclusion, e-learning seems to live up to its promises in terms of flexibility and possibly expansion of access. It is a growing activity that has, for example, significantly widened the participation in tertiary education of foreign students in some countries, such as Australia (OECD 2004b). Does E-learning Improve the Quality of Tertiary Education? The real impact of e-learning on the quality of education is difficult to measure. E-learning largely embodies two promises: improving education thanks to enhanced learning and teaching facilities and inventing and sharing of new ways of learning and teaching thanks to ICT. While the first promise is by and large becoming a reality, at least in OECD countries, the second appears farther from reach. Viewed mainly as an enhancement of on-campus education, and thus matching the reality depicted in the previous section, there is some evidence that e-learning has improved the quality of the educational experience on both faculty and student sides (not to mention enhancement of administrative management). All institutions participating in the OECD/CERI survey reported a ‘‘positive impact’’ of greater use of e-learning in all its forms on teaching and learning. However, the quality of education (with or without e-learning) is very difficult to measure, not the least because learning not only depends on the quality of teaching but also on students’ motivation and abilities and on other conditions (e.g., family, social, economic, health backgrounds). Nevertheless, the reasons explaining this positive impact on quality largely live up to the promises of e-learning to offer more flexibility of access to learners, better facilities and resources to study (e.g. e-journals and e-books), and new opportunities thanks to the relaxation of space and time constraints. Basically, they do not correspond to a significant change in class pedagogy but to a change in the overall learning experience. According to the institutions, the main drivers or components of this positive impact come from

The Impact of ICT on Tertiary Education

Facilitated access to international faculty/peers lectures or joint classes with remote students;



157

e.g., with the possibility of online

Flexible access to materials and other resources allowing students to revise a particular aspect of a class, giving more access flexibility to part-time students, or giving remote and easy access to the library materials;



Enhancement of face-to-face sessions as the availability of archived lectures online frees up faculty time to focus on difficult points and application and because the introduction of e-learning has sometimes led to a debate on pedagogy; and



Improved communication learning.



between faculty and students and increase of peer

This ‘‘positive impact’’ on the overall learning experience is, alone, a significant achievement of e-learning, even though e-learning has not radically transformed the learning and teaching processes nor really affected the classroom pedagogy. The quality of fully online learning is a more controversial question, possibly because fully online learning was once projected to achieve a higher level of quality than on-campus education (possibly including e-learning as already mentioned). The task of comparing the quality (or the beliefs about the quality) of fully online learning against traditional distance learning is difficult since traditional face-toface learning or other mixed modes of e-learning might not yield the same results: Fully online learning is indeed more readily comparable to distance learning than to on-campus education. While early adopters of fully online e-learning generally have a positive view of its (possible) impact on quality, there is little convincing evidence of the superior or inferior quality of this mode compared to other modes of tertiary education. Another question is whether fully online learning has entailed innovation in pedagogy or just replicated the face-to-face experience through other means. As noted above, ICT could indeed entail pedagogic innovations and help create a community of knowledge among faculty, students, and learning object developers that would codify and capitalize over successful innovation in pedagogy. At this stage, there is no evidence to suggest that e-learning has yielded any radical pedagogic innovations. The most successful fully online courses generally replicate virtually the classroom experience via a mix of synchronous classes and asynchronous exchanges. Arguably, they have not represented a dramatic pedagogical change. We will see below that in spite of worthwhile experiments, learning objects and open educational resources are still in their infancy. However, they hold promise for educational innovation.

158

Kurt Larsen and Ste´phan Vincent-Lancrin

The Cost Efficiency of E-learning Has e-learning lived up to its promises in terms of cost efficiency? If one considers the most radical promises, the answer is no. Virtual universities have not replaced brick and mortar and saved the cost of expensive building investments and maintenance; digital libraries have supplemented rather than replaced physical ones; the codification and standardization of teaching did not occur to save labor costs (through a smaller or less qualified faculty); and finally, ICT infrastructure and development need significant ongoing investments and represent a significant, growing cost, contrary to the insignificant marginal cost of replicating and diffusing digitized information. Moreover, the initial costs for e-learning development are often high (e.g., infrastructure, creating course material from scratch, experimentation, new kind of staff/units, and immature technologies). For most universities, cost efficiency has been a secondary goal compared to the challenge of developing innovative and high quality e-learning courses. Although the ranking of cost efficiency increased 16% between 2002 and 2004, 37% of respondents in the OBHE survey considered ‘‘cutting teaching costs long-term’’ as a key rationale (Garrett and Jokivirta 2004)—a small percentage compared to the two key rationales (yielding more than 90% of responses). Again, most universities consider e-learning materials and courses to be a supplement to traditional classroom or lecture activities rather than a substitute, so their cost is also supplementary. The predominance of the web-dependent and mixed types renders the assessment of the cost of e-learning difficult since it would need to be disentangled from the conventional face-to-face setting. It is noteworthy that sample institutions of the OECD/ CERI survey could not provide systematic data on their e-learning costs (OECD 2005). There is actually little hard evidence on the cost impact of e-learning in tertiary education institutions. The adoption of ICT for administrating tertiary education institutions has probably been the main source of cost efficiency in the tertiary education sector. E-learning investments in tertiary education can, however, be cost effective, depending on the business model, the profile and number of students, and the topics. Cost effectiveness has been demonstrated in some cases in large undergraduate science classes (Harley 2003). Following are a few examples of cost-efficient models outside the traditional colleges and universities. Among the OECD/CERI sample institutions, the Open University of Catalonia reported a cost advantage because they are not building onto a physical campus. The UK Open University, which is gradually moving from traditional distance learning courses using books, videocassettes, and CD-ROMs to online courses, reported

The Impact of ICT on Tertiary Education

159

a cost per student equal to one-third the average cost for similar on-campus programs in the United Kingdom. Fixed capital costs are lower, and it is easier to align staffing structures to e-learning processes than at traditional universities. Outside the OECD/CERI sample, the e-learning activity of the Phoenix University, a private for-profit university for mainly adult students, is also seen as cost effective. The business model in this example is based on ‘‘standardized teaching,’’ a relatively small online class size, and use of proven low-tech e-learning technologies (inducing lower costs than more sophisticated technologies). Many of the faculty staff at Phoenix University work part time and work as well in other tertiary education institutions, implying lower staff development costs than at other tertiary education institutions. On the other hand, the university invests more than average in teacher training and pedagogical research. So, has e-learning lived up to its promises? It holds true when one considers incremental improvement, including an increased access and quality of the learning experience—change whose importance should not be underestimated. As for radical innovation, the answer is instead not yet. So far, e-learning has induced a quiet rather than a radical revolution of tertiary education. An Innovation Cycle for E-learning The limited impact of e-learning to date on quality, cost, and the learning/teaching process itself might be interpreted as a sign of the immaturity of e-learning that could be overcome with time. The development of e-learning could be viewed as divided into four distinct but often overlapping adoption cycles, as Zemsky and Massy (2004) claimed. This innovation/adoption curve is used to cast light on current and future development of e-learning. The four cycles can be described as follows: 1. Enhancements to traditional course/program configurations ICT is used to inject new materials into teaching and learning processes without changing the basic mode of instruction. Examples include e-mail usage, student access to information on the Internet, and the use of multimedia software (e.g., Powerpoint) and simple simulations. 2. Use of course or learning management systems E-learning software enables faculty and students to interact more efficiently (e.g., Blackboard or WebCT) and leads to better communication with and among students, quick access to course materials, and support for study administration (examination grades, credits obtained, fee, etc.). 3. Imported course objects E-learning allows faculty to embed a rich variety of third-party materials into their courses. Examples range from compressed video

160

Kurt Larsen and Ste´phan Vincent-Lancrin

presentations through to complex interactive simulations, including the increased use of ‘‘learning objects’’—that is, electronic tools/resources that can be used, reused, and re-designed in different contexts, for different purposes, and by different academics/actors. 4. New course/program configurations At this stage, e-learning has led faculty and their institutions to reengineer teaching and learning activities in novel ways. The new configurations involve active learning and new pedagogic processes that combine face-to-face, virtual, synchronous, and asynchronous interaction. Faculty and students adopt new roles—with each other and with the technology and support staff. The overview of current e-learning adoption presented in the previous section places most tertiary education institutions in OECD countries within cycles 1 and/or 2. These first two cycles largely build upon and reinforce one another. However, the use of ICT has not fundamentally changed teaching and learning at most tertiary education institutions nor induced radical innovation in pedagogy. Cycles 3 and 4 correspond to changes that more radically remodel teaching and learning. A number of experimentations under way give us an insight into where they could head, but they are still underdeveloped. Learning Objects The third cycle involves the creation of learning objects that rely on codification and standardization of knowledge and can potentially offer an efficient approach to the development of e-learning materials (i.e., reduced faculty time, lower cost, higher quality materials), although many issues remain (e.g., copyright, lack of incentives for faculty, lack of standardization and interoperability of e-learning software). The learning objects model corresponds to a course largely assembled by or from a third-party material/course and thus to a model departing from the conventional professional process in which the individual professor develops his or her course independently—unless the learning objects are small and malleable enough to offer enough possibilities for redesign by the user. Beyond the technical and organizational challenges of developing learning objects, there are considerable pedagogic challenges in using them. Some argue that learning is so contextually based that the breaking up of the learning experience into defined objects is destructive for the learning process. Evidence from the Open Learning Initiative at Carnegie Mellon University suggests that effective e-learning courses are facilitated by having a ‘‘theme’’ that runs throughout the course, which might be difficult to obtain with the notion of decontextualized learning objects (Smith and Thille 2004). Much more research and development is needed to ensure pedagogic effectiveness of the learning objects model.

The Impact of ICT on Tertiary Education

161

The immature development of a learning object economy is illustrated by the little use that is made of publicly available learning object repositories such as MERLOT (Multimedia Educational Resource for Learning and Online Teaching). The MERLOT repository aimed to create a readily available, low-cost, webbased service to which experimenters could post their learning objects and from which interested practitioners could download objects for use in their courses, and also rate them. While the number of learning objects made available by MERLOT has grown tremendously, there has been little interest to use objects made available by other colleagues and consequently little effort to rate others’ learning objects. This might be related to the professional ethos of faculty in which autonomy and diversity stand in good place. A cultural change would be needed if faculty members are to rely on others for content. Wide use of learning objects in tertiary education would imply major changes in working habits and attitudes of faculty. This can, however, also be seen as the first step toward the construction of knowledge communities in education. Despite the immature stage of learning objects and the large number of obstacles to overcome, some standard form of learning objects will probably emerge and gain importance in the development of e-learning in tertiary education as well as in other education sectors. Open Educational Resources The fourth e-learning adoption cycle has been reached by very few institutions, if any, at least institution-wide. Some institutions are, however, experimenting with new ways of using ICT that change the traditional organization of and pedagogy in tertiary education and give us an insight into what that revolutionary future might look like. One example is the aforementioned Open Learning Initiative at Carnegie Mellon University. The use of cognitive and learning sciences to produce high quality e-learning courses and online learning practices is at the core of this initiative (Smith and Thille 2004). Because there is no generic e-learning pedagogy, the aim is to design as cognitively informed e-learning courses as possible. The establishment and implementation procedures for routine evaluation of the courses and the use of formative assessment for corrections and iterative improvements are part of the e-learning course development. The development of the e-learning courses often relies on teamwork, including faculty from multiple disciplines, web designers, cognitive scientists, project managers, learning designers, and evaluators. More generally, open educational resources (OERs) appear to be a potentially innovative practice that gives a good example of the current opportunities and

162

Kurt Larsen and Ste´phan Vincent-Lancrin

challenges offered by ICT in order to trigger radical pedagogic innovations. Digitization and the potential for instant, low-cost global communication have opened tremendous new opportunities for the dissemination and use of learning material. This has spurred an increased number of freely accessible OER initiatives on the Internet, including (1) open courseware,1 (2) open software tools2 (e.g. learning management systems), (3) open material for capacity building of faculty staff,3 (4) repositories of learning objects,4 and (5) free educational e-learning courses. OER initiatives are a relatively new phenomenon in tertiary education largely made possible by ICT. Openly sharing educational resources consists of making knowledge available to third parties, generally on noncommercial terms and sometimes in the framework of communities of practice. In such communities, users freely reveal their knowledge and work cooperatively, which results in more and better quality innovations. These communities often do not extract economic revenues directly from the knowledge and information goods they are producing, and the ‘‘sharing’’ of these goods is not steered by market mechanisms. Instead, they have specific reward systems often designed to give some kind of credit to inventors without exclusivity rights. In the case of open science, the reward system is collegial reputation, where being identified and recognized as the discoverer of a theory gives faculty incentives to publish new knowledge quickly and completely (Dasgupta and David 1994). The main motivation for people to make OER material available freely is to see this material adopted by others and, maybe, even modified and improved. As with science, reputation is the key motivation factor in OER communities. Being part of such a user community gives access to knowledge and information from others, but it also induces a ‘‘moral’’ obligation to share one’s own information. Inventors of OER can benefit from the distribution at very low marginal costs. A direct result of free revelation lies in the increase in the diffusion of the innovation compared to licensing it or keeping it secret. A widely used innovation would initiate and develop standards that could be advantageously used, even by rivals. The Sakai project, for example, partly consists in making open software tools available to many colleges and universities and has therefore set a relatively low entry contribution for additional colleges and universities wishing to have access to the software tools they are developing. In conclusion, the relatively limited impact of e-learning in the classroom to date can arguably be interpreted as a sign of its immaturity. While most tertiary education institutions are still at an early stage of e-learning adoption (cycles 1 and 2), some initiatives and experiments under way show the light of more radical innovation in tertiary education (cycles 3 and 4).

The Impact of ICT on Tertiary Education

163

Challenges for the Further Development of E-learning in Tertiary Education: What Sustainable Innovation Model? This final section aims to identify and reflect on some of the key issues that would need to be considered systematically for e-learning to develop further and become a deeper driver of innovation in tertiary education. What features should have a sustainable innovation and investment to lead the vast majority of colleges and universities to embrace the third and fourth e-learning adoption cycles? We will focus on two of them: faculty engagement and sustainability of funding. Sustainability A first challenge for further progress in e-learning lies in the development of a realistic model of investment that would be financially sustainable (while stimulating the participation of faculty and other stakeholders). During the dot-com boom, many believed that a (huge) commercial market existed for e-learning, but these predictions have proved to be false. The bulk of students have remained skeptical about this novel way of learning. Finding a business model for e-learning is thus not straightforward, and even less so given the lack of systematic knowledge on the real costs and benefits of e-learning in tertiary education. Nevertheless, in most OECD countries it is no longer a question whether or not tertiary education institutions should invest in e-learning. Because of the competition between institutions and students demanding easy access to course material and flexible learning environments, most institutions are bound to invest in e-learning. The conditions under which e-learning could become a less expensive model compared to conventional face-to-face or distance education may come from a number of different sources: substituting some online provision for on-campus learning (rather than duplicating it), facilitating increased peer/automated learning, use of standard/preexisting software, drawing on the open standards and learning objects model to increase material reuse and sharing, avoidance of duplication of effort, and greater course standardization. In any case, reorganization should involve a decrease in the cost of developing and using e-learning, a decrease in the student/staff ratio, or savings due to less facility use (e.g., classrooms). Again, only very few institutions and faculty are systematically exploring and producing reusable learning material and objects (third cycle) or have taken full advantage of ICT with a focus on active learning that combines face-to-face, virtual, synchronous, and asynchronous interaction and learning in novel ways (fourth cycle). This would plead for stronger efforts to go toward cycles 3 and 4. Unfortunately, current endeavors exemplifying cycles 3 and 4 face the same sustainability problem.

164

Kurt Larsen and Ste´phan Vincent-Lancrin

Sustainability is the key question for any project like the aforementioned Open Learning Initiative attempting a combination of open access to free content and a fee-for-service model for students using the courses in a degree-granting setting. This initiative could not have been undertaken without significant voluntary contributions from private foundations and a major research grant from the National Science Foundation. Many OER initiatives are sponsored by private foundations or public funding or paid for by the institutions themselves. Individual faculty members or institutions might not be able to sustain the costs of providing OER material freely on the Internet in the long term. It is therefore important to find revenues to sustain these activities. How? For example, by charging and taking copyrights on part of the knowledge and information activities springing out of the OER initiatives. Or by finding better ways of sharing and reusing e-learning materials (see the previous discussion on learning objects) that generate revenues. Given that e-learning is still a novel and immature activity and that it has already improved the overall student experience (first and foremost through administrative rather than pedagogic changes), there is a case for continued government funding. However, governments and institutions need to have a clearer understanding of the costs and benefits of e-learning to encourage sustainable forms that could be mainstreamed. Engaging Faculty and Other Stakeholders A second, related, challenge for further progress in e-learning lies in the engagement of faculty and students in e-learning. E-learning cannot be cost efficient and sustainable if it is actually not used by students and faculty. This implies an ongoing improvement and assessment of the quality of e-learning and the search for incentives that would lead faculty (and students) to use e-learning and participate in knowledge communities. Like in many other sectors, the cost effectiveness of e-learning investments may eventually depend on whether new organizational and knowledge management practices are adopted. More than the technological infrastructure, it is now important to provide the appropriate social, organizational, and legal framework necessary to fully embrace the advantages of e-learning in tertiary education. This might take the development of a quality assessment framework for e-learning and the reengineering of the faculty reward system. As noted above, the adoption of the third and fourth adoption cycles would require behavioral shifts of faculty and students toward technology and support staff, and between them. While ICT offers powerful new instruments for innovation, tertiary education institutions are generally decentralized institutions, where the individual faculty often has the sole responsibility for teaching courses and delivering course material. Adoption of the third and especially the fourth e-learning cycle

The Impact of ICT on Tertiary Education

165

would require changing to more collaborative ways of organizing and producing teaching material. In order to produce their course material, faculty members would in many cases have to collaborate with a whole range of new staff such as course managers, web designers, instructional/pedagogical designers, cognitive scientists, etc. This could lead to resistance from ‘‘traditional’’ faculty arguing that current teaching practices have proved their value for centuries and being dubious about the need to change to new pedagogic methods whose effectiveness remains to be proved. Moreover, promotion of faculty and funding allocations in universities are often linked to research activities rather than teaching activities, often seen as less prestigious. Faculty members therefore often have relatively few incentives to invest their time in e-learning activities. The adoption of new ways of teaching and learning at tertiary education institutions through ICTs can therefore create organizational conflicts and tensions. New organizational innovations, new knowledge management practices, and more teamwork are necessary conditions for tertiary education institutions to move to e-learning adoption cycles 3 and 4. The lack of faculty engagement indeed hinders the progress and quality increase of learning materials in OER initiatives. The attribution of the intellectual property rights attached to e-learning material developed by faculty is a key issue. Does e-learning material belong to the faculty, the institution, the technologists who helped to develop it? In many countries, including the United States, the longstanding practice in tertiary education has been to give faculty the ownership of their lecture notes and classroom presentations. This practice has not always automatically been applied to e-learning course material. Some universities have adopted policies that share revenues from e-learning material produced by faculty. Other universities have adopted policies that apply institutional ownership only when the use of university resources is substantial (American Council on Education and EDUCAUSE 2003). In any case, institutions and faculty groups must strive to maintain a policy that provides for the university’s use of materials and simultaneously fosters and supports faculty innovation. It will be interesting to analyze how proprietary versus open e-learning initiatives will develop over the coming years in tertiary education. Their respective development will depend upon how the copyright practices and rules for e-learning material will develop at tertiary education institutions, the extent to which innovative user communities will be built around OER initiatives, the extent to which learning objects models will prove to be successful, the extent to which new organizational forms in teaching and learning at tertiary education institutions will crystallize, the demand for free versus ‘‘fee-paid’’ e-learning material, the role of private companies in promoting e-learning investments, etc. It is, however, likely that proprietary elearning initiatives will not dominate or take over open e-learning initiatives or vice

166

Kurt Larsen and Ste´phan Vincent-Lancrin

versa. The two approaches will more likely develop side by side, sometimes in competition, sometimes mutually reinforcing each other through new innovations and market opportunities. Finally, the lack of faculty engagement hinders the adoption of cycles 3 and 4 because learning objects and other educational resources are not improved and quality assessed. The MERLOT learning object repository intended to have the quality and usability of the learning objects made freely available, rated by the user community. In practice, very few users have taken the time or made the effort to evaluate others’ learning objects. The lack of review process or quality assessment system is a serious issue that is hindering increased uptake and usage of OER. Given the abundance of learning material on the Internet, users will be tempted to look for existing brands and known quality if there is little or no guidance on the quality of the proposed learning material. Finding new ways for the users of OER to be ‘‘advised’’ of the quality of the learning material stored in open repositories is crucial. User commentary, branding, peer reviews, or user communities evaluating the quality and usefulness of the OER might be possible ways forward. There is indeed no golden standard or method for identifying the quality of an online learning material in tertiary education. Adapting ‘‘global OER initiatives’’ to local needs and providing a dialog between doers and users of OER are also important quality challenges. Training initiatives for users to be able to apply course material and/or software might be a way to reach potential users. Also important will be the choice (using widely agreed standards), maintenance, and user access to the technologies chosen for the OER. There is indeed a need to better understand the users of OER. Only very few and hardly conclusive surveys of OER users are available.5 Conclusion There are many critical issues surrounding e-learning in tertiary education that need to be addressed in order to fulfill objectives such as widening access to educational opportunities, enhancing the quality of learning, and reducing the cost of tertiary education. E-learning is, in all its forms, a relatively recent phenomenon in tertiary education that has not radically transformed teaching and learning practices to date nor significantly changed the access, costs, and quality of tertiary education. As we have shown, though, e-learning has grown at a rapid pace and has already enhanced the overall learning and teaching experience (mainly through administrative enhancements and quicker access to course-related information, including e-journals). While it has not lived up to its most ambitious promises to stem radical innovations in the pedagogic and organizational models of tertiary education, it has quietly

The Impact of ICT on Tertiary Education

167

enhanced and improved the traditional learning processes. Most institutions are thus currently in the early phase of e-learning adoption, characterized by important enhancements of the learning process but no radical change in learning and teaching. Like other innovations, e-learning might, however, live up to its more radical promises in the future and really lead to the invention of new ways of teaching, learning, and interacting within a knowledge community made up of learners and teachers. In order to head toward these advanced innovation cycles, a sustainable innovation and investment model will have to be developed. While a first challenge will be technical, this will also require a broad willingness of tertiary education institutions to search for new combinations of input of faculty, facilities, and technology and new ways of organizing their teaching activities. As with ICT investments in other sectors, the cost effectiveness of e-learning investments will depend on whether new organizational and knowledge management practices are adopted. Experiments are already under way that make us aware of these challenges but also of the opportunities and lasting promises of e-learning in tertiary education. Notes 1. A well-known example is the Massachusetts Institute of Technology’s Open Courseware project, which is making the course material taught at MIT freely available on the Internet. 2. An example is the so-called Sakai project in the United States where the University of Michigan, Indiana University, MIT, Stanford University, and the UPortal Consortium are joining forces to integrate and synchronize their educational software into a preintegrated collection of open software tools. 3. The Bertelsmann and the Heinz Nixdorf foundations have sponsored the e@teaching initiative aiming at advising faculty in Germany on the use of open e-learning material. 4. For example, the MERLOT learning objects repository. 5. One exception is the user survey of MIT’s Open Courseware.

References Allen, I. E., and J. Seaman (2003). Seizing the Opportunity. The Quality and Extent of Online Education in the United States, 2002 and 2003. Needham, MA: Sloan Consortium. ———, and J. Seaman (2004). Entering the Mainstream. The Quality and Extent of Online Education in the United Sates, 2003 and 2004. Needham, MA: Sloan Consortium. American Council on Education and EDUCAUSE (2003). Distributed Education: Challenges, Choices and a New Environment. Washington, DC: Author. Atkins, D. E., K. K. Droegemeier, S. I. Feldman, H. Garcia-Molina, M. L. Klein, D. G. Messerschmitt, P. Messina, J. P. Ostriker, and M. H. Wright (2003). Final Report of the NSF Blue

168

Kurt Larsen and Ste´phan Vincent-Lancrin

Ribbon Advisory Panel on Cyberinfrastructure (available at http://www.cise.nsf.gov/sci/ reports/toc.cfm). Bates, A. W. (1995). Technology, e-Learning and Distance Education. London: Routledge. Boyer, R. (2002). La croissance, de´but de sie`cle. De l’octet au gene. Paris: Albin Michel. English translation: The Future of Economic Growth: As New Becomes Old. Cheltenham, UK: Edward Elgar, 2004. Dasgupta, P., and P. A. David (1994). ‘‘Towards a New Economics of Science.’’ Research Policy 23(5). David, P. A. (2004). Toward a Cyberinfrastructure from Enhanced Scientific Collaboration: Providing its ‘‘Soft’’ Foundations May Be the Hardest Threat. Oxford: Oxford Internet Institute. Foray, D. (2004). The Economics of Knowledge. Cambridge, MA: MIT Press. Garrett, R., and L. Jokivirta (2004). ‘‘Online Learning in Commonwealth Universities: Selected Data from the 2004 Observatory Survey, Part 1.’’ Observatory on Borderless Higher Education, London (available at www.obhe.ac.uk/products/briefings.html). ———, and L. Verbik (2004). ‘‘Online Learning in Commonwealth Universities: Selected Data from the 2004 Observatory Survey, Part 2.’’ Observatory on Borderless Higher Education, London (available at: www.obhe.ac.uk/products/briefings.html). Harley, D. (2003). ‘‘Costs, Culture, and Complexity: An Analysis of Technology Enhancements in a Large Lecture Course of UC Berkeley.’’ Center for Studies in Higher Education, University of California at Berkeley, Paper CSHE3-03. Hutchins, E. (1995). Cognition in the Wild. Cambridge, MA: MIT Press. Nelson, R. (2000). ‘‘Knowledge and Innovation Systems.’’ In Knowledge Management in the Learning Society. Paris: OECD. OECD (2003). New Challenges for Educational Research. Paris: Author. OECD (2004a). Innovation in the Knowledge Economy—Implications for Education and Learning. Paris: Author. OECD (2004b). Internationalisation and Trade in Higher Education: Opportunities and Challenges. Paris: Author. OECD (2005). E-Learning in Tertiary Education—Where Do We Stand? Paris: Author. Smith, J. M., and C. Thille (2004). The Open Learning Initiative—Cognitively Informed e-Learning. London: Observatory on Borderless Higher Education. Zemsky, R., and W. F. Massy (2004). Thwarted Innovation—What Happened to e-Learning and Why. The Learning Alliance, University of Pennsylvania.

12 The Bayh–Dole Act of 1980 and University– Industry Technology Transfer: A Policy Model for Other Governments? David C. Mowery and Bhaven Sampat

Introduction The research university plays an important role as a source of fundamental knowledge and, occasionally, industrially relevant technology in modern knowledge-based economies. In recognition of this fact, governments throughout the industrialized world have launched numerous initiatives since the 1970s to link universities to industrial innovation more closely. Many of these initiatives seek to spur local economic development based on university research, e.g., by creating ‘‘science parks’’ located near research university campuses, supporting ‘‘business incubators’’ and public ‘‘seed capital’’ funds, and organizing other forms of ‘‘bridging institutions’’ that are believed to link universities to industrial innovation. Other efforts are modeled on a U.S. law, the Bayh–Dole Act of 1980, that is widely credited with improving university–industry collaboration and technology transfer in the U.S. national innovation system. In some cases, these initiatives build on long histories of collaboration between university and industry researchers that reflect unique structural features of national university systems and their industrial environment. In other cases, however, these initiatives are based on a misunderstanding of the roles played by universities in national innovation systems, as well as the factors that underpin their contributions to industrial innovation. Many of these initiatives are based in the assumption that universities support innovation in industry primarily through their production of ‘‘deliverables’’ for commercialization (e.g., patented discoveries). Moreover, the most important channels through which university–industry interaction advances industrial innovation and economic growth, in this view, are the formal channels of patent licensing and, in some cases, the formation of university ‘‘spin-off’’ firms. But for most industries university research aids innovation through its informational outputs, which in turn often reach industrial scientists and engineers through the channels of ‘‘open science,’’ such as publications, conference presentations, or the

170

David C. Mowery and Bhaven Sampat

movement of personnel between universities and industry (including the hiring by industry of university graduates). International Policy ‘‘Emulation’’: Reflections in the Funhouse Mirror The emulation by many foreign governments of the Bayh–Dole Act illustrates a phenomenon that has received too little attention in the literature on innovation policy—the efforts by policymakers to ‘‘borrow’’ policy instruments from other economies and apply these instruments in a very different institutional context. History, path dependence, and institutional ‘‘embeddedness’’ all make this type of ‘‘emulation’’ very difficult. Nonetheless, such emulation has been especially widespread in the field of technology policy, most notably in the area of collaborative R&D policies. Research collaboration was cited by U.S. and European policymakers during the 1970s and 1980s as a key policy underpinning Japan’s rapid technological advance. Accordingly, both the European Union and the United States during the 1980s implemented policies and programs to encourage such collaboration, with mixed results. One of the best-known examples of such R&D collaboration is the SEMATECH (SEmiconductor MAnufacturing TECHnology) R&D consortium established in Austin, Texas, in 1987 with public and private funding. In response to the perceived success of the SEMATECH collaboration, Japanese managers and policymakers initiated publicly and privately funded research consortia (ASET and SELETE) in the late 1990s. Japan, which initially provided the model for emulation by the United States and the European Union, now is emulating the programs that allegedly were initially based on Japanese programs. International policy emulation of this sort is characterized by two features: (1) The ‘‘learning’’ that underpins the emulation is highly selective, and (2) the implementation of program designs based on even this selective learning is affected by the different institutional landscape of the emulator. Both of these characteristics of international emulation are readily apparent in the case of SEMATECH. They are even clearer in the efforts by other industrial-economy governments to develop policies similar to the Bayh–Dole Act during the 1990s. In addition to the difficulties associated with ‘‘transplantation’’ or emulation of these U.S. models, the effects of many of these ‘‘technology commercialization’’ policies remain controversial in their nation of origin. Advocates of the Bayh–Dole Act appear to have based their support on a view of the process of university–industry technology transfer that is not well-supported by studies of the channels through which academic research influences industrial innovation (see below for further discussion). Surprisingly little attention was devoted to the magnitude of the ‘‘prob-

The Bayh–Dole Act of 1980 and University–Industry Technology Transfer

171

lem’’ to which the Bayh–Dole Act was a ‘‘solution,’’ for example, and any potentially negative consequences of the act for academic research received no attention in the debates leading to its passage. How Does Academic Research Influence Industrial Innovation? A Review of Recent Studies A number of recent studies based on interviews or surveys of senior industrial managers in industries ranging from pharmaceuticals to electrical equipment have examined the influence of university research on industrial innovation and thereby provide additional insight into the role of universities within the U.S. national innovation system. All of these studies (GUIRR 1991; Mansfield 1991; Levin et al. 1987; Cohen et al. 2002) emphasize inter-industry differences in the relationship between university and industrial innovation. The biomedical sector, especially biotechnology and pharmaceuticals, is unusual, in that university research advances affect industrial innovation more significantly and directly in this field than is true of other sectors. In these other technological and industrial fields, universities occasionally contributed relevant ‘‘inventions,’’ but most commercially significant inventions came from nonacademic research. The incremental advances that were the primary focus of the R&D activities of firms in these sectors were almost exclusively the domain of industrial research, design, problem-solving, and development. University research contributed to technological advances by enhancing knowledge of the fundamental physics and chemistry underlying manufacturing processes and product innovation, an area in which training of scientists and engineers figured prominently, and experimental techniques. The studies by Levin et al. (1987) and Cohen et al. (2002) summarize industrial R&D managers’ views on the relevance to industrial innovation of various fields of university research (Table 1 summarizes the results discussed in Levin et al. 1987). Virtually all of the fields of university research that were rated as ‘‘important’’ or ‘‘very important’’ for their innovative activities by survey respondents in both studies were related to engineering or applied sciences. Interestingly, with the exception of chemistry, few basic sciences appear on the list of university research fields deemed by industry respondents to be relevant to their innovative activities. The absence of fields such as physics and mathematics in Table 1, however, should not be interpreted as indicating that academic research in these fields does not contribute directly to technical advance in industry. Instead, these results reflect the fact that the effects on industrial innovation of basic research findings in such areas as physics, mathematics, and the physical sciences are realized only after a

172

David C. Mowery and Bhaven Sampat

Table 1 The relevance of university science to industrial technology Number of industries with ‘‘relevance’’ scores Science

b5

b6

Biology

12

3

Chemistry Geology Mathematics Physics Agricultural science

19 0 5 4 17

3 0 1 2 7

Applied math/operations research Computer science

16

2

34

10

Materials science Medical science

29 7

8 3

Metallurgy

21

6

Chemical engineering

19

6

Electrical engineering Mechanical engineering

22 28

2 9

Selected industries for which the reported ‘‘relevance’’ of university research was large (b 6) Animal feed, drugs, processed fruits/ vegetables Animal feed, meat products, drugs None Optical instruments Optical instruments, electronics? Pesticides, animal feed, fertilizers, food products Meat products, logging/sawmills Optical instruments, logging/sawmills, paper machinery Synthetic rubber, nonferrous metals Surgical/medical instruments, drugs, coffee Nonferrous metals, fabricated metal products Canned foods, fertilizers, malt beverages Semiconductors, scientific instruments Hand tools, specialized industrial machinery

Source: Previously unpublished data from the Yale Survey on Appropriability and Technological Opportunity in Industry. For a description of the survey, see Levin et al. (1987).

considerable lag. Moreover, application of academic research results may require that these advances be incorporated into the applied sciences, such as chemical engineering, electrical engineering, and material sciences. Cohen et al. (2002) report that the results of ‘‘public research’’ performed in government labs or universities were used more frequently by U.S. industrial firms (on average, in 29.3% of industrial R&D projects) than prototypes emerging from these external sources of research (used in an average of 8.3% of industrial R&D projects). A similar portrait of the relative importance of different outputs of university

The Bayh–Dole Act of 1980 and University–Industry Technology Transfer

173

Table 2 Importance to industrial R&D of sources of information on public R&D (including university research)

Information source

Percentage rating it as ‘‘very important’’ for industrial R&D

Publications and reports Informal interaction Meetings and conferences Consulting Contract research Recent hires Cooperative R&D projects Patents Licenses Personnel exchange

41.2 35.6 35.1 31.8 20.9 19.6 17.9 17.5 9.5 5.8

Source: Cohen et al. (2002).

and public-laboratory research emerges from the responses to questions about the importance to industrial R&D of various information channels (Table 2). Although pharmaceuticals again is unusual in its assignment of considerable importance to patents and license agreements involving universities and public laboratories, respondents from this industry still rated research publications and conferences as a more important source of information. For most industries, patents and licenses involving inventions from university or public laboratories were reported to be of little importance, compared with publications, conferences, informal interaction with university researchers, and consulting. Data on the use by industrial R&D managers of academic research results are needed for other industrial economies. Nonetheless, these U.S. studies highlight the different relationship in biomedical research between academic research and industrial innovation. These studies also suggest that academic research rarely produces ‘‘prototypes’’ of inventions for development and commercialization by industry— instead, academic research informs the methods and disciplines employed by firms in their R&D facilities. Finally, the channels rated by industrial R&D managers as most important in this complex interaction between academic and industrial innovation assign a low ranking to patents and licenses. Perhaps the most striking aspect of these survey and interview results is the fact that they have not informed the design of recent policy initiatives to enhance the contributions of university research to industrial innovation.

174

David C. Mowery and Bhaven Sampat

Origins and Effects of the Bayh–Dole Act In our view, much of the international ‘‘emulation’’ of the Bayh–Dole Act reflects a distorted view of the nature of the links between university research and industrial innovation as well as a failure to appreciate the lengthy history of U.S. universities’ research collaboration with industry. This section provides a brief summary of this history. We also discuss the political origins and the effects on university patenting and technology transfer of the act. Origins of the Bayh–Dole Act Collaboration between university and industrial researchers, combined with the focus of many U.S. university researchers on scientific problems with important industrial, agricultural, or other public applications, meant that a number of U.S. universities patented faculty inventions throughout the 20th century. Public universities were more heavily represented in patenting than private universities during the 1925–1945 period, both within the top research universities and more generally. Despite the adoption by a growing number of universities of formal patent policies by the 1950s, many of these policies, especially those at medical schools, prohibited patenting of inventions, and university patenting was less widespread than was true of the post-1980 period. Moreover, many universities chose not to manage patenting and licensing themselves. The Research Corporation, founded by Frederick Cottrell, a University of California faculty inventor who wished to use the licensing revenues from his patents to support scientific research, assumed a prominent role as a manager of university patents and licensing. Even in these early decades of patenting and licensing, however, biomedical technologies accounted for a disproportionate share of licensing revenues for the Research Corporation and other early university licensors, such as the Wisconsin Alumni Research Foundation. The decade of the 1970s, as much as or more so than the 1980s, represented a watershed in the growth of U.S. university patenting and licensing. U.S. universities expanded their patenting, especially in biomedical fields, and assumed a more prominent role in managing their patenting and licensing activities, supplanting the Research Corporation. Agreements between individual government research funding agencies and universities also contributed to the expansion of patenting during the 1970s. Private universities in particular began to expand their patenting and licensing rapidly during this decade. The number of universities establishing technology transfer offices and/or hiring technology transfer officers began to grow in the late 1960s, well before the passage of the Bayh–Dole Act. Although the act was followed by a wave of entry by univer-

The Bayh–Dole Act of 1980 and University–Industry Technology Transfer

175

sities into management of patenting and licensing, growth in these activities was well established by the late 1970s. Indeed, lobbying by U.S. research universities was one of several factors behind the passage of the Bayh–Dole Act in 1980. The act therefore is as much an effect as a cause of expanded patenting and licensing by U.S. universities during the post-1960 period. The Bayh–Dole Patent and Trademark Amendments Act of 1980 provided blanket permission for performers of federally funded research to file for patents on the results of such research and to grant licenses for these patents, including exclusive licenses, to other parties. The act facilitated university patenting and licensing in at least two ways. First, it replaced a web of Institutional Patent Agreements (IPAs) that had been negotiated between individual universities and federal agencies with a uniform policy. Second, the act’s provisions expressed congressional support for the negotiation of exclusive licenses between universities and industrial firms for the results of federally funded research. Supporters of Bayh–Dole asserted that university contributions to innovation were limited by difficulties in patenting the outputs of federally funded research and licensing the patents exclusively to industry. This argument was particularly salient during the competitiveness crisis in the United States during the 1970s, in spite of the fact that proponents of Bayh–Dole offered very little evidence supporting this argument, as Eisenberg (1996) points out (see also Mowery et al. 2004). Moreover, there was no discussion during the Bayh–Dole debates about any potentially negative effects of increased patenting and licensing on the other channels through which universities contribute to innovation and economic growth. The passage of the Bayh–Dole Act was one part of a broader shift in U.S. policy toward stronger intellectual property rights.1 Among the most important of these policy initiatives was the establishment of the Court of Appeals for the Federal Circuit (CAFC) in 1982. Established to serve as the court of final appeal for patent cases throughout the federal judiciary, the CAFC soon emerged as a strong champion of patentholder rights.2 But even before the establishment of the CAFC, the 1980 U.S. Supreme Court decision in Diamond v. Chakrabarty upheld the validity of a broad patent in the new industry of biotechnology, facilitating the patenting and licensing of inventions in this sector. The origins and effects of Bayh–Dole thus must be viewed in the context of this larger shift in U.S. policy toward intellectual property rights. A number of scholars have documented the role of Bayh–Dole in the growth of patenting and licensing by universities since 1980 (Henderson et al. 1998a). But Bayh–Dole is properly viewed as initiating the latest, rather than the first, phase in the history of U.S. university patenting. And this latest phase is characterized by a higher level of direct involvement by universities in management of their patenting

176

David C. Mowery and Bhaven Sampat

Figure 1 U.S. research university patents as percentage of all domestic-assignee U.S. patents, 1963– 1999.

and licensing activities, in contrast to the reluctance of many U.S. universities to become directly involved in patenting prior to the 1970s. Effects of Bayh–Dole How did the Bayh–Dole Act affect the level of patenting by U.S. universities? Since overall patenting in the United States grew during this period, indicators of university patenting need to be normalized by overall trends in patenting or R&D spending. Figures 1 and 2 present two such indicators that span the period before and after the Bayh–Dole Act. Figure 1 depicts U.S. research university patenting as a share of domestically assigned U.S. patents during 1963–1999, in order to remove the effects of increased patenting in the United States by foreign firms and inventors during the late 20th century. Universities increased their share of patenting from less than 0.3% in 1963 to nearly 4% by 1999, but the rate of growth in this share begins to accelerate before rather than after 1980. Figure 2 plots the ratio of aggregate university patenting at time t to aggregate academic R&D expenditures at time t 1, for application years 1963–1993.3 The figure reveals an increase in aggregate university ‘‘patent propensity’’ after 1981 (as pointed out by Henderson et al. 1998a), but this is the continuation of a trend that dates back at least as far as the

The Bayh–Dole Act of 1980 and University–Industry Technology Transfer

177

Figure 2 University patents per R&D dollar, 1963–1993.

early 1970s; there is no evidence of a ‘‘structural break’’ in patent propensity after Bayh–Dole.4 Another issue of interest in academic patenting is the distribution among technology fields of university patents during the pre- and post-Bayh–Dole periods. Figure 3 displays this information for U.S. research university patents during 1960–1999 and highlights the growing importance of biomedical patents in the patenting activities of the leading U.S. universities during the period. Nonbiomedical university patents increased by 90% from the 1968–1970 period to the 1978–1980 period, but biomedical university patents increased by 295%. This rapid growth in biomedical patents also reflected growth of the major biomedical funding agency’s (HEW’s) IPA program during the 1970s. The increased share of the biomedical disciplines within overall federal academic R&D funding, the dramatic advances in biomedical science that occurred during the 1960s and 1970s, and the strong industrial interest in the results of this biomedical research all affected the growth of university patenting during this period. After Bayh–Dole, universities increased their involvement in patenting and licensing, setting up internal technology transfer offices to manage licensure of university patents. Figure 4 shows the distribution of years of ‘‘entry’’ by universities into patenting and licensing, defined as the year in which the universities first devoted 0.5 FTE employees to ‘‘technology transfer activities’’ (AUTM 1998). ‘‘Entry’’ accelerated after 1980, but growth in this measure of direct involvement by universities in patenting and licensing predates Bayh–Dole.

178

David C. Mowery and Bhaven Sampat

Figure 3 Technology field of Carnegie University patents, 1960–1999.

Longitudinal data on university licensing activities are less complete, but the available data indicate that in FY 2000, U.S. universities signed more than 4,000 license agreements, representing more than a doubling since FY 1991 (AUTM 2000). Based on these trends in university patenting and licensing, many observers have argued that Bayh–Dole stimulated university–industry technology transfer. During the late 1990s and early 21st century, many commentators and policymakers portrayed the Bayh–Dole Act as the critical catalyst to growth in U.S. universities’ innovative and economic contributions. Indeed, the OECD argued that the Bayh–Dole Act was an important factor in the remarkable growth of incomes, employment, and productivity in the U.S. economy of the late 1990s.5 Implicit in many if not all of these characterizations is the argument that university patenting and licensing were necessary to these asserted increases in the economic contributions of U.S. university research.6 Similar characterizations of the effects of the Bayh–Dole Act have been articulated by the president of the Association of American Universities7 and the commissioner of the U.S. Patent and Trademark Office (Dickinson 2000).8

The Bayh–Dole Act of 1980 and University–Industry Technology Transfer

179

Figure 4 Year of ‘‘entry’’ into technology transfer activities.

These characterizations of the positive effects of the Bayh–Dole Act cite little evidence in support of their claims beyond simple counts of university patents and licenses. But growth in both patenting and licensing predates Bayh–Dole and is rooted in internationally unique characteristics of the U.S. higher education system. Nor does evidence of increased patenting and licensing by universities by itself indicate that university research discoveries are being transferred to industry more efficiently or commercialized more rapidly, as Colyvas et al. (2002) and Mowery et al. (2001) point out. Indeed, current research provides mixed support at best for a central assumption of the Bayh–Dole Act, i.e., the argument that patenting and licensing are necessary for the transfer and commercial development of university inventions. These ‘‘assessments’’ of the effects of the Bayh–Dole Act also ignore potentially negative effects of the act on U.S. university research or innovation in the broader economy. Some scholars have suggested that the ‘‘commercialization motives’’ created by Bayh–Dole could shift the orientation of university research away from ‘‘basic’’ and toward ‘‘applied’’ research (Henderson et al. 1998a), but there is little evidence of substantial shifts since Bayh–Dole in the content of academic research. Since U.S. university patenting and licensing before and after 1980 has been concentrated in a few fields of research, notably biomedical research, a field characterized by blurry lines between ‘‘basic’’ and ‘‘applied’’ research, this finding should not come as a great surprise.

180

David C. Mowery and Bhaven Sampat

A second negative effect of increased university patenting and licensing is the potential weakening of academic researchers’ commitments to ‘‘open science,’’ leading to publication delays, secrecy, and withholding of data and materials (Dasgupta and David 1994; Liebeskind 2001). Although some work on this issue suggests that the ‘‘disclosure norms’’ of academic research in specific fields have been affected by increased faculty patenting (see Campbell et al. 2002), findings thus far are not conclusive, and more work is needed. Moreover, the Bayh–Dole Act is not solely responsible for any such changes in disclosure norms. Nonetheless, given the importance assigned by industrial researchers to the ‘‘nonpatent’’ channels of interaction with universities in most industrial sectors, it is crucially important that these channels not be constricted or impeded by the intensive focus on patenting and licensing in many universities. Finally, the effects of any increased assertion by institutional and individual inventors of property rights over inputs to scientific research have only begun to receive serious scholarly attention. Patenting and restrictive licensing of inputs into future research (‘‘research tools’’) could hinder downstream research and product development (Heller and Eisenberg 1998; Merges and Nelson 1994). Although there is little compelling evidence as yet that the Bayh–Dole Act has had negative consequences for academic research, technology transfer, and industrial innovation in the United States, the data available to monitor any such effects are very limited. Moreover, such data are necessarily retrospective and in their nature are likely to reveal significant changes in the norms and behavior of researchers or universities only with a long lag. International Emulation of the Bayh–Dole Act As we noted above, recent discussions by OECD governments on the desirability of ‘‘Bayh–Dole-type’’ policies reveal little awareness of the research discussed above that highlights the variety of channels through which universities contribute to innovation and economic growth. The ‘‘emulation’’ of Bayh–Dole in other industrial economies also tends to overlook the importance and effects on university–industry collaboration and technology transfer of the many other institutions that support these interactions and the commercialization of university technologies in the United States. The evidence discussed thus far suggests that the asserted ‘‘catalytic’’ effects of the Bayh–Dole Act on university–industry technology transfer have been overstated. Nevertheless, a number of other industrial-economy governments are considering or have adopted policies emulating the act’s provisions.9 In Denmark, a 1999 law gave public research organizations, including universities, the rights to all inventions

The Bayh–Dole Act of 1980 and University–Industry Technology Transfer

181

funded by the Ministry for Research and Technology. Under Denmark’s previous policy (established in 1957), all such rights had reverted to employees (OECD 2003). The German Ministry for Science and Education in 2002 altered the ‘‘professor’s privilege,’’ which gave academic researchers primary responsibility for the decision to file for patent protection on inventions and granted them the rights to any resulting patents. The new policy requires that academic inventors inform their employers of potentially patentable inventions two months before papers disclosing such inventions are submitted for publication and grants universities four months to determine whether they wish to file for patent protection.10 In France, a 1999 law authorized the creation of technology transfer offices at universities, and in 2001 the Ministry of Research ‘‘recommended’’ that universities and public research organizations establish policies to assert their rights to employee inventions (OECD 2003). The Canadian prime minister’s Expert Panel on the Commercialization of University Research recommended in 1999 that universities retain ownership of inventions resulting from publicly funded research and ‘‘be held accountable for maximizing returns to Canada,’’ noting that ‘‘the proposed IP policy framework will inspire a transformational shift in culture within Canadian universities, as happened in the United States with the passage of the Bayh–Dole Act in 1980’’ (Advisory Council on Science and Technology 1999, p. 28).11 In varying degrees all of these initiatives cite Bayh–Dole as one justification. Nevertheless, they in fact differ significantly from the act, which sought to transfer ownership for publicly funded inventions from government agencies to universities and other nonprofits. In contrast to Bayh–Dole, all of the policies described in the previous paragraph, along with similar new policies in other European countries (e.g., Austria, Ireland, and Spain) ‘‘have focused on changing employment laws so that university professors are no longer exempted from legislation that gives employers the IP generated by employees’’ (OECD 2003, p. 11).12 Similarly, the ‘‘Japanese Bayh–Dole Act’’ of 1999 shifted ownership from individual inventors to universities (http://www.nsftokyo.org/rm04-05.html). In addition to changes in intellectual property policy and employment regulations, a number of related initiatives aim to stimulate the organization and activity of technology licensing offices. Thus, the Swedish, German, and Japanese governments (among others) have encouraged the formation of external ‘‘technology licensing organizations,’’ which may or may not be affiliated with a given university (see Goldfarb and Henrekson 2003, for a comparison of Bayh–Dole and Swedish initiatives to enhance university–industry technology transfer). As this discussion suggests, these initiatives to emulate Bayh–Dole differ from one another and from Bayh–Dole itself. The policy proposals and initiatives display the classic signs of international emulation—selective ‘‘borrowing’’ from another

182

David C. Mowery and Bhaven Sampat

nation’s policies for implementation in an institutional context that differs significantly from that of the nation being emulated. Nonetheless, these initiatives are based on the belief that university patenting was an essential vehicle for effective transfer of technology from universities to industry and that Bayh–Dole was essential to the growth of university–industry interaction in science-based industries in the United States during and after the 1980s. They focus narrowly on the ‘‘deliverable’’ outputs of university research and ignore the effects of patenting and licensing on the other, more economically important channels through which universities contribute to innovation and economic growth. But patenting and licensing were only one of many channels through which U.S. universities contributed to industrial innovation throughout the 20th century, and surveys of industrial managers suggest that these channels are not the most important ones in most technological fields. Inasmuch as patenting and licensing are of secondary importance in most fields, emulation of the Bayh–Dole Act is insufficient and perhaps even unnecessary to stimulate higher levels of university–industry interaction and technology transfer. Instead, reforms to enhance inter-institutional competition and autonomy within national university systems, as well as support for the external institutional contributors to new-firm formation and technology commercialization, appear to be more important. Indeed, emulation of Bayh–Dole could be counterproductive in other industrial economies, precisely because of the importance of other channels for technology transfer and exploitation by industry. A narrow-minded focus on licensing as the primary or only channel for technology transfer can have a chilling effect on the operation of other important channels. There are potential risks to the university research enterprise that accompany increased involvement by university administrators and faculty in technology licensing and commercialization, and uncritical emulation of Bayh–Dole in a very different institutional context could intensify these risks. Conclusion The relationship between U.S. university research and innovation in industry is a long and close one. Indeed, organized industrial research and the U.S. research university both first appeared in the late 19th century and have developed a complex interactive relationship. The unusual structure of the U.S. higher education infrastructure, which blended financial autonomy, public funding from state and local sources with federal research support, and substantial scale, provided strong incentives for university faculty and administrators to focus their efforts on research activities with local economic and social benefits. Rather than being exclusively con-

The Bayh–Dole Act of 1980 and University–Industry Technology Transfer

183

cerned with fundamental scientific principles, much of U.S. university research throughout the late 19th and 20th centuries focused on understanding and solving problems of agriculture, public health, and industry. U.S. universities have made important contributions to industrial innovation throughout the past century, not least through providing both advanced research and education. The strong links between education and research sustained a close relationship between the evolving scientific research agenda and problems of industry or agriculture while at the same time providing a powerful and effective channel (in the form of trained students) for the transfer and application of much of this knowledge to industry and other economic sectors. In addition, many university researchers in engineering and medical schools maintained close ties with the users of their research and their graduates in industry, medical practice, and agriculture. The important role of universities in industrial innovation, particularly during the post-1945 period, also relied on institutions external to the university, including venture capitalists, equity-based financing of new firms, and high levels of labor mobility between academia and industry. Based on these considerations, we believe that much of the growth in licensing and university-based ‘‘spinoffs’’ that has occurred since the passage of the Bayh– Dole Act almost certainly would have occurred in the absence of this piece of legislation. After all, as we have pointed out here and elsewhere, U.S. universities were active patenters and licensors for decades before 1980, and much of their patenting and licensing activity since 1980 has been highly concentrated in a few fields, at least some of which also have benefited from rapid growth in public research funding and significant advances in basic science. For these and other reasons, we believe that the Bayh–Dole Act was neither necessary nor sufficient for the post-1980 growth in university patenting and licensing in the United States. Moreover, given the very different institutional landscape in the national higher education systems of much of Western Europe and Japan, it seems likely that the ‘‘emulation’’ of Bayh–Dole that has been discussed or implemented in many of these economies is far from sufficient to trigger significant growth in academic patenting and licensing or university–industry technology transfer. Indeed, there is some question as to the necessity of a ‘‘patent-oriented’’ policy to encourage stronger research collaboration and technology transfer. And the potential risks associated with such policy changes have received too little attention. Acknowledgments This chapter draws on research conducted with Professors Richard Nelson of Columbia University and Arvids Ziedonis of the University of Michigan, much of which

184

David C. Mowery and Bhaven Sampat

was published in ‘‘Ivory Tower’’ and Industrial Innovation: University–Industry Technology Transfer Before and After the Bayh–Dole Act (Stanford University Press, 2004). This research was supported by the Andrew W. Mellon Foundation and the Kauffman Foundation. Notes 1. According to Katz and Ordover (1990), at least 14 congressional bills passed during the 1980s focused on strengthening domestic and international protection for intellectual property rights, and the Court of Appeals for the Federal Circuit created in 1982 has upheld patent rights in roughly 80% of the cases argued before it, a considerable increase from the pre-1982 rate of 30% for the federal bench. 2. See Hall and Ziedonis (2001) for an analysis of the effects of the CAFC and related policy shifts on patenting in the U.S. semiconductor industry. 3. Data on total academic R&D were obtained from National Science Board (2002), Appendix Table 4-4. 4. As we have pointed out elsewhere (Mowery et al. 2001) The Bayh–Dole Act did not dramatically affect the patenting and licensing activities of universities that had long been active in this area, such as Stanford University and the University of California. Indeed, the biomedical patents and licenses that dominated these institutions’ licensing revenues during the 1980s and 1990s had begun to grow before the passage of the Bayh–Dole Act. Columbia University, an institution with little experience in patenting and licensing before 1980 (and an institution that prohibited the patenting of inventions by medical faculty until 1975), also had filed for its first ‘‘blockbuster’’ patent before the effective date of the act. Nevertheless, the act did increase patenting of faculty inventions at both Stanford and the University of California, although many of these patents covered inventions of marginal industrial value and did not yield significant licensing royalties. 5. ‘‘Regulatory reform in the United States in the early 1980s, such as the Bayh–Dole Act, have [sic] significantly increased the contribution of scientific institutions to innovation. There is evidence that this is one of the factors contributing to the pick-up of US growth performance . . . ’’ (OECD 2000, p. 77). 6. ‘‘Possibly the most inspired piece of legislation to be enacted in America over the past half-century was the Bayh–Dole Act of 1980. Together with amendments in 1984 and augmentation in 1986, this unlocked all the inventions and discoveries that had been made in laboratories throughout the United States with the help of taxpayers’ money. More than anything, this single policy measure helped to reverse America’s precipitous slide into industrial irrelevance. Before Bayh–Dole, the fruits of research supported by government agencies had gone strictly to the federal government. Nobody could exploit such research without tedious negotiations with a federal agency concerned. Worse, companies found it nigh impossible to acquire exclusive rights to a government owned patent. And without that, few firms were willing to invest millions more of their own money to turn a basic research idea into a marketable product.’’ (Anonymous 2002). 7. ‘‘In 1980, the enactment of the Bayh–Dole Act (Public Law 98-620) culminated years of work to develop incentives for laboratory discoveries to make their way to the marketplace

The Bayh–Dole Act of 1980 and University–Industry Technology Transfer

185

promptly, with all the attendant benefits for public welfare and economic growth that result from those innovations. Before Bayh–Dole, the federal government had accumulated 30,000 patents, of which only 5% had been licensed and even fewer had found their way into commercial products. Today under Bayh–Dole more than 200 universities are engaged in technology transfer, adding more than $21 billion each year to the economy’’ (Hasselmo 1999, p. 3). 8. ‘‘In the 1970s, the government discovered the inventions that resulted from public funding were not reaching the marketplace because no one would make the additional investment to turn basic research into marketable products. That finding resulted in the Bayh–Dole Act, passed in 1980. It enabled universities, small companies, and nonprofit organizations to commercialize the results of federally funded research. The results of Bayh–Dole have been significant. Before 1981, fewer than 250 patents were issued to universities each year. A decade later universities were averaging approximately 1,000 patents a year.’’ 9. A recent OECD report (2003) argues that these initiatives ‘‘echo the landmark Bayh–Dole Act of 1980’’ (p. 11). 10. The new policy aims to ensure that ‘‘more inventions are brought to patent offices before they get published’’ and ‘‘is supposed to lead to active licensing transfer from university to industry and to more companies being founded on the basis of intellectual property conceived within the university environment’’ (Kilger and Bartenbach, 2002). 11. Although no uniform government policy governs the treatment of university inventions in the United Kingdom, ‘‘there is now an increasing trend for universities to claim ownership’’ over academic inventions (Christie et al. 2003, p. 71). 12. In contrast to these initiatives, Italy passed legislation in 2001 that shifted ownership from universities to individual researchers. According to Breschi et al. (2004), this policy change has ‘‘the declared intention of finally providing the right economic incentives for individual scientists to undertake ‘‘useful’’ (that is, ‘‘patentable’’) research’’ (p. 2).

References Advisory Council on Science and Technology (1999). Public Investments in University Research: Reaping the Benefits. Ottawa: Industry Canada. Anonymous (2002). ‘‘Innovation’s Golden Goose.’’ The Economist 365 (December 12): T3. Association of University Technology Managers (AUTM) (1994). The AUTM Licensing Survey: Executive Summary and Selected Data, Fiscal Years 1993, 1992, and 1991. Norwalk, CT: AUTM. ——— (1996). AUTM Licensing Survey 1996, Survey Summary. Norwalk, CT: AUTM. ——— (1998). AUTM Licensing Survey 1998, Survey Summary. Norwalk, CT: AUTM. ——— (2000). The AUTM Licensing Survey: FY 1999. Norwalk, CT: AUTM. Blumenthal, David, and Sherrie Epstein (1986). ‘‘Commercializing University Research: Lessons from the History of the Wisconsin Alumni Research Foundation.’’ New England Journal of Medicine 314: 1621–1626. ———, Eric Campbell, Melissa Anderson, Nancyanne Causino, and Karen Louis (1997). ‘‘Withholding Research Results in Academic Life Science: Evidence from a National Survey of Faculty.’’ Journal of the American Medical Association 277: 1224–1229.

186

David C. Mowery and Bhaven Sampat

Breschi, S., F. Lissoni, and F. Montobbio (2004). ‘‘Open Science and University Patenting: A Bibliometric Analysis of the Italian Case.’’ Mimeo. Campbell, E. G., B. R. Clarridge, M. Gokhale, L. Birenbaum, S. Hilgartner, M. Holtzman, A. Neil, and D. Blumenthal (2002). ‘‘Data Withholding in Academic Genetics: Evidence from a National Survey.’’ Journal of the American Medical Association 287: 473–480. Christie, A. F., S. D’Aloisio, K. L. Gaita, M. J. Howlett, and E. M. Webster (2003). ‘‘Analysis of the Legal Framework for Patent Ownership in Publicly Funded Research Institutions.’’ Commonwealth of Australia, Division of Education, Science, and Training. Cohen, W., R. Florida, L. Randazzese, and J. Walsh (1998). ‘‘Industry and the Academy: Uneasy Partners in the Cause of Technological Advance.’’ In Challenges to the Research University, R. Noll, ed. Washington, DC: Brookings Institution. Cohen, W. M., R. R. Nelson, and J. P. Walsh (2002). ‘‘Links and Impacts: The Influence of Public Research on Industrial R&D.’’ Management Science 48: 1–23. Colyvas, Jeanette, Michael Crow, Annetine Gelijns, Roberto Mazzoleni, Richard R. Nelson, Nathan Rosenberg, and Bhaven N. Sampat (2002). ‘‘How Do University Inventions Get into Practice?’’ Management Science 48: 61–72. Dasgupta, P., and P. A. David (1994). ‘‘Towards a New Economics of Science.’’ Research Policy 23: 487–521. Dickinson, Q. T. (2000). ‘‘Reconciling Research and the Patent System.’’ Issues in Science and Technology 16: 27–31. Eisenberg, R. (1996). ‘‘Public Research and Private Development: Patents and Technology Transfer in Government-Sponsored Research.’’ Virginia Law Review 82: 1663–1727. ——— (2001). ‘‘Bargaining over the Transfer of Proprietary Research Tools: Is This Market Emerging or Failing?’’ In Expanding the Bounds of Intellectual Property: Innovation Policy for the Knowledge Society, D. L. Zimmerman, R. C. Dreyfuss, and H. First, editors. New York: Oxford University Press. Etzkowitz, H. (1994). ‘‘Knowledge as Property: The Massachusetts Institute of Technology and the Debate of Academic Patent Policy.’’ Minerva 32: 383–421. Geiger, R. (1986). To Advance Knowledge: The Growth of American Research Universities, 1900–1940. New York: Oxford University Press. Geiger, Roger L. (1993). Research and Relevant Knowledge: American Research Universities Since World War II. New York: Oxford University Press. Goldfarb, B., and M. Henrekson (2003). ‘‘Bottom-Up versus Top-Down Policies Towards the Commercialization of University Intellectual Property.’’ Research Policy 32, 639–658. Government University Industry Research Roundtable (GUIRR) (1991). Industrial Perspectives on Innovation and Interactions with Universities. Washington, DC: National Academy Press. Graham, H. D., and N. Diamond (1997). The Rise of American Research Universities. Baltimore: Johns Hopkins University Press. Hall, B. H., and R. H. Ziedonis (2001). ‘‘The Patent Paradox Revisited: An Empirical Study of Patenting in the U.S. Semiconductor Industry, 1979–1995.’’ Rand Journal of Economics 32: 101–128.

The Bayh–Dole Act of 1980 and University–Industry Technology Transfer

187

Hasselmo, Nils (1999). Priorities for Federal Innovation Reform. Washington, DC: Association of American Universities. Heller, M. A., and R. S. Eisenberg (1998). ‘‘Can Patents Deter Innovation? The Anticommons in Biomedical Research.’’ Science 280: 298. Henderson, R., A. B. Jaffe, and M. Trajtenberg (1994). ‘‘Numbers Up, Quality Down? Trends in University Patenting, 1965–1992.’’ Presented at the CEPR Conference on University Goals, Institutional Mechanisms, and the ‘‘Industrial Transferability’’ of Research, Stanford University. ———, A. B. Jaffe, and M. Trajtenberg (1995). ‘‘Universities as a Source of Commercial Technology: A Detailed Analysis of University Patenting, 1965–1988.’’ National Bureau of Economic Research Working Paper W5068. ———, A. B. Jaffe, and M. Trajtenberg (1998a). ‘‘Universities as a Source of Commercial Technology: A Detailed Analysis of University Patenting, 1965–88.’’ Review of Economics & Statistics 80: 119–127. ———, A. B. Jaffe, and M. Trajtenberg (1998b). ‘‘University Patenting Amid Changing Incentives for Commercialization.’’ In Creation and Transfer of Knowledge, G. Barba Navaretti, P. Dasgupta, K. G. Ma¨ler, and D. Siniscalco, editors. New York: Springer. Jensen, R., and M. Thursby (2001). ‘‘Proofs and Prototypes for Sale: The Licensing of University Inventions.’’ American Economic Review 91: 240–258. Katz, M. L., and J. A. Ordover (1990). ‘‘R&D Competition and Cooperation.’’ Brookings Papers on Economic Activity: Microeconomics: 137–192. Kilger, Chrisitian, and Kurt Bartenbach (2002). ‘‘New Rules for German Professors.’’ Science 298: 1173–1175. Levin, R. C., A. Klevorick, R. R. Nelson, and S. Winter (1987). ‘‘Appropriating the Returns from Industrial Research and Development.’’ Brookings Papers on Economic Activity 3: 783–820. Levine, David O. (1986). The American College and the Culture of Aspiration, 1915–1940. Ithaca, NY: Cornell University Press. Liebeskind, Julia (2001). ‘‘Risky Business: Universities and Intellectual Property.’’ Academe 87. Available at http://www.aaup.org/publications/Academe/01SO/so01lie.htm. Lowe, Robert A. (2001). ‘‘The Role and Experience of Start-Ups in Commercializing University Inventions: Start-Up Licensees at the University of California.’’ In Entrepreneurial Inputs and Outcomes, G. Libecap, editor. Amsterdam: JAI Press. Mansfield, Edwin (1991). ‘‘Academic Research and Industrial Innovations.’’ Research Policy 20: 1–12. Mazzoleni, R., and R. Nelson (1998). ‘‘The Benefits and Costs of Strong Patent Protection: A Contribution to the Current Debate.’’ Research Policy 27: 274–284. Merges, Robert, and Richard Nelson (1994). ‘‘On Limiting or Encouraging Rivalry in Technical Progress: The Effect of Patent Scope Decisions.’’ Journal of Economic Behavior and Organization 25: 1–24. Mowery, D. C. (2002). ‘‘The Changing Role of Universities in the 21st Century U.S. R&D System.’’ In AAAS Science and Technology Policy Handbook, A. H. Teich, S. D. Nelson,

188

David C. Mowery and Bhaven Sampat

and S. J. Lita, editors. Washington, DC: American Association for the Advancement of Science. ———, and B. N. Sampat (2001a). ‘‘Patenting and Licensing University Inventions: Lessons from the History of the Research Corporation.’’ Industrial and Corporate Change 10: 317– 355. ———, and B. N. Sampat (2001b). ‘‘University Patents, Patent Policies, and Patent Policy Debates, 1925–1980.’’ Industrial and Corporate Change 10: 781–814. ———, and B. N. Sampat (2004). ‘‘Universities in National Innovation Systems.’’ In Oxford Handbook of Innovation, J. Fagerberg, D. C. Mowery, and R. R. Nelson, editors. Oxford, UK: Oxford University Press. ———, R. R. Nelson, B. N. Sampat, and A. A. Ziedonis (2001). ‘‘The Growth of Patenting and Licensing by U.S. Universities: An Assessment of the Effects of the Bayh–Dole Act of 1980.’’ Research Policy 30: 99–119. ———, R. R. Nelson, B. N. Sampat, and A. A. Ziedonis (2004). ‘‘Ivory Tower’’ and Industrial Innovation: University–Industry Technology Transfer Before and After the Bayh–Dole Act. Stanford, CA: Stanford University Press. ———, B. N. Sampat, and A. A. Ziedonis (2002). ‘‘Learning to Patent: Institutional Experience and the Quality of University Patents.’’ Management Science 48: 73–89. National Research Council (1997). Intellectual Property Rights and Research Tools in Molecular Biology. Washington, DC: National Academy Press. National Science Board (2002). Science and Engineering Indicators: 2002. Washington, DC: U.S. Government Printing Office. OECD (2000). A New Economy? Paris: OECD. ——— (2002). Benchmarking Science–Industry Relationships. Paris: OECD. ——— (2003). Turning Science into Business: Patenting and Licensing at Public Research Organizations. Paris: OECD. Office of Technology Transfer (1997). ‘‘Annual Report: University of California Technology Transfer Program.’’ Oakland, CA: University of California Office of the President. Rai, A. T., and R. S. Eisenberg (2001). ‘‘The Public and the Private in Biopharmaceutical Research.’’ Presented at the Conference on the Public Domain, Duke University. Rai, Arti, and Rebecca Eisenberg (2003). ‘‘Bayh–Dole Reform and the Progress of Biomedicine.’’ American Scientist 91: 52–59. Reimers, Niels (1998). ‘‘Stanford’s Office of Technology Licensing and the Cohen/Boyer Cloning Patents.’’ Oral History compiled in 1997 by Sally Smith Hughes, Ph.D., Regional Oral History Office. Berkeley, CA: The Bancroft Library, UC Berkeley. Rosenberg, N. (1992). ‘‘Scientific Instrumentation and University Research.’’ Research Policy 21: 381–90. ———, and R. R. Nelson (1994). ‘‘American Universities and Technical Advance in Industry.’’ Research Policy 23: 323–348. Roush, Wade, Eliot Marshall, and Gretchen Vogel (1997). ‘‘Publishing Sensitive Data: Who’s Calling the Shots?’’ Science 276: 523–526.

The Bayh–Dole Act of 1980 and University–Industry Technology Transfer

189

Sampat, B. N., and R. R. Nelson (2002). ‘‘The Emergence and Standardization of University Technology Transfer Offices: A Case Study of Institutional Change.’’ Advances in Strategic Management 19. ———, and A. A. Ziedonis (2003). ‘‘Cite-Seeing: Patent Citations and Economic Value.’’ Presented at the Conference on Empirical Economics of Innovation and Patenting, Centre for European Economic Research, Mannheim Germany, March. Sampat, Bhaven, David Mowery, and Arvids Ziedonis (2003). ‘‘Changes in University Patent Quality After the Bayh–Dole Act: A Re-Examination,’’ International Journal of Industrial Organization. Slaughter, Sheila, and Larry L. Leslie (1997). Academic Capitalism: Politics, Policies, and the Entrepreneurial University. Baltimore: Johns Hopkins University Press. Stokes, D. E. (1997). Pasteur’s Quadrant: Basic Science and Technological Innovation. Washington, DC: Brookings Institution. Swann, John (1988). Academic Scientists and the Pharmaceutical Industry: Cooperative Research in Twentieth-Century America. Baltimore, MD: Johns Hopkins University Press. Thursby, J., and M. Thursby (2002). ‘‘Who Is Selling the Ivory Tower? Sources of Growth in University Licensing.’’ Management Science 48: 90–104. ———, R. Jensen, and M. Thursby (2001). ‘‘Objectives, Characteristics and Outcomes of University Licensing: A Survey of Major U.S. Universities.’’ Journal of Technology Transfer 26: 59–72. Trajtenberg, M., R. Henderson, and A. B. Jaffe (1997). ‘‘University Versus Corporate Patents: A Window on the Basicness of Inventions.’’ Economics of Innovation and New Technology 5: 19–50. Trow, M. (1979). ‘‘Aspects of Diversity in American Higher Education.’’ In On the Making of Americans, H. Gans, editor. Philadelphia: University of Pennsylvania Press. ——— (1991). ‘‘American Higher Education: ‘Exceptional’ or Just Different.’’ In Is America Different? A New Look at American Exceptionalism, B. E. Shafer, editor. New York: Oxford University Press. U.S. Congress Joint Economic Committee (1999). ‘‘Entrepreneurial Dynamism and the Success of U.S. High-Tech: Joint Economic Committee Staff Report.’’ Washington, DC: U.S. Government Printing Office. U.S. Patent and Trademark Office (1998). ‘‘US Colleges and Universities—Utility Patent Grants, 1969–1998.’’ Walsh, J. P., A. Arora, and W. M. Cohen (2003). ‘‘Research Tool Patenting and Licensing and Biomedical Innovation.’’ In The Patent System in the Knowledge-Based Economy, W. M. Cohen and S. Merrill, editors. Washington, DC: National Academies Press. Zucker, Lynne, Michael Darby, and Jeff Armstrong (1994). ‘‘Inter-Institutional Spillover Effects in the Commercialization of Bioscience.’’ ISSR Working Papers in Social Science 6.3, UCLA.

IV Knowledge and Place

13 The Changing Dynamics of the Global Market for the Highly Skilled Andrew Wyckoff and Martin Schaaper

During the 1990s, the United States was able to sustain rapid growth in skillintensive industries such as software, IT, and R&D without running into severe shortages of scientists and engineers that would have slowed the expansion. This period of growth has made the United States the exemplar of the ‘‘knowledge-based economy’’ and the benchmark by which many other countries compare themselves. This success is in spite of the fact that for decades, experts have warned that U.S. competitiveness is threatened by the poor performance of its schools and the weakness of its students in fundamentals such as reading, math, and science. A prime explanation to this paradox is that the United States has been able to attract the highly skilled from abroad. This chapter analyses the global market for the highly skilled. It marshals available data, albeit of a mixed quality, to describe the state of this market at the turn of the 21st century and then analyzes the dynamics that are likely to affect this market in the near term. A number of policy implications are outlined, especially regarding the location of research and the related infrastructure. A Nation at Risk? Twenty-two years ago an influential study entitled ‘‘A Nation at Risk’’ was delivered to Washington’s policy makers and analysts. The basic premise of the report was that a key factor undermining America’s competitiveness was the failure of its schools to produce properly skilled workers. Our nation is at risk. Our once unchallenged pre-eminence in commerce, industry, science, and technological innovation is being overtaken by competitors throughout the world . . . . The world is indeed one global village. We live among determined, well-educated, and strongly motivated competitors. We compete with them for international standing and markets, not only with products but also with the ideas of our laboratories and neighbourhood workshops. America’s position in the world may once have been reasonably secure with only a few exceptionally well-trained men and women. It is no longer. (NCEE 1983)

194

Andrew Wyckoff and Martin Schaaper

Identifying the Japanese, South Koreans, and Germans as competitors who were taking over key sectors such as autos, steel, and machine tools, the report linked these developments to a ‘‘ . . . redistribution of trained capability throughout the globe.’’ The report called for a reform of the U.S. educational system so as to ‘‘ . . . keep and improve on the slim competitive edge’’ the United States still retained, noting that ‘‘Learning is the indispensable investment required for success in the ‘information age’ we are entering.’’ (NCEE 1983) The report outlined a number of steps that were needed to be undertaken to address the problem, including the strengthening of high school graduation requirements, raised standards for admissions to colleges and universities, a refocusing on the basics in schools, and greater accountability on the part of educators and elected officials. Fast-forward to 2005 and many of the same criticisms of the U.S. school system remain (NSB 2004; NAE forthcoming). Recent international tests of the achievement of 15-year-olds in science and mathematics placed the United States 22nd and 28th, respectively, among 40 countries participating in the test (OECD 2004d, Table 2.5c, p. 356), roughly equal to Poland, Spain, and the Russian Federation. At the top of the international ranks were Finland, Japan, Korea, and Hong Kong, China. While significant changes have occurred and a number of efforts to improve U.S. schools have been launched, it appears that after two decades the relative performance of U.S. schools is little changed. The United States as the Benchmark This mediocre position in the international school achievement rankings stands in stark contrast to the position most observers attribute to the United States as a fiercely competitive and innovative country that represents the target other countries use to assess their innovativeness, R&D performance, and productivity. This was especially true during the 1990s, when the U.S. rate of productivity growth doubled in the second half of the 1990s relative to the average obtained for the previous two decades. This led many to question whether the United States had a new economy, where growth was based on a different formula of science, technology, human capital, and managerial expertise. Various international organizations and researchers launched projects to assess these assertions and generally concluded that while no one factor could be identified that explained the differences in performance—in fact the United States did outperform most of its large OECD competitors (OECD 2000 and 2003b). Two of the most widely quoted rankings of ‘‘competitiveness’’—IMD’s World Competitiveness Rankings and the World Economic Forum (WEF) Competitiveness Index

The Changing Dynamics of the Global Market for the Highly Skilled

195

Rankings—rank the United States numbers one and two, respectively, unchanged from 2003. Japan and Germany, the key challengers to the United States in the 1980s, are listed at numbers nine and 13, respectively, by the WEF and 23 and 21 by IMD (WEF 2004; IMD 2004). How could the United States achieve this status even though its schools are still relatively unchanged from the time when ‘‘A Nation at Risk’’ was written? The answer lies, in part, because the authors focused on the average achievement of the domestic population—the hump in the bell curve distribution of skills—when since the 1980s, the United States has been very good at increasing the tail of its skills distribution thanks to its very good universities and its ability to attract foreign talent, especially in the fields of science and engineering. The 1990s: The Tail of the Distribution of Skills The international flows of the highly skilled are not a new phenomenon (OECD 1970). In the postwar period the flows were largely from Europe to the United States and were due to the ‘‘pull’’ of a greatly expanded S&T postwar and coldwar U.S. system and the ‘‘push’’ of a relatively less advanced and resourced European system. The beneficiaries of these flows in the United States were largely academia and government laboratories, which were engaged in fundamental, basic research. The 1990s marked a significant change in these flows as the primary source shifted from Europe to Asia. The United States was still the main destination but the beneficiaries had broadened; now the industry that is engaged in applied, developmental research was the main employer of the foreign-born highly skilled. The push and pull factors have changed, where the key port of entry for attracting the highly skilled is through the opportunity to study at world-class colleges and universities. In 2000 there were 1.5 million foreign students enrolled in higher education institutions in OECD countries, about double the level of 20 years ago (OECD 2004a). The United States, the United Kingdom, Australia, and Canada together represent the destination for more than half of all foreign students in the OECD area, while Asia accounts for almost half (43%) of all the international tertiary-level students in the OECD area. China (including Hong Kong, China) alone accounts for 10% of all these international students in the OECD area. These trends are reinforced when the focus shifts to the graduate level. In absolute terms, the United States receives more foreign doctorate-level students than all other OECD countries combined, with the number of doctorate degrees in science and engineering awarded to foreign students more than doubling between 1985 and 1996 in the United States. Every year between 1992 and 2001, almost 10,000

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

Total foreign citizenship North America (total) South America (total) Europe (total) East Asia (total) West Asia (total) Pacific/Australasia (total) Africa (total) Country unknown

8,926 525 387 972 4,486 1,670 216 500 170

9,475 515 394 950 4,865 1,891 220 510 130

9,754 506 394 1,104 5,010 1,903 227 474 136

10,542 547 408 1,150 5,484 2,077 231 583 62

10,502 505 359 1,254 5,486 2,181 231 423 63

10,815 530 454 1,260 5,597 2,140 232 442 160

9,779 462 388 1,276 4,555 1,914 196 335 653

9,790 545 411 1,477 4,531 1,750 192 339 545

8,888 532 420 1,460 4,069 1,595 156 327 329

9,057 605 433 1,494 4,249 1,559 178 347 192

Source: From Barrere et al. (2004), based on data from SESTAT, National Science Foundation.

Andrew Wyckoff and Martin Schaaper

Country of citizenship

196

Table 1 Non–U.S. citizens awarded doctorates in the sciences and in engineering, by country of citizenship and year of doctorate, 1991– 2000

The Changing Dynamics of the Global Market for the Highly Skilled

197

Figure 1 Number of S&E doctorates awarded to foreign citizens in the United States. Source: OECD, based on data from U.S. National Science Foundation, 2003.

non-U.S. citizens were awarded doctoral degrees in the sciences or in engineering (S&E) in the United States (Table 1). In 2001, this number stood at 9,188, of which a little more than a quarter were Chinese citizens (see Figure 1). The second largest group of foreigners were Koreans (9.4%), followed by Indians (8.8%) and students from Chinese Taipei (5.9%). Asian students therefore represent the bulk of S&E doctorates awarded to foreigners in the United States. Holders of temporary visas represented 86% of these foreign doctorate recipients, a trend that increased during the 1990s (see Figure 1). Having studied in the United States, about two-thirds of the foreign recipients of U.S. S&E doctorates (1998–2001) have ‘‘firm plans’’ to stay in the United States, a rate that is up significantly from 1994 to 1997, when only 57% indicated an intention to stay (NSF 2004). Stay rates differ by nationality of the student: About 50% of foreign students from France and Italy indicated firm plans to stay, while 67% of those from China and 73% from India did. For selected fields such as mathematics/ computer science, the stay rate of large pools of foreign doctorate recipients from India and China was even higher (71% of Chinese and almost 80% of Indians). Accurate global data on the flows of people, especially the highly skilled, are not available, but new international work that compiles censuses taken in about 2000 provides a more reliable picture of immigrant populations (Dumont and Lemaıˆtre 2005). Calculations based on the U.S. decennial census of 2000 estimate that there was a stock of more than 8 million ‘‘highly qualified’’1 foreign-born immigrants in

198

Andrew Wyckoff and Martin Schaaper

Figure 2 Stock of highly skilled immigrants in OECD countries. Source: Dumont and Lemaıˆtre (2004).

the United States. As can be seen in Figure 2, the U.S. stock accounted for more than the next eight largest OECD member countries combined. The NSF analysis of the 2000 census shows that 22% of the college graduates in S&E occupations are foreign born. As the focus shifts to doctorates the percentage increases to 37% and goes even higher in some select fields: about 45% for mathematical and computer science occupations and more than 50% for engineering occupations (NSF 2004). This inflow of foreign students who stay on and the arrival of scholars and highly educated during the 1990s is one of the factors that enabled the explosive growth of the ICT sector, particularly the software segment, where human capital is a key input. Saxenian shows that nearly a third of Silicon Valley’s 1990 workforce was composed of immigrants, two-thirds of them from Asia, primarily China or India (Saxenian 1999). Between 1995 and 1998, Chinese and Indian engineers started 29% of Silicon Valley’s technology companies, up from 13% in the 1980–1984 period. While brain drain and gain is a phenomenon that has existed for centuries, the sharp increase in demand from the United States effectively meant that this part of the labor force, or at least key segments of it, had become global. Although the source of these highly skilled professionals was global, in fact the demand was narrowly based in the United States. The New Millennium The factors that combined to create a global market for the highly skilled in the 1990s have changed with the arrival of the new millennium, where the position of

The Changing Dynamics of the Global Market for the Highly Skilled

199

the United States as the main driving force generating demand has diminished and the opportunities in the main supply countries have improved. Combined, the market has become more truly global, posing a significant challenge for the United States and its system of innovation and an opportunity for other countries. The Post–9/11 Climate Many have noted that the new security precautions put in place after the terrorist attacks of 2001 have made the United States less welcoming to foreigners, including the highly skilled (Buderi 2003; Nye 2004; Mahroum 2002b). The problems associated with increased airport security, refused visas, and months spent trying to gain entry circulate quickly in this wired circle, and they have begun to affect behavior. Foreign student enrollments in U.S. higher education institutions dropped for the first time in 30 years in 2003–2004 (IIE 2004), and high-skill–related work visa applications are down 19.4% from 2001 to 2003, with the refusal rate of these applications increasing from 9.6% to 17.8% (NSF 2004). A 2003 report by the American Institute of Physics reported a 15% drop in international students entering U.S. physics programs, and about 20% of those who were admitted were unable to start because of visa problems (Armstrong 2003). These trends have made the headlines, generated U.S. congressional hearings, and led to new, more streamlined procedures (Science and Government Report 2004; AIP 2004). In short, there is evidence that those in control are trying to strike the right balance between improving security and making adjustments to keep this important flow coming; but what many miss is that concurrent with this less-friendly environment (the ‘‘pull’’ of the United States) is a significant change abroad in the forces that ‘‘pushed’’ these people to the United States in the first place. European Union Aside from the United States, the single largest regional block of R&D and concentration of researchers is in Europe. In this sense, Europe is the other key buyer in the global market for the highly skilled. During the course of the 1990s, the European research area was a collection of national systems, with very little coordination. Gradually that has changed as the European Commission have exerted more influence, as central funding for research has increased, and as it becomes clear that to compete on the world stage with players like the United States, a more cohesive European strategy is needed (European Commission 2003). This change in mind-set became most apparent at the Lisbon Summit of Heads of State in 2000, which set the goal for Europe ‘‘ . . . to become the most competitive and dynamic knowledgebased economy region in the world,’’ and the follow-up summit in Barcelona in 2002 that set a quantitative target of increasing the EU’s R&D intensity (total

200

Andrew Wyckoff and Martin Schaaper

R&D/GDP) from about 1.9% to a level approaching 3.0% by 2010 (European Commission 2002). Since at least half of R&D spending goes to pay the wages of researchers, increasing R&D by this amount will require a significant increase in the number of researchers. Estimates of the number of additional researchers needed to meet R&D spending targets depend on the assumptions made, but if it is assumed that R&D spending per researcher in the European Union begins to look more like that of the United States and that the annual rate of growth in GDP is the same as it has been for the past decade (2%), then the European Union would need approximately 500,000 more researchers by 2010 to meet its 3% target, an increase of more than 50% above 2000 levels (Sheehan and Wyckoff 2003). The European Commission itself estimates a need for an additional 700,000 researchers to reach the goal (European Commission 2003). Adding another 500,000 to 600,000 researchers to the E.U. workforce by 2010 will present a challenge to Europe and a potential bottleneck for satisfying the goal of 3% R&D intensity (Sheehan and Wyckoff 2003). Few analysts believed this goal could be reached when it was set in 2000 and 2002, and even fewer think it is possible in 2005 since the R&D intensity of the European Union is still hovering at 1.9%. But it is a mistake to interpret this goal simply on an analytical basis—it is a political goal and in this sense it has already begun to succeed. Innovation policy is now high up on the policy agenda of Europe, to the point where prime ministers and ministers of finance are now concerned about the problem (Brown 2005). A key element of this concern focuses on the highly skilled. Europe faces a dual problem, as do most developed countries: While they try to attract more researchers to bolster their R&D activity, they are faced with a rapidly aging population and an increasing rate of retirement of many researchers. This combination creates a need to produce more researchers from the native population, especially from relatively untapped reservoirs such as women, while at the same time becoming more aggressive about attracting the highly skilled from abroad and stemming the outward migration of Europe’s brains. A number of policy initiatives have been recently launched to achieve these objectives (OECD 2004b). Japan After the United States, Japan has the second-largest national S&T system in the world as measured by absolute R&D expenditures and number of researchers. This effort has been sustained from an indigenous supply of highly skilled scientists, engineers, and technicians, but as Japan’s population ages and as shortages in select areas such as software engineers arise, Japan is beginning to enact policies aimed at attracting the highly skilled from abroad, particularly from India and China: The

The Changing Dynamics of the Global Market for the Highly Skilled

201

general immigration law was relaxed in 1989 to ease the entry of the highly skilled, the Ninth Basic Plan on Employment Measures (2002) extended the period of residence for a variety of immigration statuses from one to three years, exemptions from immigration restrictions have been expanded for IT technicians and foreign researchers, mutual accreditation of IT technicians with India has been launched, and a postdoctoral fellowships program for foreign researchers has been established (METI 2003; OECD 2004b). Japan has more than doubled (from 537 in 1996 to 1,225 in 2000) the number of postdocs provided to foreign, mainly Asian, scientists through the Japan Society for the Promotion of Science (JSPS) Fellowship Program (Mahroum 2002a). While the flows of foreign highly skilled people to Japan are still relatively modest, they are beginning to grow, reflecting another change in the global market for the highly skilled. In 1992, the number of foreigners with ‘‘special and/or technical skills’’ registered in Japan for purposes of employment was about 85,000; this climbed modestly to about 98,000 in 1996, after which the level grew more rapidly, reaching about 118,000 in 1998 and 169,000 in 2001—nearly double the 1992 level (METI 2003). Some of the fastest growing occupations include professor, researcher, and engineer. More than half of all foreign engineers in Japan come from China—a near doubling of the absolute number between 1994 and 2001. Korea is second, accounting for about 10%. India accounts for only about 7% of foreign engineers in Japan in 2001, but this represents a 53% increase from 2000. China As Saudi Arabia and Kuwait are to world oil supply, China and India are to the international flows of brains. The huge demographic size of these countries means that the tails of the skills distribution are large. Until very recently, they had few domestic opportunities and had to go abroad to pursue an education and career. In fact, government policy in both countries had an explicit goal of promoting the diaspora (OECD 2001), and the United States was the overwhelming destination of these people. While the flows of the highly skilled Chinese to the United States are still considerable, there are several signs that the various factors that ‘‘pushed’’ the highly skilled Chinese away from China are changing as the Chinese S&T system grows and opportunities to study, conduct research, and work in a high-tech company expand. These developments suggest that increasingly China will become a competitor for the highly skilled, especially for its indigenous supply. Changes in the flow of students The pursuit of higher education is an important channel for attracting the highly skilled since many stay in their host country after

202

Andrew Wyckoff and Martin Schaaper

Figure 3 Number of Chinese students enrolled in tertiary education in the United States, Japan, and the European Union (thousands). Source: OECD, education database (February 2005).

graduation. Thus, changes in the international flow of students are an early indicator of likely changes in the international mobility of the highly skilled. Two changes in the flow of Chinese students are appearing. The first is that the dominance of the United States as a host country to Chinese students is decreasing. In 2002, about 63,000 of the Chinese students enrolled in OECD countries were studying in the United States, equivalent to 35% of the total number of Chinese students enrolled in OECD countries (Figure 3). While this marks an increase in the absolute number of students, it represents a decrease in the overall share as the European Union attracted increasing numbers of Chinese students, almost doubling its share during the period 1998–2002. The second change is that since 1999, China has greatly expanded the enrollment of students in its own universities. The trend began with the passing of the Great Cultural Revolution in the late 1970s and the initiation of examinations for entrance to institutions of higher learning, but it was not until the rapid economic growth and the demand for highly skilled labor in the 1990s that significant numbers of bachelor, masters, and doctoral students began to matriculate. The number of bachelors and masters degrees conferred in 1999–2003 was nearly double the 1982–1989 level, while the number of doctorates increased by a factor of 12, from about 5,000 to 67,000 (Song and Xuan 2004) (Figure 4). This rapid growth is likely to continue since the number of doctoral students admitted has rapidly increased, jumping from about 14,500 in 1998 to 48,700 in 2003. The majority of the doctoral degrees earned between 1992 and 2003 in China

The Changing Dynamics of the Global Market for the Highly Skilled

203

Figure 4 Doctoral degrees awarded in China. Source: Song and Xuan (2004).

were for engineering (38% of the total), natural sciences (22%), and medicine (15%). In comparison, the NSF reports that about 200 Chinese students earned S&E doctorates in the United States in 1986, a number that increased to almost 3,000 in 1996, after which it declined in 1997 and rose slightly in 1998—a change in trend attributed to the possibility of increased capacity for graduate education in China (Johnson 2001). As can be seen from Table 1, U.S. doctorate awards in S&E to East Asians started to decline in 1996—significantly before the terrorist attacks in the United States and just as the Chinese awards of doctorates began to surge (Figure 4). The expanding capacity of Chinese science & technology Concurrent with this shift in the flows of Chinese students is the growth of the Chinese science and technology system. This is evident from measures such as the number of researchers, the level of R&D being performed, and the establishment of business R&D centers. Researchers In most economies, the number of researchers has been growing steadily during the past decade (Figure 5). The data for China show slow growth between 1991 and 1997, followed by a drop in 1998 and a slight recovery in 1999. Since 1999, however, the figure has soared from around 531,000 in 1999 to around 811,000 in 2002. Part of this growth can be attributed to improved measurement, but it is also associated with an explicit policy to increase the national R&D effort significantly during the 10th Five-Year Plan (2001–2005). While differences in terms of quality may exist, China now counts more researchers than Japan (approximately 676,000 in 2001) and is quickly approaching the level of the European Union (1 million in 2001) (OECD 2004c).

204

Andrew Wyckoff and Martin Schaaper

Figure 5 Number of researchers (thousands of FTE). Note: There is a break in series for China between 1999 and 2000 because of improved measurement; for more details, see Schaaper (2004). Source: OECD, MSTI database.

R&D expenditures Since about half of all R&D expenditures goes to pay wages of researchers, the level of Chinese R&D has grown significantly as well. During the 1990s, the R&D effort increased on average by 15.2% annually in real terms2 (Figure 6), with recent growth increasing by 20.6% in real terms annually. Even after normalizing for the size of the Chinese economy by looking at the R&D intensity (total R&D as a percentage of GDP), the Chinese R&D effort has rapidly increased, from 0.7% of GDP in 1998 to 1.3% in 2002—about half the 2003 U.S. intensity of 2.7%. R&D centers In an effort to stimulate business innovation, China continues to privatize its R&D institutes, converting more than 1,000 centers in 2002 (OECD 2004b). Associated with this has been the construction of more than 60 industrial parks, with the intent of luring home from overseas highly skilled Chinese. Accompanying this has been an inflow of multinational enterprises (MNEs) establishing R&D centers in China, where a wide range of technology-intensive firms such as DuPont, Ford, GM, Lucent Technologies, Motorola, IBM, Intel, Microsoft, Oracle, Siemens, GE, Nokia, Cisco, and Philips have contributed to the increase in Chinese R&D. In total, estimates of the number of MNE research labs range from 300 to 600, with the bulk of them opened in the past few years (Buckley 2004). U.S. foreign affiliates in China performed USD 506 million worth of R&D in 2000 com-

The Changing Dynamics of the Global Market for the Highly Skilled

205

Figure 6 Growth of R&D expenditure, annual average growth rate 1991–2001 (based on national currencies in constant prices). Note: There is a break in series for China between 1999 and 2000 because of improved measurement; for more details, see Schaaper (2004).

pared to only USD 7 million in 1994 (Moris 2004). A similar trend is forming for European, Japanese, and Korean firms, even if North America remains a key destination for R&D-related FDI. Increased Chinese demand for the highly skilled These developments coupled with an explicit policy by the Chinese Ministry of Personnel of encouraging highly skilled overseas Chinese to return to China has led to a return flow of highly skilled Chinese that has grown on average by 13% a year in the 1990s, albeit significantly below the rate of increase in the outflow (OECD 2004b). The Chinese estimate that of the 450,000 Chinese students abroad, 150,000 have returned. One indicator of this growing Chinese demand for S&T personnel is the bidding up of wages that is occurring. The average labor cost for R&D personnel (FTE) in China increased by 30% (nominal terms) from 2000 to 2002.3 This corresponds to more anecdotal information of shortages of qualified engineers that has led to wages doubling in the past five years, equivalent to an annual increase of 15% (Marsh 2004). Thus, while a huge wage differential exists between China and most OECD member countries for ordinary manufacturing workers, the gap for engineers and scientists is narrowing quickly. India With China, India represents the other major source of supply of the internationally mobile highly skilled. Like China, the supply of talent from India, especially for IT and health professionals, is in demand from a growing number of countries, including

206

Andrew Wyckoff and Martin Schaaper

India itself. Coupled with this is the growing capacity of Indian institutions of higher learning to educate Indians at home, the development of high-tech industries fueled by foreign direct investment, and the expansion of research opportunities in India that collectively reduce the ‘‘push’’ that used to send the highly skilled abroad in search of rewarding careers. In fact, there is increasing evidence that Indians who went abroad are returning home. As the second large source country for the highly skilled, these changes within India will have repercussions for the global market for the highly skilled. Students Like China, India is increasing the enrollment of students in Indian higher-education institutions, albeit at a slower rate. During the course of the 1990s, India increased its student enrollment by 47%, climbing from 5.2 million to 7.7 million students. In absolute terms, the largest increase was in the humanities, which added more than a million students, but the natural sciences increased by more than 400,000, representing about one-fifth of the total enrollment in 1999– 2000. At the doctorate level, the role of the natural sciences is more prominent, accounting for more than a third of the doctorate degrees in 1998–1999 (Khadria 2004). Thus, while more than 800 Indians earned science and engineering doctorate degrees in the United States in 2001, the Indian system itself produced more than five times that number (ibid.). Growing capacity of the Indian S&T system In the past, many of these highly skilled graduates from Indian institutions would migrate abroad in search of work in science and technology careers (OECD 2001). While this continues, a significant number remain in India, causing the stock of S&T personnel to increase by 60% above that of the 1990s (Khadria 2004). Almost half of these people were graduates from science studies (such as mathematics, chemistry, biology, and physics), while the largest growth rates were reported for engineers. By the end of the 1990s, the stock of S&T personnel, a basic element to the strengthening of the S&T capacity of India, had reached nearly 8 million. This figure includes a mix of various degrees across a range of fields, making exact comparisons to other countries difficult, although using a more restricted definition of ‘‘researchers,’’ India has approximately the same absolute number as Canada or Korea (95,000), albeit with a much lower per-capita density (OECD 2003a). A key factor behind the advancement of the Indian S&T system has been the development of the IT sector, particularly the software and computer services sectors. The employment of IT professionals in India has increased almost tenfold during the past 10 years (Khadria 2004). This trend is likely to continue as information and communication technologies enable firms in developed countries to ‘‘offshore’’

The Changing Dynamics of the Global Market for the Highly Skilled

207

Table 2 Net migration, cumulative stock, and annual flow estimates of IT labor (software) supply in India (thousands) 2000– 2001 Existing stock (excluding ITES professionals) India: New IT labor / Number of IT professionals leaving India (onsite work) Number of IT professionals returning to India Number of IT professionals

360

2001– 2002

2002– 2003

2003– 2004

2004– 2005

360

429

542

675

133 64

158 64

173 64

192 21

—

20

24

29

429

542

675

875

Note: The above supply summary excludes ITES (IT-enabled services) professionals. Source: From Khadria (2004) and NASSCOM (2002).

work that can be digitized, creating competitive pressures that force other firms to follow suit. An area of particular strength for India is software design and development, which has attracted ‘‘offshored’’ work from a variety of software firms including IBM, Thompson, CAP Gemini Ernst & Young, and Google. The growth of the Indian IT sector is also attributable to the IT downturn in the United States, which forced many Indians on temporary visas to return home. These people left the United States with skills and know-how obtained from working in places such as Silicon Valley as well as with contacts to venture capitalists, U.S. firms, and the broader IT community, especially other fellow Indians who remain in the United States. While this return flow to India is still relatively modest, as opportunities grow in India, it is increasing to the point where India’s NASSCOM (National Association of Software & Service Companies) estimates that the return flow of IT professionals now offsets the outflow (Table 2). R&D investment Thanks to its large pool of talent and relative low cost, India has ramped up its R&D as multinational enterprises such as SAP, Oracle, HewlettPackard, Texas Instruments, Cadence, Analog Devices, Cisco, GE, IBM, Intel, Motorola, DaimlerChrysler, Electrolux, Google, and GE have set up labs (Rai 2004). During 2000–2001, the government performed nearly three-quarters of the estimated USD 19.4 billion (in current PPP) spent on R&D, placing India slightly ahead of Canada and behind Korea in terms of absolute R&D effort, although comparisons of absolute efforts are plagued by the choice of a common currency. Aside from IT the Indian government has focused on biotechnology, where private sources indicate that from a small base, the total investment (including R&D)

208

Andrew Wyckoff and Martin Schaaper

has tripled between 1999 and 2002, with more than half of this investment being directed toward health applications (Chaturvedi 2005). Policy Challenges The market for the highly skilled has transformed from one where demand originated largely from a single buyer, the United States, in the 1990s to one where demand is now more differentiated across buyers, including the European Union, Japan, Canada, and Australia as well as the large supply countries themselves— China and India. This shift is just beginning and will probably move in fits and starts, but several factors suggest that it will continue and strengthen, leading to the formation of a global market for the highly skilled (Harris 2004). This evolution of the market could have a wide range of implications for individual national innovation systems, macroeconomic policy, the generation and flows of knowledge, and correspondingly the shape and operation of the network through which knowledge is shared. National Innovation Systems As the structure of economies shifts toward more knowledge-intensive activities and countries seek to strengthen their innovative capacity, the demand for the highly skilled, particularly scientists and engineers, will increase. In the short term, where the next generation of researchers is already enrolled in graduate school, there is no choice but to compete in this global market for the highly skilled, representing a new dimension in the national innovation systems as they adapt to these new market dynamics. Shifts in these markets create an opportunity for nimble countries to enter and seize part of the market for the highly skilled that has been dominated by the United States. While a potential problem for the United States, which depends on the inflow of the highly skilled from abroad, it also creates a challenge for continental Europe and Japan, which lack sufficient flexibility in their higher-education and S&T systems and a social environment that is geared toward accommodating the highly skilled from abroad. In this sense, immigration-based countries such as Australia and Canada as well as the United Kingdom may be best positioned to take advantage of these changes in the global market. This will require investments on their part to ensure that their research capabilities, at least in a few select fields, are at the technological and scientific frontier. As countries address some of the issues that have pushed their highly skilled abroad, increasingly pull factors will play a larger role in the dynamics of the international flows of the highly skilled (Khadria 2001). A key pull factor for attracting the highly skilled from abroad is world-class universities. This necessitates a change

The Changing Dynamics of the Global Market for the Highly Skilled

209

in mind-set for many countries that tend to view their universities as being a purely national resource and not part of an increasingly global science and technology network and a source for innovative activity (OECD 2004a). This change in mind-set requires the adoption of a wide range of policies, including enacting accommodating immigration laws, establishing a supportive social structure for the foreign students, and providing financial aid for these students. But most fundamentally, it is the quality of the education and research that pulls in the top students, especially in science and engineering fields. The role of universities as an educational institutional for citizens, centers of research, and a pole that attracts the highly skilled underscores the need for greater support to universities as part of the overall national system of innovation. In the longer term, another response to this increasingly global market is to increase the indigenous supply of the highly skilled. As the supply from large source countries such as India and China decreases as the push to leave these countries declines, the need to attract citizens into science and technology becomes more imperative and requires a focus on the entire supply pipeline, from primary and secondary schooling to university education and doctoral training. During the past few years, countries have implemented a range of initiatives to stimulate the domestic supply of graduates and improve the attractiveness of research careers. Some of the key initiatives include raising the interest in and awareness of science, especially among youths aged seven to 12; improving teacher training so that these subjects are better taught and are of greater interest; revising the curricula to make programs more responsive to student needs and demands from industry; recruiting women and other underrepresented populations; increasing funding for doctoral students and postdocs; and creating more autonomous research positions (OECD 2004b). The two large challenges that confront policy makers as they try to attract more people into science and technology careers are how to attract and retain the interest of girls and young women and how to improve the typical career path for scientists and engineers, which compared to other professions is increasingly unattractive both financially and socially. As the global market for the highly skilled develops, it may exert downward pressure on wages, further fueling this problem. The three sectors that have generated much of the demand for the highly skilled in the OECD member countries—ICT, aerospace, and biotech—are adjusting to a much slower pace of growth than existed in the 1990s, and many of the growing markets for the output of these sectors lie abroad. This will lead to the further integration of national systems of innovation into a global system as products are tailored for foreign markets, foreign affiliates are set up to service these markets, and researchers are encouraged to become more internationally mobile and more global in their outlook (NRC 2004).

210

Andrew Wyckoff and Martin Schaaper

As these trends develop, the notion of a national system of innovations may become less and less relevant. Increasingly, the best national outcomes may be obtained from tapping into talent and expertise abroad, especially in the precompetitive phase of research, where the ability to appropriate the results is difficult. Rather than trying to preserve some second-rate national effort or reach some nationally or regionally based target, it is better to gain access or join international efforts with an eye toward using the results to enhance national competitiveness (Griffith et al. 2004). This necessitates more of a global perspective in the development of national S&T policy, an openness and receptivity to ideas and innovations abroad, and active participation in the global network. Macroeconomic Effects A likely scenario for the future is that the global demand for the highly skilled increases as supply stays roughly constant due to the retirement of baby-boom researchers, offsetting possible additions from both indigenous sources (women, minorities) and developing countries. Under these conditions, the price for the highly skilled should begin to equalize globally and will then increase over the longer term. The relative low cost the United States enjoyed in the 1990s when there was little global competition for this cadre of talent is over, and it will cost more to attract and retain these people. These costs are broader than merely wages but include advancement opportunities, research funding, and in general the quality of work. Thus, the salutatory effect the inflow of the highly skilled had in sustaining the U.S. IT boom of the 1990s will not be as pronounced in the future. This said, the development of a truly global market for this segment of the labor market will be beneficial to the economic performance of smaller countries (e.g., Australia, Canada, Finland), which on their own could not transform this segment of the labor force into a global market. As the cost of labor increases, those involved in the business of research will try to minimize these costs through the substitution of capital. This will further fuel the trend of more intensive use of information and communication technologies to increase the productivity of innovative activities and force an upgrading to research facilities, both public and private, that could put pressure on budgets and government spending. These costs to the developed world should be more than offset by gains achieved by the developing world, especially China and India, as their integration into the global science and technology community continues, and will have beneficial economic effects on their domestic economic development as well as creating huge new markets for international trade, which should stimulate global economic growth. The broadening of innovative activities across the globe will lead to an overall increase in the rate of innovation, further stimulating growth and productivity.

The Changing Dynamics of the Global Market for the Highly Skilled

211

Changes in the global market for the highly skilled will also help to diffuse and push best practices, forcing policy in a number of areas to be more accommodating. This is already evident in India, where Indians returning home after having been exposed to Silicon Valley have been instrumental in influencing Indian government policy in areas such as venture capital, telecom communications deregulation, and preferential tax treatment (Saxenian 2002). These pressures should be useful for breaking down sheltered enclaves, injecting new ideas, and fostering more policy experiments for improving the innovative climate. As the S&T system becomes more global, this will further link the economies globally, necessitating even more global coordination of economic policies. To support this improved coordination, the development of better data series that track the evolution, breadth, and depth of the interdependencies is needed. This is a huge task covering a wide range of data series, but two priority areas are the need to improve global understanding of the activity of multinational enterprises and the measures of the highly skilled, especially their international flows.4 A Global Knowledge Network The expanding global market for the highly skilled and the establishment of increasingly important scientific and technical expertise in places such as China and India will necessitate a reconfiguration of the knowledge network, extending its geographic spread and shifting its locus away from the United States, and developed OECD countries in general, to a broader set of partners. Technically, this will require an integration of these new actors into the global network, but more fundamentally it will require a social inclusion of these new partners that could represent a challenge, given their different cultural and economic positions. As the number of global locations for innovation increases and their sophistication develops, the physical movement of the highly skilled could be increasingly replaced with the global movement of ideas and knowledge through a cyberinfrastructure, while the people increasingly remain in the same place.5 For example, an MNE faced with a growing market in China and an increasing supply of highly educated researchers in China might respond by establishing an R&D center in China, rather than bringing the highly skilled Chinese, on a permanent basis, to its more expensive U.S. or E.U. R&D establishment. As the difference in cost of doing research in OECD countries versus China or India declines over the medium term, MNEs are likely to take steps to encourage the flow of knowledge via temporary postings of their staff (moving in both directions) and other short-term methods or create project-based teams that draw on people from all over the world, but again on a temporary basis. In this scenario, the globalization of innovation will lead to a short-term increase in the circulation of the highly skilled, followed by a long-term decline as the highly skilled in the key source

212

Andrew Wyckoff and Martin Schaaper

countries in Asia are increasingly less attracted to moving to the United States or the European Union on a long-term basis. This observation is supported by the current behavior seen in the United States or Japan, where a very low percentage of the highly skilled live abroad or have plans to do so, suggesting that scientists would rather stay at home if suitable opportunities exist (Burrelli 2004). If the future lies in a decline in the physical circulation of the highly skilled and an increase in the circulation of knowledge, it has several implications for policy. It underscores the need to develop a supply of highly skilled human capital from the indigenous population, and thus the need to fuel an interest in science among current elementary and secondary students. Immigration systems will need to adapt to the needs of shortterm (under two years) transfers of the highly skilled. Developed countries will need to tap into methods that ensure the effective transfer of ideas and knowledge and create forums for their circulation. Close linkages built through mechanisms such as shared research programs and scholar exchanges between universities in China and India will be increasingly important. These new locations of innovative activity should be integrated into the global science community through the extension of the cyberinfrastructure that links scientists and engineers, which is predominantly configured around an OECD research community. As David (forthcoming) points out, technically this extension is relatively easy compared to the social and cultural challenges. This integration could be eased through tapping into the ‘‘transnational technical communities’’ that increasingly link the countries of the world (Saxenian 2002). To achieve this will require a better understanding of the social structure and interaction of these various communities. As the key player funding and performing R&D, the home to many of the world’s premier universities, the headquarters for many of the world’s innovation-intensive MNEs, and the architect of the information network that links the global research community and as a country that has been at one time or another home to much of the world’s diaspora, the United States has a pivotal role to play in this transition and the formation of this more global knowledge network. Conclusion Changes since 2000 have redrawn the boundaries and points of exchange in the global science and technology system, which is best depicted by the human capital that supports this system. Various events have led to changes that are causing the dynamics of these flows of human capital to change, where other developed countries such as Japan and Europe as well as the large source countries of China and India now compete more directly in the global market for the highly skilled.

The Changing Dynamics of the Global Market for the Highly Skilled

213

These changes are important for the global science and technology system as the foci of knowledge and research broaden on a more global scale, requiring adjustments to accommodate the new nodes and move away from a less concentrated system, where the United States played a key role. It would be an exaggeration to say that the United States is at risk because of these changes in the global market. The U.S. S&T system enjoys many advantages and has shown an ability to adapt to changing circumstances, which suggest that it will react to these new challenges. Nevertheless, these changes do require attention and a reaction by policy makers, something that can be difficult to achieve when there are many immediate policy issues that are competing for the same attention and resources. The changes in the global market for the highly skilled have implications for the global science and engineering community and any country that is trying to bolster its innovative capacity by building its S&T workforce. These changes are important and have a momentum that will carry them forward for some time. For this reason and the fact that this trend toward a more global system has many very positive attributes, national policy makers should not resist this development. Rather, they should embrace it and react to it by modifying their domestic policies (e.g., immigration policies, education systems) to be more accommodating and alter their foreign policies to reflect an S&T system that is more global and less concentrated in a few leading countries. Notes 1. ‘‘Highly-qualified’’ is defined in terms of achieving an educational attainment equal to ‘‘the first stage of tertiary education (not leading directly to the award of an advanced research qualification)’’ (ISCED 5) or ‘‘the second stage of tertiary education (leading directly to the award of an advanced research qualification)’’ (ISECD 6). See http://www.uis.unesco.org. 2. A break in series due to improved measurement methods occurs between 1999 and 2000, inflating the growth rate during this period. 3. Personal communication with Bangwen Chen, professor at the Management Institute of the Huazhong University of Science & Technology, 9 April 2004. 4. Initial international work in these has begun. See Barnabe (2003) and Auriol (2004). 5. This observation and its implication for policy was made in a comment received from Anthony Arundel of Merit.

References AIP (American Institute of Physics) (2004). ‘‘House Science Committee Reviews Visa Process.’’ Bulletin of Science Policy News 24 (March 1). Armstrong, John A. (2003). ‘‘The Foreign Student Dilemma.’’ Issues in Science and Technology (Summer).

214

Andrew Wyckoff and Martin Schaaper

Auriol, Laudeline (2004). ‘‘Proposal for an OECD Pilot Project to Develop Surveys of Careers of Doctorate Holders (CDH).’’ Presented at the OECD Working Party of Experts on Science and Technology Indicators, June 21–23. Barnabe, Richard (2003). ‘‘Seeing the Whole of the Elephant: A Proposed Experiment on Measuring the Activities of Multinational Enterprises.’’ Presented at the Conference of European Statisticians, Geneva, June 10–12, http://www.unece.org/stats/documents/ces/2003/13.e .pdf. Barrere, Rodolfo, Lucas Luchilo, and Julio Raffo (2004). ‘‘Highly Skilled Labour and International Mobility in South America.’’ STI Working Paper 2004/10, www.oecd.org/sti/workingpapers. Brown, Gordon (2005). ‘‘Putting Britain at the Forefront of Global Trade.’’ The Financial Times (February 4). Buckley, Chris (2004). ‘‘Let a Thousand Ideas Flower: China Is a New Hotbed of Research.’’ The New York Times (September 13). Buderi, R. (2003). ‘‘Technological McCarthyism.’’ Technology Review (July/August). Burrelli, Joan (2004). ‘‘Emigration of US-born S&E Doctorate Recipients.’’ SRS InfoBrief (June). Chaturvedi, Sachin (2005). ‘‘Dynamics of Biotechnology Research and Industry in India: Statistics, Perspectives and Key Policy Issues.’’ STI Working Paper No. 2005/6, http://www.oecd .org/sti/working-papers. David, Paul (forthcoming). ‘‘Towards a Cyberinfrastructure for Enhanced Scientific Collaboration.’’ Presented at Advancing Knowledge and the Knowledge Economy Conference, National Academy of Science, Washington, DC, January 10–11, 2005, http://www .advancingknowledge.com. Dumont, Jean-Christophe, and Georges Lemaıˆtre (2005). ‘‘Counting Immigrants and Expatriates: A New Perspective.’’ OECD, Social Employment and Migration Working Paper No. 25, http://www.oecd.org/dataoecd/27/5/33868740.pdf. European Commission (2002). ‘‘Presidency Conclusions: Barcelona European Council, 15 and 16 March 2002.’’ SN 100/02, Brussels. ——— (2003). ‘‘Investing in Research: An Action Plan for Europe,’’ COM(2003)226 final, and ‘‘Europe Must Take Action to Compete in Global Market for Researchers.’’ CORDIS News (November 25), http://www.cordis.lu.era/mobility.htm. Griffith, R., Rupert Harrison, and John Van Reenen (2004). ‘‘How Special Is the Special Relationship? Using the Impact of US R&D Spillovers on UK Firms as a Test of Technology Sourcing.’’ Centre for Economic Performance Discussion Paper dp0659. Harris, Richard G. (2004). ‘‘Labour Mobility and the Global Competition for Skills: Dilemmas and Options.’’ Prepared for Roundtable on International Labour Mobility, Industry Canada, Ottawa, February 27, 2004. IIE (Institute of International Education) (2004). ‘‘Open Doors Report.’’ November, http:// opendoors.iienetwork.org. IMD (2004). World Competitiveness Yearbook 2004. http://www02.imd.ch/wcc/.

The Changing Dynamics of the Global Market for the Highly Skilled

215

Johnson, Jean (2001). ‘‘Human Resource Contributions to US Science and Engineering from China.’’ SRS Issue Brief, January. Khadria, Binod (2001). ‘‘Shifting Paradigms of Globalisation in the 21st Century.’’ International Migration 39, no. 5, Special Issue 1. ——— (2004). ‘‘Human Resources in Science and Technology in India and the International Mobility of Highly Skilled Indians.’’ STI Working Paper 2004/7, http://www.oecd.org/sti/ working-papers. Mahroum, Sami (2002a). ‘‘Europe and the Prospect of Brain Drain.’’ The IPTS Report 66, July, http://www.jrc.es/pages/iptsreport/vol66/english/STR1E666.html. ——— (2002b). ‘‘US Science and the Fear of a Backlash: The Possible Fallout of September 11th on the Immigration of Scientists and Engineers to the US.’’ GaWC (Globalization and World Cities Study Group and Network) Research Bulletin 79, http://www.lboro.ac.uk/ gawc/rb/rb79.html. Marsh, Peter (2004). ‘‘World’s Manufacturers March into China.’’ The Financial Times (June 21): 11. METI (Ministry of Economy, Trade and Industry) (2003). Japanese White Paper on International Trade, ‘‘Section 2: Utilisation of Excellent Overseas Human Resources.’’ Moris, Francisco (2004). ‘‘US–China R&D Linkages: Direct Investment and Industrial Alliances in the 1990s.’’ SRS InfoBrief (February). NAE (National Academy of Engineering) (forthcoming). ‘‘Engineering Research and America’s Future: Meeting the Challenges of a Global Economy.’’ http://www.nap.edu. NASSCOM (National Association of Software and Service Companies) (2002). ‘‘Knowledge Professionals.’’ In Strategic Review 2002, Chap. 5, pp. 63–82. New Delhi. NCEE (National Commission on Excellence in Education) (1983). A Nation at Risk: The Imperative for Educational Reform. http://www.ed.gov/pubs/NatAtRisk/risk.html. NRC (National Research Council) (2004). Preparing Chemists and Chemical Engineers for a Globally Oriented Workforce. Washington, DC: National Academies Press, www.nap.edu. NSB (National Science Board) (2004). The Science and Engineering Workforce: Realizing America’s Potential. Washington, DC. NSF (National Science Foundation) (2004). Science and Engineering Indicators 2004. Arlington, VA. Nye, J. (2004). ‘‘You Can’t Get Here from There.’’ The New York Times (November 29). OECD (1970). Gaps in Technology. Paris: OECD. ——— (2000). A New Economy? The Changing Role of Innovation and Information Technologies. Paris: OECD. ——— (2001). International Mobility of the Highly Skilled. Paris: OECD. ——— (2003a). Science, Technology and Industry Scoreboard. Paris: OECD, http://www .oecd.org/sti/scoreboard. ——— (2003b). The New Economy: Beyond the Hype. Paris: OECD. ——— (2004a). Internationalisation and Trade in Higher Education. Paris: OECD.

216

Andrew Wyckoff and Martin Schaaper

——— (2004b). Science, Technology and Industry Outlook. Paris: OECD. ——— (2004c). Main Science and Technology Indicators. Paris: OECD. ——— (2004d). Learning for Tomorrow’s World: First Results from PISA 2003. Paris: OECD. Rai, Saritha (2004). ‘‘From India, Genius on the Cheap.’’ International Herald Tribune (December 15): 12. Saxenian, Anna Lee (1999). ‘‘Silicon Valley’s Skilled Immigrants: Generating Jobs and Wealth for California.’’ Research Brief No. 21, Public Policy Institute of California, San Francisco. ——— (2002). ‘‘Transnational Communities and the Evolution of Global Production Networks: The Cases of Taiwan, China and India.’’ Industry and Innovation, Special Issue on Global Production (December). Science & Government Report (2004). ‘‘New Steps Are Urged to Ease Entry of Foreign Scientists and Students to the US.’’ June 1. Schaaper, Martin (2004). ‘‘An Emerging Knowledge-Based Economy in China? Indicators from OECD Databases.’’ STI Working Paper 2004/4, http://www.oecd.org/sti/workingpapers. Sheehan, Jerry, and Andrew Wyckoff (2003). ‘‘Targeting R&D: Economic and Policy Implications of Increasing R&D Spending.’’ STI Working Paper 2003/8, http://www.oecd.org/sti/ working-papers. Song, Weiguo, and Zhaohui Xuan (2004). ‘‘Preliminary Analysis of China’s Doctor Education.’’ OECD Careers of Doctorates Workshop, September 2004. WEF (World Economic Forum) (2004). ‘‘Global Competitiveness Report, 2003–04.’’ http:// www.weforum.org.

14 Knowledge in Space: What Hope for the Poor Parts of the Globe? Jan Fagerberg

Introduction The topic of this chapter, the role knowledge plays in catch-up/development (including possible policy implications), has been a controversial one for several decades. Already more than a century ago Karl Marx pointed to the richest countries at the time as role models for the poor parts of the world. Writing during the first half of the previous century, the highly unorthodox economist Thorstein Veblen presented an intriguing analysis of the facilitating role played by (modern forms of) knowledge in German catch-up toward the then world leader, the United Kingdom. Fifty years ago the economic historian Alexander Gerschenkron returned to the topic with a somewhat less optimistic approach, emphasizing in particular the stringent requirements for its successful exploitation and the derived need role of policy to help overcome the obstacles for knowledge exploitation and catch-up. However, in spite of the emphasis placed on knowledge by Veblen, Gerschenkron, and others, many analyses by economists of cross-country differences in growth and development have not had much to say about knowledge. One important reason for this is that it has been common among economists to regard knowledge as a so-called ‘‘public good,’’ e.g., something that is freely available to everyone everywhere (and hence cannot logically be invoked in explanations of, say, cross-country differences in development). The relevance of this public-good approach to knowledge is discussed in the next section. Arguably, much economic reasoning on growth and development operates with a much too narrow understanding of knowledge and the economic processes in which knowledge take part, and therefore fails to understand the role of knowledge in catch-up and development. A broader perspective is needed, and the purpose of this chapter is to contribute toward that aim. We start by discussing how this issue has been dealt with by some classic studies of catch-up. First we consider the contributions by Thorstein Veblen, Alexander Gerschenkron, and others on European catch-up prior to the First World War. The

218

Jan Fagerberg

main point of interest here is the interpretation of the German catch-up with the United Kingdom and the role of knowledge, policy, and institutions in this context. Second, there is a large literature on Asian catch-up, particularly Japan, but increasingly also on Korea, Taiwan, and other countries that to a varying degree have attempted to follow the Japanese route, which we also consider. The argument that an activist, ‘‘developmental state’’ has been an efficient means to successful technological catch-up has been a central focal point in much of this literature. On the basis of these studies we discuss some attempts that have been made to provide a more general framework for the study of technological catch-up. Concepts such as ‘‘social capability’’ and ‘‘absorptive capacity’’ have been central in this literature. It is suggested that work in this area, and policy design, would benefit from a clearer distinction between the capacity to generate new knowledge and the capacity to exploit it commercially. An example of how this may be done in practice, based on data for 100 countries in the past decade, is presented, and the implications of this for development are discussed. Knowledge, Growth, and Development: Received Wisdom Reconsidered Intuitively, most people easily accept the idea that knowledge and economic development are intimately related. However, this is not the way different levels of development used to be explained by economists. From the birth of the so-called ‘‘classical political economy’’—a term invented by Karl Marx—two centuries ago, what economists have focused on when trying to explain differences in income or productivity is accumulated capital per worker. Similarly, differences in economic growth have been seen as reflecting different rates of capital accumulation. This perspective arguably reflects the important role played by ‘‘mechanization’’ as a mean for productivity advance during the so-called (first) industrial revolution, the period during which the frame of reference for much economic reasoning was formed. Closer to our own age Robert Solow adopted this perspective in his so-called ‘‘neoclassical growth theory’’ (Solow 1956). The theory predicted that, under otherwise similar circumstances, investments in poor countries (e.g., those with little capital) would be more profitable than in the richer ones, so that the former would be characterized by higher investment and faster economic growth than the latter. As a consequence of this logic, a narrowing of the development gap (so-called ‘‘convergence’’) should be expected. Based on another argument borrowed from the classical political economists (reflecting their opposition toward mercantilist politics and feudal privileges), such convergence was by many economists deemed all the more probable, the less the state interfered with working of the ‘‘free’’ market. This gave birth to a particular approach to development policy, termed the ‘‘market friendly’’

Knowledge in Space

219

approach associated, advocated by international agencies such as the IMF and the World Bank (see, for instance, World Bank 1993). The prediction that global capitalist dynamics would be accompanied by a convergence in income and productivity between initially poor and rich countries was based on a liberal and optimistic view on global economic development. However, it is rare to see a prediction that is so completely rejected by the evidence as this one is. In fact, the history of capitalism from the industrial revolution onward is one of increasing differences in productivity and living conditions across different parts of the globe. According to one source, 250 years ago the difference in income or productivity per head between the richest and poorest country in the world was approximately 5:1, while today this difference has increased to 400:1 (Landes 1998). But in spite of this long-run trend toward divergence in productivity and income, there are many examples of (initially) backward countries that—at different times—have managed to narrow the gap in productivity and income between themselves and the frontier countries, in other words, to ‘‘catch up.’’ Japan in the decades before and after the Second World War and the ‘‘Asian tigers’’ more recently are obvious examples. How to explain this diversity in patterns of development? Is it related to a superior ability to develop and/or exploit knowledge in the successful countries, as many perhaps would suspect? What role—if any—did policy play in this context? As noted in the introduction, theoretical work for a long time tended to ignore the role of knowledge in development. This was caused not only by the fact that economists’ focus for historical reasons was elsewhere. It also had to do with a particular view on knowledge that came to dominate economics, that is, knowledge as a socalled ‘‘public good’’ or a body of information, freely available to all interested, that can be used over and over again (without being depleted). Arguably, if this is what knowledge is about, it should be expected to benefit everybody all over the globe to the same extent and hence cannot be invoked as an explanation of differences in growth performance. Hence, following the logic, the real reasons behind such differences must rest elsewhere. Moreover, if everybody benefits to the same extent, why should anybody care to provide it? For a long time many economists found this question so perplexing that they chose to ignore knowledge altogether (i.e., regard it as a factor that is alien to economic reasoning, or ‘‘exogenous’’ as it is conventionally expressed). More recently economists such as Paul Romer have put an end to this practice by suggesting that knowledge, in the above ‘‘public good’’ sense, is a byproduct of investments that firms undertake in order to develop new products and services (Romer 1990). The reason why, following this view, firms find it profitable to do so is that intellectual property rights (patents, etc.) give them sufficient protection

220

Jan Fagerberg

to secure a healthy private return on their investments. The social returns are, at least on average,1 assumed to be even higher, enhancing the pool of public, freely available knowledge and spurring growth. If such pools of knowledge can be assumed to be ‘‘national’’ in character, models based on this perspective (so-called ‘‘new growth theory’’) might yield predictions consistent with the observed longrun tendency toward divergence in GDP per capita (with large countries—with large ‘‘national’’ knowledge stocks—in a particularly good position). However, such an assumption would be extremely difficult to justify, given the perspective on knowledge underlying the approach (a body of information). Indeed, the logic of the argument clearly suggests that such freely available knowledge would not be bound to (geographical) context and hence should be expected to benefit all countries. So what is wrong? Should we accept that knowledge is not an important factor behind the vast differences in income across different parts of the globe? Or is something fundamentally wrong with the way knowledge is conceived by the theoreticians? We put our bets on the latter. When Robert Solow and others started to model growth more than fifty years ago, there was not a lot of work available on knowledge and innovation in firms. However, during the past two decades we have seen a proliferation of work in this area, with several big surveys, numerous case studies, and a lot of interpretative work, and we now know a good deal more about how firms search for, develop, and use new knowledge. Surprisingly, this new ‘‘knowledge on knowledge’’ does not seem to have been exploited much by the theoreticians in their attempts to construct models of knowledge-based growth. If they had, they would have found that the type of knowledge on which they focus, e.g., codified information that is patented and traded in markets (or not patented and hence provided for free), is only one among several types of economically relevant knowledge (albeit an important one). In fact, there is now a large body of research showing that firms generally do not regard patenting as an important way to protect their knowledge, nor do they see universities and public research institutes as very important sources of information and/or knowledge (Eurostat 2004; Foray 2004; Granstrand 2004). This does not imply that there may not be segments within certain sectors or industries that are different in these respects (the biotechnology industry is the prime example). But the general picture is a different one. The truth of the matter is that in most areas of knowledge, there is a long way from scientific discoveries to commercial exploitation. Lags of several decades or more are not uncommon (Rogers 1995; Fagerberg 2004). Technological activities of firms seldom take abstract scientific principles as point of departure and search for commercial applications (although that may happen). The general pattern is that of a perceived need among customers, a problem that needs to be solved, which generates a search for relevant knowledge. Research emphasizes that, in most cases,

Knowledge in Space

221

firms have only imperfect knowledge on the relevant options in front of them and that they tend to be myopic, searching internally at first, then in the neighborhood of their existing competence/network (Nelson and Winter 1982; Dosi 1988; Cohen and Levinthal 1990; van der Ven et al. 1999). Consistent with this, the most highly valued external sources are typically customers and suppliers. While it is true that knowledge codification has been on the increase for centuries and that this continues at an accelerated rate in the present ICT era, much economically relevant knowledge is not of this form. For a firm to profit from knowledge, whether through exploitation of existing or creation of new knowledge, what is required is the ability to combine many different kinds of knowledge/capabilities, of which some may not be codified and have little to do with science or technology in the received sense. This challenge holds for developed and developing country firms alike, but it is an especially tough one for the latter, and we return to this later in the chapter. Lessons from the Study of European Catch-Up The discussion of continental Europe catch-up illustrates nicely some of the central issues in the catch-up literature.2 Veblen (1915), who initiated the discussion, put forward the argument that recent technological changes had altered the conditions for industrialization in latecomer economies. In earlier times, he argued, diffusion of technology had been hampered by the fact that technology was mostly embodied in persons, so that migration of skilled workers was a necessary prerequisite for the spread of technology across different locations. However, with the advent of the ‘‘machine technology,’’ this logic had changed (ibid., p. 191). In contrast to the conditions that had prevailed previously, Veblen argued, this new type of knowledge ‘‘can be held and transmitted in definite and unequivocal shape, and the acquisition of it by such transfer is no laborious or uncertain matter’’ (ibid.). Although Veblen did not use the terminology that is now commonly applied to the process he described, it is pretty clear what he had in mind. Effectively, what he was arguing is that while technology was previously ‘‘tacit’’ and embodied in persons, it later became more ‘‘codified’’ and easily transmittable. Hence, catch-up should be expected to be relatively easy and was under ‘‘otherwise suitable circumstances’’ largely ‘‘a question of the pecuniary inducement and . . . opportunities offered by this new industry’’ (ibid., p. 192). Veblen mentions factors such as the ‘‘funds available for investment’’ (ibid., p. 186), a sufficient supply of ‘‘educated men’’ (ibid., p. 194), as well as a ‘‘sufficiently well-instructed force of operative workmen’’ (ibid., p. 192), which, however, did not have to be particularly well educated or trained (ibid., p. 188). Since the latecomers could take over the new technology ‘‘ready-made,’’

222

Jan Fagerberg

without having to share the costs of its development, this might be expected to be a very profitable affair (ibid., p. 249). This being the case, Veblen predicted that other European countries, e.g., France, Italy, and Russia, would soon follow suit (he also mentioned the case of Japan). While in Veblen’s interpretation German catch-up was a relatively easy affair, the economic historian Alexander Gerschenkron (1962) took a different view, emphasizing the difficulty of the matter. While, he argued, technology when Britain industrialized had been small scale, and hence institutionally not very demanding, these conditions were radically altered in the 19th century when Germany started to catch up. What Gerschenkron particularly had in mind was the seemingly inbuilt tendency of modern technology to require ever larger and more complex plants (static and dynamic economies of scale), with similarly changing requirements with respect to the physical, financial, and institutional infrastructure. He argued that, because of the high potential rewards from successful entry and the heavy transformation (modernization) pressure on the rest of the economy it helped to generate, it was of paramount importance for the latecomer to target such progressive, dynamic industries and compete globally through investing in the most modern equipment/plants: ‘‘To the extent that industrialization took place, it was largely by the application of the most modern and efficient techniques that backward countries could hope to achieve success, particularly if their industrialization proceeded in the face of competition from the advanced country’’ (ibid., p. 9).3 However, to succeed in this endeavor, catching-up countries had in Gerschenkron’s view to build up new ‘‘institutional instruments for which there was little or no counterpart in the established industrial country’’ (ibid., p. 7). The purpose of these institutional instruments would be to mobilize resources to undertake the necessary changes at the new and radically enlarged scale that modern technology required. His favorite example was the German investment banks (and similar examples elsewhere in Europe), but he also admitted that, depending on the circumstances, other types of institutional instruments, such as for instance the government (in the Russian case), might conceivably perform the same function.4 Surprisingly, perhaps, he did not (nor did Veblen) put much emphasis on the achievements made by Germany in other areas, such as the educational sector, and in pioneering the development of an R&D infrastructure. With hindsight it is clear that Germany pioneered the development of a new, science-based model of industrial development, above all in the chemical industry, in which science, R&D in firms, cooperative R&D, and public–private sector cooperation came to play an important if not decisive role (Bruland and Mowery 2004). This historically new system, which fitted well with Gerschenkron’s emphasis on new ‘‘institutional instruments for which there was little or no counterpart in the established industrial

Knowledge in Space

223

country,’’ was later to be imitated by the other major catching-up country of the time, the United States. It also greatly influenced the policies of other catching-up economies, notably Japan. It is tempting to use Veblen’s and Gerschenkron’s highly different accounts of German catch-up to make a tentative classification of the relationship between catch-up and knowledge. The type described by Veblen assumes that knowledge (technology) is easily available/transferable, that it is not very demanding in terms of skills or infrastructure, that social resistance/inertia is not a major problem, and that market forces are able to take care of the necessary coordination without largescale involvement of external ‘‘change agents.’’ In contrast there is the Gerschenkronian case, in which knowledge (technology) transfer is so demanding in terms of skills/infrastructure, or social resistance/inertia so massive, that market forces, if left alone, are considered unlikely to lead to success, and some degree of active intervention in markets by outsiders, being private organizations or parts of government, is consequently deemed necessary. The Asian Experience Similar arguments to those advanced by Veblen and Gerschenkron have also played a role in the discussions of Asian catch-up in the post–Second World War period. The primary examples, in addition to Japan, are Korea, Singapore, and Taiwan. Although some observers have attempted to classify these as Veblen-type catch-up stories (World Bank 1993), there is by now an abundant literature showing that the catch-up strategies applied are much closer to the Gerschenkronian scheme (Johnson 1982; Amsden 1989; Wade 1990; Shin 1996). There are many accounts of Japanese catch-up. The so-called Meiji-restoration in 1868 provides a natural starting point. What happened in 1868 was that a fraction of the ruling elite established a new regime, with the explicit purpose of strengthening the economy (through catching-up) and the military strength of the state, which at the time was strongly challenged by Western imperialism (Beasley 1990). ‘‘A rich society and a strong army’’ was the slogan of the day. Since Japan lacked other ‘‘modernization’’ agents, the government (bureaucracy) took on the challenge. It modernized the legal system, the physical infrastructure, and the educational system, initiated new businesses (that later on were privatized) in industries that were deemed strategically important, etc. Universities, colleges, and research centers were also founded, often with a bent toward engineering and applied science. While the public sector played a vital role, particularly in the initial phase, private initiatives and cooperation between public and private actors became gradually more important. Much of the initiative came to rest with a number of emerging family-owned

224

Jan Fagerberg

business groups, the Zaibatsus, in interaction with the bureaucracy and the military. During the First World War and the period that followed, the Japanese economy underwent a rapid transformation, with machinery and other ‘‘heavy industries’’ taking over from food processing and textiles as leading sectors. R&D activity also soared, partly for military needs, and was, according to one source (Odagiri and Goto 1996) well above 1% of GDP in the early 1940s. The defeat of Japan in the Second World War changed the power structure in Japanese society by eliminating two of the three contending power centers in society, the military and the (owners of the) Zaibatsus, hence giving a boost to the bureaucracy that once more took on the challenge of gearing the economy and the society at large toward economic catch-up with the West. The sequence of events from the late 19th century somehow repeated itself, with a very important role for the state (and—in particular—the Ministry for Trade and Industry, MITI) in the early phase and a growing role for private initiatives (and business groups) as the economy grew stronger. The exact role of the government versus private actors in the various phases of Japanese economic growth is a matter of considerable controversy. Suffice it to say that government/bureaucracy intervention, through activist economic, industrial, and trade policy (protectionism), was very important, especially in early phases. Although not everything it touched turned into ‘‘gold’’ and sometimes its interventions were strongly resisted by private business (and for perfectly good reasons), it is no doubt that it contributed significantly to gear the attention of private business to catch up with the West. In this way Japanese industry soon rose to the productivity frontier in its chosen fields, first in the steel industry and shipbuilding and later in cars and (consumer) electronics. Although Japanese innovation in the catch-up phase also included a large number of product innovations, especially of the minor type (adaptations to demand), the main emphasis was on process innovations, particularly of the organizational type (the ‘‘just-in-time system,’’ for instance), that allowed for simultaneous exploitation of scale economics and flexibility, leading to high throughput, efficient inventory management, high quality/reliability, and a proven ability to adjust to the needs of the end user. The Japanese experience generated a lot of interest in other developing countries, particularly in Asia, that considered the policies and practices pursued by the Japanese as a possible model for their own catch-up towards Western levels. The prime examples are, as noted, Korea, Singapore, and Taiwan. What these countries have in common is that they have caught up very rapidly, underwent extensive structural change, and—finally—established themselves as among the major producers (and exporters) internationally in the most technologically progressive industry of the

Knowledge in Space

225

day, electronics (broadly defined). The government appears to have played a very important role in these processes. Everywhere a lot of emphasis has been placed on the expansion of education, particularly of engineers (Lall 2000). In the early phases, governments in Korea and Taiwan intervened heavily with tariff protection, quantitative restrictions, financial support, etc. to benefit the growth of indigenous industries in targeted sectors. Singapore is a special case, since its government has relied heavily on inward foreign direct investment (FDI) in its industrialization efforts, and targeting hence has had to be achieved through selective FDI policies (Lall 2000). However, in all countries, targeting production for exports and rewarding successful export performance was very important. More recently all countries have put a lot of emphasis on policies supporting R&D and innovation. With respect to the Gerschenkronian scheme, the experiences of these Asian economies fit well with the emphasis of targeting the technologically most progressive industries. On a general level, in all four countries the state (bureaucracy) played a very important role at an early stage. However, as noted above, this was done in different ways. For instance, in both Japan and Korea credit rationing by the state (so-called ‘‘directed credit’’) was extensively used to persuade private business to go along with the government’s objectives, while this mechanism played virtually no role in Taiwan (which underwent a financial liberalization early on). In the Taiwanese case the government had to rely on other instruments such as stateowned firms (which came to play an important role) and, in particular, heavily supported ‘‘intermediate institutions’’ (R&D infrastructure etc.) with mixed public/ private sector participation. Moreover, while industrialization in Japan, and in the United States and Germany before it, was mainly geared toward the home market, exports played a similar role in the catch-up strategies of the three ‘‘tigers.’’ This may, arguably, have to do with the fact that the domestic markets in the latter in many cases were too small to support large-scale industrialization efforts, but the gradual reduction in barriers to trade during the post–World War II period also played an important role (Abramovitz 1994). A Conceptual Framework? Another strand of catch-up research focuses on the macro level. It asks questions of the type, to what extent has catch-up or convergence actually occurred and how may this be explained. As mentioned in the introduction, an important finding in this literature is that the long-run trend since the British industrial revolution points to divergence, not convergence, among capitalist economies. It has also been shown that these trends differ a lot between time periods. For example, one such period in

226

Jan Fagerberg

which the conditions for catch-up appear to have been especially favorable (and during which many countries managed to narrow the gap in productivity and income vis-a`-vis the leader) was the decades following the end of the Second World War, what Moses Abramovitz has called ‘‘the postwar catch-up and convergence boom’’ (Abramovitz 1986, 1994). Abramovitz suggested that such differences in performance over time and across countries might to some extent be explained with the help of two concepts, technological congruence and social capability.5 The first concept refers to the degree to which leader and follower country characteristics are congruent in areas such as market size, factor supply, etc. For instance, the technological system that emerged in the United States around the turn of the century was highly dependent on access to a large, homogenous market, something that hardly existed in Europe at the time, which may help explain its slow diffusion there. The second concept points to the capabilities that developing countries have to develop in order to catch up, such as improving education (particularly technical) and the business infrastructure (including the financial system). Abramovitz explained the successful catch-up of Western Europe in relation to the United States in the first half of the postwar period to be the result of both increasing technological congruence and improved social capabilities. As an example of the former he mentioned how European economic integration led to the creation of larger and more homogenous markets in Europe, facilitating the transfer of scale-intensive technologies initially developed for U.S. conditions. Regarding the latter, he pointed to such factors as the general increases in educational levels, the rise in the share of resources devoted to public and private sector R&D, and how good the financial system became in mobilizing resources for change, among other things. Abramovitz’s work has been criticized for not being sufficiently historically specific (Shin 1996). Although Abramovitz’ emphasis on technological congruence points to an awareness of the importance of changes in technological dynamics over time, he did little to substantiate it empirically. In this, he clearly sided with Gerschenkron, who also focused almost exclusively on catch-up in scale-based technologies. There is, however, no scarcity of contributions that argue that the dynamics of the scale-based system is waning and that the requirements that catching-up countries have to phase in therefore may have changed accordingly (Nelson and Wright 1992; Fagerberg et al. 1999). If so, this may have implications for ‘‘social capabilities’’ as well. Are the social capabilities that helped countries exploiting knowledge and catch-up in the past the same as those that matter for developing countries today? However, in spite of the concept’s general appeal (it is hard to find an applied paper on cross-country growth that does not reference it), it is, as Abramovitz himself admitted, ‘‘vaguely’’ and ‘‘poorly’’ defined (Abramovitz 1994,

Knowledge in Space

227

pp. 24 and 36), so this is a difficult question to address. In fact, in many practical applications it appears to be a ‘‘catch all’’ concept that the researcher may interpret to her liking. Another popular concept in the applied literature on growth and development that touches on some of the same issues as ‘‘social capability’’ is ‘‘absorptive capacity.’’ Wesley Cohen and Daniel Levinthal, who suggested the term, defined it as ‘‘the ability of a firm to recognize the value of new, external information, assimilate it, and apply it to commercial ends’’ (Cohen and Levinthal 1990, p. 128). They saw it as largely dependent on the firm’s prior related knowledge, which in turn was assumed to reflect its cumulative R&D. Although their focus was on firms, many of the same considerations seem to apply at more aggregate levels, such as regions or countries, and the concept has won quite general acceptance. It should be noted, however, that this concept by definition reduces three different processes to one, namely, (1) search, (2) assimilation (or absorption) of what is found, and (3) its commercial application. So in reality it refers not only to ‘‘absorption’’ in the received meaning of the term but to the ability to identify, exploit, and create knowledge more generally. Hence, this is a ‘‘catch all’’ concept by construction. The authors, being well aware of this, defend their position by arguing—with due reference to relevant psychological literature—that the ability to assimilate existing knowledge and the ability to create new knowledge are so similar that there is no point in distinguishing between them (ibid., p. 130). However, while it is easy to agree with the view that these three processes are related, it is not entirely obvious that it, for certain purposes at least, cannot be advantageous to consider them separately. This has to do with the fact that while all firms by definition assimilate information/knowledge, not all firms put the same emphasis on creation of new knowledge (or innovation). For some this may happen only accidentally or perhaps not at all, while for others it is the outcome of strategic decisions and resource commitment. The same may hold for countries. Arguably, such differences in strategic orientation may affect both the allocation of resources to and the output from these processes, and as students of this we would like to have a framework that allows us to study this in some detail. Technology and Capacity Competitiveness In the following6 we present recent empirical evidence of the two main aspects discussed above, namely The capacity of the firms of a country to compete through creation of new technology (what they call ‘‘technology competitiveness’’) and



228

Jan Fagerberg

The capacity of the firms of a country to exploit existing knowledge, independently of where it is created (what they call ‘‘capacity competitiveness’’).



Both of these aspects are clearly multidimensional in character, hence composite indicators are called for. In order to construct such indicators the researcher needs to identify the most important dimensions, find reliable indicators from as many countries as possible, express these in a comparable format, and weigh them together.7 Technology (or technological) competitiveness refers to the ability to create new goods and services (e.g., to innovate). There is, however, no available data source that measures this directly. Instead, what we have are different data sources reflecting different aspects of the phenomenon. R&D expenditures, for instance, measure some (but not all) of the resources that go into developing new goods and services. Patent statistics, on the other hand, measure the output of (patentable) inventions, but the propensity to patent varies considerably across industries, and many innovations are not patentable. So many innovations would not be accounted for by using this indicator only. Taking into account both indicators clearly gives a more balanced picture. We also include a measure of the quality of the science base on which innovation activities depend (as reflected in articles published in scientific and technical journals). Regarding capacity competitiveness, we focus on three dimensions: investments in human capital, efforts directed toward diffusion of technology, and relevant policy aspects. The importance of a well-developed human capital base for exploiting technological opportunities goes without saying; here we focus on secondary and tertiary education (as reflected in gross enrolment rates) in particular. Efforts directed toward technology diffusion are captured by gross fixed capital formation (a measure of new technology embodied in machinery and equipment) and the spread of computers across the population, since a well-developed ICT infrastructure is generally acknowledged as critical for the ability to benefit from new technology. Finally, we acknowledge that there may be factors of a political nature that influence the capacity to exploit technological opportunity. Although such factors often defy measurement, at least on a broad cross-country/cross-temporal basis, there exist survey data on factors such as the quality of governance and adherence to human rights across countries, which it might be relevant to consider. Figure 1 summarizes the changes over the past decade (1993–2002) in technology and capacity competitiveness for a sample consisting of 100 countries at different levels of development. The differences across country groups are striking. As for technology competitiveness, there is a clear divide between the advanced countries, with healthy and continuing increases, and the rest of the world, which are stagnant at best (with a partial exception for East Asia). The Asian tigers stand out with the

Knowledge in Space

229

Figure 1 Change of technology and capacity competitiveness (1993–2002). Source: Fagerberg et al. (2004).

best performance. A divide of a different sort is clearly visible along the capacity dimension. In this case there actually is some catch-up along one dimension, human capital, particularly by the new E.U. members, but also for many other developing countries. This, however, is more than counteracted by an increasing digital divide (ICT infrastructure), caused by much faster diffusion of computers in the already developed economies and among the Asian tigers than elsewhere. What light do the indicators presented here throw on the various dimensions of catch-up emphasized by the previous literature? Taking literally ‘‘absorptive capacity’’ should reflect cumulative R&D. If so there is strong divergence along this dimension, with a number of developed countries (including the Asian tigers) increasing their lead, while most low and middle income countries continue to fall farther behind.

230

Jan Fagerberg

Capacity competitiveness comprises elements of both ‘‘technological congruence’’ and ‘‘social capability.’’ It seems reasonable to identify the former with ICT diffusion, since this obviously is a necessary prerequisite for exploiting the dominant technologies of our times, and the latter with the other indicators, particularly education. It then follows that in terms of technological congruence, there is a process of divergence going on between rich (including the Asian tigers) and poor countries. There appears to be more convergence along the social capability dimension, with many developing countries (particularly middle income) catching up relatively rapidly. But some of the poorest countries, especially in sub-Saharan Africa, are losing out along this dimension as well. With respect to the Gerschenkronian scheme, it seems clear that the countries that have followed that route also stand out in terms of the indicators. In fact, on average the Asian tigers have the highest recorded increase on both technology and capacity competitiveness of all country groupings (Figure 1). Concluding Remarks This chapter has addressed the role of knowledge in development. We have argued that much economic theorizing operates with a perspective of knowledge that prevents rather than facilitates a good understanding of the subject. On the one hand knowledge tends to be regarded as a body of codified information that is either provided for free or easily accessible on the market at a fair price. On the other hand the potential users of this knowledge, e.g., the firms, are portrayed as omnipotent, endowed with perfect insights on the options in front of them, including what the relevant knowledge is and how to access it. If this were how the global economy works, knowledge should indeed be expected to be a source of strong convergence in productivity and income across different parts of the globe, particularly in an era of ICT in which large bodies of information can be distributed across the globe in almost no time (and at almost no cost). But the real world is not like this. Firms seldom buy ready-made product designs off the shelf (or download them from the Internet). Problems or needs come first, and then firms try to provide solutions. At the outset, they often lack a clear view of how to do it, what the relevant knowledge is, and where it can be found. To remedy this, the firm starts searching. These characteristics are, of course, common for firms in both developed and developing countries, but being far from both the technology frontier and the potential market greatly accentuate these problems for the latter. Moreover, the developing country firm may, to a much larger extent than developed country firms, be constrained by its environment: It may have a wish (and perhaps even the capability) to introduce a new product or process, but the

Knowledge in Space

231

possibility to do so may depend on capabilities in other firms or skills that are simply not there (or that require substantial investments to occur). Hence, to avoid being stuck along an inferior path and never catching up, new ‘‘institutional instruments’’ may be needed to compensate for some of these ‘‘latecomer disadvantages,’’ to use a Gerschenkronian term. Arguably, much of what firms and governments in catching-up countries have done can be understood from this perspective. For instance, the diversified business groups that developed in Japan and Korea might be seen as ‘‘institutional instruments’’ fulfilling some of those needs (Shin 1996). The OEM (original equipment manufacture) system that has developed in the electronics industries of East Asia may also be seen as an ‘‘institutional instrument’’—or ‘‘organizational innovation’’ (Hobday 2000)—geared toward improving links with the technology frontier and the market simultaneously. Similarly, attracting inward FDI may be seen as a ‘‘functional equivalent’’ to OEM, which, however, judged by the empirical evidence, seems to be less favorable for indigenous innovation. Other, more demanding, but perhaps also more rewarding ways—since they allow the latecomer firm to reap a larger share of the profit generated—include technology licensing, investments in own brands (OBM), etc. Improving the supply of needed skills has, of course, been a central preoccupation of many latecomer governments, as illustrated above. What can the extraordinary success that some catching-up countries have had, and the failure of others, teach present-day developing countries? There are, arguably, big potential rewards to following the Gerschenkronian strategy of targeting technologically progressive sectors, as part of a broader attempt to transform the economy and stimulate learning and the creation of new skills (or assets). However, not every country is equally well equipped with the capabilities necessary to pursue such a strategy, and investing in education may be a good place to start for countries that have not succeeded in catering for those needs already. But this will hardly be sufficient. How to overcome the increasing digital divide is another hurdle that policy makers in developing countries need to think hard about. Moreover, as productivity and income per capita increase and the gap vis-a`-vis the frontier countries becomes smaller, the requirements for continuing success become more stringent. In fact, the countries that have been most successful in catching up during the past 40 years or so, the Asian tigers, are now among the top performers worldwide when it comes to technology and capacity competitiveness. In contrast, some developing countries, of which China is the most spectacular example, have for a while managed to catch up (from a very low level) mainly by exploiting low labor costs. However, there is a danger that unless appropriate actions are taken, such countries may soon find their economic performance constrained by lagging technology and capacity competitiveness (Dahlman and Aubert 2001).

232

Jan Fagerberg

Notes 1. Since new technology displaces old technology, and hence makes investments made in the latter obsolete, social returns may also in some cases be less than (the sum of) the private returns (see Aghion and Howitt 1998). We will not discuss this possibility further here. 2. The following discussion (on European and Asian catch up and conceptual issues) draws on Fagerberg and Godinho (2004) by permission of Oxford University Press. 3. However, it should be pointed out that he did not rule out that there may be other paths to successful industrialization. For instance, he pointed to Denmark as an example of a country that managed to catch up without targeting the progressive industries of his time and explained this by its close links to the rapidly growing British market for agricultural products. 4. Gerschenkron’s work is often identified with his focus on the role of banks in industrialization, although as pointed out by Shin (1996), it is possible to see it as an attempt to arrive at a more general theory about catch-up, focusing on certain requirements that need to be met for successful catch-up to take place and different though ‘‘functionally equivalent’’ institutional responses (or catch-up strategies). 5. The term ‘‘social capability’’ comes from Ohkawa and Rosovsky (1973). 6. This section draws on recent work by Fagerberg et al. (2004). 7. See Fagerberg et al. (2004) for detailed information on sample, variables, and method. Whenever possible, indicators were defined as activities measured in quantity or constant prices, deflated by population. The indicators were standardized before aggregating them in the composites as follows: actual value mean value : standard deviation The same mean and standard deviation (derived from pooled data) were used for both periods). This means that changes over time in the volume of the activities measured by the individual indicators (increasing use of ICTs for instance) were allowed to spill over to the composite indicator (along with the changes caused by shifts in the position of countries on each individual indicator). A principal component analysis was carried out to determine the weights used in the calculation of the composite indicator.

References Abramovitz, M. (1986). ‘‘Catching Up, Forging Ahead, and Falling Behind.’’ Journal of Economic History 46: 386–406. Abramovitz, M. (1994). ‘‘The Origins of the Postwar Catch-Up and Convergence Boom.’’ In The Dynamics of Technology, Trade and Growth, J. Fagerberg, B. Verspagen, and N. von Tunzelmann, editors, pp. 21–52. Aldershot: Edward Elgar. Aghion, Philippe, and P. Howitt (1998). Endogenous Growth Theory. Cambridge, MA: MIT Press. Amsden, A. H. (1989). Asia’s Next Giant: South Korea and Late Industrialization. New York: Oxford University Press.

Knowledge in Space

233

Beasley, W. G. (1990). The Rise of Modern Japan. New York: St. Martin’s Press. Bruland, K., and D. C. Mowery (2004). ‘‘Innovation Through Time.’’ Chapter 13 in Oxford Handbook of Innovation, J. Fagerberg, D. C. Mowery, and R. R. Nelson, editors. Oxford: Oxford University Press. Chang, Ha-Joon (2002). Kicking Away the Ladder: Development Strategy in Historical Perspective. London: Anthem Press. Cohen, W., and Levinthal, D. (1990). ‘‘Absorptive Capacity: A New Perspective on Learning and Innovation.’’ Administrative Science Quarterly 35: 123. Dahlman, C. J., and J.-E. Aubert (2001). China and the Knowledge Economy: Seizing the 21st Century. WBI Development Studies, Washington, DC: The World Bank. Dosi, G. (1988). ‘‘Sources, Procedures, and Microeconomic Effects of Innovation.’’ Journal of Economic Literature 26: 1120–1171. Eurostat (2004). Innovation in Europe: Results for the EU, Iceland, and Norway. Luxembourg: Office for Official Publications of the European Communities. Fagerberg, J. (2004). ‘‘Innovation: A Guide to the Literature.’’ Chapter 1 in Oxford Handbook of Innovation, J. Fagerberg, D. C. Mowery, and R. R. Nelson, editors. Oxford: Oxford University Press. ———, and M. M. Godinho (2004). ‘‘Innovation and Catching-Up.’’ Chapter 19 in Oxford Handbook of Innovation, J. Fagerberg, D. C. Mowery, and R. R. Nelson, editors. Oxford: Oxford University Press. ———, P. Guerrieri, and B. Verspagen, editors (1999). The Economic Challenge for Europe: Adapting to Innovation-Based Growth. Aldershot: Edward Elgar. ———, M. Knell, and M. Srholec (2004). ‘‘The Competitiveness of Nations.’’ Paper prepared for the Second Globelics Conference, Beijing, China, October 18–20. Foray, D. (2004). The Economics of Knowledge. Cambridge, MA: MIT Press. Gerschenkron, A. (1962). Economic Backwardness in Historical Perspective. Cambridge, MA: Belknap Press. Granstrand, O. (2004). ‘‘Innovation and Intellectual Property Rights.’’ Chapter 10 in Oxford Handbook of Innovation, J. Fagerberg, D. C. Mowery, and R. R. Nelson, editors. Oxford: Oxford University Press. Hobday, M. (2000). ‘‘East versus Southeast Asian Innovation Systems: Comparing OEMand TNC-led Growth in Electronics.’’ In Technology, Learning, & Innovation: Experiences of Newly Industrializing Economies, L. Kim and R. Nelson, editors, pp. 129–169. Cambridge: Cambridge University Press. Johnson, C. A. (1982). MITI and the Japanese Miracle: The Growth of Industrial Policy, 1925–1975. Stanford, CA: Stanford University Press. Lall, S. (2000). ‘‘Technological Change and Industrialization in the Asian Newly Industrializing Economies: Achievements and Challenges.’’ In Technology, Learning, & Innovation: Experiences of Newly Industrializing Economies, L. Kim and R. Nelson, editors, pp. 13–68. Cambridge: Cambridge University Press. Lam, A. (2004). ‘‘Organizational Innovation.’’ Chapter 5 in Oxford Handbook of Innovation, J. Fagerberg, D. C. Mowery, and R. R. Nelson, editors. Oxford: Oxford University Press.

234

Jan Fagerberg

Landes, D. (1998). The Wealth and Poverty of Nations. London: Abacus. Nelson, R. R., and S. G. Winter (1982). An Evolutionary Theory of Economic Change. Cambridge, MA: Harvard University Press. ———, and G. Wright (1992). ‘‘The Rise and Fall of American Technological Leadership: The Postwar Era in Historical Perspective.’’ Journal of Economic Literature 30: 1931– 1964. Odagiri, H., and A. Goto (1996). Technology and Industrial Development in Japan. Oxford: Clarendon Press. Ohkawa, K., and H. Rosovsky (1973). Japanese Economic Growth. Stanford: Stanford University Press. Rogers, E. (1995). Diffusion of Innovations, 4th ed. New York: Free Press. Romer, P. M. (1990). ‘‘Endogenous Technological Change.’’ Journal of Political Economy 98: S71–S102. Shin, Jang-Sup (1996). The Economics of the Latecomers: Catching-Up, Technology Transfer, and Institutions in Germany, Japan, and South Korea. London: Routledge. Solow, R. M. (1956). ‘‘A Contribution to the Theory of Economic Growth.’’ Quarterly Journal of Economics 70: 65–94. Van de Ven, A., D. E. Polley, R. Garud, and S. Venkataraman (1999). The Innovation Journey. New York: Oxford University Press. Veblen, T. (1915). Imperial Germany and the Industrial Revolution. New York: Macmillan. Wade, R. (1990). Governing the Market: Economic Theory and the Role of Government in East Asian Industrialization. Princeton: Princeton University Press. World Bank (1993). The East Asian Miracle: Economic Growth and Public Policy. New York: Oxford University Press.

V New Models of Innovation

15 Democratizing Innovation: The Evolving Phenomenon of User Innovation Eric von Hippel

When researchers say that innovation is being democratized, we mean that users of products and services—both firms and individual consumers—are increasingly able to innovate for themselves.1 User-centered innovation processes offer great advantages over the manufacturer-centric innovation development systems that have been the mainstay of commerce for hundreds of years. Users that innovate can develop exactly what they want, rather than relying on manufacturers to act as their (often very imperfect) agents. Moreover, individual users do not have to develop everything they need on their own: They can benefit from innovations developed and freely shared by others. User-centered innovation processes are very different from the traditional, manufacturer-centric model, in which products and services are developed by manufacturers in a closed way, with the manufacturers using patents, copyrights, and other protections to prevent imitators from free-riding on their innovation investments. In the manufacturer-centric model, a user’s only role is to have needs, which manufacturers then identify and fill by designing and producing new products. This traditional model does fit some fields and conditions. However, a growing body of empirical work shows that users are the first to develop many and perhaps most new industrial and consumer products. Further, there is good reason to believe that the importance of product and service development by users is increasing over time. The trend toward democratization of innovation applies to information products such as software and also to physical products and is being driven by two related technical trends: (1) the steadily improving design capabilities (innovation toolkits) that advances in computer hardware and software make possible for users and (2) the steadily improving ability of individual users to combine and coordinate their innovation-related efforts via new communication media such as the Internet. The ongoing shift of innovation to users has some very attractive qualities. It is becoming progressively easier for many users to get precisely what they want by designing it for themselves. Innovation by users also provides a very necessary

238

Eric von Hippel

complement to and feedstock for manufacturer innovation. And innovation by users appears to increase social welfare. At the same time, the ongoing shift of product development activities from manufacturers to users is painful and difficult for many manufacturers. Open, distributed innovation is ‘‘attacking’’ a major structure of the social division of labor. Many firms and industries must make fundamental changes to long-held business models in order to adapt. Further, governmental policy and legislation sometimes preferentially supports innovation by manufacturers. Considerations of social welfare suggest that this must change. The workings of the intellectual property system are of special concern. But despite the difficulties, a democratized and user-centric system of innovation appears well worth striving for. Today a number of innovation process researchers are working to develop our understanding of user-centered innovation processes. In this chapter, I offer a review of some collective learnings on this important topic to date. Importance of Innovation by Users Users, as I use the term, are firms or individual consumers that expect to benefit from using a product or a service. In contrast, manufacturers expect to benefit from selling a product or a service. A firm or an individual can have different relationships to different products or innovations. For example, Boeing is a manufacturer of airplanes, but it is also a user of machine tools. If one were examining innovations developed by Boeing for the airplanes it sells, Boeing would be a manufacturer-innovator in those cases. But if one were considering innovations in metal-forming machinery developed by Boeing for in-house use in building airplanes, those would be categorized as user-developed innovations and Boeing would be a user-innovator in those cases. Innovation user and innovation manufacturer are the two general ‘‘functional’’ relationships between innovator and innovation. Users are unique in that they alone benefit directly from innovations. All others (here lumped under the term ‘‘manufacturers’’) must sell innovation-related products or services to users, indirectly or directly, in order to profit from innovations. Thus, in order to profit, inventors must sell or license knowledge related to innovations, and manufacturers must sell products or services incorporating innovations. Similarly, suppliers of innovation-related materials or services—unless they have direct use for the innovations—must sell the materials or services in order to profit from the innovations. The user and manufacturer categorization of relationships between innovator and innovation can be extended to specific functions, attributes, or features of products and services. When this is done, it may turn out that different parties are associated with different attributes of a particular product or service. For example, house-

Democratizing Innovation

239

holders are the users of the switching attribute of a household electric light switch— they use it to turn lights on and off. However, switches also have other attributes, such as ‘‘easy wiring’’ qualities, that may be used only by the electricians who install them. Therefore, if an electrician were to develop an improvement to the installation attributes of a switch, it would be considered a user-developed innovation. Both qualitative observations and quantitative research in a number of fields clearly document the important role users play as first developers of products and services later sold by manufacturing firms. Adam Smith (1776) was an early observer of the phenomenon, pointing out the importance of ‘‘the invention of a great number of machines which facilitate and abridge labor, and enable one man to do the work of many.’’ Smith went on to note that ‘‘a great part of the machines made use of in those manufactures in which labor is most subdivided, were originally the invention of common workmen, who, being each of them employed in some very simple operation, naturally turned their thoughts towards finding out easier and readier methods of performing it.’’ Rosenberg (1976) explored the matter in terms of innovation by user firms rather than individual workers. He studied the history of the U.S. machine tool industry, finding that important and basic machine types like lathes and milling machines were first developed and built by user firms having a strong need for them. Textile manufacturing firms, gun manufacturers, and sewing machine manufacturers were important early user-developers of machine tools. Quantitative studies of user innovation document that many of the most important and novel products and processes in a range of fields have been developed by user firms and by individual users. Thus, Enos (1962) reported that nearly all the most important innovations in oil refining were developed by user firms. Freeman (1968) found that the most widely licensed chemical production processes were developed by user firms. Von Hippel (1988) found that users were the developers of about 80% of the most important scientific instrument innovations, and also the developers of most of the major innovations in semiconductor processing. Pavitt (1984) found that a considerable fraction of invention by British firms was for inhouse use. Shah (2000) found that the most commercially important equipment innovations in four sporting fields tended to be developed by individual users. Empirical studies also show that many users—from 10% to nearly 40%—engage in developing or modifying products (Table 1). About half of these studies do not determine representative innovation frequencies; they were designed for other purposes. Nonetheless, when taken together, the findings make it very clear that users are doing a lot of product modification and product development in many fields. Studies of innovating users (both individuals and firms) show them to have the characteristics of ‘‘lead users’’ (Urban and von Hippel 1988; Herstatt and von Hippel 1992; Olson and Bakke 2001; Lilien et al. 2002). That is, they are ahead of the

240

Eric von Hippel

Table 1 Studies of user innovation frequency

Innovation area Industrial products 1. Printed circuit CAD software(a) 2. Pipe hanger hardware(b) 3. Library information systems(c)

4. Medical surgery equipment(d) 5. Apache OS server software security features(e) Consumer products 6. Outdoor consumer products(f) 7. ‘‘Extreme’’ sporting equipment(g) 8. Mountain biking equipment(h)

Number and type of users sampled

Percentage developing and building product for own use

136 user firm attendees at a PC-CAD conference Employees in 74 pipe hanger installation firms Employees in 102 Australian libraries using computerized OPAC library information systems 261 surgeons working in university clinics in Germany 131 technically sophisticated Apache users (webmasters)

24.3

153 recipients of mail order catalogs for outdoor activity products for consumers 197 members of 4 specialized sporting clubs in 4 ‘‘extreme’’ sports 291 mountain bikers in a geographic region known to be an ‘‘innovation hot spot’’

9.8

36 26

22 19.1

37.8

19.2

Sources: (a) Urban and von Hippel (1988); (b) Herstatt and von Hippel (1992); (c) Morrison et al. (2000); (d) Lu¨thje (2003); (e) Franke and von Hippel (2003b); (f ) Lu¨thje (2004); (g) Franke and Shah (2003); (h) Lu¨thje et al. (2002).

majority of users in their populations with respect to an important market trend, and they expect to gain relatively high benefits from a solution to the needs they have encountered there. The correlations found between innovation by users and lead user status are highly significant, and the effects are very large (Franke and Shah 2003; Lu¨thje et al. 2002; Morrison et al. 2000). Since lead users are at the leading edge of the market with respect to important market trends, one can guess that many of the novel products they develop for their own use will appeal to other users too and so might provide the basis for products

Democratizing Innovation

241

Figure 1 User-innovators with stronger ‘‘lead user’’ characteristics develop innovations having higher appeal in the general marketplace (data from Franke and von Hippel 2003b).

manufacturers would wish to commercialize. This turns out to be the case. A number of studies have shown that many of the innovations reported by lead users are judged to be commercially attractive and/or have actually been commercialized by manufacturers. Research provides a firm grounding for these empirical findings. The two defining characteristics of lead users and the likelihood that they will develop new or modified products have been found to be highly correlated (Morrison et al. 2004). In addition, it has been found that the higher the intensity of lead user characteristics displayed by an innovator, the greater the commercial attractiveness of the innovation that that lead user develops (Franke and von Hippel 2003a). In Figure 1, the increased concentration of innovations toward the right indicates that the likelihood of innovating is higher for users having higher lead user index values. The rise in average innovation attractiveness as one moves from left to right indicates that innovations developed by lead users tend to be more commercially attractive. (Innovation attractiveness is the sum of the novelty of the innovation and the expected future generality of market demand.)

242

Eric von Hippel

Why Many Users Want Custom Products Why do so many users develop or modify products for their own use? Users may innovate if and as they want something that is not available on the market and are able and willing to pay for its development. It is likely that many users do not find what they want on the market. Meta-analysis of market-segmentation studies suggests that users’ needs for products are highly heterogeneous in many fields (Franke and Reisinger 2003). Mass manufacturers tend to follow a strategy of developing products that are designed to meet the needs of a large market segment well enough to induce purchase by and capture significant profits from a large number of customers. When users’ needs are heterogeneous, this strategy of ‘‘a few sizes fit all’’ will leave many users somewhat dissatisfied with the commercial products on offer and probably will leave some users seriously dissatisfied. In a study of a sample of users of the security features of Apache web server software, Franke and von Hippel (2003b) found that users had a very high heterogeneity of need and that many had a high willingness to pay to get precisely what they wanted. Nineteen percent of the users sampled actually innovated to tailor Apache more closely to their needs. Those who did were found to be significantly more satisfied. Users’ Innovate-or-Buy Decisions Even if many users want ‘‘exactly right products’’ and are willing and able to pay for their development, we must understand why users often do this for themselves rather than hire a custom manufacturer to develop a special just-right product for them. After all, custom manufacturers specialize in developing products for one or a few users. Since these firms are specialists, it is possible that they could design and build custom products for individual users or user firms faster, better, or cheaper than users could do this for themselves. Despite this possibility, several factors can drive users to innovate rather than buy. In the case of both user firms and individual user-innovators, agency costs play a major role. In the case of individual user-innovators, enjoyment of the innovation process can also be important. With respect to agency costs, consider that when a user develops its own custom product, that user can be trusted to act in its own best interests. When a user hires a manufacturer to develop a custom product, the situation is more complex. The user is then a principal that has hired the custom manufacturer to act as its agent. If the interests of the principal and the agent are not the same, there will be agency costs. In general terms, agency costs are (1) costs incurred to monitor the agent to ensure that it (or he or she) follows the interests of the principal, (2) the cost incurred by the

Democratizing Innovation

243

agent to commit itself not to act against the principal’s interest (the ‘‘bonding cost’’), and (3) costs associated with an outcome that does not fully serve the interests of the principal (Jensen and Meckling 1976). In the specific instance of product and service development, a major divergence of interests between user and custom manufacturer does exist: The user wants to get precisely what it needs, to the extent that it can afford to do so. In contrast, the custom manufacturer wants to lower its development costs by incorporating solution elements it already has or that it predicts others will want in the future—even if by doing so it does not serve its present client’s needs as well as it could. A user wants to preserve its need specification because that specification is chosen to make that user’s overall solution quality as high as possible at the desired price. For example, an individual user may specify a mountain-climbing boot that will precisely fit his unique climbing technique and allow him to climb Everest more easily. Any deviations in boot design will require compensating modifications in the climber’s carefully practiced and deeply ingrained climbing technique—a much more costly solution from the user’s point of view. A custom boot manufacturer, in contrast, will have a strong incentive to incorporate the materials and processes it has in stock and expects to use in future even if this produces a boot that is not precisely right for the present customer. For example, the manufacturer will not want to learn a new way to bond boot components together even if that would produce the best custom result for one client. The net result is that when one or a few users want something special they will often get the best result by innovating for themselves. A model of the innovate-or-buy decision (von Hippel 2005) shows in a quantitative way that user firms with unique needs (in other words, a market of one) will always be better off developing new products for themselves. It also shows that development by manufacturers can be the most economical option when n or more user firms want the same thing. However, when the number of user firms wanting the same thing lies between 1 and n, manufacturers may not find it profitable to develop a new product for just a few users. In that case, more than one user may invest in developing the same thing independently, owing to market failure. This results in a waste of resources from the point of view of social welfare. The problem can be addressed by new institutional forms, such as the user innovation communities that will be mentioned later. It is important to note that an additional incentive can drive individual userinnovators to innovate rather than buy: They may value the process of innovating because of the enjoyment or learning that it brings them. It might seem strange that user-innovators can enjoy product development enough to want to do it themselves—after all, manufacturers pay their product developers to do such

244

Eric von Hippel

work! On the other hand, it is also clear that enjoyment of problem solving is a motivator for many individual problem solvers in at least some fields. Consider for example the millions of crossword-puzzle aficionados. Clearly, for these individuals enjoyment of the problem-solving process rather than the solution is the goal. One can easily test this by attempting to offer a puzzle solver a completed puzzle—the very output he or she is working so hard to create. One will very likely be rejected with the rebuke that one should not spoil the fun. Pleasure as a motivator can apply to the development of commercially useful innovations as well. Studies of the motivations of volunteer contributors of code to widely used software products have shown that these individuals too are often strongly motivated to innovate by the joy and learning they find in this work (Hertel et al. 2003; Lakhani and Wolf 2005). Users’ Low-Cost Innovation Niches An exploration of the basic processes of product and service development shows that users and manufacturers tend to develop different types of innovations. This is due in part to information asymmetries: Users and manufacturers tend to know different things. Product developers need two types of information in order to succeed at their work: need and context-of-use information (generated by users) and generic solution information (often initially generated by manufacturers specializing in a particular type of solution). Bringing these two types of information together is not easy. Both the need information and the solution information are often very ‘‘sticky’’—that is, costly to move from the site where the information was generated to other sites (von Hippel 1994; Ogawa 1998). As a result, users generally have a more accurate and more detailed model of their needs than manufacturers have, while manufacturers have a better model of the solution approach in which they specialize than the user has. When information is sticky, innovators tend to rely largely on information they already have in stock. One consequence of the information asymmetry between users and manufacturers is that users tend to develop innovations that are functionally novel, requiring a great deal of user-need information and use-context information for their development. In contrast, manufacturers tend to develop innovations that are improvements on well-known needs and that require a rich understanding of solution information for their development. This sticky information effect is visible in studies of innovation. Riggs and von Hippel (1994) studied the types of innovations made by users and manufacturers that improved the functioning of two major types of scientific instruments. They found that users tended to develop innovations that enabled the instruments to do qualitatively new types of things for the first time. In contrast, manufacturers tended

Democratizing Innovation

245

Table 2 Source of innovations by nature of improvement effected Type of improvement provided by innovation 1. New functional capability 2. Sensitivity, resolution or accuracy improvement 3. Convenience or reliability improvement Total

Innovation developed by Users (%)

Users

Manufacturers

Total

82 48

14 11

3 12

17 23

13

3

21

24 64

Source: Riggs and von Hippel (1994).

to develop innovations that enabled users to do the same things they had been doing, but to do them more conveniently or reliably (Table 2). For example, users were the first to modify the instruments to enable them to image and analyze magnetic domains at submicroscopic dimensions. In contrast, manufacturers were the first to computerize instrument adjustments to improve ease of operation. Sensitivity, resolution, and accuracy improvements fall somewhere in the middle, as the data show. These types of improvements can be driven by users seeking to do specific new things or by manufacturers applying their technical expertise to improve the products along known general dimensions of merit, such as accuracy. If we extend the information-asymmetry argument one step further, we see that information stickiness implies that information on hand will also differ among individual users and manufacturers. The information assets of some particular user (or some particular manufacturer) will be closest to what is required to develop a particular innovation, and so the cost of developing that innovation will be relatively low for that user or manufacturer. The net result is that user innovation activities will be distributed across many users according to their information endowments. With respect to innovation, one user is by no means a perfect substitute for another. Why Users Often Freely Reveal Their Innovations The social efficiency of a system in which individual innovations are developed by individual users is increased if users somehow diffuse what they have developed to others. Manufacturer-innovators partially achieve this when they sell a product or a service on the open market (partially because they diffuse the product incorporating the innovation but often not all the information that others would need to fully understand and replicate it). If user-innovators do not somehow also diffuse what they

246

Eric von Hippel

have done, multiple users with very similar needs will have to independently develop very similar innovations—a poor use of resources from the viewpoint of social welfare. Empirical research shows that users often do achieve widespread diffusion by an unexpected means: They often ‘‘freely reveal’’ what they have developed. When we say that an innovator freely reveals information about a product or service it has developed, we mean that all intellectual property rights to that information are voluntarily given up by the innovator, and all interested parties are given access to it— the information becomes a public good (Harhoff et al. 2003). The empirical finding that users often freely reveal their innovations has been a major surprise to innovation researchers. On the face of it, if a user-innovator’s proprietary information has value to others, one would think that the user would strive to prevent free diffusion rather than help others to a free ride on what it has developed at private cost. Nonetheless, it is now very clear that individual users and user firms—and sometimes manufacturers—often freely reveal detailed information about their innovations. The practices visible in ‘‘open source’’ software development were important in bringing this phenomenon to general awareness. In these projects it was clear policy that project contributors would routinely and systematically freely reveal code they had developed at private expense (Raymond 1999). However, free revealing of product innovations has a history that began long before the advent of open source software. Allen, in his 1983 study of the eighteenth-century iron industry, was probably the first to consider the phenomenon systematically. Later, Nuvolari (2004) discussed free revealing in the early history of mine pumping engines. Contemporary free revealing by users has been documented by von Hippel and Finkelstein (1979) for medical equipment, by Lim (2000) for semiconductor process equipment, by Morrison et al. (2000) for library information systems, and by Franke and Shah (2003) for sporting equipment. Henkel (2003) has documented free revealing among manufacturers in the case of embedded Linux software. Innovators often freely reveal because it is often the best or the only practical option available to them. Hiding an innovation as a trade secret is unlikely to be successful for long: Too many generally know similar things, and some holders of the ‘‘secret’’ information stand to lose little or nothing by freely revealing what they know. Studies find that innovators in many fields view patents as having only limited value (Harhoff et al. 2003). Copyright protection and copyright licensing are applicable only to ‘‘writings,’’ such as books, graphic images, and computer software. Active efforts by innovators to freely reveal—as opposed to sullen acceptance— are explicable because free revealing can provide innovators with significant private benefits as well as losses or risks of loss. Users who freely reveal what they have done often find that others then improve or suggest improvements to the innovation,

Democratizing Innovation

247

Table 3 User innovation is widely distributed: A few users developed more than one major commercialized innovation Number of innovations each user developed User samples

1

2

3

6

NA

Sample (n)

Scientific Instrument users(a) Scientific Instrument users(b) Process equipment users(c) Sports equipment users(d)

28 20 19 7

0 1 1 0

1 0 0 0

0 1 0 0

1 0 8 0

32 28 29 7

Source: von Hippel (2005), Table 7-1. Data from (a) von Hippel (1988), Appendix: GC, TEM, NMR Innovations; (b) Riggs and von Hippel (1994), Esca, and AES; (c) von Hippel (1988), Appendix: Semiconductor and pultrusion process equipment innovations; (d) Shah (2000), Appendix A: Skateboarding, Snowboarding and Windsurfing Innovations Developed by Users.

to mutual benefit (Raymond 1999). Freely revealing users also may benefit from enhancement of reputation, from positive network effects due to increased diffusion of their innovation, and from other factors. Being the first to freely reveal a particular innovation can also enhance the benefits received, and so there can actually be a rush to reveal, much as scientists rush to publish in order to gain the benefits associated with being the first to have made a particular advancement. Innovation Communities Innovation by users tends to be widely distributed rather than concentrated among just a very few very innovative users (Table 3). As a result, it is important for userinnovators to find ways to combine and leverage their efforts. Users achieve this by engaging in many forms of cooperation. Direct, informal user-to-user cooperation (assisting others to innovate, answering questions, and so on) is common. Organized cooperation is also common, with users joining together in networks and communities that provide useful structures and tools for their interactions and for the distribution of innovations. Innovation communities can increase the speed and effectiveness with which users and also manufacturers can develop and test and diffuse their innovations. They also can greatly increase the ease with which innovators can build larger systems from interlinkable modules created by community participants. Free and open source software projects are a relatively well developed and very successful form of Internet-based innovation community. However, innovation communities are by no means restricted to software or even to information products,

248

Eric von Hippel

and they can play a major role in the development of physical products. Franke and Shah (2003) have documented the value that user innovation communities can provide to user-innovators developing physical products in the field of sporting equipment. The analogy to open source innovation communities is clear. The collective or community effort to provide a public good—which is what freely revealed innovations are—has traditionally been explored in the literature on ‘‘collective action.’’ However, behaviors seen in extant innovation communities fail to correspond to that literature at major points. In essence, innovation communities appear to be more robust with respect to recruiting and rewarding members than the literature would predict. The reason for this appears to be that innovation contributors obtain some private rewards that are not shared equally by free riders (those who take without contributing). For example, a product that a user-innovator develops and freely reveals might be perfectly suited to that userinnovator’s requirements but less well suited to the requirements of free riders. Innovation communities thus illustrate a ‘‘private-collective’’ model of innovation incentive (von Hippel and von Krogh 2003). Adapting Policy to User Innovation Is innovation by users a ‘‘good thing?’’ Welfare economists answer such a question by studying how a phenomenon or a change affects social welfare. Henkel and von Hippel (2005) explored the social welfare implications of user innovation. They found that, relative to a world in which only manufacturers innovate, social welfare is very probably increased by the presence of innovations freely revealed by users. This finding implies that policy making should support user innovation, or at least should ensure that legislation and regulations do not favor manufacturers at the expense of user-innovators. The transitions required of policy making to achieve neutrality with respect to user innovation versus manufacturer innovation are significant. Consider the impact on open and distributed innovation of past and current policy decisions. Research done in the past 30 years has convinced many academics that intellectual property law is sometimes or often not having its intended effect. Intellectual property law was intended to increase the amount of innovation investment. Instead, it now appears that there are economies of scope in both patenting and copyright that allow firms to use these forms of intellectual property law in ways that are directly opposed to the intent of policy makers and to the public welfare (Foray 2004). Major firms can invest to develop large portfolios of patents. They can then use these to create ‘‘patent thickets’’—dense networks of patent claims that give them plausible grounds for threatening to sue across a wide range of intellectual property. They may do this to prevent others from introducing a superior innovation and/or to

Democratizing Innovation

249

demand licenses from weaker competitors on favorable terms (Shapiro 2001; Bessen 2003). Movie, publishing, and software firms can use large collections of copyrighted work to a similar purpose (Benkler 2002). In view of the distributed nature of innovation by users, with each tending to create a relatively small amount of intellectual property, users are likely to be disadvantaged by such strategies. It is also important to note that users (and manufacturers) tend to build prototypes of their innovations economically by modifying products already available on the market to serve a new purpose. Laws such as the (U.S.) Digital Millennium Copyright Act, intended to prevent consumers from illegally copying protected works, also can have the unintended side effect of preventing users from modifying products that they purchase (Varian 2002). Both fairness and social welfare considerations suggest that innovation-related policies should be made neutral with respect to the sources of innovation. It may be that current impediments to user innovation will be solved by legislation or by policy making. However, beneficiaries of existing law and policy will predictably resist change. Fortunately, a way to get around some of these problems is in the hands of innovators themselves. Suppose many innovators in a particular field decide to freely reveal what they have developed, as they often have reason to do. In that case, users can collectively create an information commons (a collection of information freely available to all) containing substitutes for some or a great deal of information now held as private intellectual property. Then user-innovators can work around the strictures of intellectual property law by simply using these freely revealed substitutes (Lessig 2001). This is essentially what is happening in the field of software. For many problems, user-innovators in that field now have a choice between proprietary, closed software provided by Microsoft and other firms and open source software that they can legally download from the Internet and legally modify as they wish to serve their own specific needs. Policy making that levels the playing field between users and manufacturers will force more rapid change onto manufacturers but will by no means destroy them. Experience in fields where open and distributed innovation processes are far advanced show how manufacturers can and do adapt. Some, for example, learn to supply proprietary platform products that offer user-innovators a framework upon which to develop and use their improvements (Jeppesen 2004). Diffusion of User-Developed Innovations Products, services, and processes developed by users become more valuable to society if they are somehow diffused to others that can also benefit from them. If user innovations are not diffused, multiple users with very similar needs will have to invest to (re)develop very similar innovations, which, as was noted earlier, would be a

250

Eric von Hippel

poor use of resources from the social welfare point of view. In the case of information products, users have the possibility of largely or completely doing without the services of manufacturers. Open source software projects are object lessons that teach us that users can create, produce, diffuse, provide user field support for, update, and use complex products by and for themselves in the context of user innovation communities. In physical product fields, the situation is different. Users can develop products. However, the economies of scale associated with manufacturing and distributing physical products give manufacturers an advantage over ‘‘do-ityourself’’ users in those activities. How can or should user innovations of general interest be transferred to manufacturers for large-scale diffusion? We propose that there are three general methods for accomplishing this. First, manufacturers can actively seek innovations developed by lead users that can form the basis for a profitable commercial product. Second, manufacturers can draw innovating users into joint design interactions by providing them with ‘‘toolkits for user innovation.’’ Third, users can become manufacturers in order to widely diffuse their innovations. We discuss each of these possibilities in turn. To systematically find user-developed innovations, manufacturers must redesign their product development processes. Currently, almost all manufacturers think that their job is to find a need and fill it rather than to sometimes find and commercialize an innovation that lead users have already developed. Accordingly, manufacturers have set up market research departments to explore the needs of users in the target market, product development groups to think up suitable products to address those needs, and so forth. In this type of product development system, the needs and prototype solutions of lead users—if encountered at all—are typically rejected as outliers of no interest. Indeed, when lead users’ innovations do enter a firm’s product line, they typically arrive with a lag and by an unconventional and unsystematic route. For example, a manufacturer may ‘‘discover’’ a lead user innovation only when the innovating user firm contacts the manufacturer with a proposal to produce its design in volume to supply its own in-house needs. Or sales or service people employed by a manufacturer may spot a promising prototype during a visit to a customer’s site. Modification of firms’ innovation processes to systematically search for and further develop innovations created by lead users can provide manufacturers with a better interface to the innovation process as it actually works and so provide better performance. A natural experiment conducted at 3M illustrates this possibility. Annual sales of lead user product ideas generated by the average lead user project at 3M were conservatively forecast by management to be more than eight times the sales forecast for new products developed in the traditional manner—$146 million

Democratizing Innovation

251

versus $18 million per year. In addition, lead user projects were found to generate ideas for new product lines, while traditional market research methods were found to produce ideas for incremental improvements to existing product lines. As a consequence, 3M divisions funding lead user project ideas experienced their highest rate of major product line generation in the past 50 years (Lilien et al. 2002). Toolkits for user innovation custom design involve partitioning product development and service development projects into solution information–intensive subtasks and need information–intensive subtasks. Need-intensive subtasks are then assigned to users along with a kit of tools that enable them to effectively execute the tasks assigned to them. In the case of physical products, the designs that users create using a toolkit are then transferred to manufacturers for production (von Hippel and Katz 2002). Toolkits make innovation cheaper for users and also lead to higher customer value. Thus, Franke and Piller (2004) in a study of a consumer wrist watches found the willingness to pay for a self-designed product was 200% of the willingness to pay for the best-selling commercial product of the same technical quality. This increased willingness to pay was due to both the increased value provided by the selfdeveloped product and the value of the toolkit process for consumers engaging in it (Schreier and Franke 2004). Manufacturers that offer toolkits to their customers can attract innovating users into a relationship with their firm and so get an advantage with respect to producing what the users develop. The custom semiconductor industry was an early adopter of toolkits. In 2003, more than $15 billion worth of semiconductors were produced that had been designed using this approach (Thomke and von Hippel 2002). Innovations developed by users sometimes achieve widespread diffusion when those users become manufacturers—setting up a firm to produce their innovative product(s) for sale. Shah (2000) showed this pattern in sporting goods fields. In the medical field, Lettl et al. (2004) have shown a pattern in which innovating users take on many of the entrepreneurial functions needed to commercialize the new medical products they have developed but do not themselves abandon their user roles. New work in this field is exploring the conditions under which users will become entrepreneurs rather than transfer their innovations to established firms (Hienerth 2004; Shah and Tripsas 2004). Democratizing Innovation I summarize this overview article by again saying that users’ ability to innovate is improving radically and rapidly as a result of the steadily improving quality of computer software and hardware, improved access to easy-to-use tools and components for innovation, and access to a steadily richer innovation commons. Today, user

252

Eric von Hippel

firms and even individual hobbyists have access to sophisticated programming tools for software and sophisticated CAD design tools for hardware and electronics. These information-based tools can be run on a personal computer, and they are rapidly coming down in price. As a consequence, innovation by users will continue to grow even if the degree of heterogeneity of need and willingness to invest in obtaining a precisely right product remain constant. Equivalents of the innovation resources described above have long been available within corporations, to a few. Senior designers at firms have long been supplied with engineers and designers under their direct control and with the resources needed to quickly construct and test prototype designs. The same is true in other fields, including automotive design and clothing design: Just think of the staffs of engineers and model makers supplied so that top auto designers can quickly realize and test their designs. But if, as we have seen, the information needed to innovate in important ways is widely distributed, the traditional pattern of concentrating innovation support resources on a few individuals is hugely inefficient. High-cost resources for innovation support cannot efficiently be allocated to ‘‘the right people with the right information’’: It is very difficult to know who these people may be before they develop an innovation that turns out to have general value. When the cost of high-quality resources for design and prototyping becomes very low (the trend we have described), these resources can be diffused very widely, and the allocation problem diminishes in significance. The net result is a pattern of increasing democratization of product and service innovation—a pattern that will involve significant changes for both users and manufacturers. Note Originally published in Journal fu¨r Betriebswirtschaft (2005). 1. Readers interested in exploring the evolving phenomenon of democratizing innovation in more depth and detail may wish to read Eric von Hippel, Democratizing Innovation (Cambridge, MA: MIT Press, 2005). In addition to the printed version, an electronic version will be available for cost-free download from the MIT Press website (MITPress.mit.edu) under a Creative Commons license.

References Allen, R. C. (1983). ‘‘Collective Invention.’’ Journal of Economic Behavior and Organization 4(1): 1–24. Benkler, Y. (2002). ‘‘Intellectual Property and the Organization of Information Production.’’ International Review of Law and Economics 22(1): 81–107.

Democratizing Innovation

253

Bessen, J. (2003). ‘‘Patent Thickets: Strategic Patenting of Complex Technologies.’’ Research on Innovation and Boston University School of Law Working Paper. Enos, John Lawrence (1962). Petroleum Progress and Profits: A History of Process Innovation. Cambridge, MA: MIT Press. Foray, D. (2004). Economics of Knowledge. Cambridge, MA: MIT Press. Franke, N., and F. Piller (2004). ‘‘Value Creation by Toolkits for User Innovation and Design: The Case of the Watch Market.’’ Journal of Product Innovation Management 21(6): 401–415. ———, and H. Reisinger (2003). ‘‘Remaining Within Cluster Variance: A Meta Analysis of the ‘Dark’ Side of Cluster Analysis.’’ Vienna Business University Working Paper. ———, and S. Shah (2003). ‘‘How Communities Support Innovative Activities: An Exploration of Assistance and Sharing Among End-Users.’’ Research Policy 32(1): 157–178. ———, and E. von Hippel (2003a). ‘‘Finding Commercially Attractive User Innovations.’’ MIT Sloan School of Management Working Paper No. 4402-03. ———, and E. von Hippel (2003b). ‘‘Satisfying Heterogeneous User Needs via Innovation Toolkits: The Case of Apache Security Software.’’ Research Policy 32(7): 1199–1215. Freeman, C. (1968). ‘‘Chemical Process Plant: Innovation and the World Market.’’ National Institute Economic Review 45 (August): 2957. Harhoff, D., J. Henkel, and E. von Hippel (2003). ‘‘Profiting from Voluntary Information Spillovers: How Users Benefit by Freely Revealing Their Innovations.’’ Research Policy 32(10): 1753–1769. Henkel, J. (2003). ‘‘Software Development in Embedded Linux: Informal Collaboration of Competing Firms.’’ In Proceedings der 6. Internationalen Tagung Wirtschaftsinformatik 2003, W. Uhr, W. Esswein, and E. Schoop, editors, Vol. 2, pp. 81–99. Heidelberg: Physica. ———, and E. von Hippel (2005). ‘‘Welfare Implications of User Innovation.’’ Journal of Technology Transfer (forthcoming). Herstatt, C., and E. von Hippel (1992). ‘‘From Experience: Developing New Product Concepts Via the Lead User Method: A Case Study in a ‘Low Tech’ Field.’’ Journal of Product Innovation Management 9(3): 213–222. Hertel, G., S. Niedner, and S. Herrmann (2003). ‘‘Motivation of Software Developers in Open Source Projects: An Internet-Based Survey of Contributors to the Linux Kernel.’’ Research Policy 32(7): 1159–1177. Hienerth, Christoph (2004). ‘‘The Commercialization of User Innovations: Sixteen Cases in an Extreme Sporting Industry.’’ Vienna University of Economics and Business Administration Working Paper. Jensen, M. C., and W. H. Meckling (1976). ‘‘Theory of the Firm: Managerial Behavior, Agency Costs, and Ownership Structure.’’ Journal of Financial Economics 3(4): 305–360. Jeppesen, L. B. (2004). ‘‘Profiting from Innovative User Communities: How Firms Organize the Production of User Modifications in the Computer Games Industry.’’ Department of Industrial Economics and Strategy, Copenhagen Business School, Working Paper WP-04. Lakhani, K. R., and B. Wolf (2005). ‘‘Why Hackers Do What They Do: Understanding Motivation and Effort in Free/Open Source Software Projects.’’ In Perspectives on Free and Open

254

Eric von Hippel

Source Software, J. Feller, B. Fitzgerald, S. Hissam, and K. R. Lakhani, editors. Cambridge, MA: MIT Press. Lessig, L. (2001). The Future of Ideas: The Fate of the Commons in a Connected World. New York: Random House. Lettl, C., C. Herstatt, and H. Gemu¨nden (2004). ‘‘The Entrepreneurial Role of Innovative Users: Evidence from Radical Innovations in the Field of Medical Technology.’’ Technical University Berlin Working Paper. Lilien, Gary L., Pamela D. Morrison, Kathleen Searls, Mary Sonnack, and Eric von Hippel (2002). ‘‘Performance Assessment of the Lead User Idea Generation Process.’’ Management Science 48(8) (August): 1042–1059. Lim, K. (2000). ‘‘The Many Faces of Absorptive Capacity: Spillovers of Copper Interconnect Technology for Semiconductor Chips.’’ MIT Sloan School of Management Working Paper #4110. Lu¨thje, C. (2003). ‘‘Customers as Co-Inventors: An Empirical Analysis of the Antecedents of Customer-Driven Innovations in the Field of Medical Equipment.’’ In Proceedings from the 32nd EMAC Conference 2003, Glasgow. ——— (2004). ‘‘Characteristics of Innovating Users in a Consumer Goods Field: An Empirical Study of Sport-Related Product Consumers.’’ Technovation (forthcoming). ———, C. Herstatt, and E. von Hippel (2002). ‘‘The Dominant Role of Local Information in User Innovation: The Case of Mountain Biking.’’ MIT Sloan School Working Paper #437702. Morrison, P. D., J. H. Roberts, and D. F. Midgley (2004). ‘‘The Nature of Lead Users and Measurement of Leading Edge Status.’’ Research Policy 33(2): 351–362. ———, J. H. Roberts, and E. von Hippel (2000). ‘‘Determinants of User Innovation and Innovation Sharing in a Local Market.’’ Management Science 46(12): 1513–1527. Nuvolari, A. (2004). ‘‘Collective Invention During the British Industrial Revolution: The Case of the Cornish Pumping Engine.’’ Cambridge Journal of Economics 28(3): 347–363. Ogawa, S. (1998). ‘‘Does Sticky Information Affect the Locus of Innovation? Evidence from the Japanese Convenience-Store Industry.’’ Research Policy 26(7–8): 777–790. Olson, Erik L., and Geir Bakke (2001). ‘‘Implementing the Lead User Method in a High Technology Firm: A Longitudinal Study of Intentions versus Actions.’’ Journal of Product Innovation Management 18(2) (November): 388–395. Pavitt, K. (1984). ‘‘Sectoral Patterns of Technical Change: Towards a Taxonomy and a Theory.’’ Research Policy 13(6): 343–373. Raymond, E. (1999). The Cathedral and the Bazaar: Musings on Linux and Open Source by an Accidental Revolutionary. Sebastopol, CA: O’Reilly. Riggs, William, and Eric von Hippel (1994). ‘‘The Impact of Scientific and Commercial Values on the Sources of Scientific Instrument Innovation.’’ Research Policy 23 (July): 459– 469. Rosenberg, Nathan (1976). Perspectives on Technology. Cambridge: Cambridge University Press.

Democratizing Innovation

255

Schreier, M., and N. Franke (2004). ‘‘Tom Sawyer’s Great Law in Action: Why Users are Willing to Pay to Design Their Own Products via Toolkits for User Innovation and Design.’’ Vienna University of Economics and Business Administration Working Paper. Shah, S. (2000). ‘‘Sources and Patterns of Innovation in a Consumer Products Field: Innovations in Sporting Equipment.’’ MIT Sloan School of Management Working paper #4105. ———, and M. Tripsas (2004). ‘‘When Do User-Innovators Start Firms? Towards a Theory of User Entrepreneurship.’’ University of Illinois Working Paper #04-0106. Shapiro, C. (2001). ‘‘Navigating the Patent Thicket: Cross Licenses, Patent Pools, and Standard Setting.’’ In Innovation Policy and the Economy, A. Jaffe, J. Lerner, and S. Stern, editors, Vol. 1, pp. 119–150. Cambridge, MA: MIT Press. Smith, A. (1776). An Inquiry into the Nature and Causes of the Wealth of Nations. 1776; 5th ed. 1789; Modern Library Edition, edited by Edwin Cannan. New York: Random House, 1937. Thomke, S. H., and E. von Hippel (2002). ‘‘Customers as Innovators: A New Way to Create Value.’’ Harvard Business Review 80(4): 74–81. Urban, G. L., and E. von Hippel (1988). ‘‘Lead User Analyses for the Development of New Industrial Products.’’ Management Science 34(5): 569–582. Varian, H. R. (2002). ‘‘New Chips Can Keep a Tight Rein on Consumers.’’ New York Times (July 4). von Hippel, E. (1988). The Sources of Innovation. New York: Oxford University Press. ——— (1994). ‘‘Sticky Information and the Locus of Problem Solving: Implications for Innovation.’’ Management Science 40(4): 429–439. ——— (2005). Democratizing Innovation. Cambridge, MA: MIT Press. ———, and S. N. Finkelstein (1979). ‘‘Analysis of Innovation in Automated Clinical Chemistry Analyzers.’’ Science & Public Policy 6(1): 24–37. ———, and R. Katz (2002). ‘‘Shifting Innovation to Users Via Toolkits.’’ Management Science 48(7): 821–833. ———, and G. von Krogh (2003). ‘‘Open Source Software and the ‘Private-Collective’ Innovation Model: Issues for Organization Science.’’ Organization Science 14(2): 209–223.

16 Innovation, Experimentation, and Technological Change Stefan Thomke

Introduction At the heart of every company’s ability to innovate lies a process of experimentation that enables the organization to create and refine its products and services.1 In fact, no product can be a product without it first having been an idea subsequently shaped through experimentation. Today, a major development project involves literally thousands of experiments, all with the same objective: to learn, through rounds of organized testing, whether the product concept or the proposed technical solution holds promise for addressing a need or problem. The information derived from each round is then incorporated into the next set of experiments, until the final product ultimately results. In short, innovations do not arrive fully fledged but are nurtured—through an experimentation process that takes place in laboratories and development organizations. But experimentation has often been expensive in terms of the time involved and the labor expended, even as it has been essential in terms of innovation. What has changed, particularly given new technologies available, is that it is now possible to perform more experiments in an economically viable way while accelerating the drive toward innovation. Not only can more experiments be run today, the kinds of experiments possible is expanding. Never before has it been so economically feasible to ask ‘‘what-if’’ questions and generate preliminary answers. New technologies enable organizations to both challenge presumed answers and pose more questions. They amplify how innovators learn from experiments, creating the potential for higher R&D performance and new ways of creating value for firms and their customers. At the same time, many companies do not fully unlock that potential because how they design, organize, and manage their approach to innovation gets in the way. That is, even deploying new technology for experimentation, these organizations are not organized to capture its potential value—in experimentation, in innovation.

258

Stefan Thomke

‘‘Experimentation’’ encompasses success and failure; it is an iterative process of understanding what doesn’t work and what does. Both results are equally important for learning, which is the goal of any experiment and of experimentation overall. Thus, a crash test that results in unacceptable safety for drivers, a software user interface that confuses customers, or a drug that is toxic can all be desirable outcomes of an experiment—provided these results are revealed early in an innovation process and can be subsequently reexamined. Because few resources have been committed in these early stages, decision making is still flexible, and other approaches can be ‘‘experimented with’’ quickly. In a nutshell, experiments that result in failure are not failed experiments—but they frequently are considered that when anything deviating from what was intended is deemed ‘‘failure.’’ Herein lies a managerial dilemma that innovators face. A relentless organizational focus on success makes true experimentation all too rare. In fact, the book In Search of Excellence noted years ago: The most important and visible outcropping of the action bias in the excellent companies is their willingness to try things out, to experiment. There is absolutely no magic in the experiment. It is simply a tiny completed action, a manageable test that helps you learn something, just as in high-school chemistry. But our experience has been that most big institutions have forgotten how to test and learn. They seem to prefer analysis and debate to trying something out, and they are paralyzed by fear of failure, however small. (Peters and Waterman 1982, pp. 134–135)

Because experiments that reveal what doesn’t work are frequently deemed ‘‘failures,’’ tests may be delayed, rarely carried out, or simply labeled verification, implying that only finding out what works is the primary goal of an experiment. If there is a problem in the experiment, it will, under this logic, be revealed very late in the game. But when feedback on what does not work comes so late, costs can spiral out of control; worse, opportunities for innovation are lost at that point. By contrast, when managers understand that effective experiments are supposed to reveal what does not work early, they realize that the knowledge gained then can benefit the next round of experiments and lead to more innovative ideas and concepts— early ‘‘failures’’ can lead to more powerful successes faster. IDEO, a leading product development firm, calls this ‘‘failing often to succeed sooner.’’ But organizing for rapid feedback coming more frequently—as powered by these new technologies—is not trivial. Multiple issues can arise, for instance, the ‘‘problem’’ of greater experimental capacity. What do we do with the opportunity to experiment ‘‘more’’? Consider the attempted integration of computer modeling and simulation in the automotive industry. Car companies have spent hundreds of millions of dollars on computer-aided technologies and employ many engineers and specialists to improve the performance of their complex development processes. By

Innovation, Experimentation, and Technological Change

259

replacing expensive physical testing with virtual models, management hopes not only to save costs and time but also to streamline decision making and coordination among team members. Managers would admit that the potential of new technologies is great, but they don’t fully recognize that the problems encountered relate not to ‘‘the technology’’ but to how it must be integrated into product development activities and organizations surrounding them, and, more significantly, what existing expectations are. One does not simply ‘‘swap in’’ a new technology and hope that both behavior and economics will magically change. Thus, new technologies for experimentation pose new challenges and require new strategies for the organization of innovation. Experimentation and Innovation The pursuit of knowledge is the rationale behind experimentation, and all experiments yield information that comes from understanding what does, and does not, work. For centuries, researchers have relied on systematic experimentation, guided by their insight and intuition, as an instrumental source of new information and the advancement of knowledge. Famous experiments have been conducted to characterize naturally occurring processes, to decide among rival scientific hypotheses about matter, to find hidden mechanisms of known effects, to simulate what is difficult or impossible to research: in short, to establish scientific laws inductively. Some of the most famous series of experiments have led to scientific breakthroughs or radically new innovations from which we still benefit today. Louis Pasteur’s discovery of artificial vaccines is one example (Hare 1981, p. 106). Pasteur had been struggling for years to understand the course of disease, in this case cholera, and ran extensive experiments to accumulate a knowledge base to help him make sense of what experiments in his laboratory were yielding. In 1879, he returned from a summer vacation not realizing that chicken broth cultures, part of one ongoing experiment, had become infected. He thus injected his hens with the infected culture and followed that with injections of fresh, virulent microbes. What he discovered in this process was that the mild disease the infected cultures gave rise to forestalled the deadly form from occurring. Pasteur was able to compare the results of previous experiments with recent ones and thereby draw accurate conclusions based on the knowledge accumulated over the course of all these experiments. Nearly a century later, the discovery of 3M’s Post-It adhesive demonstrates the role of experimentation in the discovery of both technical solutions and new market needs. The story began in 1964, when 3M chemist Spencer Silver started a series of experiments aimed at developing polymer-based adhesives. As Silver recalled:

260

Stefan Thomke

The key to the Post-It adhesive was doing the experiment. If I had sat down and factored it out beforehand, and thought about it, I wouldn’t have done the experiment. If I had limited my thinking only to what the literature said, I would have stopped. The literature was full of examples that said that you can’t do this. (Nayak and Ketteringham 1997, p. 368)

Although Silver’s discovery of a new polymer with adhesive properties departed from predictions of current theories about polymers, it would take 3M at least another five years before a market was determined for the new adhesive. Silver kept trying to sell his glue to other departments at 3M, but they were focused on finding a stronger glue that formed an unbreakable bond, not a weaker glue that only supported a piece of paper. Market tests with different concepts (such as a sticky bulletin board) were telling 3M that the Post-It concept was hopeless—until Silver met Arthur Fry. Fry, a chemist and choir director, observed that members of his choir would frequently drop bookmarks when switching between songs. ‘‘Gee,’’ wondered Fry, ‘‘if I had a little adhesive on these bookmarks, that would be just the ticket.’’ This ‘‘eureka moment’’ launched a series of experiments with the new polymer adhesive that broadened its applicability and ultimately led to a paper product that could be attached and removed, without damaging the original surface. In other words, repeated experimentation was instrumental in finding the now obvious solution, once the ‘‘eureka moment’’ occurred. While such ‘‘eureka moments’’ make for memorable history, they do not give a complete account of the various experimentation strategies, technologies, processes, and history that lead to scientific or innovative breakthroughs. After all, such moments are usually the result of many failed experiments and accumulated learning that prepare the experimenter to take advantage of the unexpected. ‘‘Chance,’’ noted Louis Pasteur, ‘‘favors only the prepared mind.’’ Consider what the authors of a careful study of Thomas Alva Edison’s invention of the electric light bulb concluded: This invention [the electric light], like most inventions, was the accomplishment of men guided largely by their common sense and their past experience, taking advantage of whatever knowledge and news should come their way, willing to try many things that didn’t work, but knowing just how to learn from failures to build up gradually the base of facts, observations, and insights that allow the occasional lucky guess—some would call it inspiration—to effect success. (Friedel and Israel 1987, p. xiii)

When firms aim for breakthrough innovations, however, senior management cannot rely on luck or even lucky guesses alone; experimentation must be organized and managed as an explicit part of a strategy for pursuing innovation itself. At the same time, the serendipitous may be more likely when an effective experimentation strategy is in place and new experimentation technologies are integrated into it. The serendipitous is also more likely when experimenters are clear that understanding what does not work is as important to learning as knowing what does.

Innovation, Experimentation, and Technological Change

261

If we attempt to add up all the significant experiments that have been carried out since the Greeks began systematic scientific studies around 400 BCE up until the 19th century, we can probably say that the number is in the millions. If we then include experiments initiated in industrial R&D laboratories since the 19th century, the number perhaps reaches several hundred million. That number, in turn, will be dwarfed by the billions or trillions of experiments we will run with computers, combinatorial technologies, and other methods in the coming decade alone, fundamentally challenging how innovation will happen. The sheer quantity of inexpensive experimentation possible with these new technologies, along with the knowledge gained from them, will make the ‘‘lucky guess’’ much more likely as long as companies are willing to fundamentally rethink how they research and develop new products and create value for their customers. Managing Uncertainty All experimentation—whether conducted in Ancient Greece, in Edison’s laboratory, or in the presence of simulation or other sophisticated technology today—generates knowledge. That knowledge, however, comes as much from failure as it does from success. Innovators learn from failure: Again, understanding what doesn’t work is as important as understanding what does. The next round of experimentation should benefit equally from either result. Further, knowledge of either failure or success itself can be stockpiled, providing a resource that, if not applicable to one set of experiments, can be used for subsequent inquiries. The fact is, when pharmaceutical companies such as Eli Lilly launch new drugs or automotive firms like BMW introduce new cars, these products are the result of as many failed experiments as successful ones. An innovation process, overall, should ensure the gradual accumulation of new knowledge that will guide the path of development itself. This new knowledge, however, is at least partially based on ‘‘accumulated failure’’ that has been carefully understood. The reason why experiments inevitably fail as part of product development effort has to do with the uncertain nature of the innovation process itself. When teams undertake the development of products or services—particularly novel or complex ones—they rarely know in advance whether a particular concept will work as intended. That means they have to find ways of rapidly discarding dysfunctional ideas while retaining others that show promise. At the same time, the ‘‘dysfunctional ideas’’ themselves have generated knowledge and should, as such, be captured. Edison understood this very well when he noted that ‘‘Just because something doesn’t do what you planned it to do doesn’t mean it’s useless. Reverses should be an incentive to great accomplishment. Results? Why, man, I have gotten lots of results! If I find 10,000 ways something won’t work, I haven’t failed. I am not discouraged,

262

Stefan Thomke

because every wrong attempt discarded is just one more step forward.’’ A century later, academic research on R&D organizations showed these insights to be more relevant than ever: Project teams spent an average of 77% of their time on experimentation and related analysis activities to resolve uncertainty (Allen 1977). Not all uncertainty is alike, however. Technical uncertainty arises from the exploration of solutions (e.g., materials) that have not been used before, or have not been combined in ‘‘this’’ way before, or miniaturized in such a way before. As such, it often relates to product functionality and can be managed through rigorous prototype testing throughout development. Production uncertainty exists when we do not know if a technical solution that works well in prototypes can also be produced cost effectively. What may work in small quantities may not be feasible when production ramps up: The entire manufacturing process itself may need to be revised. At every stage of R&D, technical and production uncertainty exists and needs to be managed, in part through a systematic process of experimentation. Beyond technical and production uncertainty, rapidly changing customer demands create need uncertainty, another critical reason for rigorous experimentation. Customers are rarely able to fully specify all of their needs because they either face uncertainty themselves or cannot articulate their needs on products that do not yet exist. If they have neither seen nor used such a product before, they themselves will have to experiment before arriving at a recommendation. Finally, when innovations are ‘‘disruptive,’’ as research has shown, market uncertainty can be so significant that firms are reluctant to allocate sufficient resources to the development of products for those markets (Christensen 1997). In such cases, the composition and needs of new markets evolve themselves and are either difficult to assess or change so quickly that they can catch good management by surprise. To successfully harness the opportunities of disruptive change, successful managers rely in part on experimentation (Garvin 2002). According to the research, such managers ‘‘planned to fail early and inexpensively in the search for the market for a disruptive technology. They found that their markets generally coalesced through an iterative process of trial, learning, and trial again’’(Christensen 1997, p. 99). An effective experimentation strategy addresses learning opportunities in all four areas: technical, production, need, and market uncertainty. Learning by Experimentation Central to experimentation is the use of models, prototypes, controlled environments, and computer simulations that allow innovators to reflect, improvise, and evaluate the many ideas that are generated in organizations: in short, to learn by trying things out (Simon 1969). In an ideal experiment, managers and engineers sepa-

Innovation, Experimentation, and Technological Change

263

rate an independent variable (the ‘‘cause’’) and dependent variable (the ‘‘effect’’) and then manipulate the former to observe changes in the latter. The manipulation, followed by careful observation and analysis, then gives rise to learning about relationships between cause and effect that, ideally, can be applied to or tested in other settings. In real-world experimentation, environments are constantly changing, linkages between variables are complex and poorly understood, and often the variables are uncertain or unknown themselves. The result is iteration: Innovators make progress through iterative experimentation that is guided by some insight where a solution might lie. In fact, all experimentation involves iteration, sooner or later. When all relevant variables are known, formal statistical techniques and protocols allow for the most efficient design and analysis of experiments. These techniques are used widely in many fields of process and product optimization today and can be traced to the first half of the 20th century when the statistician and geneticist Sir Ronald Fisher first applied them to agricultural and biological science (Fisher 1921). However, when independent and dependent variables themselves are uncertain, unknown, or difficult to measure, experimentation itself is much more informal or tentative. Iterative experimentation goes on all the time and is so much an integral part of innovation processes that it has become like breathing—we do it but are not fully aware of the fact that we are really experimenting. Moreover, good experimentation goes well beyond the individual or the experimental protocols but has implications for firms in the way they manage, organize, and structure innovation processes. It isn’t just about generating information by itself but about how firms can learn from trial and error and structured experimentation. The rate of learning possible is influenced by a number of factors, some affecting the process and others how it is managed. The following factors impact how learning through experimentation occurs (or does not occur): fidelity, cost, feedback time, capacity, sequential and parallel strategies, signal-to-noise, and type all enhance the power of experimentation (Thomke 2003, chapter 3). It is through these factors that managers can change how their organizations learn from experiments. The Changing Economics of Experimentation Traditionally, testing has been relatively expensive, so companies had to be parsimonious with the number of experimental iterations. To overcome this, managers essentially have had two choices available to them: change the fundamental economics of experimentation through new technologies and process innovation or try to get more out of each experiment itself—make experiments more efficient. ‘‘Design of experiments’’ that employs sophisticated statistical methods has focused primarily on the latter and, as already mentioned, has had a significant impact on

264

Stefan Thomke

how R&D is done (Montgomery 1991). By manipulating multiple variables in a single experiment, while maintaining integrity in its data analysis, scientists and engineers have been able to get more learning out of their experiments than in the past. Experiments per se can often be made more efficient, of course: This chapter does not deny the importance of the many excellent methods derived from statistics theory. Alternatively, new technologies that slash experimentation cost and time not only bring much needed capacity but can also make possible ‘‘what-if’’ experiments that, up to now, have been either too expensive or nearly impossible to run. What if an airplane, a car, a drug, or a business were designed in a particular way? By employing new experimentation technologies, it’s possible to explore the assumptions that underlie a design, how they could be changed, and what the consequences would be, positive and negative. Further, if positive, we can see implications for product, process, and system improvements. Indeed, these technologies hold the possibility of improving themselves, as has happened in the integrated circuit industry. For example, new technologies such as computer modeling and simulation, rapid prototyping, and combinatorial chemistry allow companies to create more learning more rapidly; we see how that knowledge can be incorporated in more experiments at less expense. Indeed, new information-based technologies have driven down the marginal costs of experimentation, just as they have decreased the marginal costs in some production and distribution systems. Moreover, an experimental system that integrates new information-based technologies does more than lower costs; it also increases the opportunities for innovation. That is, some technologies can make existing experimental activities more efficient, while others introduce entirely new ways of discovering novel concepts and solutions (see Figure 1). The multibillion dollar integrated circuit industry has been transformed by new technologies more than once. In fact, the exponential performance gains of integrated circuits, accompanied by better models, have fueled dramatic advances in computer simulation and tools for design automation used in many fields. These advances have come full circle: Today’s complex chips would be impossible to design and manufacture without the tools that they helped to create. The changes have thus far affected businesses with high costs of product development, such as the pharmaceutical, automotive, semiconductor, and software industries. My ten years of research in these industries suggest that as the cost of computing and other combinatorial technologies keeps falling—thereby making complex calculations faster and cheaper—and as new combinatorial technologies and our knowledge of building models emerge, virtually all companies will discover that they have a greater capacity for rapid experimentation to investigate diverse concepts. Financial institutions, for example, now use computer simulations to test

Innovation, Experimentation, and Technological Change

265

Figure 1 Advances in problems being solved by computer simulation. Growth of computer technology since 1955, showing advances in average commercial performance and milestone events (redrawn with modifications from Brenner 1996). Problems that are solvable in reasonable times at the indicated level of computer performance are shown in brackets. Approximate system prices are shown in dollars at the time.

266

Stefan Thomke

new financial instruments. In fact, the development of spreadsheet software itself has forever changed financial modeling; even novices can perform many sophisticated what-if experiments that were once prohibitively expensive. The enthusiasm for these technologies gets a dose of reality when we turn to how they are used in innovation processes. When knowing what does and doesn’t work can happen much more rapidly and frequently, the promise these technologies hold is often diluted when they are not used to maximize the power of experimentation: Organizational ‘‘realities’’ get in the way of more effective experimentation. Drawing on years of research in the global automotive industry has shown how the introduction of computer-aided design and engineering (CAD and CAE) technologies has suffered from several predicaments. In particular, processes and people limit the impact of new technologies, organizational interfaces can get in the way of experimentation cycles, and technologies often change faster than human and organizational behavior. Shifting the Locus of Experimentation to Customers So far, we have discussed primarily the management of experimentation within companies: how to change the processes, organization, and management of innovation to tap into the power of experimentation and the opportunities that new technologies provide for. That in itself will challenge most organizations, but the strong evidence from industries as diverse as automotive, financial, semiconductors, and software has shown that it can be done. The exciting part is that the changes are not just about raising productivity but also fundamentally changing the kinds of products and services that are created, leading to innovations that simply weren’t possible before. What would happen if managers took experimentation to another level—beyond organizations—where they can change the way companies create new products and services with customers and suppliers?2 Specifically, by putting experimentation technologies into the hands of customers, managers can tap into possibly the largest source of dormant experimentation capacity. Not only can shifting experimentation to customers result in faster development of products that are better suited to their needs, but their experimentation could also result in innovations we simply cannot imagine today. Some companies have abandoned their efforts to understand exactly what products their customers want and have instead equipped them with tools to design and develop their own new products, ranging from minor modifications to major new innovations. The user-friendly tools, often integrated into a ‘‘toolkit’’ package, deploy new technologies (e.g., computer simulation and rapid prototyping) to make

Innovation, Experimentation, and Technological Change

267

innovation faster, less expensive, and, most importantly, better as customers run ‘‘what-if’’ experiments themselves. A variety of industries have started to use this approach. Bush Boake Allen (BBA), a global supplier of specialty flavors that has recently been acquired by International Flavors and Fragrances (IFF), has developed a toolkit that will enable its customers to develop their own flavors, which BBA then manufactures. In the materials field, General Electric provides customers with Web-based tools for designing better plastic products. In software, a number of companies allow people to add customdesigned modules to their standard products and then commercialize the best of those components. Indeed, shifting experimentation and innovation to customers has the power to completely transform industries. Product development is often difficult because the ‘‘need’’ information (what the customer wants) resides with the customer, and the ‘‘solution’’ information (how to satisfy those needs) lies with the manufacturer (see Figure 2). Traditionally, the onus has been on manufacturers to collect the customer need information through various means, including market research and information gathered from the field. The process can be costly and time consuming because customer needs are often complex, subtle, and fast changing. Frequently, customers don’t fully understand their needs until they try out prototypes to explore exactly what does—and doesn’t—

Figure 2 Moving information between supplier and customers. Note: Traditionally, ‘‘need’’ information is primarily collected and moved from customers to suppliers via market research methods (left side). This process can be costly and time consuming when needs are unique, complex, and fast-changing. In the new model, the supplier’s ‘‘solution’’ information is embodied in innovation toolkits that are moved to customers so that they can experiment and design their own products (right side).

268

Stefan Thomke

work. Many companies are familiar with customers’ reactions when they see and use a finished product for the first time: ‘‘This is exactly what I asked you to develop, but it is not what I want.’’ In other words, customers learn about their needs through informal experimentation while using new products or services. Not surprisingly, traditional product development is a drawn-out process of trial and error, often ping-ponging between manufacturer and customer. First, the manufacturer develops a prototype based on information from customers that is incomplete and only partially correct. The customer then tries out the product, finds flaws, and requests corrections. This ‘‘learning by experimentation’’ cycle repeats until a satisfactory solution is reached, frequently requiring many costly and time-consuming iterations. When companies have to work with hundreds of customers with different needs, each requiring market research and a well-managed iteration process, one can see how those firms become overwhelmed and thus focus only on the largest and most profitable customers. Innovation Toolkits for Experimenters: Custom Integrated Circuits The reason why some companies may want customers to experiment with alternative design solutions has to do with the fact that it can be a win–win proposition for both. Instead of moving ‘‘need’’ information from customers to a supplier, ‘‘solution’’ information is moved from a supplier to customers via innovation toolkits. This puts experimentation power into the hands of users, who become an integral part of a company’s innovation process. The manufacturer can focus on developing better solution platforms that are customized through user-friendly toolkits in the hands of customers. The customer can experiment and get feedback more rapidly, control intellectual property on the application-specific part of a design, and, most importantly, find a solution that closely matches her needs. It is important to note that shifting product development activities to customers does not eliminate learning by experimentation—nor should it. What it does is make traditional product development better and faster—for several reasons. First, a company can bypass the expensive and error-prone effort to understand customer needs in detail. Second, the trial-and-error cycles that inevitably occur during product development can progress much more quickly because the iterations will be performed solely by the customer. To understand the major impact that shifting design and experimentation to customers can have, consider the integrated circuit industry. Its history holds several profound lessons about how the right toolkit can turn a market on its ear. During the late 1970s, a typical customer of specialized semiconductors, such as a toy manufacturer that needed circuitry to operate its new robotic dog, might have hired a

Innovation, Experimentation, and Technological Change

269

chip supplier to develop a custom design. Because that process was complicated and costly, the chip company could afford to undertake projects only for high-volume customers. Smaller customers were turned away and had no choice but to use standard circuit components. That, in turn, limited the extent to which they could develop products that were smaller, better in performance, lower in cost, or simply more innovative. In other words, there was a very large unfilled demand for custom chips because the dominant suppliers couldn’t economically serve smaller customers. Then a handful of start-ups turned everything upside down. New firms like LSI Logic and VLSI Technology provided both large and small customers with do-ityourself tools that enabled them to design their own specialized chips.3 As Wilf Corrigan, LSI’s CEO and principal founder, recalled: Having decided to start a semiconductor company, I spent quite a few months visiting people I knew in the computer industry, both in the United States and in Europe. A pattern started to emerge: Most of these companies needed custom circuits. I had flashbacks of when I was at Fairchild and good customers would ask me, ‘‘Would you do a custom program for us?’’ At Fairchild, I always said no, but in retrospect, I should have recognized that over the previous few years there was an opportunity bubbling up in custom circuits . . . . Custom circuits—the big guys don’t want to mess with them, the customers seem to need them, yet there don’t seem to be any viable sources in spite of increasing demand. (Walker 1992, p. 47)

Corrigan’s inability to respond while working at Fairchild, a leading semiconductor firm in the 1960s and 1970s, was certainly not because he lacked influence: He was its CEO from 1974 to 1980. In fact, many members of LSI’s founding team and its design tool expertise came from Fairchild Semiconductor, which had abandoned its custom IC efforts because it was losing money for the firm. With LSI’s new development system, customers could benefit by getting what they wanted through their own experimentation, and the fledgling chip firms could profit by manufacturing those customer designs. The win–win solution was right on the money. Between the 1980s and 2000, the market for such custom integrated circuits has soared from virtually nothing to more than $15 billion, with the number of customers growing from a handful of high-volume buyers to hundreds of thousands of firms with very diverse end-user applications. One of the keys to that market is the toolkit technology. In principle, outsourcing custom design to customers can help slash development times and costs, but customers are not experts in a supplier’s R&D or production process. So how could customers possibly be expected to create custom designs that are producible on a manufacturer’s sophisticated process equipment? The answer to that was found in a major shift that had been taking place in the semiconductor industry. Traditionally, specialized information used by a manufacturer to design and build custom products has been locked in the minds of the company’s development engineers.

270

Stefan Thomke

This knowledge accumulates over decades of experience. In recent years, companies have been able to incorporate a considerable amount of this human expertise into computer-based tools. These computer-aided design, engineering, and manufacturing programs (CAD/ CAE/CAM) have grown increasingly sophisticated, and many now contain libraries of tested and debugged modules that people can simply plug into a new design. The best tools also enable rapid testing through computer simulation and provide links to automated equipment that can build prototypes quickly. This leading-edge technology, which manufacturers had been using internally to increase R&D productivity and innovation, has become the basic building block for toolkits geared to customers. When LSI was founded in 1981, R&D engineers at large semiconductor companies were already using many elements of the customer toolkit, but there was no integrated system that less-skilled customers would be comfortable with. So LSI bought some of the pieces, made them customer-friendly by adding graphical user interfaces, and integrated them into a package called the LSI Development System (LDS). The result was a packaged toolkit that enabled customers to design their own chips with little support from LSI. The insight that made possible a toolkit for less-skilled customers was that the design of the chip’s fundamental elements, such as its transistors, could be standardized and could incorporate the manufacturer’s solution information of how semiconductors are fabricated. Then, all the information the customer needed about how the chip would function could be concentrated within the electrical wiring that connects those fundamental elements. In other words, this new type of chip, called a ‘‘gate array,’’ had a novel architecture created specifically to separate the manufacturer’s solution information from the customer’s need information. As a result, all customers had to do was use a toolkit that could interconnect a gate array based on their specific needs. For its part, LSI had to rethink how to make its production processes more flexible so that it could manufacture the custom chips at low cost. Customer toolkits based on gate-array technology offer the four major capabilities. They contain a range of tools, including those to test a design, that enable users to create their own prototypes via trial and error. They are customer-friendly in that they use Boolean algebra, which is the design language commonly taught to electrical engineers. They contain extensive libraries of pretested circuit modules. And they also contain information about production processes so that users can test their designs to ensure that they can be manufactured. Interestingly, more recent technology—integrated circuits called field programmable logic devices (FPLDs)—enables the customer to become both the designer and the manufacturer. FPLDs are one of a family of programmable chip technologies

Innovation, Experimentation, and Technological Change

271

where links between components are either created, ‘‘melted’’ in the case of fuse technologies, or programmed in the case of field-programmable gate arrays (FPGAs). Supplier such as Xilinx, Inc., or Altera Corporation prefabricate these chips and sell them to customers who use their design and simulation software and equipment to program chips for themselves. Suppliers do not have to be involved in the design process and physical prototypes can be prepared by customers at little cost or time. Benefits and Challenges of Toolkits Well-designed customer toolkits, such as those developed for the design of custom semiconductor chips, offer several major advantages. First, they are significantly better at satisfying subtle aspects of customer need because customers know what they need better than manufacturers do. Second, designs will usually be completed much faster because customers can experiment at their own site, with minimal supplier involvement. Third, if customers follow the rules embedded in a toolkit (and if all the technological bugs have been worked out), their designs can be manufactured the first time around. And there are ancillary benefits. Toolkits enable a company to retain small customers that might have been prohibitively expensive to work with before, thus expanding the accessible market—and the number of product innovations. By serving these smaller clients, toolkits also reduce the pool of unserved, frustrated potential customers who might turn to competitors or to new entrants into the market. Furthermore, they allow companies to better serve their larger, preferred customers. That’s a benefit most suppliers wouldn’t expect, because they’d assume that their bigger customers would prefer the traditional hand-holding to which they’re so accustomed. Experience shows, however, that such customers are often willing to use a toolkit, especially when fast product turnaround is crucial. Of course, toolkits will not satisfy every type of customer. For one thing, they are generally not able to handle every type of design. Also, they create products that are typically not as technically sophisticated as those developed by experienced engineers at a manufacturer using conventional methods. So manufacturers may continue to design certain products (those with difficult technical demands) while customers take over the design of others (those that require quick turnarounds or a detailed and accurate understanding of the customer’s need). And if homogenous markets require standard products, the traditional approach of deep market research will probably work better. The business challenges of implementing a toolkit can be daunting. Turning customers into innovators requires no less than a radical change in management mind-set. Pioneers LSI Logic and VLSI Technology were successful because they

272

Stefan Thomke

abandoned a principle that had long dominated conventional management thinking at leading companies like IBM, Intel, and Fujitsu. For many years, these companies had assumed that their interests would best be served by keeping design expertise, tools, and technologies away from customers. In contrast, LSI and the other industry upstarts understood that they needed to do just the opposite by putting robust, user-friendly toolkits into customers’ hands. Such a dramatic shift in mind-set required a thorough rethinking of wellentrenched business practices. In essence, a company that shifts experimentation to customers is outsourcing a valuable service that was once proprietary, and the change can be traumatic if that capability has long been a major source of competitive advantage. For example, a common problem is resistance from sales and marketing departments, which have traditionally been responsible for managing relationships with customers and providing first-class service to them. With toolkits, computer-to-computer interactions replace intense person-to-person contact during product development. In other words, customers who design products themselves have little need for a manufacturer’s sales or marketing department to determine what they need. If this change affects the compensation of sales representatives in the field, it could easily derail any efforts to alter the company’s business model. As a result, senior management needs to face these issues head-on—for example, by determining how the sales and marketing functions should evolve and by using specific incentives to induce employees to support the transformation. Creating and Capturing Value: Industry Effects Perhaps the most important lesson to be learned from our research is that a company that wants to shift experimentation and innovation to customers must adapt its business accordingly. In fact, adopting the new approach is neither easy nor straightforward. We have found that because the value of customer toolkits tends to migrate, a company must continually reposition itself to capture that value. When a supplier introduces a toolkit, the technology first tends to be company specific: The designs can only be produced in the factory of the company that developed the toolkit. This creates a huge short-term advantage for the pioneering supplier, which can reduce its custom design costs because they are partially outsourced to customers. That, in turn, enables the supplier to serve more customers. And because the customers’ designs must be produced on the supplier’s system, the supplier doesn’t risk losing any business. But the loss of leverage by customers represents a fundamental shift. Traditionally, in the field of specialized industrial products, companies interested in a customer’s business must develop a custom design and submit it for evaluation. The

Innovation, Experimentation, and Technological Change

273

customer picks the proposal from one supplier, and the others are saddled with a loss for their time and investment. A toolkit tied to a single supplier changes that dynamic: A customer who develops a design using the toolkit cannot ask for competing quotes because only one company can manufacture it. Of course, customers would prefer the advantages of a toolkit without the associated loss of leverage. In the long run, this type of solution tends to emerge: Customer pressure induces third parties to introduce toolkits that can create designs to fit any supplier’s manufacturing process. Or, in a slight variation, customers complain until a company that owns a dominant toolkit is forced to allow a spin-off to evolve the technology into a supplier-neutral form. Then, customers are free to shop their designs around to competing manufacturers. In other words, the long-term result of customer toolkits is that manufacturers lose a portion of the value they have traditionally delivered. But if the conditions are ripe for the technology to emerge in a given industry and if customers will benefit from it—and our research shows that they will—then suppliers really don’t have a choice. Some company will eventually introduce a toolkit and reap the short-term advantages. Then, others must follow. In the case of custom integrated circuits, the lessons are striking. Fujitsu initially resisted making its in-house design technology available to customers, thinking the move was too risky. Aspects of this dilemma are neatly captured by a conversation between Wilf Corrigan, LSI’s cofounder and CEO, and Mr. Yasufuku, a senior executive at Fujitsu. Corrigan explained LSI’s strategy of shifting design and experimentation to customers to Yasufuku. (At the time of this conversation, LSI was a startup and Fujitsu was an established major player in the custom integrated circuit market with a major market share.) ‘‘We are going into the gate array business with software tools which our customers will have in their own hands.’’ Yasufuku responded: ‘‘That is a brilliant strategy. If you do that and the software is good, you will win.’’ Corrigan then asked, ‘‘Why don’t you do that?’’ (Fujitsu had developed an excellent set of internal design tools.) The answer: ‘‘Our software is so valuable that if we expose it to outsiders, they will steal it.’’ Fujitsu hadn’t even transferred their tools to its U.S. subsidiary because they were afraid of losing control of it (Walker 1992, pp. 79–90). But the outcome in the custom integrated circuits market showed that the choice facing firms in that marketplace really was only one of timing. After LSI introduced its toolkit and design methodology to the marketplace, customers showed that they greatly preferred the option of ‘‘doing it themselves’’ by moving business to LSI. Faced with eroding market share, Fujitsu and other established suppliers were forced to also adopt the toolkits approach. The cost of delay: LSI had been given a window to grow from an insignificant startup into a major player and, together with competitors such as VLSI Technology, captured much value of the underserved

274

Stefan Thomke

market for custom chips. With continued developments such as field programmable technologies offered by firms such Xilinx Inc. and Altera Corporation, LSI’s move transformed the industry and created billions of dollars’ worth of new value. Conclusion Tapping into the inventiveness and imagination of customers—not just R&D departments—can indeed generate tremendous value, but capturing that value is hardly a simple or straightforward process. Not only must companies develop the right design toolkit, they must also revamp their business models as well as their management mind-sets. When companies relinquish a fundamental task—such as designing a new product—to customers, the two parties must redefine their relationship, and this change can be risky. With custom computer chips, for instance, companies traditionally captured value by both designing and manufacturing innovative products. Now, with customers taking over more of the design task, companies must focus more intently on providing the best custom manufacturing and design tools. In other words, the location where value is created and captured changes, and companies must reconfigure their business models accordingly. Notes 1. The material in this chapter is adapted (with modifications) from Thomke 2003. 2. The research for the remainder of the chapter was done jointly with Eric von Hippel (see Thomke and von Hippel 2002). 3. The LSI Logic history is based on interviews published in Walker (1992).

References For a much more exhaustive bibliography on experimentation and innovation, see Thomke 2003. Allen, T. (1977). Managing the Flow of Technology. Cambridge, MA: MIT Press. Brenner, A. (1996). ‘‘The Computer Revolution and the Physics Community.’’ Physics Today 46: 24–39. Christensen, C. (1997). The Innovator’s Dilemma: When New Technologies Cause Great Firms to Fail. Boston: Harvard Business School Press. Clark, K., and T. Fujimoto (1991). Product Development Performance. Boston: Harvard Business School Press. Fisher, R. (1921). ‘‘Studies in Crop Variation: I. An Examination of the Yield of Dressed Grain from Broadbalk.’’ Journal of Agricultural Science 11: 107–135.

Innovation, Experimentation, and Technological Change

275

Friedel, R., and P. Israel (1987). Edison’s Electrical Light: Biography of an Invention. Piscataway, NJ: Rutgers University Press. Garvin, D. (2002). ‘‘A Note on Corporate Venturing and New Business Creation.’’ Note 302-091. Boston: Harvard Business School. Hare, R. (1981). Great Scientific Experiments. Phaidon Press. Iansiti, M. (1997). Technology Integration: Making Critical Choices in a Turbulent World. Boston: Harvard Business School Press. Leonard-Barton, D. (1995). Wellsprings of Knowledge. Boston: Harvard Business School Press. Millard, A. (1990). Edison and the Business of Innovation. Baltimore, MD: Johns Hopkins University. Montgomery, D. (1991). Design and Analysis of Experiments. New York: John Wiley & Sons. Nayak, P., and J. Ketteringham (1997). ‘‘3M’s Post-It Notes: A Managed or Accidental Innovation?’’ In The Human Side of Managing Technological Innovation, R. Katz, editor. Oxford, UK: Oxford University Press. Nelson, R., and S. Winter (1982). An Evolutionary Theory of Economic Change. Boston: Belknap Press of Harvard University Press. Peters, T., and R. Waterman (1982). In Search of Excellence. New York: Harper & Row. Pisano, G. (1997). The Development Factory. Boston: Harvard Business School Press. Simon, H. (1969). The Sciences of the Artifical. Cambridge, MA: MIT Press. Thomke, S. (2003). Experimentation Matters. Boston: Harvard Business School Press. ———, and E. von Hippel (2002). ‘‘Customers as Innovators: A New Way to Create Value.’’ Harvard Business Review (April). von Hippel, E. (1988). The Sources of Innovation. Oxford, UK: Oxford University Press. Walker, Rob (1992). Silicon Destiny: The History of Application Specific Integrated Circuits and LSI Logic Corporation. Milpitas, CA: C.M.C. Publications.

17 Knowledge, Platforms, and the Division of Labor W. Edward Steinmueller

Introduction Modern economics makes its contribution to understanding a complex social world through the use of assumptions that smooth the wrinkles produced by the operation of real world markets and institutions. Economists interested in the study of knowledge and technology have devoted considerable attention to reconsidering one of these simplifying assumptions—that knowledge about available production and exchange possibilities is widely dispersed among economic agents.1 Two different approaches have been employed in reconsidering the ‘‘widespread knowledge’’ assumption. The first approach retains the common assumption in economics that knowledge and information are equivalent and focuses on the implications of assuming that rather than identical endowments, economic agents have asymmetric endowments of information. The second approach rejects the assumption that information and knowledge are equivalent and investigates how knowledge might be generated, reproduced, and exchanged if it is not equivalent to information. Both of these approaches offer important insights about a central theme of modern economics—the interfirm and international division of labor. The internationalization or globalization of production involves interfirm division of labor in the creation of products and services, the outsourcing of component and subsystem production, and the generation and exchange of knowledge about product and service design and implementation. Analyzing how this interfirm division of labor occurs for specific technologies from both the endowment and from the generation and exchange viewpoints provides a way to account for marked differences in industrial structure—the division of labor between firms and the global ‘‘reach’’ of companies. Because much of the knowledge relevant to economic activity is of a specialized nature, many institutions that are relevant to knowledge generation and exchange operate within the more delimited context of a specific industry—for example, strategic technological groupings (Mowery et al. 1996), organizations responsible

278

W. Edward Steinmueller

for compatibility and reference standards making and enforcement (David and Greenstein 1990), and industrial associations (Procassini 1995). The institutional connections that bind together interorganizational cooperation in the generation and exchange of knowledge may be examined in terms of market exchange (Arora et al. 2001), or the governance of these relationships may include more complicated social and contractual issues. Taking the latter approach leads to the idea of ‘‘sectoral’’ systems of innovation that complement and operate in parallel with market exchange (Edquist 1997; Malerba 2004). Proceeding further toward a more disaggregated unit of analysis, it is possible to consider interorganizational institutional arrangements that link networks of firms to achieve coordination and complementarities in the production of components that may be integrated into system products (Steinmueller 2003a). This chapter focuses on the market and institutional ties that bind together networks of firms involved in the production of what we will call ‘‘platforms’’— complete product or service offerings built from components produced by a variety of different companies.2 The aim of this chapter is to identify and discuss the implications of both positive (how does it work?) and normative (how should it work?) features of the division of labor involved in the construction of product and service platforms comprised of components that are mostly produced by different companies. The term ‘‘platform’’ is used to suggest some degree of flexibility in the combination of these components and to convey the interdependence between the ‘‘platform integrator,’’ who usually bears the principal responsibility for the market promotion of the platform, and the suppliers of components that can be integrated in a platform. The degree of flexibility in combining inputs is the primary feature distinguishing platforms from ‘‘modular systems’’—the latter are defined in terms of strict interoperability and the (not always reliable) assurance that components can be freely interchanged. Examples of platforms include personal computers, automobiles, airplanes, buildings, travel and tourism services, and large retail companies. The first three examples are systems that principally involve the integration of physical components, and only the first of these, personal computers, is dominated by modular components.3 The fourth example, building construction, involves the integration of physical components and a collection of service components within a variety of different application contexts—with varying levels of compatibility. For the last two examples, although physical artifacts are used, the successful integration of service components is of central importance to their success. The variety of the examples cited highlights the idea of the platform employed in this chapter—the platforms to be discussed not only involve systems with interoperable components that can be freely interchanged but also include systems in which component suppliers can reuse the

Knowledge, Platforms, and the Division of Labor

279

knowledge gained in serving one platform producer by using it to address the needs of other platform producers. In other words, platforms are systems where integration involves a broader set of understandings about ‘‘compatibilities’’ between the components employed. The extension of the concept of ‘‘compatibilities’’ comes at some cost in the precision of definition but helps to address the positive and normative questions of this chapter. For example, it provides a way to consider whether there are important distinctions between ‘‘compatibilities’’ in different industries that serve as limits or boundaries to the division of labor in the production of components. In particular, it suggests the need to make distinctions between industries producing systems defined by interoperable components with technological interface compatibility standards such as electronic or communication systems and other systems defined by interfirm agreements (institutions) about the integration of components such as buildings or travel and retail services. These differences provide a basis for evaluating the limits to interfirm and international division of labor. The contemporaneous context of this progressive division of labor is especially interesting for two reasons. First, many of the means for coordinating the planning and execution of business processes whereby components are planned, designed, produced, delivered, and integrated into the platform rely upon the Internet, whose commercial use has gathered momentum only during the past decade. Second, many of the techniques that are used to manage the integration of components on platforms employ relatively recent developments in information technology, which provide a precise means of specifying the relationship between these components; in some cases, they also provide a means of simulating the performance or other features of the results of integration. This chapter is organized in two major sections—the first is concerned with analyzing the nature of platforms and the role of technical compatibility standards and related institutions in enabling the creation and extension of platforms. A particular focus of this section is on the limits to the growth of ‘‘modular’’ architectures from the electronics to other industries—‘‘modular’’ architectures are based on technical compatibility standards. The second major section addresses the value of achieving a higher degree of ‘‘compatibility’’ and actions that might be taken to promote and govern the movement toward more ‘‘standardized’’ components in the construction of platforms. Modularity and Platforms Production of systems with heterogeneous components requires innovation in organizational and management techniques and particularly in the ways that knowledge

280

W. Edward Steinmueller

is accumulated by the coordinators or integrators of such systems.4 A key feature of the new methods of control devised specifically to coordinate networks of component suppliers is the explicit specification of organizational and technological ‘‘interfaces’’ that establish the relationships between the suppliers of system components and the components that are employed in systems. The concern with interfaces allowed the development of a modular approach to systems integration. The value of modularity was clearly recognized at an early stage by those involved with computer sciences (Simon 1969, 1996). In this context, modularity was of fundamental importance in localizing defects or undesirable behavior in large systems. Modularity also provided a framework for the division of labor in the construction of large systems and hence a ‘‘discipline of design’’ that could be extended from computers and information systems to other system contexts. Specifically, in the design of computer systems, explicit definition of interfaces provides a means to achieve interoperability (mutual compatibility) between components of a system. This modular approach has been extended from large computer design to a vast array of other electronic designs. By the mid-1970s, the technical literature was explicitly suggesting that electronics designers should embrace technical compatibility standards as the way forward to meeting the demands for variety that they faced—standard integrated circuits could be viewed as ‘‘general purpose’’ logic devices from which a wide variety of electronic systems could be created (Blakeslee 1975). The power of this approach was augmented by the steep trajectory of increase in integrated circuit complexity commonly referred to as Moore’s law, which has served as a map for technological opportunity in the integrated circuit and electronics industry for almost four decades. System modularity employing integrated circuit components has steadily expanded through a process of creating ever more capable ‘‘modules’’ and the proliferation of systems designs implementing increasingly complex networks of these modules. Is the experience of the electronics industry an ‘‘early indicator’’ of the potentials for a broader paradigm of industrial organization based on ‘‘modularity’’? This claim is developed by Baldwin and Clark (1997, 2000). In their approach, modularity provides vast new opportunities for the division of labor across organizational boundaries because where technological and organizational interfaces can be effectively defined, opportunities for the entry and specialization of producers create a powerful engine for innovation and technological advance. While drawing heavily on the experience of the computer industry, the vision presented by Baldwin and Clark extends to other industries and offers a ‘‘paradigm’’ or way of thinking about how to improve upon the performance of any industry in which ‘‘systems’’ play an important role.

Knowledge, Platforms, and the Division of Labor

281

This is a powerful vision deserving of careful critical scrutiny. Modularity involves distributing knowledge throughout an industry, a precise definition of the interfaces (physical or organizational) between the components in a system, and an industrial structure with considerable intercompany division of labor and specialization. These features provide a basis for comparative assessment. Why are modular architectures not more broadly deployed in the construction of platforms? One answer is that there may be important and persistent obstacles to achieving higher degrees of compatibility. Although the electronics industry is able to overcome these roadblocks, other industries are unlikely to be able to do so in the same ways or to the same extent. Since the electronics industry is both the origin of the modularity vision and the industry in which the approach is most advanced, it is appropriate to examine obstacles in the realization of the full potential of modularity that have persisted in this industry and whether they can be more effectively addressed.5 All of these problems relate to the interaction between knowledge and the organizational and technological ‘‘interfaces.’’ The features of these problems are introduced in the following subsections. Specifying the Platform—The Role of Standards In order to understand the alternative methods for specifying a platform, it is useful to consider two examples—the Cuisinart food processor and the IBM-compatible or WINTEL personal computer.6 The Cuisinart food processor is a system employing a variety of components (blades, ice cream bowls, dough hooks, blenders, etc.) that, in principle, could be imitated by alternative suppliers to create a ‘‘Cuisinartcompatible’’ market for components. Such a market does not appear to exist. Carl G. Sontheimer, the inventor of the Cuisinart, and his colleagues made a strategic choice to identify their product from the beginning as a ‘‘food processor’’ to support trademark rights in the name ‘‘Cuisinart’’ and were granted patents for features of their food processor design.7 The trademark and patents, however, did not prevent other companies such as Philips, Sanyo, and Kenwood from producing competing brands of food processors with components that are incompatible with one another. The IBM personal computer, introduced in 1981, provides a different model for platform definition. The hardware interface standards defining the IBM personal computer platform involved ‘‘card slots,’’ a parallel printer port, two serial ports, and a video output port. Devices that observed the technical standards for one of these interfaces could be electronically connected to the IBM PC and, with appropriate software, it was possible to receive and transmit data through these interfaces for purposes such as data capture, communication, and display. One of the most remarkable features of this platform definition was its degree of ‘‘inclusivity’’ or

282

W. Edward Steinmueller

‘‘openness’’—several of the interfaces (e.g., the parallel and serial port definitions) were based upon industry standards that were available for general use, and for the interfaces that IBM did control, such as the expansion card connectors, IBM did not initially adopt an exclusive strategy. Moreover, even though proprietary, the definition of the software for manipulating these communication interfaces and communicating data within the IBM PC (the BIOS) was possible to imitate. In other words, IBM had created a platform that rival platform producers could duplicate or ‘‘clone,’’ leading to a progressive ‘‘opening’’ of the IBM PC and direct competition with products from other companies. Although the example of the IBM PC is sometimes used as an illustration of the danger of ‘‘losing control’’ of the standards for platform definition, it can also be argued that IBM’s strategy greatly accelerated the rate of adoption of personal computers and that IBM benefited substantially from the resulting market expansion, which stimulated demand for all sorts of other hardware and software that IBM produced. In the early part of this process, IBM nearly extinguished its primary rival platform producer (Apple Computer Inc.), which eventually regained some of its position by adopting a platform compatibility strategy.8 As the process of ‘‘opening’’ has proceeded, however, IBM has been displaced as the primary manufacturer of personal computers by a variety of PC ‘‘compatible’’ producers, and the component producers, Microsoft and Intel, have become the principal companies defining the future of the WINTEL personal computer standard.9 There are several possible ways to account for the differences in the interfirm division of labor between the food processor and personal computer platforms. For the present purposes, however, the key lies in the ‘‘how’’ rather than the ‘‘why.’’ This key point is that the standards defining the components of food processors and personal computers are controlled by different mechanisms.10 In the case of the food processor, these standards are proprietary and closed to access by others, while in the case of the personal computer, standards that were originally intended to be proprietary and closed were ‘‘opened’’ through imitative competition. However, there is another possibility. Networks or ‘‘clubs’’ of companies may define technical compatibility standards that exclude rivals but that they negotiate among themselves. In many complex product industries, the system integrator plays a key role in codifying, managing, and certifying the knowledge necessary to ensure compatibility between components (Steinmueller 2003a). In these industries (e.g., building construction or aircraft and aerospace), however, it is common for a considerable amount of the knowledge to be dispersed among component producers and for the compatibility standards to be negotiated between the system integrator

Knowledge, Platforms, and the Division of Labor

283

and component producers. In other words, it is possible for technical standards to be neither proprietary (exclusively controlled by a single company, as in the case of Cuisinart) nor relatively open and nonexclusively owned by industry participants (as in the case of the WINTEL personal computer). Despite considerable discussion of open standards in recent years, there remains considerable confusion about what constitutes an open standard. As suggested by the discussion so far, most standards for the compatibility and interoperability of components within platforms are interfirm standards defined either by a single company or by negotiation among a group of companies. Nonetheless, there is a large class of standards that are published by public and private standards organizations. These standards are generally shielded from duplication by some type of intellectual property protection—patents, copyrights, or commercial (trade) secrets. One definition of ‘‘open standards’’ is that they are published and, to the extent that they incorporate intellectual property that is essential for the standard’s implementation, this intellectual property is available on fair, reasonable, and nondiscriminatory terms and conditions. This approach has been adopted by most of the world’s standards making bodies, which operate with explicit or implicit government endorsement.11 An alternative definition is that ‘‘open’’ standards can be accessed without restriction, without having to ask permission or to negotiate the terms of access. To achieve these aims requires the use of specific terms in the license for the use of intellectual property. The latter concept of openness is derived from efforts to define such licenses by the World Wide Web Consortium (W 3 C) and the free/open source software (F/LOSS) communities.12 In considering the differences between the Cuisinart and the WINTEL personal computer, the key issue is whether it is possible and desirable to ‘‘open’’ the compatibility standards that define the platform to allow other producers to supply components and perhaps to produce competing versions of the platform using the same components. By foreclosing this market development, Cuisinart avoids a Cuisinartcompatible and a Cuisinart-clone market with interoperable components but faces a market in which many other home appliance producers create rival food processor systems. The producers of WINTEL personal computers and components have a common interest in maintaining a number of ‘‘open’’ standards to ensure that their products will be complements. In doing so, they enable each other as well as new rivals to compete in producing components or entire systems. The WINTEL personal computer is a platform with an amalgam of proprietary and ‘‘open’’ standards. Three categories of compatibility standards control processes can be discerned: (1) single company controlled, (2) ‘‘open’’ (possibly established or endorsed by a public standards making organization), and (3) negotiated. Each has different implications

284

W. Edward Steinmueller

for the processes of generating and exchanging the knowledge necessary to specify a platform. Single company controlled standards, by definition, involve the platform producer taking principal responsibility for generating the knowledge necessary to set standards. Knowledge about the technological opportunities and costs of alternatives must be exchanged with potential suppliers, who may have their own preferences for the deployment of their resources. By moving toward a negotiated standards-defining process, companies may be able to achieve a richer flow of knowledge concerning technological possibility and cost, although in so doing they must invest in the partnership that is being negotiated. If platform producers see even further market growth opportunity, they may elect to participate in ‘‘open’’ standards and in doing so obligate themselves to a substantial investment in knowledge acquisition and access in order to avoid being supplanted by rival platform producers, or by component suppliers becoming platform producers. From the component supplier’s perspective, it is not obvious which standards making process is to be preferred. In the case of single company controlled standards, it is possible that the components supplier may have better knowledge regarding cost and technological opportunity, which allows it to profit from the platform producer’s demands for a specific component. The probability of this being the case increases with the extent of cospecialization of assets (the degree to which the supplier is producing a particular component for a platform producer) (Teece 1986), a feature that helps explain why negotiated standards in networks or ‘‘clubs’’ of companies are emerging as an alternative to single company control of standards. In the case of open standards, the market opportunities available to suppliers provide the principal incentive for knowledge investment. Maintaining the Standard—Is the Electronics Industry Unique? The potential of modularity for industry growth and competition is illustrated very well by the WINTEL personal computer. Although it is possible to assemble many examples from the electronics industry in which technical compatibility standards serve to create dynamic and competitive markets, platforms in other industries appear more commonly to be based upon single company controlled or negotiated technical compatibility models. Is there some reason to believe that the electronics industry is unique? Maintaining a platform, particularly in technologically dynamic industries, does require a continuous process of platform redefinition. This subsection considers the difficulties of maintaining compatibility between platform components. There are two aspects to this problem. The first relates to the commercial damage or legal liability that the platform producer may face as the result of the decisions made by components producers. The second set of issues relates to the technological prob-

Knowledge, Platforms, and the Division of Labor

285

lems of maintaining compatibility in complex systems where interface standards are incomplete or inadequate to the task of maintaining the platform. Compared to many industries, the electronics industry, and the personal computer industry in particular, appears to have a guardian angel protecting it from the perils of product liability litigation. PCs and PC software are sold without warranties regarding their fitness for any purpose, and few legislatures have attempted to grapple with whether such warranties should be imposed in the interests of consumer protection. Nonetheless, the information and communication technology industries do suffer from ‘‘commercial damage’’ (damage to reputation, reluctance to purchase based upon negative experiences in the installation or use of the product, etc.) stemming from problems with the interoperability of components. In plain language, commercial damage is likely to arise when a customer is unable to properly install or use a component that is supposed to ‘‘work with’’ a particular platform. Whose fault is it? The customer is likely to assign blame to the component supplier rather than the platform. From the component supplier’s viewpoint, however, compatibility uncertainties are a particularly thorny problem. The platform producer may have provided a specification for the interface that is defective or incomplete, or there may be interaction effects with the products of other component suppliers. Even if willing to take responsibility for the problem, the component producer may not have or be able to acquire knowledge about how to rectify the problem. After rechecking the technical compatibility rule books, what is the next step? ‘‘Finger pointing’’ is likely to occur—the situation where the components maker blames the platform producer, who blames either the complaining components producer or another components producer, who in turn blames. . . . These problems in the personal computer industry tip the balance toward greater control by the platform producer and, in other industries, suggest limits to the desirability of ‘‘unbundling’’ systems to a competitive market of suppliers. There are almost always possibilities for interaction between components in a platform that are undiscovered. It is simply not practical to develop ‘‘testbeds’’ that reflect all of the possible states of even a modestly complex system. Knowledge is generated in use, and the capture and management of this knowledge emphasizes the role of the platform producer. Outside of the electronics and software industries, where product liability standards such as implicit fitness for purpose apply, platform producers face important legal liability risks in choosing a strategy of standardization for components to be integrated into their platforms. These problems would not exist if it were possible to define technical compatibility in a complete and unambiguous way or to effectively divide product liability risk between platform and component producers. Neither is a realistic option, and the maintainability of platforms is reduced. Instead of

286

W. Edward Steinmueller

open standards, platform producers are likely to choose to maintain control of standards or negotiate them with component suppliers in agreements that explicitly deal with the assignment of risk for product liability and take account of the commercial damage that may ensue when technical compatibility standards fail to work as advertised. Increasing complexity in the design of products and services further highlights the asymmetric interdependence between platform producers and component suppliers. For example, there is considerable concern about system reliability in the case of systems with embedded software such as automobile antilocking braking systems (ABS) or medical devices that employ electromagnetic radiation or particle emission (Lee 1999, 2000). A basic approach to the design of such systems is wherever possible not to automate those operations that pose even a remote threat to operators or to other people. There are limits to this design approach, however—it may be necessary, for example, to automate certain potentially hazardous operations because ‘‘manual’’ approaches are ineffective or, as in the case of ABS, because the whole point of the system is to bypass manual control. Complex systems with embedded software are also likely to involve a host of electronic sensors, data communication circuits, memory, and processor elements. Each of these elements is subject to faults. The central issue here, as in the discussion of product liability, is the ‘‘robustness’’ of compatibility standards. It is possible to imagine from an ideal perspective that compatibility standards would permit complete modularity in which failures or defects were detected and either resolved automatically or used to trigger a shutdown of the system that would limit or completely avoid negative consequences. Practical implementation of such solutions has often proven difficult. Similarly, actions involving the recall of products, providing ‘‘patches’’ or ‘‘fixes’’ to components within the platform, or issuing advisories about hazards and conditions of use have been difficult even in the personal computer industry and are an even larger problem in other industries. The way that these ameliorative actions are undertaken further emphasizes the value of platform-producer or small-group control of the knowledge. For example, in the case of systems embedded in automobiles, service and maintenance networks are directly linked to the platform (automobile) manufacturer and information about product safety hazards flows on a regular basis. Although the electronics industry is not uniquely able to benefit from modularity, this section has emphasized product liability and system reliability issues that limit the application of the modularity paradigm to other industries. A highlight of this discussion that is taken forward to the next section is the problem of defining robust interfaces, which depends in a fundamental way on abilities to generate and manage knowledge about how components interact within a platform.

Knowledge, Platforms, and the Division of Labor

287

Is Help on the Way? Simulating the Platform The reliability and functionality of platforms depends upon the robustness of the interfaces linking together their components. The modeling and simulation of platforms and the various supply chains and knowledge exchanges that contribute to their integration are of increasing significance for many industries as well as to those where such modeling and simulation are indispensable aspects of platform creation and maintenance. Thus, while it would be completely impossible to design an integrated circuit containing more than a million transistors without the aid of computers and software, it is becoming as difficult to create and maintain a large retail store network or a large construction project without similar tools. The modeling and simulation of platform products and services is a central purpose of some of the world’s largest software companies such as Oracle, SAP, and Microsoft. These companies’ enterprise resource planning systems and the supply of project management, computer-aided design and engineering, and other more specialized software is of central importance to the possibilities for platform design and implementation. They provide the basis for ‘‘virtual’’ models of the platform that are informed by a flood of real world data acquired from the growing array of inventory, point of sale, and ordering terminals. One of the major forces driving platform organization is the ability to represent the various knowledge flows involved in constructing the platform, coordinating suppliers, and delivering the platform. The means of knowledge representation are heavily reliant not only on advances in information and communication technologies but also on the design of software information systems and the collection of organizational procedures that support the gathering and updating of data. Ironically, the profitability and productivity gains from using these processes in platforms on which components are compatible with them may be higher than in modular platforms based on interoperable components, where these processes are indispensable. Historically, industries as diverse as aircraft manufacture, financial services, and travel services have had to produce integrated information systems in order to survive the growing complexity of their nonmodular systems. The delay in other industries’ adoption of such methods appears to be related to the scale of organizational change required and the modest gains that could be achieved through any individual step. It is true, for example, that many service companies are implementing customer relation management systems in a step-by-step manner with only modest confidence in their eventual financial returns but an increasing perception that such systems will eventually be necessary to maintain competitive position (Steinmueller 2003b). In other words, virtual models of the platform linking the virtual model to day-to-day operations may be particularly useful in industries that have not previously undergone major ICT-related organizational change.

288

W. Edward Steinmueller

There are, however, limits to the improvements that can be expected from such virtual models and to their ability to substitute for other forms of knowledge generation and exchange. A principal limitation is that achieving a relative competitive gain over rivals through the use of such models is difficult (Bresnahan 1986). As new applications come into service and demonstrate their value, rival firms adopt the same techniques. After a round of adoptions has occurred, relative competitive position will depend upon how other aspects of the business are managed. It is only in cases where leadership in successive waves of improved systems can be made part of the business model that such models are likely to have a sustained competitive impact. Assessing the potential of virtual models of platforms to substitute for other forms of knowledge generation and exchange is more complicated. These models do allow for experimentation through simulation and, therefore, offer the capability to experiment with different arrangements of the components of the platform or the means by which it is integrated. Whether this experimentation will provide better outcomes than can be obtained from localized knowledge and decision-making is not so clear. For example, despite their widespread promotion, it is difficult to find independent assessments of the value of ‘‘data warehouses.’’ Even more detailed levels of modeling and simulation, such as employing details about the engineering of products, involves the problem of product obsolescence—there would be little value in having the millions of pages of information that were written about the Apple II personal computer in a ‘‘data warehouse,’’ and detailed information about contemporaneous products may have as little value in only a few years. Despite these limitations, modeling and simulation methods are becoming more important in relation to individual aspects of platform design and integration. Computer aided design and engineering techniques are now used in almost all manufacturing industries, enterprise resource planning systems are employed in both manufacturing and services, and the use of project management software is ubiquitous. The consequence of these developments is that an ever-growing stream of data is being generated, with much of it exchanged between companies to support the coordination and maintenance of component supply chains. While, in principle, modeling and simulation techniques provide a framework for integrating these data, enormous cognitive and organizational problems remain in the translation of these data into useful knowledge. The extent of these problems within organizations has been documented in a very rapidly expanding literature (of which D’Adderio 2001 and 2004 are examples). Many of the issues identified in this section could be resolved if virtual models and simulations of the platform could be successfully engineered. The data that are increasingly generated as a by-product of the operations of component and platform

Knowledge, Platforms, and the Division of Labor

289

producer activities could, in principle, be integrated into these models and provide the basis for platform specification and maintenance and identification and resolution of the problems of incompatibility and complexity. However, we are not yet to this point of development. More coherent models and simulations are likely to arise from more closely coordinated and monitored relationships between platform producers and component suppliers such as those occurring where negotiated standards are employed. In addition to the greater extent of codevelopment that is likely to exist in these relationships, the need to monitor adherence to compatibility standards and, indeed, the very definition of compatibility are likely to be more fine-grained than in the case of ‘‘open’’ standards. Thus, again, single company controlled and negotiated standards making processes have an advantage, even though it is now foreseeable that modeling and simulation techniques may provide greater scope for open standards in the future. Governing Platforms The previous sections considered how platforms are created and maintained and, in the process, considered the institutional framework for standards setting that governs the relationship between platform producers and component suppliers. This discussion identified the mechanisms governing standards—single company controlled, open, and intercompany negotiated standards—in an assessment of the prospects for the extension of the modular paradigm to other industries than electronics and software. This section takes up two questions with policy relevance. First, can technical compatibility standards be used anticompetitively, and if so, what might be done to avoid this possibility? Second, is there a social welfare justification for promoting the modularity paradigm, and if so, what policies might support movement toward the more open standards that this paradigm requires? Although the answers to these two questions suggest specific government interventions, we will also note the potential for the concerted actions of platform customers achieving similar objectives. Exploiting Standards to Build Market Power The risks that standards may increase market power and diminish social welfare are twofold. Standards that are controlled by a single company or a group of companies may allow capture of the definition of the technological trajectory: the direction and, to some degree, the rate at which future technological progress is made. This possibility is examined in the first half of this subsection. Another risk is that the platform producer may be able to extend the market power offered by control of standards. This may occur if the practice of offering fair, reasonable, and

290

W. Edward Steinmueller

nondiscriminatory terms and conditions for access to the intellectual property essential to standards use is not followed, i.e., where the framework of public standards making bodies is not followed either by single companies controlling standards or in negotiated standards settings.13 This is considered in the second half of this section. In platform markets where there is rapid technological progress, there are powerful incentives to adopt open standards as a means to heighten competition among component suppliers and accelerate the formation of an industry standard or dominant platform. The relative difficulty of achieving a persistent technological advantage in information and communication technology markets provides an incentive for firms to adopt strategies that aim at shaping the rate and direction of technological progress. Many of these strategies are discussed by Shapiro and Varian (1998) and need not be replicated here. The basic objective of these strategies is to provide platform producers with an advantage over rivals in defining and implementing the next incremental step in technological advance. The strategy must be chosen such that the incremental benefits are sufficient to encourage an economically significant number of users to upgrade to the new generation of technological solution. The value of this strategy will be amplified if the new technology is subject to ‘‘bandwagon effects’’ (the formation of expectations about who will be the market winner based upon shares of early adopters captured) or produces ‘‘network externalities’’ (the value of adopting the increases due to the number of prior adopters). To implement such strategies, platform producers need to take into account the costs to the user of making the adoption decision, and may reduce the costs of learning and adaptation relative to rivals. In some circumstances, the incumbent platform producer may be able to sequence the introduction of new products with sufficient regularity to reduce the space available for competitive entry, in effect defining the dominant design in a dynamic fashion, based upon the firm’s next product offering. Strictly speaking, either bandwagon or network externality effects are necessary conditions to deny rivals with equal or superior platform offerings an entry into the market. Under these conditions, it does not make a great deal of difference whether the associated technical compatibility standards are proprietary or open. In either case, the dominant firm is likely to retain its position because of its ability to define how the next generation of the product will be extended and developed. It is important to note that this control is not proof of the technological or market superiority of the path or trajectory that the dominant firm chooses to follow; the dominant firm retains its dominance by shaping or directing the trajectory in ways that best serve its advantages against rivals. In this situation of dynamic market control, it is tempting to suggest that mandating earlier disclosure of a dominant firm’s intentions might increase competition.

Knowledge, Platforms, and the Division of Labor

291

Establishing rules to achieve this objective, however, has costs and other consequences. Early disclosure would, in effect, commit a firm to particular design choices, which would not allow last minute changes that would be possible were the design still under its control. Allowing the firm flexibility allows last minute changes to be used as a strategic tool to raise rivals’ costs. But not allowing it to make last minute changes may reduce the rate of technical progress. The second risk addressed in this section is whether a dominant platform producer can extend its market power either by (1) employing a ‘‘dynamic market control’’ strategy or (2) explicitly limiting competition through the control of a standard such as denying rivals (or their suppliers) access to essential intellectual property necessary to compete in the platform market. The feasibility of these options depends upon how the ‘‘standardized’’ product or service is combined with other inputs to produce a final product. A platform producer that selects ‘‘open’’ standards benefits from a network of suppliers of related products and services and the marketing efforts of those selling the platform to customers. Attempting to exploit these firms’ dependencies on the dominant platform producer increases the possibility that they would defect to an alternative platform supplier, deposing the dominant platform producer. The management of a dominant position involves the selective exercise of control, excluding only those firms that are likely to create a viable coalition in favor of a rival standard and entrant. This can be a difficult task to achieve. Nonetheless, such extension strategies can succeed for two reasons. First, the process of creating a viable alternative network of suppliers and marketing agents is costly and time consuming. Efforts to do so will meet with competitive responses from the dominant firm or firms, making it difficult to maintain the commitment from the new coalition’s participants. Also, there is little basis for policy intervention in this situation—suppliers may simply be trying to extract better terms from the platform producer by threatening to defect to an alternative coalition. Strengthening the position of suppliers may lead to an unproductive fragmentation of the market. Second, in many markets, the suppliers may have relatively specialized capabilities and/or be relatively young firms in no position to defect from the dominant coalition. Here, there is a case for limiting the range of ‘‘competitive’’ responses available to the dominant platform producer by questioning exclusive supply contracts, e.g., by using competition or antitrust policy advice to equate such agreements (subject to a rule of reason test) with anticompetitive intent. While it may be argued that such a policy would limit opportunities for some small and medium sized enterprises, the possibilities for abuse in the context of either dynamic control strategies or direct exploitation of essential intellectual property are substantial.

292

W. Edward Steinmueller

The feasibility of and rewards from a dominant platform producer extending its market power through open standards are no greater than are available from proprietary standards. However, the possibility of becoming a dominant platform producer may be enhanced through the use of open standards. The analysis of dynamic market control in this section clearly indicates the fallacy of assuming that ‘‘open’’ standards are inherently procompetitive. The modularity paradigm does not dispense with the need for competition policy or antitrust analysis. Industry structures based on platform and component supplier networks make such analyses more difficult because market control and power is dispersed through a network of firms in which anticompetitive coalitions and strategies may be much more difficult to detect or address. It is important, however, to note that open standards provide a basis for rapidly reconstructing a market if the dominant firm should falter in its efforts to maintain the pace of technological advance and may be very useful in markets where there is no dominant platform producer. This is the principal reason why, other things being equal, it is a desirable outcome from a social welfare viewpoint to favor the creation of open standards. Supportive Policies In considering issues of intervention in support of open standards, it is of vital importance to recognize at the outset that the concerted actions of users may play a vital role in supporting open standards and overcoming of the anticompetitive issues identified in the preceding section. Users may either produce or endorse common open standards. The relative infrequency of user-authored standards is an indication of the difficulty of mobilizing users and, in some cases, of users’ heterogeneous needs. Developing methods for mobilizing users would offer an important alternative to platform producer control and, while uncommon in the past, may have potential for future developments. Markets where technological change either (1) is slow and diffuse in nature or (2) occurs in large incremental steps with substantial uncertainty provide opportunities for proactive government policy to accelerate the rate of market development. There are four policy instruments for doing this. First, it is possible to support R&D efforts in the area aimed at creating a larger body of technical knowledge relevant to the supply side innovation. The availability of such knowledge allows both producers and users to better gauge or anticipate broad features in future market developments and earlier exploit their potential. The formation of expectations about the future development of technologies is an important influence in achieving market coordination, and research that results in convincing vision or scenarios concerning future developments can help align the

Knowledge, Platforms, and the Division of Labor

293

investment behaviors of the private sector. This knowledge also provides an opportunity for entrepreneurial firms to anticipate enabling or complementary technological developments that will be needed as the market develops. The historical (1950–1970) role in the United States of military and space research was to uncover information about technological opportunities and thus to encourage earlier market development than would have occurred solely through commercial exploration of (more immediate) frontier opportunities. Second, policies supporting an increase in the publicly available knowledge about likely trajectories or paths of future development can hasten the development of provisional standards aimed at earlier deployment of technologies. It is important, however, to recall the past failures of ‘‘anticipatory’’ standards such as ISDN (David and Steinmueller 1990). The availability of advance information reduces the fear of suppliers that they will commit themselves to a technology that will be obsolete before it can be deployed. The Winchester hard disk market is an example of such a process, albeit one where a large firm (IBM) rather than public policy produced the advance knowledge. IBM’s Almaden laboratory in San Jose, California, accelerated the development of magnetic drives with the aim of strengthening IBM’s competitive advantage. Some of the basic principles of the Winchester disk became general industry knowledge. By identifying the key bottlenecks and constraints to technological advance, the Almaden laboratory encouraged entrepreneurs (in particular Alan Shugart, one of the researchers) to reduce the technology to practice sooner rather than later. In particular, the market for personal computer hard disks proved to be far larger than for the most advanced and highest performance drives. Third, in some circumstances public policy can accelerate market development through procurement policy. Government purchase of actual products encourages the full reduction to practice of research developments and, thus, can launch a market earlier than it might have developed solely through market forces. Procurement policy is an expensive and uncertain method for accelerating market development. A principal example of successful procurement policy is integrated circuits, which were produced for U.S. military purchasers and whose initial prices were entirely uncompetitive with discrete transistors. The early development of the technology, though, provided the United States with a decade of commercial advantage in the market. It would, however, have been very difficult to commit public funds for this procurement for any purpose other than national defense. Government procurement of other advanced technologies such as breeder reactors and coal gasification plants has led to the squandering of huge amounts of public resources, an illustration of the risks of procurement policies. Fourth, there are several opportunities for helpful public research policies. First, it would be useful to fund more research on means for modeling and simulating the

294

W. Edward Steinmueller

operation of product and service platforms. As noted in this chapter, these methods offer the best hope of overcoming some of the persistent problems in extending open standards and the modular paradigm. Government policy aimed at improving the social science research foundations for understanding the role of information and communication technologies in modern enterprises—particularly in the choice between proprietary and open standards—would also be helpful. These suggestions for policy intervention are necessarily tentative due to the persistent uncertainty surrounding the creation of new platforms and standards for their implementation. Conclusion The vision of an economic world governed by open standards supporting either modular or compatible assembly of components onto platforms with an inclusive division of labor embracing entrants with a better idea or a higher level of efficiency is extremely attractive. The vision offers a new and inclusive possibility for the international division of labor, a formula for preserving competition in industries where increasing returns might otherwise limit market competition, and a means to reduce the wasted time and deferred benefits of markets that do not develop because they are fragmented by standards wars. This chapter has explored the limits of this vision by considering the problems of standards for compatibility and modularity that have persisted in the information and communication technology industries. Many of the persistent problems afflicting the information and communication technology industries are amplified when we consider other industries. In addition, many of these industries suffer from severe knowledge coordination problems due to the asymmetries in component suppliers’ and system producers’ knowledge. A principal reason that strategies of modularity and compatibility have been so effective in the electronics and software industries is the fortuitous combination of rapid market expansion and enormous technological opportunity. There is, nonetheless, some prospect for the spread of platform strategies based upon open compatibility standards in other industries due to the market opportunities offered by the international division of labor combined with the use of information and communication technologies for managing knowledge flows and for planning, designing, and implementing product and service platforms. The advantages of open compatibility standards platforms do not, however, eliminate the risk that sponsored or negotiated standards making processes may eventually lead to social welfare losses through the control of the evolution of technology

Knowledge, Platforms, and the Division of Labor

295

or the extension of market power into other markets. These possibilities should be taken seriously: Actions can be taken to reduce their likelihood. Finally, research policies have been identified that would provide support for moving toward increased use of open standards in constructing product and service platforms. These policies largely involve providing better information about emerging technological opportunities and supporting those technologies that are particularly useful in achieving more inclusive division of labor in the production of platforms and their components. Notes 1. The textbook assumption is ‘‘perfect information,’’ i.e., that all economic agents have access to the full range of production and exchange possibilities, which is a useful filter for separating students willing to make strong economic assumptions from those who are not. It is quite sufficient to assume that those with the potential to take advantage of information have ready access to it. 2. The English language does not have a generic word for the elements that might comprise a complete service. In this chapter, we will use the somewhat awkward term ‘‘service component’’ for this purpose. In this terminology, the sommelier at a restaurant provides a ‘‘service component.’’ 3. The engines of both automobiles and aircraft are highly integrated, and although some of the specific components employed in a specific engine may be produced by a component supplier, there is little opportunity to exchange components between different engines. Moreover, every engine producer offers several different types of engine with mutually incompatible parts. It is nonetheless true that a components manufacturer can transfer some of the knowledge about the production of components for one type of engine to another. 4. Specific company histories such as those of science and engineering at AT&T (Millman 1983, 1984), for example, provide important source material for identifying the emergence of these issues but offer little insight into how they were resolved. 5. An alternative approach, which begins with systems that have proven difficult to modularize and proceeds to identify reasons that this might be true, is employed in Brusoni et al. (2001). 6. The dominant personal computer design was initially based upon an IBM-defined standard and has subsequently come to be defined by the links between Microsoft’s Windows operating system (hence the WIN) and a microprocessor architecture defined by Intel (hence TEL). 7. Pierre Verdun’s patent 3,892,365 is cited in the background section of Sontheimer’s patent 3,985,304. 8. Microsoft supported this strategy by making many of its application software packages available for operation under Apple’s proprietary operating system. 9. In 2005, IBM sold its personal computer division to Lenovo, a Chinese manufacturer prepared to focus on manufacturing costs.

296

W. Edward Steinmueller

10. Standards employed in defining the ‘‘interfaces’’ between components of a system may be defined as ‘‘technical compatibility standards’’ as a way of differentiating such standards from those defining physical properties or qualities, which may be defined as ‘‘reference standards’’ (David and Greenstein 1990). 11. The business models of these organizations is now largely based upon the sale of standards documentation and contributions from technical experts whose participation in standards making is typically sponsored by the companies that employ them. 12. There are important differences in further interpretation of ‘‘free’’ and ‘‘open’’ within the latter community; see the Open Source Initiative (http://www.opensource.org/) and the GNU Project (http://www.gnu.org/) for discussions of these differences. A central point of difference is the terms under which components of software may be integrated into a system that is offered to others. 13. It may be argued that public standards-making organizations are more inclusive than their industry consortia counterparts because the former have procedural requirements ensuring that affected stakeholders have an ability to participate. Industrial consortia, however, may adopt similar rules and need not allow their members to bar access to the intellectual property essential for standards implementation.

References Arora, A., A. Fosfuri, and A. Gambardella (2001). Markets for Technology: The Economics of Innovation and Corporate Strategy. Cambridge, MA: MIT Press. Baldwin, C. Y., and K. B. Clark (1997). ‘‘Managing in an Age of Modularity.’’ Harvard Business Review 75(5): 84–93. ———, and K. B. Clark (2000). Design Rules: Volume 1—The Power of Modularity. Cambridge, MA: MIT Press. Blakeslee, T. R. (1975). Digital Design with Standard MSI and LSI. New York: John Wiley and Sons. Bresnahan, T. F. (1986). ‘‘Measuring the Spillovers from Technical Advance: Mainframe Computers in Financial Services.’’ American Economic Review 76(4) (September): 742–755. Brusoni, S., A. Prencipe, and K. Pavitt (2001). ‘‘Knowledge Specialisation, Organisational Coupling, and the Boundaries of the Firm: Why Do Firms Know More Than They Make?’’ Administrative Science Quarterly 26(4) (December): 597–621. D’Adderio, L. (2001). ‘‘Crafting the Virtual Prototype: How Firms Integrate Knowledge and Capabilities Across Organisational Boundaries.’’ Research Policy 30(9): 1409–1424. ——— (2004). Inside the Virtual Product: How Organizations Create Knowledge Through Software. Cheltenham: Edward Elgar. David, P. A., and S. Greenstein (1990). ‘‘The Economics of Compatibility Standards: An Introduction to Recent Research.’’ Economics of Innovation and New Technology 1(1): 3–41. ———, and W. E. Steinmueller (1990). ‘‘The ISDN Bandwagon Is Coming—But Who Will Be There to Climb Aboard?: Quandaries in the Economics of Data Communication Networks.’’ Economics of Innovation and New Technology 1(1–2): 43–62.

Knowledge, Platforms, and the Division of Labor

297

Edquist, C., editor (1997). Systems of Innovation: Technologies, Institutions and Organizations. London: Pinter. Lee, E. A. (1999). ‘‘Embedded Software—An Agenda for Research.’’ Berkeley, CA: Electronics Research Laboratory, University of California, Berkeley. ——— (2000). ‘‘What’s Ahead for Embedded Software?’’ IEEE Computer (September): 18–26. Malerba, F., Ed. (2004). Sectoral Systems of Innovation. Cambridge: Cambridge University Press. Millman, S., editor (1983). A History of Engineering and Science in the Bell System: Physical Sciences, 1925–1980. AT&T Bell Laboratories. ———, editor (1984). A History of Engineering and Science in the Bell System: Communication Sciences, 1925–1980. AT&T Bell Laboratories. Mowery, D. C., J. E. Oxley, and B. S. Silverman (1996). ‘‘Strategic Alliance and Interfirm Knowledge Transfer.’’ Strategic Management Journal 17: 77–91. Procassini, A. (1995). Competitors in Alliance: Industrial Associations, Global Rivalries, and Business–Government Relations. Westport, CT: Quorum Books. Shapiro, C., and H. Varian (1998). Information Rules: A Strategic Guide to the Network Economy. Cambridge, MA: Harvard Business School Press. Simon, H. A. (1969). The Sciences of the Artificial. Cambridge, MA: MIT Press. ——— (1996). The Sciences of the Artificial, 3rd ed. Cambridge, MA: MIT Press. Steinmueller, W. E. (2003a). ‘‘The Role of Technical Standards in Coordinating the Division of Labour in Complex System Industries.’’ In The Business of System Integration, A. Davies, Andrea Prencipe, and Michael Hobday, editors, pp. 133–151. Oxford: Oxford University Press. ——— (2003b). ‘‘Assessing European Developments in Electronic Customer Relations Management in the Wake of the dot.com Bust.’’ In Industrial Dynamics of the New Digital Economy, J. F. Christensen and P. Maskell, editors, pp. 233–262. Cheltenham: Edward Elgar. ——— (2004). ‘‘The European Software Sectoral System of Innovation.’’ In Sectoral Systems of Innovation, F. Malerba, editor, pp. 193–242. Cambridge: Cambridge University Press. Teece, D. J. (1986). ‘‘Profiting from Technological Innovation: Implications for Integration, Collaboration, Licensing and Public Policy.’’ Research Policy 15: 285–305.

18 Between ‘‘Knowledge’’ and ‘‘The Economy’’: Notes on the Scientific Study of Designs Carliss Y. Baldwin and Kim B. Clark

Introduction Designs are the instructions based on knowledge that turn resources into things that people use and value. Behind every innovation lies a new design. Tangible products and production processes, intangible services and experiences, corporate strategies, organizations, methods of contracting, governance, and dispute resolution—all of these things have designs.1 Thus, ‘‘knowledge economies,’’ which are based on continuous innovation and competition between old and new things, must produce a never-ending stream of new designs.2 Designs are created through purposeful human effort. A design process is a set of activities that starts with someone’s problem and then devises an artifact to solve the problem. The outcome of this process is the design of a particular thing that is a solution to the problem.3 The solution may be tangible (a good) or intangible (a process or a service) or a combination of the two. Conceptually, designs can be thought of as lying between ‘‘knowledge’’ and ‘‘the economy,’’ as depicted in Figure 1. At any point in time, knowledge about the world exists in the heads of various people, in libraries, and in social and organizational networks. Of itself, though, as historian of technology Joel Mokyr has argued, such ‘‘propositional knowledge’’ doesn’t do anything. To affect the world, propositional knowledge must be converted into ‘‘prescriptive knowledge,’’ that is, ‘‘designs and instructions . . . like a piece of software or a recipe.’’4 Thus, it is only through the agency of designs that knowledge can become the basis of real goods and services. Furthermore, to create complex goods and services, the process of converting propositional knowledge to prescriptive knowledge must itself be organized. Thus, designs fall into two categories: design architectures, which are used to organize design processes, and complete designs, which are the end result of such processes. Design architectures are the starting point, hence the ‘‘forward-looking’’ or ‘‘future-oriented’’ aspect of design processes. A design architecture creates a sensible

300

Carliss Y. Baldwin and Kim B. Clark

Figure 1 Designs link knowledge to the economy.

subdivision of the tasks involved in designing a large system. The architect sets up the design rules for the system: He or she divides a to-be-designed system into parts, sets up interfaces between those parts, and specifies ways of verifying the properties and testing the performance of the components and the system.5 Just as physical architectures both create and constrain opportunities for movement in physical spaces, design architectures both create and constrain opportunities in the so-called ‘‘design spaces’’ wherein the search for new designs takes place.6 Complete designs are the end result of design processes. A complete design is the ‘‘information shadow’’ of an artifact. It can be made into something real and valuable: a product or a service. The economy in turn is based on the production and consumption of products and services. Long ago, most goods were produced without first creating a separate design. Today, much of the economy is devoted to the creation of designs and the subsequent production of artifacts based on those designs. Once they are created, design architectures and complete designs can be added to the stock of knowledge, as the backward arrows in Figure 1 show. They can be used again and again. Preexisting designs also serve as the starting point for new design processes. Each generation of designs builds on the previous one, so that a series of design processes can result in cumulative design improvement or burgeoning design variety.7 Clearly, designs are an important source of economic value, consumer welfare, and competitive advantage for individuals, companies, and countries. They have also been the focus of scientific research in a number of fields, including engineering, computer science, architecture, and management.8 But despite their pervasive influence and the large amount of academic research that has been done, designs as drivers of innovation and wealth creation are not much discussed by social scientists, senior managers, or policy-makers. More often than not, to nonspecialists designs appear to be esoteric objects, which can only be understood and evaluated by experts in the design’s particular domain. We believe it is time to integrate the study of designs across disciplines and make them the focus of unified scientific research in their own right. The structure and value of designs, as well as what designs ‘‘need’’ in the way of organizations and social policies, are all topics that can be investigated scientifically and in a unified

Between ‘‘Knowledge’’ and ‘‘The Economy’’

301

way across disciplines. These topics belong on the agenda of research that seeks to understand how knowledge creates wealth in modern economies. Such research in turn may allow engineers to construct more valuable designs and design architectures. It can also help senior managers organize their enterprises more productively, assist investors in allocating resources, inform public debate, and serve as the basis of rational public policy. How does the scientific study of designs differ from other ways of studying innovation and technology? Many scholars are already seeking to explain the dynamics of technological change and innovation, drawing on economics, organizational behavior, sociology, strategy, and other academic disciplines. What does the scientific study of designs offer that is new? How can it improve on the excellent work already being done? In essence, the scientific study of designs as a general phenomenon offers a new level at which to observe technologies and how they change. Social scientists especially have struggled for some time with the problem of how to characterize ‘‘technologies’’ and measure them in meaningful ways. But they have tended to approach ‘‘technology’’ at quite a high level of abstraction.9 We and others who study designs scientifically think that there is critical, observable structure below the level of an abstract ‘‘technology’’ and indeed often within a single design. As we explain below, we believe it is important, and sometimes crucial, to analyze designs at the level of decisions and dependencies. Only by understanding designs at this more microscopic level can one ascertain their potential to evolve, their economic value, and their probable future trajectories. In the rest of this chapter, we describe recent work that contributes to a scientific understanding of designs across a range of fields. First, drawing on economics, we list the properties of designs and compare them to other types of goods. Next, drawing on engineering and computer science, we discuss design structure. We argue that there are general and useful ways to map design structure: To support this argument, we describe one set of methods, the so-called design structure matrix (DSM) mapping technique. Methods such as these now make it possible to study designs as a general phenomenon, as opposed to within particular domains of engineering, architecture, and management. Returning to economics, we then describe the ‘‘net option value’’ (NOV) method of valuing designs and their architectures and discuss the challenges of applying this method in practice. Finally, we explain how designs both require and give rise to incentives, rewards, and resource allocation mechanisms that, taken as a whole, amount to a system of institutions. We recount two cases from the 1980s and 1990s in which the observable institutions changed substantially, apparently in response to changes in underlying design structure and value.

302

Carliss Y. Baldwin and Kim B. Clark

We are neither the first nor the only ones to contend that designs are worthy of scientific study. Herbert Simon did so most eloquently in a series of essays and lectures in the 1960s, and many others have followed in his footsteps. Simon even laid out the subfields of inquiry for a general ‘‘science of design.’’ We have purposely not called the studies described below the beginnings of a general science. They are simply a set of scientific studies, widely scattered and yet related in their focus on designs as a general phenomenon. At the end of this essay, we will speculate as to why there is as yet no general science of design and consider whether such a field might emerge in the future. Critical Properties of Designs We begin by describing what we believe are the critical properties of designs. Listing properties allows us to treat designs as general conceptual objects as opposed to objects within a particular field or discipline. A list of properties can also serve as an axiomatic base on which to build formal theories and models. Given a set of axioms, one can derive testable hypotheses by considering how the properties of designs interact with various external factors, such as constraints on resources, the presence of property rights, and assumptions about incentives and human behavior. The critical properties of designs are as follows:10 Designing requires effort, hence designs are costly. (Here we are referring to the cost of creating the design, not the cost of making the artifact from the design.)




Designs cannot be consumed directly: Their value is derived from the functions performed by the artifacts they describe. In most cases, designs must be reified (realized or implemented) in order to be valuable.11 Reification means that the design instructions are carried out and become embodied in a physical object, a service, or an experience. The description of the process by which the design is reified is part of the design.




Designs are ‘‘non-rival’’; that is, one person’s utilization of a design does not prevent another’s use of the same design.







Ex ante, the outcomes of design processes are uncertain.




In a formal sense, new designs are options.




Ex post, some designs are rankable within a category.




Designs have a structure made up of decisions and their dependencies.

Based on this list, we can compare designs to other types of things that people need or value, as shown in Table 1. The fact that designs are costly means that they are economic goods in scarce supply. A not-yet-complete design must therefore

Table 1 Comparison of designs to other types of goods Types of goods

Properties

Designs

Information goods (music, books)

Tangible goods (food, clothing)

Physical assets (buildings, machinery)

Financial assets (stocks, bonds)

Costly to complete

X

X

X

X

X

Cannot be consumed directly

X

O

O

X

X

X

X

O

O

O

Uncertain behavior and value

X

S

S

S

S

Optional

X

S

S

S

S

Rankable

S

S

S

S

S

Structure of decisions/dependencies

X

O

O

O

O

Key: X ¼ has the property O ¼ does not have the property S ¼ sometimes has the property

Between ‘‘Knowledge’’ and ‘‘The Economy’’

Nonrival

303

304

Carliss Y. Baldwin and Kim B. Clark

offer (someone) enough economic value to cover the cost of completing it and of making the artifact. However, the value offered by the design does not have to be denominated in money, nor does it have to be exchanged. Eric von Hippel and his colleagues have demonstrated that user-innovators may complete designs because they anticipate direct benefit from use of the corresponding artifact.12 Even so, the decision to devote time and effort to completing a design is an allocation of scarce resources, hence an economic action. Although designs are economic goods, Table 1 also shows that they are not exactly like any other major types of good. The main differences are highlighted by the two panels. We discuss them below. First of all, because designs are only a description or ‘‘shadow’’ of a thing, they cannot be consumed directly. In this sense, they are not like tangible goods such as food or clothing, nor like other types of information such as ‘‘baseball scores, books, databases, magazines, movies, music, stock quotes, and Web pages.’’13 All of these things can be consumed directly or used in a production process. In contrast, a design must be turned into something—the thing specified by the design— in order to be useful. Indeed all the other goods mentioned in the table, including the information goods, have designs—that is, each has a set of instructions that specifies how the good will be produced. Designs are a kind of asset and thus may be compared to physical assets and financial assets. Physical assets such as buildings and equipment supply a flow of services, which can be consumed or used in a production process. Financial assets such as stocks and bonds produce a stream of cash in the future: The cash cannot be consumed directly but can be converted into other things. A (complete) design provides the ability to make something in the future: In this sense it supplies a flow of ‘‘design services.’’ However, the analogy between physical or financial assets and designs is not perfect. A physical asset provides specific services; a financial asset provides general purchasing powers; a design provides instructions for making one specific thing. Designs can be represented as a stream of symbols, communicated in symbolic form and translated from one language or medium to another. Thus, designs are ‘‘information.’’ Like other forms of information, designs are ‘‘nonrival.’’ This means that the use of a design by one person does not preclude another from using it too.14 In general, information cannot be ‘‘consumed’’ in the sense of ‘‘used up.’’ Therefore, a design survives its own use, although it may be lost or forgotten. The outcome of a design process is always uncertain. If the content of a design were known, the design would already exist and the design process would be finished.15 Because design processes are uncertain, the behavior of a newly designed artifact is not perfectly predictable, and the ways users will react to it are not pre-

Between ‘‘Knowledge’’ and ‘‘The Economy’’

305

dictable either. Therefore, the ultimate value of a design—the value users will ascribe to the artifact less the cost of making it—is uncertain while the design process is under way. This means that design processes, unlike production processes, cannot be algorithmic progressions with well-controlled, guaranteed-to-be-correct outcomes (Whitney 1990). Uncertainty in turn makes options valuable. Technically, an option is ‘‘the right but not the obligation’’ to take a particular action.16 When a new design is created, users can accept it or reject it. They have ‘‘the right but not the obligation’’ to solve some problem in a new way. Formally, therefore, all new designs are options. Other types of goods also provide options: For example, an investor may purchase an option to buy a financial security at a fixed price. A flexible production line incorporates options to change inputs or outputs in response to price fluctuations. Hypertext gives readers options as to what information to seek next. But while other goods sometimes provide options, new designs are always options. In addition to being uncertain and optional, designs are sometimes rankable within a category. If so, most people will agree that a particular design is best for some purpose.17 When designs are rankable, their ‘‘optional’’ and ‘‘nonrival’’ properties interact in a powerful way. The best design can be used by everyone (the nonrival property), and the inferior designs can be discarded (the optional property). As a result, competition among rankable designs will be characterized by ‘‘winnertake-all’’ payoffs and serial obsolescence. Only the best designs in any cohort will be rewarded, and new and better designs (and artifacts) will replace older ones over time. In contrast, when designs are not rankable, many designs will be rewarded, and many will survive, serving different needs in different niches. The above-named properties, which apply to whole designs, are sufficient for some types of analysis. But in other cases it is necessary to look below the level of the whole and investigate design structure in more detail. Looking at structure is especially important when one is trying to establish the boundaries of designs for purposes of valuation and in order to understand their evolutionary behavior. And as we discuss in the next section, the structure of designs is determined by a pattern of decisions and dependencies. Design Structure Much of science involves the study of how observable structure affects behavior. Thus, without a structure to observe, scientific inquiry cannot begin. All of the goods listed in Table 1 have underlying structures. Tangible goods are made up of atoms and molecules. Financial assets are made up of contractual promises and contingencies. ‘‘Ordinary’’ information goods are made up of content (e.g., baseball

306

Carliss Y. Baldwin and Kim B. Clark

stories and scores) and templates for arranging content (e.g., the sports pages of a newspaper). The structural elements of a design are different from any of these. In investigations of design structure, there is an emerging consensus that the fundamental units of design—the smallest building blocks—are decisions.18 Design decisions yield the instructions and parameters that determine the final form of the artifact.19 Design structure in turn is determined by dependencies that exist between (or among) decisions. Speaking informally, decision B depends on decision A if a change in A might require a change in B. In this case, B’s decision-maker needs to know what has been decided about A in order to choose B appropriately. Even small designs may have thousands of associated decisions. To avoid getting bogged down in details, design decisions that are highly interdependent may be grouped into clusters corresponding to the components of the design. The pattern of dependencies between any two components in turn may be independent, modular, or integral. Two components are independent if anything in the first can change without any impact on the second and vice versa. The designs of a laptop computer and an automobile are essentially independent in this sense. Two components are integrally related, or simply integral, if almost every decision about either one depends—directly or indirectly—on decisions about the other. Finally, two components are modularly related, or simply modular, if they are (almost) independent of each other but work together on the basis of a common set of design rules.20 The significance of these categories is that with independent or modular designs, design decisions can be divided among several autonomous or semiautonomous groups. In contrast, integral designs require close coordination, so their decisions cannot be easily divided up. In this fashion, design structure directly affects organizational structure—that is, how work gets done in the economy. Independent, modular, and integral are three basic patterns of design structure. Other patterns are possible too, and a large design may display different patterns in different places.21 This is why we think it is essential to map design decisions and dependencies. One useful mapping technique is the so-called design structure matrix (DSM) mapping method. We will discuss this method in some detail to give readers a sense of what design mapping involves and what it can reveal.22 To apply the DSM mapping method, a design is first characterized by listing the design decisions or components of the system. (As indicated, a component is a group or cluster of decisions.) The components are then arrayed along the rows and columns of a square matrix. The matrix is filled in by checking—for each component—which decisions about other components affect it and which in turn are affected by it. For example, if a decision about component A affects some decision about B, then we put a mark ‘‘x’’ in the cell where the column of A and the row

Between ‘‘Knowledge’’ and ‘‘The Economy’’

307

Figure 2 Design structure matrix map of a laptop computer. Source: McCord and Eppinger (1993). Reprinted by permission.

of B intersect. We repeat this process until we have recorded all the dependencies. The result is a map showing the locations of the dependencies. Figure 2 presents a DSM map of the dependencies in the design for a laptop computer system circa 1993. The map shows that the laptop computer design has four blocks of very tightly interrelated design parameters corresponding to the drive system, the main board, the LCD screen, and the packaging of the machine. There is also a scattering of dependencies (‘‘x’s’’) outside the blocks. The dependencies arise both above and below the main diagonal blocks; thus, the blocks are interdependent. Because each component depends on every other one, directly or indirectly, the overall design structure is integral. Herbert Simon (1962) and Christopher Alexander (1964) appear to have been the first to represent the dependencies of a complex system using a square matrix. Donald Steward (1981) came independently to the same representation and identified the rows and columns with design decisions (or components). Daniel Whitney

Figure 3 DSM of an industrial gas turbine circa 2002. Source: Sharman et al. (2002a). Reprinted by permission.

Between ‘‘Knowledge’’ and ‘‘The Economy’’

309

(1990) argued that Steward’s matrices could be used in ‘‘designing the design process’’ of a complex artifact. Whitney, together with Steven Eppinger and his colleagues,23 have extended Steward’s framework and used it to construct maps of numerous engineering design processes and complex artifacts. We built on this prior work in developing our concepts of modularity and design rules (Baldwin and Clark 2000). More recently, Yuanfang Cai and Kevin Sullivan (2005) formalized the notion of pair-wise dependency among design variables in terms of a constraint-based representation of design spaces. Because the fundamental elements of a DSM are decisions and dependencies, DSMs can be constructed for any design or design architecture. Some examples are presented below. Figure 3 shows the components and dependencies of a 10megawatt industrial gas turbine, a large physical artifact. Like the laptop computer, the structure of this design is (essentially) integral. Figure 4 presents DSMs for two software codebases. In contrast to the laptop computer and the gas turbine, these design structures are (essentially) modular. (One should not generalize from these examples. Tangible artifacts do not always have integral designs and codebases do not always have modular designs.) The modularity of the software DSMs in Figure 4 can be seen from the fact that each has a set of almost independent block components in its lower right quadrant plus one or more vertical columns, representing external variables and design rules, running down the left-hand side. The DSMs also reveal an important property known as ‘‘information hiding.’’24 In these designs, external conditions, which are outside the designer’s control yet might change, do not interact with the design rules. This is evidenced by the fact that in the DSMs, the blocks labeled ‘‘external parameters,’’ ‘‘basic concerns,’’ and ‘‘crosscutting concerns’’ have no crossdependencies with the ‘‘design rules’’ blocks. Information hiding was proposed as a desirable property of software designs by David Parnas in 1972. Sullivan et al. (2001) were the first to include enviromental variables in a DSM and to characterize an information-hiding modularization as one in which the design rules are invariant to the environmental variables.25 In effect, an information-hiding modularization, whose presence can be verified by DSM mapping techniques, protects the ‘‘skeleton’’ of the design structure from outside disruption but allows change to take place in the so-called ‘‘hidden modules.’’ In this fashion, information hiding tends to localize the impact of external change on the design and thus enhance the evolvability of the system as a whole.26 As a final example, Figure 5 presents DSMs for two states of a codebase known as Mozilla. These differ from the previous DSMs in several ways. First of all, these maps are considerably larger and more detailed than the previous ones: each has more than 1500 rows and columns, while those in the previous figures had less

Figure 4 Two software DSMs: Winery locator and hypercast. Sources: Winery locator DSM: Lopes and Bajracharya (2005); hypercast DSM: Sullivan et al. (2005).

Figure 5 Call graph DSMs for two versions of the Mozilla browser. Source: MacCormack et al. (2004).

312

Carliss Y. Baldwin and Kim B. Clark

than 50 each. It was feasible to construct these larger and finer-scale maps because the decisions and dependencies were automatically extracted from the artifact itself (the codebase). Source files were used as a proxy for (clusters of) decisions and function calls were used as a proxy for dependencies.27 With automated mapping, DSM techniques can be applied to much larger systems than was previously possible. However, automatic extraction can be problematic. Some dependencies, which may have influenced a design process, do not leave ‘‘tracks’’ in the finished artifact. Nevertheless, many dependencies do show up in automated maps, and the ones that do may be the ones most likely to affect the future evolvability of the design. More work clearly is needed to develop and to assess the strengths and weaknesses of automated design mapping techniques. Figure 5 also illustrates an important point about design structure: The functions of a design do not totally determine its structure.28 The two codebases shown here were two versions of the same browser and were (for practical purposes) functionally equivalent. Yet, as the figure shows, their design structures are dramatically different. The codebase depicted on the left was developed within a company (Netscape) using rapid-cycle methods.29 When Netscape ran into financial difficulties, the codebase was released under an open source licence. But in the open source environment, this design structure was found to be unsatisfactory: Open source developers did not want to maintain or contribute to the codebase because (among other things) it was too unwieldy for their methods.30 A small team of designers then spent half a year ‘‘refactoring’’ the browser code to make its structure more modular. The resulting change in design structure is evident in Figure 5. These and other studies have shown quite conclusively that for complex designs, function does not wholly determine structure. The architects of complex designs have degrees of freedom and can satisfy functional requirements in different ways. That is good news for consumers and entrepreneurs because it means that there is room for improvement even of very successful designs. But it is bad news for those who would like to predict the future of a technology without delving into the details of design structure. Just observing ‘‘what the technology does’’ is not enough. It is also necessary to look at how the underlying designs are put together—their structure of decisions and dependencies and their so-called ‘‘technical potential’’—to figure out what the future may hold. This fact again points up the need for comparative studies of design structure spanning a range of disciplines. The development of DSMs and other general design mapping methods during the past several decades now makes it possible to study designs as a general phenomenon. Until now, most designs had to be studied within the ‘‘silos’’ of specific engineering disciplines. Great theorists of design such as Herbert Simon or Allen Newell could see unity in the phenomena and begin to sketch the outlines of a

Between ‘‘Knowledge’’ and ‘‘The Economy’’

313

science. But without a lingua franca that could span disciplinary boundaries, there was no way to capitalize on their insights. DSMs and other mapping methods offer a common set of building blocks and a general way to represent designs. These maps can be a lingua franca that cuts across disciplines and unifies the scientific study of designs. At the same time, all maps have limitations. Indeed, the problems inherent in DSM mapping (for example) illustrate the difficulties that are endemic in all mapping efforts. First of all, some patterns of dependency do not lend themselves to a flat, two-dimensional representation. For example, David Sharman, Ali Yassine, and Paul Carlile (2002a) have shown that some designs (including the gas turbine of Figure 3) are better represented in three dimensions than two. However, all mapping involves the projection of higher dimensional phenomena onto lower dimensional representations. Thus, while it is important to know what may be hidden behind the projections, the fact that things are lost (or obscured) in a mapping does not mean that the map itself is worthless. Second, observing designs at the level of decisions and dependencies is truly daunting: Even a small design may involve hundreds of decisions and tens of thousands of dependencies. Thus, the clustering of design decisions into ‘‘components’’ or ‘‘protomodules’’ is essential. In fact, we have never seen a DSM that was constructed or observed at the level of single decisions. All DSMs aggregate decisions in some fashion, but methods of aggregation are still largely intuitive and ad hoc. Nevertheless, one of the strengths of this methodology is that, in practice, one can identify a dependency without isolating the exact decision(s) that created it. Thus, a dependency can be attributed to a cluster of decisions (a component) without understanding the structure of the cluster in detail. This in turn means that maps of dependencies can be ‘‘bootstrapped.’’ The mapmaker can start with a coarse representation of decisions and dependencies and work toward finer representations, stopping when the cost of greater detail outweighs the benefit. The bottom line is that we need to know more about what different maps of design structure do and do not show. The only way we can learn more is to work systematically with the maps we have, criticize them, improve them, and experiment with new mapping methods. Such work is a quintessentially scientific undertaking. Design Value A design process is a costly venture into the unknown. Each step in the process is expensive: formulating an architecture, completing the design, and reifying the design once it is complete. Because the path is uncertain, there are also often costly cycles, loops, and even blind alleys. How then can one come to an informed

314

Carliss Y. Baldwin and Kim B. Clark

judgment that the search is worthwhile? How can one know if the expected benefits are likely to exceed the inevitable costs? And when confronted with different ways of organizing the design process—different design architectures—how can one decide which is likely to result in a better design at the end of the day? The questions just posed are all about the comparative value of different alternatives. The valuation of alternatives in turn is the focus of the branch of applied mathematics that deals with decision-making under uncertainty. During the past fifty years, this field has grown in many directions. One of its main subfields, which has developed within the discipline of economics, deals with the valuation of options. As we said earlier, every new design embodies at least one option and some involve many. Thus, option theory is highly relevant to decisions about whether to undertake and how to organize design search processes.31 Just as DSMs offer a general way to represent designs across a range of fields, option theory offers a general way to value designs at any point in their existence, from prearchitecture to postreification. In fact, design representation and design valuation are inseparable. The application of option theory requires that the boundary and scope of the individual options in a design be crisply delineated. If, as is often the case, the design embeds multiple options, they must be enumerated. The boundary and scope of options in turn are determined by the underlying pattern of dependencies. Basically, designs that have many independent (or quasi-independent) components contain more options and are likely to have higher option value. And the boundaries of options correspond to the ‘‘thin crossing points’’ in a map of dependencies.32 Modular designs require components to be (almost) independent of one another, linked only by design rules. Because they are (almost) independent, the module designs can be ‘‘mixed and matched’’ and can evolve along separate paths independently of one another.33 In this fashion, modular designs create options (hence option value) in the later stages of a design process. Coordination across modules is accomplished via design rules, which all groups must obey and in turn can expect others to obey. Although subject to design rules, each module embodies a separate and distinct set of options. In practice, this means that the design of a module can change and improve over time without regard to what is happening in other modules and without harming the rest of the system. A general formula for the value of a complex design can be written as a sum of the value of a ‘‘minimal system’’ plus the values of individual modules. The value of individual modules in turn is determined by the functions they perform for the end user or for the system.34 By convention, this method is known as the NOV (net option value) approach to design valuation. The NOV approach has been applied to several actual designs, including those depicted in Figures 3 and 4. But this

Between ‘‘Knowledge’’ and ‘‘The Economy’’

315

approach to design valuation is still in its infancy. As is common with early-stage work, each application of the methodology has raised as many questions as it has settled. The most salient questions are: (1) How does one translate achieved functionality into value, and (2) from what probability distribution(s) are the uncertain design outcomes being drawn? We know from both logic and observation that designs are valued because of the functions that ‘‘their’’ artifacts perform, and that design outcomes are uncertain. Furthermore, designers and architects regularly make qualitative judgments about the potential of different designs and design architectures to achieve functionality and deliver value in return for effort. Nevertheless, as of today, we have no data to support statistical estimation of the relevant probability distributions, and there is also no theory to tell us what those distributions ‘‘should’’ look like.35 In essence, design valuation today is in a state similar to that of insurance contracting 350 years ago. At that point in time, life and property insurance contracts were being bought and sold, but statistics on mortality and property losses were not available to the buyers and sellers. As a result, the pricing of insurance contracts was a helter-skelter, catch-as-catch-can affair. Many mistakes were made, and many frauds were perpetrated because of the lack of objective data on which to base projections of future claims.36 With respect to design valuation today, we have a promising framework based on robust mathematical and economic logic. But we have insufficient data on hand to support formal hypothesis testing or statistical inference of the key parameters describing functions or value. Making matters worse, design valuation is more complex than insurance valuation, because the functions of artifacts are far more diverse than the functions of insurance contracts. However, research is even now being done to address these gaps. The work mainly focuses on open source codebases: These are promising sites for scientific work because they are accessible and because they often have well-documented design histories. Currently, two separate research efforts are under way that aim to correlate codebase changes with achieved functionality and value in order to assess option value.37 These studies represent important first steps toward building up useful data on design functions and outcomes, which can support more objective and quantitative methods of design valuation in the future. Design Games and the Institutions of Innovation The intrinsic difficulty of design valuation is compounded by the fact that in modern economies, companies and entrepreneurs play complex, competitive ‘‘value-capture

316

Carliss Y. Baldwin and Kim B. Clark

games’’ within design architectures.38 As a result of these ‘‘games,’’ the value created by an evolving set of designs does not always stay in the same hands. This is good news for society but bad news for science, which must track design value as it moves around and seek explanations for its movement. Sometimes the first to introduce a new architecture captures the lion’s share of its value. At other times, value is captured by those who focus on a small set of modules. In the marketplace of personal computers, for example, Intel Corporation represents the first type of success; Dell Computer Company represents the second. IBM is an example of an architect-firm that failed to capture long-term value from its PC architecture; Compaq Computer first succeeded and then failed at competition focused on modules of the PC architecture. The complexity of value-capture games means that the scientific study of designs must distinguish between achieved functionality (a property of a complete design) and the financial success of the design’s creators, owners, or sponsors. Achieved functionality is necessary, but not sufficient, for financial success. The Internet, for example, is a triumph of achieved functionality, but it is not ‘‘owned’’ by anyone. It has not made its creators rich in proportion to the value it has created for others. The distinction between value created and value captured points to another topic in the general scientific study of designs: the study of ‘‘what designs need’’ from the economy and from society. As we said above, designing is a costly activity. For a complex design, several stages of cost must be incurred before a (hopefully) valuable artifact or system can come into existence. The economy and society must therefore structure incentives and rewards and provide resource allocation mechanisms to support these stages from beginning to end. Taken as a whole, the incentives, rewards, and mechanisms that support the creation and reification of designs constitute a system of institutions in the formal sense defined by Masahiko Aoki (2001). According to Aoki, institutions can be viewed as equilibria of linked games with self-confirming beliefs. As a result, the properties of institutions can be derived from the formal specification of a game. The properties of designs, design structure, and design value in turn can be part of the formal specification. Thus, from Aoki’s theory of institutions, it is possible for the first time to develop a formal and comprehensive theory of the institutional systems needed to support the creation of new designs. These systems perforce are institutions of innovation. Using Aoki’s methods, studies of the institutions of innovation can be based on the twin foundations of design structure and design value. Design structure constrains the form and organization of the institutions; design value supplies the fuel (in the form of incentives and rewards) and channels it (via resource allocation mechanisms) to different points in the design structure. However, as we have seen,

Between ‘‘Knowledge’’ and ‘‘The Economy’’

317

scientific studies of design structure and design value are just getting started: thus, it may be too early for formal studies of institutions to get off the ground. Still, the need for this type of work is clear. If we want to understand how designs affect the world from a scientific perspective, we must look at how they affect and are in turn affected by institutions. Because the formal study of institutions through the lenses of design structure and design value is brand new, there is hardly any ‘‘recent work’’ to report yet.39 Instead, in the remainder of this section, we will describe two recent cases wherein the institutions changed—visibly and radically—apparently because the structure and value of the underlying designs changed.40 Because we understand so little, it is appropriate to frame these two cases as ‘‘puzzles.’’ Solving these ‘‘puzzles,’’ we believe, requires research on how design structure and value together create a nexus in which new institutional forms can arise and flourish. Puzzle #1—Vertical-to-Horizontal Industry Transitions and Modular Clusters In 1995, Andy Grove described a vertical-to-horizontal transition in the computer industry.41 In a now-famous picture (Figure 6), he described the transformation of that industry from a set of vertically integrated ‘‘silos,’’ e.g., IBM, DEC, Sperry Univac, and Wang, to a large number of firms spread out among a set of horizontal layers: specifically, the chip layer, the computer layer, plus the operating system, application software, and sales and distribution layers. Grove did not know exactly what had caused this transition. Intuitively, he felt it was spurred by changes in the cost of components and the recombinant possibilities of the underlying designs, that is, by changes in design structure and value:

Figure 6 The vertical-to-horizontal transition in the computer industry. Source: Adapted from Grove (1996), p. 44.

318

Carliss Y. Baldwin and Kim B. Clark

A consumer could pick a chip from the horizontal chip bar, pick a consumer manufacturer from the computer bar, choose an operating system . . . grab one of the several ready-to-use applications off the shelf . . . and take the collection of these things home. . . . He might have trouble making them work, . . . but for $2000 he had just bought a computer system. . . .42

But though the causes were unclear, Grove believed the consequences of the transition were profound: Going into the eighties, the old computer companies were strong, growing and vital. . . . But by the end of the eighties, many large vertical computer companies were in the midst of layoffs and restructuring. . . . [A]t the same time, the new order provided an opportunity for a number of new entries to shoot into preeminence.43

Grove’s horizontal industry structure is distinguished by the fact that most firms in it make modules that are in turn parts of larger systems. For this reason, we call this industry structure a ‘‘modular cluster.’’ Modular clusters can be made up of hundreds or even thousands of firms operating in many ‘‘submarkets,’’ i.e., different but complementary product categories. The ‘‘modular cluster’’ form of industry structure is—probably—an institution of innovation, meaning that its form responds to the structure and value of an underlying set of designs. This form emerged in the computer industry between 1975 and 1990 during a time when computer design architectures were becoming increasingly modular and ‘‘open,’’ giving rise to the ‘‘mix-and-match’’ property Grove described. The consequences of the transition were indeed vast—in terms of value created for consumers, value created (and destroyed) for investors, and turbulence in participation and market shares. At least one other industry—mortgage banking—has gone through a vertical-to-horizontal transition, as documented by Michael Jacobides (2005). Other industries such as telecom and pharmaceuticals are allegedly moving in the same direction and may become modular clusters in the process. But the causes of these transitions—in particular their roots in the underlying design structures and values—remain quite mysterious.44 Puzzle #2: Open Source Development of Linux Our second institutional puzzle is the emergence of the open source development process in the 1990s. Before 1990, it was widely believed that software code above a certain level of complexity had to be designed and built by a tightly knit team of dedicated experts. Eric Raymond described his own views at that time as follows: [ I] believed there was a certain critical complexity above which a more centralized, a priori approach was required. I believed that the most important software (operating systems and really large tools like the Emacs programming editor) needed to be built like cathedrals, carefully created by individual wizards or small bands of mages working in splendid isolation, with no beta released before its time.45

Between ‘‘Knowledge’’ and ‘‘The Economy’’

319

The rationale for these beliefs was convincingly set forth in Fred Brooks’ classic, The Mythical Man-Month, published in 1975. Brooks was one of the chief architects of IBM’s System/360, the first ‘‘truly modular’’ computer system. But when he attempted to partition the design of the System/360’s system software46 into discrete modules (as had been done with the hardware), the attempt failed. Reflecting later on this (and other) software engineering projects he had led, he formulated Brooks’ Law: Adding manpower to a late software project makes it later.47 In a nutshell, Brooks argued that when new people are added to a project, the extra tasks of training them and repartitioning the work would drag down the performance of the group. Accordingly, Brooks advocated the ‘‘small sharp team’’ approach to software design and development: [O]ne wants the system to be built by as few minds as possible.48 [T]he entire system also must have conceptual integrity, and that requires a system architect to design it all, from the top down.49

Against the backdrop of Brooks’ Law, Linux and its development process emerged in the mid-1990s as an anomaly. According to Raymond: Linus Torvalds’ style of development—release early and often, delegate everything you can, be open to the point of promiscuity—came as a surprise. No quiet, reverent cathedral building here—rather the Linux community seemed to resemble a great, babbling bazaar of differing agendas and approaches . . . out of which a coherent and stable system could seemingly arise only by a succession of miracles. . . . [But] the Linux world not only didn’t fly apart in confusion, [it] seemed to go from strength to strength. . . .50

In fact, Linux was only one, albeit the most visible, of a group of open source codebases that came into public view during the 1990s. These codebases were developed, debugged, and maintained by self-described communities of user-developers. In contrast to Brooks’ notion of ‘‘small, sharp teams,’’ open source methods seemed to be anarchic. Tens, hundreds, or even thousands of people would participate in the creation and evolution of a codebase on a voluntary, as-needed basis. Open source development communities are also—probably—institutions of innovation. It appears that some design structures can support and benefit from this form of organization, while others cannot. (Linux is an example of the former type; the first-released Mozilla codebase, depicted on the left-hand side of Figure 5, is an example of the latter type.) In related work, we have argued that design structure and value can explain the scale of effort that will be drawn into an open source development process.51 But as with modular clusters, there is still much work to be done to place this argument on a firm scientific footing.

320

Carliss Y. Baldwin and Kim B. Clark

Scientific Studies of Designs vs. Simon’s Science of Design In this chapter we have argued that designs are worthy of investigation as a general phenomenon and can be the object of scientific study across disciplines. Herbert Simon said the same thing more than forty years ago. Never given to understatement, he sought to rearrange Alexander Pope’s famous dictum, saying: The proper study of mankind has been said to be man. . . . If I have made my case, then we can conclude that, in large part, the proper study of mankind is the science of design.52

Reality has not lived up to Simon’s vision, however. At present, most scientific work on designs takes place in widely separated, often noncommunicating fields. The study of designs has made no dent on the natural or social sciences. There is no recognized field called the ‘‘science of design.’’ Putting Simon’s bluster aside, why has so little happened? Why did his compelling vision fail to materialize? Historian of science Peter Galison (1987) has argued that scientists will go where their tools of observation and analysis take them, but can go no further. We think that Simon, with characteristic optimism, greatly underestimated the complexity of actual designs and overestimated the capacity of our tools to measure, sort, categorize, and compare designs across different domains. He assumed that designs would be easily accessible to ‘‘full inspection and analysis.’’ This is simply not the case. A design can be made up of a million different instructions. Such an object cannot be categorized, taxonomized, or compared to others very easily. Yet for purposes of conducting science, the ability to observe an object in its raw state is not enough. One also needs tools that can convert raw observations into useful summaries, projections, and views—and do so efficiently. Designs have proved to be much more complicated than Simon perceived in the late 1960s. Moreover, the abstractions needed to support different design processes across a range of fields are not very similar, even when they are all expressed in digital formats. The CAD files describing a building are not easily compared to the source code of an operating system. Both are expressed in computer-readable languages, but the translation from one to the other is difficult and tedious. Simon did not foresee such high barriers to integration across different fields of design. Thus today, while there are many places where designs are studied scientifically, there is no unified ‘‘science of design’’ of the type Simon envisioned. Given that unification has not happened yet, it is hard to be optimistic about the possibility of a truly general science of design emerging in the near future. Nevertheless, as we have tried to show, in the forty-plus years since Simon delivered his manifesto, there has been significant progress in building tools that can be applied to the scientific study of designs as a general phenomenon. As a result, the ‘‘Galison gap’’ that existed in the 1960s may have shrunk somewhat. The most important tools, we

Between ‘‘Knowledge’’ and ‘‘The Economy’’

321

believe, address the three areas of inquiry identified above: structure, value, and institutions. In support of design structure analysis, there are DSMs and other mapping methodologies, which can be a lingua franca of design structure. In support of design valuation, there is option theory, functional valuation, and the net option value (NOV) method. And in support of institutional studies, there are the methods of comparative institutional analysis pioneered by Masahiko Aoki (2001). Significantly, the new tools are compatible and complementary. They have a common mathematical base: search and decision-making under uncertainty in complex design spaces. Hence, with these new tools, three previously separate areas of inquiry—design structure, design value, and institutions of innovation—can be integrated in mutually supportive ways. Indeed, many of the works cited above have already done so with preliminary but exciting results. In these works one can see how design structure affects value, but value helps to predict the evolution of structure. One can see how design structure constrains institutional forms, but institutions also influence changes in design structure over time. One can see that design value matters because it both predicts and rewards behavior, while institutions are important because they filter value. These and other insights are being verified and amplified as the work proceeds. In conclusion, new tools of observation and analysis now make it possible for widely scattered studies of design structure, value, and institutions to come together and begin to build upon one another. If that were to happen, the separate scientific studies of design would coalesce into a general science of design and Simon’s vision would become a reality. At this point in time, the barriers to integration are still very high, but they are coming down. Thus, with new tools in hand, we are— cautiously—optimistic. Acknowledgments Our special thanks to Christoph Hienerth, Peter Murmann, David Sharman, Marcin Strojwas, Kevin Sullivan, Eric von Hippel, Tony Wasserman, Daniel Whitney, and Jason Woodard for commenting on earlier drafts of this paper. Thanks also go to Sushil Bajracharya, Cristina Videira Lopes, John Rusnak, Alan MacCormack, Joachim Henkel, Michael Jacobides, Nitin Joglekar, Gregor Kiczales, Karim Lakhani, Sonali Shah, Mary Shaw, and Edwin Steinmuller for sharing key data and insights. Finally, we would like to thank participants in the NSF Science of Design Workshop, the MIT–University of Munich Innovation Workshop, and the Conference on Advancing Knowledge and the Knowledge Economy for conversations that contributed to this chapter in significant ways. We alone are responsible for errors, oversights, and faulty reasoning.

322

Carliss Y. Baldwin and Kim B. Clark

Notes A previous, shorter version of this paper was entitled ‘‘Designs and Design Architecture: The Missing Link between ‘Knowledge’ and the ‘Economy.’ ’’ 1. Simon (1981), p. 129. 2. Baumol (2002). 3. Simon op cit., pp. 6–8; Alexander (1964), pp. 55–70. 4. Mokyr (2002), pp. 4–21. 5. Baldwin and Clark (2000), pp. 76–77. Note that it is possible for the design of a complex system to be created without the agency of a design architect. In that case, the system itself will have an architecture in the sense of ‘‘an abstract description of the entities of a system and the relationships between those entities’’ (ESD Architecture Committee 2004). However, the system architecture will be undesigned or (in the language of complexity theory) ‘‘emergent.’’ 6. A ‘‘design space’’ consists of all possible variants of the design of an artifact. A complete design is a point within a design space. ‘‘Value’’ is a mapping of a mathematical function onto a design space; the process of design can be thought of as a search through a design space for high points in a ‘‘value landscape’’ (Simon 1981, pp. 136–144; Baldwin and Clark 2000, pp. 24–28, 232–234). In computer science, the concept of a design space was pioneered by Gordon Bell and Allen Newell (1971) and has been used extensively in the fields of automated design and artificial intelligence. The concept also appears in many fields of engineering. For example, in software engineering, one early example of the explicit use of design spaces was Garlan and Notkin (1991). The concept was generalized by Thomas Lane under the supervision of Mary Shaw and David Garlan (Shaw and Garlan 1996, pp. 97–113) and recently has been formalized by Cai and Sullivan (2005). Analogous concepts of search spaces and (fitness) landscapes arise in evolutionary biology and complexity theory. 7. Improvement (adaptation) and variety (radiation into niches) are two aspects of design evolution. 8. The community of scholars researching ‘‘design theory and methods’’ is estimated to be roughly on the order of 500 to 1000 people (Daniel Whitney, private communication). 9. See, for example, recent work on so-called ‘‘general purpose technologies’’ in economics, e.g., Bresnahan and Trajtenberg (1995), Helpman and Trajtenberg (1998), David and Wright (2003). 10. This list is open to debate and discussion. Also, different subsets of these axioms may be useful for different purposes. 11. Unreified designs may have educational or artistic value, but these are the exception, not the rule. 12. Von Hippel (1988, 2005); Franke and Shah (2003); Hienerth (2004). 13. This list of information goods is taken from the introduction of the influential book Information Rules (Shapiro and Varian 1999, p. 3). 14. Property rights, e.g. patents or copyrights, can prevent others from using the design. However, property rights are a feature of the institutional environment (see below), not an intrinsic property of designs.

Between ‘‘Knowledge’’ and ‘‘The Economy’’

323

15. Clark (1985). 16. Merton (1998). 17. Note that goods are rankable if and only if their designs are rankable. Saying that ‘‘this computer is better than that one (for gaming)’’ or ‘‘this coat is better than that one (for warmth)’’ is the same thing as saying ‘‘this design is better than that one (for some purpose).’’ 18. See, for example, two recent papers from different fields: Yu et al. (2003) and Cai and Sullivan (2005). 19. Note that a ‘‘decision’’ is both a ‘‘task’’ (viewed ex ante) and an ‘‘outcome’’ (view ex post). Both views are relevant, though one or the other may dominate in different types of analysis. 20. The design rules specify what the modules must do in order to work together as a system. At a minimum, the rules must prescribe the interfaces between interacting components, common protocols, and conformance standards (Baldwin and Clark 2000, p. 77). Ideally, modules are strictly independent of each other except for their common dependence on design rules. In practice, however, some intermodular dependencies can be tolerated. For a somewhat different definition of modularity, based on the encapsulation of functions, see Ulrich (1995). 21. Baldwin and Clark (2000), pp. 49–62. 22. Other domain-independent mapping techniques include layered views of designs, design hierarchy diagrams, and 3-dimensional (molecular) views of design dependencies. 23. Eppinger (1991); McCord and Eppinger (1993); Eppinger et al. (1994). See also the entries at http://www.dsmweb.org/publications_year.htm. 24. Parnas (1972, 2001). 25. In subsequent work, Cai and Sullivan (2005) formalized the concept of ‘‘informationhiding modularity’’ within a particular class of mathematically represented design spaces. 26. Basically, information hiding is a strategy of encapsulation in the sense described by Kirschner and Gerhart (1998). Information hiding localizes the impact of particular environmental changes and thus prevents them from ramifying throughout the system. Kirschner and Gerhart argue that encapsulation is a general property of evolvable systems. 27. Rusnak (2005). 28. This statement is embarrassingly obvious to designers, engineers, and architects, but its implications are often overlooked by managers, policymakers, and social scientists. 29. MacCormack (2001). 30. Raymond (1999). 31. Robert Merton was the first to put forward a general theory of option valuation based on the principles of dynamic decision-making under uncertainty (Merton 1973). Design options differ from financial options in two important ways: First, they are ‘‘real options,’’ meaning that their exercise affects the world, and second, there is usually no underlying asset to be replicated, thus Black-Scholes replication does not apply. Despite these differences, design options fall within Merton’s general framework (Merton 1998). 32. Baldwin and Clark (2003); Gomes and Joglekar (2004). 33. In contrast, the design rules, including the interfaces, must remain relatively fixed.

324

Carliss Y. Baldwin and Kim B. Clark

34. The formula is as follows: System Value ¼ S0 þ NOV1 þ NOV2 þ    þ NOVj ; where S0 is the value of a minimal system and the NOVi are the values of each module. Each module’s value in turn can be written as NOVi ¼ max½si ðni Þ 1=2 Qðki Þ  Cðni Þki  Zi ; ki

where ki is the number of experimental trials conducted on the ith module; si ðni Þ 1=2 Qðki Þ is the expected value of the best of ki designs, Cðni Þki is the cost of the experiments, and Zi measures the degree to which the module is ‘‘hidden’’ from others. 35. As a working assumption, the NOV method assumes that design outcomes have normal, mean-zero, i.i.d. distributions. The differences between distributions for different modules then come down to differences in the parameter of ‘‘technical potential,’’ denoted s. 36. Hacking (1975). The problems of insurance valuation were a key driver in the development of modern probability theory and statistics. 37. Bajracharya and Ngo (2005); Karim Lakhani and Neil Conway (private communication). 38. Brandenburger and Nalebuff (1996). 39. There is, of course, a large and valuable literature that looks at the institutions of innovation from other perspectives. First and foremost is Richard Nelson and Sidney Winter’s path-breaking book An Evolutionary Theory of Economic Change, which has stimulated an enormous amount of scholarly research since first published in 1982. Much of the work in this line deals implicitly with the impact of new designs on corporate and institutional structures and vice versa. See, in particular, seminal papers by Langlois and Robertson (1992), Garud and Kumaraswamy (1995), Sanchez and Mahoney (1996), and Schilling (2000) on modular designs and organizational forms, as well as recent contributions by Brusoni and Prencipe (2001), Sturgeon (2002), and Jacobides (2005). Murmann’s study of the coevolution of chemical engineering science and institutions in Britain and Germany in the late 19th century is especially revealing of the interaction between the ‘‘needs’’ of a set of new product designs and the institutional structures that were developed to fulfill those ‘‘needs’’ (Murmann 2003). Other related work evaluates search strategies on abstract value landscapes: see, for example, Levinthal (1997), Rivkin (2000), and Rivkin and Siggelkow (2003). What is new today is the opportunity to integrate explicit characterizations of design structure and design value with Aoki’s game-theoretic approach to institutions. 40. Strojwas (in progress) considers another case in which the institutions of innovation changed in response to changes in design structure and value. 41. Grove (1996). 42. Ibid., pp. 41–42. 43. Ibid., p. 45. 44. Jason Woodard (in progress) is conducting computational experiments designed to shed light on how and why modular cluster form and how such clusters evolve. 45. Raymond (1999), p. 29. 46. ‘‘System software’’ is now called the computer’s operating system. 47. Brooks (1995), p. 25. Italics in original.

Between ‘‘Knowledge’’ and ‘‘The Economy’’

325

48. Ibid. p. 30. 49. Ibid. p. 37. 50. Raymond, op. cit. p. 30. 51. Baldwin and Clark (in press). 52. Simon (1981), p. 159. Italics added. Pope’s lines, from An Essay on Man, are Know then thyself, presume not God to scan; The proper study of Mankind is Man.

References Alexander, Christopher (1964). Notes on the Synthesis of Form. Cambridge, MA: Harvard University Press. Aoki, Masahiko (2001). Towards a Comparative Institutional Analysis. Cambridge, MA: MIT Press. Bajracharya, Sushil K., and Trung Chi Ngo (2005). ‘‘Characterizing the Technical Potential of a Software Module’’ [manuscript] (March). ———, Trung Chi Ngo, and Cristina V. Lopes (2005). ‘‘On Using Net Options Value as a Value Based Design Framework’’ [manuscript] (March). Baldwin, Carliss Y., and Kim B. Clark (2000). Design Rules, Volume 1: The Power of Modularity. Cambridge, MA: MIT Press. ———, and Kim B. Clark (2003). ‘‘Where Do Transactions Come From? A Perspective from Engineering Design.’’ Working Paper 03-031, Harvard Business School, Boston, MA. ———, and Kim B. Clark (in press). ‘‘The Architecture of Participation: Does Code Architecture Mitigate Free Riding in the Open Source Development Model?’’ Management Science. Baumol, William J. (2002). The Free-Market Innovation Machine. Princeton, NJ: Princeton University Press. Bell, C. Gordon. and Allen Newell (1971). Computer Structures: Readings and Examples. New York: McGraw-Hill. Brandenburger, Adam M., and Barry J. Nalebuff (1996). Co-opetition, New York: Doubleday. Bresnahan, Timothy J., and Manuel Trajtenberg (1995). ‘‘General Purpose Technologies: ‘Engines of Growth’?’’ Journal of Econometrics 95: 83–108. Brooks, Frederick P. (1995). The Mythical Man–Month: Essays on Software Engineering, 20th Anniversary Edition. Reading, MA: Addison-Wesley. Brusoni, Stefano, and Andrea Prencipe (2001). ‘‘Unpacking the Black Box of Modularity: Technologies, Products and Organizations.’’ Industrial and Corporate Change 10(1): 179– 205. Cai, Yuanfang, and Kevin Sullivan (2005). ‘‘A Value-Oriented Theory of Modularity in Design, Viewed as a Decision-Making Activity.’’ University of Virginia, Charlottesville (April). Clark, Kim B. (1985). ‘‘The Interaction of Design Hierarchies and Market Concepts in Technological Evolution.’’ Research Policy 14(5): 235–251.

326

Carliss Y. Baldwin and Kim B. Clark

David, Paul A., and Gavin Wright (2003). ‘‘General Purpose Technologies and Productivity Surges: Historical Reflections on the Future of the ICT Revolution.’’ In The Economic Future in Historical Perspective, P. A. David and G. Wright, eds. Oxford: Oxford University Press for the British Academy. Eppinger, Steven D. (1991). ‘‘Model-Based Approaches to Managing Concurrent Engineering.’’ Journal of Engineering Design 2: 283–290. ———, D. E. Whitney, R. P. Smith, and D. A. Gebala (1994). ‘‘A Model-Based Method for Organizing Tasks in Product Development.’’ Research in Engineering Design 6(1): 1–13. ESD Architecture Committee (2004). ‘‘The Influence of Architecture in Engineering Systems.’’ Engineering Systems Monograph, MIT, Cambridge, MA (March). Franke, Nikolaus, and Sonali Shah (2003). ‘‘How Communities Support Innovative Activities: An Exploration of Assistance and Sharing Among End-Users.’’ Research Policy 32(1): 157–178. Galison, Peter (1987). How Experiments End. Chicago: University of Chicago Press. Garlan, David, and David Notkin (1991). ‘‘Formalizing Design Spaces: Implicit Invocation Mechanisms.’’ In Proceedings of the 4th International Symposium of VDM Europe on Formal Software Deveolpment, Vol. 1, pp. 31–44. New York: Springer-Verlag. Garud, Raghu, and Arun Kumaraswamy (1995). ‘‘Technological and Organizational Designs to Achieve Economies of Substitution.’’ Strategic Management Journal 17: 63–76. Gomes, Paulo J., and Nitin R. Joglekar (2004). ‘‘The Costs of Coordinating Distributed Software Development Tasks’’ [manuscript] (October). Grove, Andrew S. (1996). Only the Paranoid Survive. New York: Doubleday. Hacking, Ian (1975). The Emergence of Probability. Cambridge, UK: Cambridge University Press. Helpman, Elhanan, and Manuel Trajtenberg (1998). ‘‘Diffusion of General Purpose Technologies.’’ In General Purpose Technologies and Economic Growth, E. Helpman, ed., pp. 85– 119. Cambridge, MA: MIT Press. Hienerth, Christoph (2004). ‘‘The Commercialization of User Innovations: Sixteen Cases in an Extreme Sporting Industry.’’ In Proceedings of the 26th R&D Management Conference, Sesimbra, Portugal. Jacobides, Michael G. (2005). ‘‘Industry Change Through Vertical Dis-Integration: How and Why Markets Emerged in Mortgage Banking.’’ Academy of Management Journal (June). Kirschner, Marc, and John Gerhart (1998). ‘‘Evolvability.’’ Proceedings of the National Academy of Sciences USA 95: 8420–8427. Langlois, Richard N., and Paul L. Robertson (1992). ‘‘Networks and Innovation in a Modular System: Lessons from the Microcomputer and Stereo Component Industries.’’ Research Policy 21: 297–313. Levinthal, Daniel A. (1997). ‘‘Adaptation on Rugged Landscapes.’’ Management Science 43: 934–950. Lopes, Cristina V. (2005). ‘‘On the Nature of Aspects: Principles of Aspect-Oriented Design.’’ Submitted to ACM Transactions of Software Engineering.

Between ‘‘Knowledge’’ and ‘‘The Economy’’

327

———, and Sushil K. Bajracharya (2005). ‘‘An Analysis of Modularity in Aspect-Oriented Design.’’ In AOSD ’05: Proceedings of the 4th International Conference on Aspect-Oriented Software Development, pp. 15–26. ACM Press. MacCormack, Alan D. (2001). ‘‘Product-Development Practices That Work: How Internet Companies Build Software.’’ Sloan Management Review 42(2): 75–84. ———, John Rusnak, and Carliss Baldwin (2004). ‘‘Exploring the Structure of Complex Software Designs: An Empirical Study of Open Source and Proprietary Code.’’ Working Paper 05-016, Harvard Business School, Boston, MA (September). McCord, Kent R., and Steven D. Eppinger (1993). ‘‘Managing the Iteration Problem in Concurrent Engineering.’’ Working Paper 3594-93-MSA, MIT, Cambridge, MA (August). Merton, Robert C. (1973). ‘‘Theory of Rational Option Pricing.’’ Bell Journal of Economics and Management Science 4 (Spring): 141–183; reprinted in Continuous Time Finance, Oxford, UK: Basil Blackwell, 1990. Merton, Robert C. (1998). ‘‘Applications of Option-Pricing Theory: Twenty-Five Years Later’’ [Nobel Lecture]. American Economic Review 88(3): 323–349. Mokyr, Joel (2002). The Gifts of Athena: Historical Origins of the Knowledge Economy. Princeton, NJ: Princeton University Press. Murmann, Johann Peter (2003). Knowledge and Competitive Advantage: The Coevolution of Firms, Technology, and National Institutions. Cambridge, UK: Cambridge University Press. Nelson, Richard R., and Sidney G. Winter (1982). An Evolutionary Theory of Economic Change. Cambridge, MA: Harvard University Press. Parnas, David L. (1972). ‘‘On the Criteria to Be Used in Decomposing Systems into Modules.’’ Communications of the ACM 15: 1053–1058. ——— (2001). Software Fundamentals: Collected Papers by David L. Parnas, D. M. Hoffman and D. M. Weiss, eds. Boston, MA: Addison-Wesley. Raymond, Eric S. (1999). The Cathedral and the Bazaar. Sebastopol, CA: O’Reilly & Associates. Rivkin, Jan W. (2000). ‘‘Imitation of Complex Strategies.’’ Management Science 46: 824– 844. ———, and Nicolaj Siggelkow (2003). ‘‘Balancing Search and Stability: Interdependencies Among Elements of Organizational Design.’’ Management Science 49: 290–311. Rusnak, John (2005). ‘‘The Design Structure Analysis System: A Tool to Analyze Software Architecture.’’ PhD thesis [unpublished], Harvard University, Division of Engineering and Applied Sciences, Cambridge, MA. Sanchez, Ronald A., and Joseph T. Mahoney (1996). ‘‘Modularity, Flexibility and Knowledge Management in Product and Organizational Design.’’ Strategic Management Journal 17: 63–76. Schilling, Melissa A. (2000). ‘‘Toward a General Systems Theory and Its Application to Interfirm Product Modularity.’’ Academy of Management Review 25(2): 312–334. Shah, Sonali K. (2003). ‘‘Community-Based Innovation & Product Development: Findings from Open Source Software and Consumer Sporting Goods.’’ PhD thesis [unpublished], Sloan School of Management, MIT, Cambridge, MA.

328

Carliss Y. Baldwin and Kim B. Clark

Shapiro, Carl, and Hal R. Varian (1999). Information Rules: A Strategic Guide to the Network Economy. Boston, MA: Harvard Business School Press. Sharman, David (2002). ‘‘Valuing Architecture for Strategic Purposes.’’ MS thesis [unpublished], Engineering Systems Division, MIT, Cambridge, MA. ———, Ali Yassine, and Paul Carlile (2002a). ‘‘Characterizing Modular Architectures.’’ Proceedings of the ASME 14th International Conference on Design Theory & Methodology, DTM-34024, Montreal, Canada (September). ———, Ali Yassine, and Paul Carlile (2002b). ‘‘Architectural Optimization Using Real Options Theory and Dependency Structure Matrices.’’ Proceedings of the ASME 28 Design Automation Conference, DAC-34119, Montreal, Canada, (September). Shaw, Mary, and David Garlan (1996). Software Architecture: An Emerging Discipline. Upper Saddle River, NJ: Prentice-Hall. Simon, Herbert A. (1962). ‘‘The Architecture of Complexity.’’ Proceedings of the American Philosophical Society 106: 467–482; repinted in idem. (1981) The Sciences of the Artificial, 2nd ed., pp. 193–229. Cambridge, MA: MIT Press. ——— (1981). The Sciences of the Artificial, 2nd Ed. Cambridge, MA: MIT Press. Steward, Donald V. (1981). ‘‘The Design Structure System: A Method for Managing the Design of Complex Systems.’’ IEEE Transactions on Engineering Management EM-28(3): 71– 74 (August). Strojwas, Marcin (in progress). ‘‘Form and Functionality: The Impact of Organizational Form on Product Development Performance in the Semiconductor Industry.’’ PhD thesis [unpublished], Information, Technology & Management, Harvard University, Cambridge, MA. Sturgeon, Timothy (2002). ‘‘Modular Production Networks: A New American Model of Industrial Organization.’’ Industrial and Corporate Change 11(3): 451–496. Sullivan, Kevin, William G. Griswold, Yuanfang Cai, and Ben Hallen (2001). ‘‘The Structure and Value of Modularity in Software Design.’’ SIGSOFT Software Engineering Notes 26(5): 99–108. ———, William G. Griswold, Yuanyuan Song, Yuanfang Cai, Macneil Shonle, Nishit Tewari, and Hridesh Rajan (2005). ‘‘On the Criteria for Decomposing Systems into Aspects.’’ Submitted to ESEC/FSE (April). Ulrich, Karl (1995). ‘‘The Role of Product Architecture in the Manufacturing Firm.’’ Research Policy 24: 419–440. von Hippel, Eric (1988). The Sources of Innovation, Oxford: Oxford University Press. ——— (2005). Democratizing Innovation. Cambridge, MA: MIT Press. Whitney, Daniel E. (1990). ‘‘Designing the Design Process.’’ Research in Engineering Design 2: 3–13. Woodard, C. Jason (in progress). ‘‘Architectural Strategy and Design Evolution in Complex Engineered Systems.’’ PhD thesis [unpublished], Information, Technology & Management, Harvard University, Cambridge, MA. Yu, Tian-li, Ali Yassine, and David E. Goldberg (2003). In Proceedings of the ASME 2003 International Design Engineering Technical Conferences, 15th International Conference on Design Theory & Methodology, DETC2003/DTM-48657, Chicago, Illinois (September).

VI Models of Control and Cooperation

19 Patent Quantity and Quality: Trends and Policy Implications Dietmar Harhoff

That reminds me to remark, in passing, that the very first official thing I did, in my administration—and it was on the very first day of it, too—was to start a patent office; for I knew that a country without a patent office and good patent laws was just a crab, and couldn’t travel any way but sideways or backways. —Mark Twain, A Connecticut Yankee in King Arthur’s Court

Introduction It is probably a great gain for society that Mark Twain became a writer rather than an inventor—his inventions and investments in other inventors were the cause of Twain’s private misfortune, while his writing earned him the adoration of his readers and a considerable amount of money. Twain was fascinated with inventions and on Sept. 9, 1871, he applied for a patent on an ‘‘Improvement in Adjustable and Detachable Straps for Garments,’’ which was granted to Samuel L. Clemens (the author’s original name) on Dec. 19, 1871, as USPTO patent No. 121,992. The invention did not become a commercial success, and later in his life, Mark Twain lost his considerable fortune when he invested it in other people’s inventions.1 But the ‘‘crab’’ statement had been inserted into the literature already and was there to stay. While Mark Twain’s unambiguously positive assessment of patent systems is still shared by numerous policy-makers and patent office representatives, many economists have grown increasingly skeptical that the currently operating patent systems always have the immensely beneficial effect ascribed to them by Mark Twain. There is good reason for skepticism. During the past decade, various aspects of the patent system have been studied with new vigor. It is no overstatement to say that the ‘‘black box’’ of the patent system has finally been opened by economists, and that in the course of the analysis a number of interesting and puzzling facts have emerged. One is the discovery of the so-called ‘‘patent paradox’’—the surprising

332

Dietmar Harhoff

coincidence of increasing numbers of patent applications and decreasing importance of patents in the eyes of R&D managers (Hall and Ziedonis 2001). Another is the suggestion that the share of ‘‘questionable patents’’ issued by patent offices has been increasing (FTC 2003) and that patents may have become major obstacles to innovation in some sectors. Moreover, there is evidence of an anticommons problem (Heller and Eisenberg 1998)—an impediment to innovation arising from the fact that patent rights are nowadays overlapping and too dispersed to support efficient investment activity. There are industrial sectors such as chemicals and pharmaceuticals where many observers still attribute an important function to the patent system, but for technical fields such as software, the answer is much more controversial. Some U.S. studies have suggested that the negative impact of software patents outweighs the positive incentive effects created by them. Both in Europe and the United States, the debate about the contribution of the patent system has finally reached the public, and the discussion is likely to become more controversial over time. Much attention has focused on developments in the U.S. system, and it is often implicitly assumed that the rest of the world will follow in due course. This chapter seeks to contribute to the evolving discussion by analyzing some aspects of the European situation and by outlining possible improvements in the EPO system. Europe has seen its own increase in patenting activity. Yet, the superficial similarities between Europe and the United States may be deceiving. Are the ‘‘diseases’’ really the same in all patent systems? Or is there evidence that would call for a more differentiated approach in determining what form of cure should be administered? Major patent reform does not come easily, and while it would be tempting to attribute all problematic phenomena in Europe to the migration of U.S. practice, the evolution of patent systems and their close linkage to the respective legal systems puts each country under the spell of path dependence. Europe has not shown the same dysfunctional developments that have become apparent in the United States. Examination quality and postgrant review institutions appear to work considerably better than comparable elements of the U.S. system. But threats to patent quality have become apparent and need to be addressed. Patent Numbers and Patent Quality The Developments in the United States There is considerable agreement among academics about the actual trends that have materialized in patent systems during the past two decades, but some disagreement about the causes behind these trends. The annual number of USPTO patent grants was essentially constant until the mid-1980s and then grew at an annual rate of

Patent Quantity and Quality

333

about 5%. Real R&D increased only at a rate of 2.4% per year in that time period. Hence, patent grants grew considerably faster than R&D inputs. Econometric tests performed by Hall (2005) identify 1984 as the year in which a structural break occurred. Moreover, Hall (2005) suggests that the high increase in patenting is mainly due to applicants in the United States. Structural breaks have occurred in all technology fields except for pharmaceuticals and chemicals. Changes are driven by firms mainly active in the electrical, computing, and instruments industries. Their activities spill over into the aggregate statistics of other sectors in which these firms are also active and show the same patenting behavior as in their primary industry. Hall and Ziedonis (2001) interpret the rise in patent applications in the semiconductor industry as a patent arms race in which all parties seek to amass a large portfolio of patents that can be used in court if any of their rivals choose to attack. The root of the patent flood is then that even marginal applications can enhance the threat of counterattack—even if it is not meant to protect products or processes used by the enterprise. What concerns economists is that the increasing private benefit from patenting may not be accompanied by social returns that enrich society as a whole. Instead, patenting is turning more and more into an activity that can be called ‘‘market-stealing’’ rather than market-extending or market-creating. Some evidence points to the fact that in parallel to the increasing patenting activity, patent litigation in the United States has been increasing, both in absolute and relative terms. Lanjouw and Schankerman (2001) find that the rate of patent litigation increased somewhat from the 1978–1984 period to the 1991–1995 period, from 19 suits per thousand patents in the first period to 21 suits per thousand patents in the second, with some variation across technology areas. Somaya (2002) argues that the litigation rate rose in the late 1990s. In a recent study of patent litigation focusing on cases that terminated in 1998–2000, Allison et al. (2003) estimate a litigation rate of approximately 32 suits per thousand patents. The rate of patent litigation activity may have been increasing for a number of reasons—one is that there is a higher share of patents that are badly delineated (from each other and from the nonpatented state of the art). Uncertainty about the boundaries of patent rights then leads to legal controversies. A second explanation is that in the presence of a court system that allows parties to drive up each other’s costs and treats patent holders preferentially, patents are a valuable instrument for the extortion of payments—irrespective of their ultimate validity status. Some of the welfare consequences of these developments have been discussed in a report prepared by the Federal Trade Commission (2003). The report comes to the devastating conclusion that patents have become major obstacles to innovation in some sectors.2 Given the prospect of such effects, it is not surprising that patent system reform has appeared prominently on the public policy agenda in the United

334

Dietmar Harhoff

States. A major academic study was produced by the National Academies (National Research Council 2004), recommending changes in examination and postexamination stages of the patenting process as well as in the patent litigation system. The U.S. Patent and Trademark Office (2003) has developed a strategic plan to deal with the challenges. In their call for patent reform, these institutions have been joined by a number of large corporations.3 The European Situation As in the United States, both patent applications and patent grants at the EPO4 have increased much faster than R&D inputs in OECD countries.5 This statement holds even if applications with U.S. origin and U.S. R&D expenditures are excluded from the comparison. From 1990 to 2000, EPO patent applications grew from 70,955 to 145,241 (an average growth rate of 7.4% per annum) while OECD R&D inputs (in 1995 real terms) increased from $398 to $555 billion, which reflects an average annual growth of 3.4%. To provide a comparative figure for Germany: Patent applications filed at the DPMA, WIPO, or EPO with designation Germany grew at an annual growth rate of 5.9% from 1988 to 1998.6 At the EPO, the growth in applications was not dominated by any particular applicant nation—the distribution of patent origin remained remarkably stable from 1985 to 2000. The actual grant rate (the share of patent applications leading to a patent grant) remained almost constant at about 65% for patents with application years from 1978 to 1995 (Harhoff and Wagner 2005). Some observers have interpreted this figure as an indicator of a rather selective examination process when compared to the USPTO (Quillen et al. 2002). It is important to point out that even this figure is the result of a two-step selection process—about 95% of all EPO filings are submitted after the applicants have made a first priority filing at the respective national offices.7 Many applicants let only their most promising and relevant applications proceed to the EPO. Hence, a grant rate of 65% reflects the selection from a set of applications that have already been screened. In terms of growth rates of applications and grants, however, the actual growth of ‘‘patent quantities’’ is very similar to that in the United States. Given the current U.S. debate and the similarity in application growth, some questions come to mind immediately. Has the quality of applications submitted decreased with the increase in the absolute number of applications? Has the European patent system become too permissive in the sense that the EPO has issued ‘‘too many’’ patent grants? If the number of applications has increased strongly, does the roughly constant grant tell us that there are more economically relevant patents relative to R&D expenditures? After all, the applications have been examined by what has been taken to be a fairly strict standard of novelty and inventive step. Or is the constant grant rate the effect of habitual patent office behavior?

Patent Quantity and Quality

335

These questions are notoriously difficult to answer, but some particularities of the European system allow us to study the evolution of a number of important patent characteristics. The first indicator I use here to characterize long-term changes in applications is the number of claims, which I interpret as a measure of patent complexity. The number of claims is also an important determinant of patent office workload. Because examiners have to assess patent applications with particular attention to the claims, additional claims add to the examiners’ work effort and have been shown to increase pendencies. Complex claim structures can also be used to pursue strategic objectives. An impressive example is a recent WO publication (WO 2005/046746 A2) that contains 10,247 claims. The EPO has rendered a ‘‘no search’’ decision for this application (meaning that it will not issue a search report on the application in its current form), but the applicant is entitled to maintain the priority rights stemming from the application until the claim structure has become sufficiently transparent that a search can be undertaken. The applicant may use this situation to delay both the search and examination process. Whatever the specific motivation of the applicant, ‘‘claim flooding’’ of this type can hardly be interpreted as a harmless exploitation of loopholes in the Patent Cooperation Treaty. It is clear that strategic behavior of this type should be met with appropriate sanctions, possibly even the loss of any priority rights. But currently (and in particular under the PCT), patent offices can do little to counter dysfunctional applicant behavior of this type.8 Even if one is willing to exclude extreme cases such as this one, it is clear that the complexity of patent applications has been increasing in general. On average, the number of claims in incoming EPO applications (including PCT filings) has increased from 10.1 to 16.9 in the period from 1980 to 2000. Figure 1 demonstrates that this development has been similar for different types of applications at the EPO even if some differences persist. However, PCT applications from U.S. applicants exhibit a particularly strong increase in the number of claims. A simple multivariate analysis demonstrates that this development is not due to the technology mix of U.S. origin patents—presumably, it reflects drafting styles and strategic motives. Furthermore, PCT applications are generally more complex than non-PCT applications, presumably because they are subject to fewer restrictions and cost penalties than applications filed directly at the EPO or the USPTO. Both the USPTO and EPO have adopted cost rules that ask for a claim fee for any claims exceeding a certain threshold (10 at the EPO, 20 at the USPTO). An analysis of claims data demonstrates that applicants are sensitive to these fees. Leaving aside examiner incentives, simple statistics indicate that the increase in the number of claims in applications was to some degree, but not fully, translated into a larger number of claims in the actual patent grants. From 1988 to 1998, the

336

Dietmar Harhoff

Figure 1 Average number of claims. Source: Author’s computations based on data from www.epoline .org.

number of claims in applications increased by 33.9%, from 11.5 to 15.4, but during the same period, the number of claims in granted patents increased by only 18.6% (from 10.2 to 12.1 claims).9 To explore this development further, it is necessary to take a look at the quality of incoming applications. Fortunately, the EPO search process generates such an indicator. In EPO search reports, all patent and nonpatent documents used to reference prior art are classified as A-, X-, or Y-type references.10 X-type references characterize documents that by themselves are harmful to the novelty or inventive step of the application under review.11 In cases in which an application’s search report includes an X reference, this indicates that from the search officer’s point of view, at least one claim in the application does not meet the requirements of novelty or inventive step because a previous document (marked as the X reference) by itself casts doubt on the novelty or inventive step of the claim. This may not preclude the invention from being patented after changes have been made to the application (e.g., by striking out particular claims or limiting their scope), but the share of X references is a reasonable measure of the quality of the initial application. Figure 2 displays how the share of X-type references has developed over time. The increase in the number of these references could be explained by more marginal claims being included by applicants in their filings. Recall that applicants have

Patent Quantity and Quality

337

Figure 2 Average share of X-type references. Source: Author’s computations based on data from www.epoline.org.

increased the number of claims in their applications. Let us assume for a moment that each claim has a constant probability of generating an X-type reference and that claims have independent effects. In this case, an increase in the number of claims, as demonstrated in Figure 1, would also lead to an increase in the share of X-type references. However, if one calculates the average number of X-type references per claim in the application document, then there were 7.2 X-type references per hundred claims in 1988. That ratio increased to 10.3 X-type references per hundred claims in 1998. This suggests that the average quality of claims deteriorated over the time period considered here. As I pointed out before, neither the decreasing quality of incoming applications nor the increasingly complex claim structures have led to lower overall grant rates at the EPO. The typical logic put forth by patent practitioners to explain this phenomenon is that applicants who wish to obtain a patent can do so if they are willing to see the number of claims reduced or their patent narrowed in some other way. But the distribution of bargaining power in this negotiation process will depend crucially on the examiner’s willingness and opportunities to ultimately refuse a patent grant. As few as 5.1% of patent applications at the EPO with application years 1978–1995 were effectively refused, but in most of the cases not leading to a patent

338

Dietmar Harhoff

grant (27.4% of all applications in 1978–1995), the applicant withdrew the application at some point, often once the correspondence with the examiner showed the extent to which the examiner would allow the original claims to be included in the final grant (see Harhoff and Wagner 2005, Table 1, for details). These statistics demonstrate that examiner incentives are important and need to be analyzed in more detail. I turn to an important third indicator of developments at the EPO. The EPO system (and many national patent systems in Europe) includes mechanisms that allow third parties to challenge patent grants in a low-cost proceeding. Indeed, the EPO opposition system is often discussed as a model for a postgrant review system in the U.S. patent system. All of the reform proposals mentioned earlier include the recommendation to introduce such a postgrant review system. The analysis by Hall et al. (2003) and the recent analysis by Graham and Harhoff (2005) suggest that the introduction of such a postgrant review mechanism would be beneficial to the United States in terms of welfare gains. Opposition activity serves as a second, intensive screening of particularly valuable patents. Since errors in the delineation of these patent rights could cause particularly serious welfare losses, third parties who initiate opposition proceedings provide two types of important information. First, they select the more valuable patents, which are then subject to another round of review. Second, they typically provide information that is important for an objective assessment of the validity of the patent. The impact of such information is easily documented—about one-third of the patents attacked in opposition proceedings are ultimately revoked, and another third of the opposed patents are amended, i.e., the patent is narrowed. A detailed analysis of opposition at the EPO (and at the DPMA, the German Patent and Trademark Office) shows that opposition activity relative to the total number of patents granted has actually decreased considerably during the past 20 years in most, but not all, technical fields. This development is displayed in Figure 3, which plots the opposition rate by grant year for five major technical fields. Opposition activity has decreased in all of the aggregate technical fields. But the reduction has been most pronounced in fields such as electrical engineering and information technology, for which Hall (2005) has identified particularly strong increases in patenting activity. Unfortunately, there is no systematic evidence regarding patent litigation activity over time in Europe. Representative statistics on court cases are hard to collect, since most European countries do not have electronic case registration systems. Comprehensive data on annulment cases resolved by the German Patent Court are currently being analyzed by this author, and the absolute number of judgments shows a significant increase over time.12 However, the relative number of filings (as a share of patent grants) appears to have been roughly constant.

Patent Quantity and Quality

339

Figure 3 Opposition rate by technical field. Source: Author’s computations based on data from www.epoline.org.

As of yet there is no convincing explanation of why opposition activity in Europe has been decreasing. The cost of opposition does not appear to have increased sharply in the past 20 years. Therefore, the reasons for the considerable decline in opposition rates are likely to be linked to the returns that determine the incentives of a potential opponent. I suggest that three potential arguments are of particular interest for the current discussion. First, opposition activity may have been decreasing because the average patent granted by the EPO may now be less valuable than it used to be. As patents become less valuable, the value of having a patent revoked will be reduced as well, and fewer patents will be attacked. This explanation is consistent with simple theoretical models of opposition (e.g., Harhoff and Reitzig 2004). A second possible explanation runs as follows: The distribution of the patent’s value to the patent owner may not have shifted. But incentives for potential opponents may have changed, nonetheless. Patents may now be impediments to a larger

340

Dietmar Harhoff

number of rivals than before—and to each rival, the value of revoking the patent has decreased. The spread between social and private incentives to initiate an opposition may have increased. After all, bringing a patent to be revoked is a public good, and only players that receive a sufficiently high benefit from doing so will file an opposition.13 Note also that this particular explanation introduces a potential interaction between the first hurdle (examination) and the second hurdle (opposition) at the EPO. If examination standards have been relaxed or if current examination practice favors patents with low inventive step and high breadth, then these may be less likely to be challenged, even though these patents could cause welfare losses. A third potential explanation would follow the argument of Hall and Ziedonis (2001), who show that in the semiconductor industry, players have shifted to more cooperative behavior, as demonstrated by increased cross-licensing activity, which helps to avoid costly litigation. This argument may also apply to opposition activity in a broad range of industries. The change in the behavior of potential opponents and patent holders may have been brought about by the increased number of applications and patent grants. Although the European opposition mechanism offers a low-cost instrument for opponents, a regime of mutual nonaggression may have become relatively attractive in some industries. Obviously, this hypothesis (as well as the other two potential explanations) requires careful study. But the three candidates have in common that increasing patent quantities and (potentially) lower grant quality threaten an important element of the European patent system. Incentive Structures in a Vicious Cycle? What has led to the current development in Europe? While many of the developments mimic U.S. trends, what could account for the stunning drop in opposition activity and for the apparent stability of patent litigation? Any attempt at assessing the recent developments should be able to explain why Europe has been doing better than the United States in some areas (e.g., examination quality and litigation), why it has shared with the United States some of the trends in patent application quantities, and why the corrective effect of opposition has been decreasing lately. Here, I present a ‘‘story’’ that appears to be capable of explaining these developments. Applicants, in the United States as well as in Europe, do not seek to maximize the social returns to innovation—they are motivated by private concerns. Filing a marginal patent application more makes good business sense in an environment where everybody else is doing so, too. Irrespective of the deeper reasons for changing applicant behavior, it is clear that starting in the 1980s, patent applicants in the United States and to some degree in Europe have changed their patenting behavior. EPO patent applications have become more complex, and claims have had lower quality

Patent Quantity and Quality

341

than in the past. A potential explanation for such behavior could be that applicants had to use more complex claim structures in order to position their applications in the increasingly crowded patent space. Patent offices have shown different responses to the challenge of increasing workloads. The USPTO appears to have focused on fast processing of patents and on ‘‘patent-granting.’’ The EPO has allowed longer pendencies to occur, but it is now under increasing pressure to reduce these, even in the face of increasing application numbers. As pressure has increased to cut down on the backlogs of applications, the criteria by which patents have been evaluated in Europe may have slipped, giving even greater incentives to applicants for filing additional applications. A lowering of the examination hurdle may have resulted in greater demand for patents. In this process, applications are becoming even more complex and contain more claims, which are inserted in order to maintain options in an increasingly complex environment. A vicious cycle of deteriorating quality and of increasing numbers of applications may thus have developed. At this point, it is difficult to assess the quality of examined patents conclusively, but the described combination of reduced quality of applications and the reduction in opposition frequency are troubling news. It is conceivable that this dynamic could be stopped by a reduction in grant rates, but at the EPO, the grant rate has remained constant even in the face of strongly increasing application numbers and falling application quality. If the quality of patent applications has been decreasing, why should patent grant rates have remained at the same, roughly constant level? While some national offices (e.g., the DPMA) have reduced their grant rates, there may be few incentives for policy-makers in the EPO system to pursue or recommend such a move. The current institutional framework at the European Patent Office is presumably more conducive to support a pro-quantity than a pro-quality policy. EPC member nations (typically through their patent offices) are represented in the Administrative Council of the EPO, which is the institution’s highest decision-making body. The council has to approve major policy changes that affect, for example, the fee structure, search and examination policies, and the office’s budget. Incentives favoring a pro-quantity policy come into play, since EPC member nations (respectively, their national offices) receive half of the renewal fees for EPO-granted patents in the respective designated country. Thus, the individual member nations have an incentive to increase patent grants in order to receive higher revenues. This is particularly true for the smaller national offices, which have been elevated to some status of relevance only by the establishment of a European application and examination path. It is difficult to see how the dominance of such motives in the Administrative Council can be countered in a political environment in which politicians regularly refer to the increasing number of patent applications as evidence of real innovation activity. Patent numbers have an

342

Dietmar Harhoff

almost hypnotic effect on many decision-makers—and it is all too tempting for patent office officials to enjoy the praise rather than putting the record straight. Finally, as has been pointed out before, the EPO system differs significantly from the USPTO by allowing for more third-party involvement (for details, see Hall and Harhoff 2004). But the reduction in the opposition rate from about 10% in the early 1980s to less than 5% in 2004 is a troubling development. I suggest that this is largely due to shifts toward lower overall patent values, which has weakened the incentives for potential opponents in many, but not in all, technical fields. In chemicals, pharmaceuticals, and biotechnology—where innovation is often based on a discrete invention—no or little reduction in opposition has occurred. It is in complex or systems technology industries (information technology, electronics, computers, etc.) that the reduction in opposition has been most pronounced. Opposing marginal patents that have been granted may indeed have become less attractive for single opponents. Clearly, there is more analysis to be conducted before arriving at a definite conclusion, but the reduction in opposition may indeed be related to the granting of questionable patent rights. This ‘‘story’’ does not deny that the overall situation at the EPO is considerably better than at the USPTO. Indeed, in a bilateral comparison Europe can be shown to have the better system with respect to many dimensions. One should also keep in mind that in some areas, such as software and business methods, the European patent systems have been considerably less permissive than the USPTO (or than the USPTO had to be under the influence of recent court decisions), and the effect appears to have been a positive one. This author does not want to suggest that patent examiners (either at the USPTO or the EPO) are neglecting quality intentionally. But subtle shifts in incentives and the political framework may have had a strong cumulative effect within patent offices, and these may have worked to the detriment of the European patent system. Steps Toward Improvements While one may be more optimistic about the state of affairs in Europe than about the U.S. patent system, some aspects of the European system warrant attention in order to maintain and improve its effectiveness. I list a number of suggestions here, some of which clearly warrant further discussion. Debunk the Patent-Quantity View Some representatives of governments, businesses, and patent offices do not get tired of restating the view that the increasing number of patent applications or patent grants bears witness to an increase in innovation. This leads to the naive notion

Patent Quantity and Quality

343

that more patents are unambiguously better for an economy. This view is the foundation and the cornerstone of the patent-quantity mentality outlined before. It is counterproductive because it distracts policy-makers and the public from taking a hard look at the problems of the patent system. It is essential to challenge and debunk this view. Patents per se are irrelevant. Their impact on innovation and investment counts. Refine Criteria for Granting Patent Rights From an economic perspective, patent rights should be granted if invention and investment would not occur without the creation of these property rights. Otherwise, such rights will simply increase the cost of innovation to other parties in the economy. Patent rights should not become primary instruments for the distribution of wealth that would be generated without them—they should support the generation of new economic activities and thus of new wealth. As many practitioners have pointed out, this is easily said but difficult to translate into an operational rule that can be used by patent examiners on a day-to-day basis. The prescription ‘‘do not grant patents on obvious things’’ should lead to debate as to what is obvious or nonobvious in the United States, or in Europe to the debate as to what is inventive or not inventive. The difficulty of these discussions should be recognized—but it is also clear that the criteria used in the past 20 years need to be reconsidered.14 Bill Patent Applicants for Strategic Complexity of Applications and for Strategic Delays In a previous section, I referred to a patent application with 10,247 claims (WO 2005/046746 A2). This is not merely an extreme aberration (or conscious exploitation) of applicant behavior. Strategic complexity occurs nowadays on a routine basis. Each additional claim adds to the examination burden that patent offices face. Fee rules should be adjusted so that the cost of patenting would be lowered for applicants who do not exploit the system strategically by filing an abundance of claims while costs increase sharply for those who take the system for a ride. Moreover, patent offices should consider increasing the fees for divisional applications and delays in responding to patent office communication. Some patent applicants seek to avoid an early conclusive outcome of the examination process in order to maximize the option value of their patent. If there is evidence of such behavior, patent offices should be allowed to sanction it. The option of delaying examination by up to seven years that exists in some systems (e.g., at the German Patent and Trademark Office) should be limited (as was recently done at the JPO). A further option for reducing patent office workload and applicant opportunism might be to introduce a kind of preexamination prior to search that will allow examiners to

344

Dietmar Harhoff

communicate serious defects on an application, e.g., excess number of claims, complexity, etc. Strengthen Examiners’ Incentives and Rights to Impose Sanctions and to Refuse Patent Grants Leaving aside that some cost rules could be adjusted to deter dysfunctional behavior, the role of patent examiners needs to be rethought. During the past two decades, examiners have been tied more and more to procedural rules that seek to limit degrees of freedom on the examiner’s side. Simultaneously, the behavior of some applicants has become more opportunistic—a behavior that patent examiners frequently recognize as abusive but that they cannot sanction effectively. Examiners often see that applicants delay proceedings, try to add to the original application, or seek to position the patent ex post into the investment path of other players, and they know relatively well which of the applicants write low-quality applications with complicated claim structures meant to confuse rivals or the examiner. These behaviors impose high social costs on other participants in the system, and examiners should be given stronger rights to impose sanctions on applicants who abuse the system in these and other ways. Obviously, in order to maintain the fairness of the process, applicants should be allowed to file an appeal or have access to other kinds of recourse. But there needs to be a real threat against abuse of the system. The EPO examiners are highly qualified public sector employees—there should be more trust in their capabilities, and less confidence in the formulation of institutionalized routines. Moreover, the in-house incentives of patent offices should ensure that patent examiners are given adequate effort recognition for refusals and grants. In the EPO system, there appear to be strong disincentives for EPO examiners to block a patent grant as desired by the applicant.15 In private discussions, some examiners estimate the time effort for a refusal to be at least twice as high as the effort for a patent grant. The controlling system at the EPO counts both outcomes equally with respect to the examiner’s workload. It should not. Refusals that take more effort than granting a patent right should be given greater weight in the assessment of an examiner’s workload.16 Encourage Participation by Informed Third Parties Patent office examiners will generally not have all the information needed to find an optimal balance between the breadth of patent protection and incentive effects. They are likely to know a lot about the underlying technology, but possibly very little about the applicant’s market position and other business-related aspects. The EPO system includes various institutions that allow third parties to step forth and inform

Patent Quantity and Quality

345

the patent office about these aspects. Since such actions are associated with private costs for the informed party, these institutions need to be inexpensive for the contributing party. At the same time, the proceedings associated with them should be resolved quickly so that the strategic incentive to file frivolous attacks against a patent grant can be minimized. Obviously, the two objectives just mentioned are always in conflict. But the current resource allocation between examination and postgrant institutions may need adjustment. It is well known by now that patent value is distributed in a very skewed form—about 10% of all patents make up 90% of the total value.17 A fast resolution of opposition cases may make opposition more attractive in cases where the patent is questionable; at the same time, legal certainty will be resolved faster for patent owners whose patent has been granted correctly. Shifting some resources to resolving opposition cases may slow down the examination process for the average patent marginally, but incentives associated with the most important patent grants are likely to improve significantly. Do Not Destroy Working Patent Litigation Systems in the Pursuit of European Harmonization The hypothesis that ‘‘friendly courts’’ have contributed to the rise of (bad) patenting incentives has received much attention in the literature. But the hypothesis would presumably fail in Europe, where no such systematic rise in patent litigation has been observed and patent opposition has actually been falling. While these differences need to be analyzed in more depth, it seems clear that European patent courts have been working decently. Most European court systems do not allow parties to manipulate the costs of their adversaries for strategic purposes. The British rule of cost allocation strengthens the party with a strong case. Conversely, patent litigation in the United States has largely had the effect of strengthening patent owners. The European patent litigation system has not been characterized by such a preference for patent holders. Taken together, these factors may have made a decisive difference—they may have helped European countries to avoid the worst excesses that the United States has experienced. Indeed, one of the most controversial developments in the United States—the rise of the patent troll extortion model—has not taken hold in Europe as of yet. The harmonization of patent litigation systems is now one of the priorities of the European Community. But paradoxically, in the current context, the creation of a harmonized litigation system with stronger patent enforcement may have the unintended effect of increasing the incentives to take out questionable patents. The impact of such a step has to be assessed carefully— pursuing them without taking the wider ramifications into account might lead to a deterioration of the system.

346

Dietmar Harhoff

Technical expertise in the courtroom is often indispensable when it comes to decisions about patent infringement. Technical expertise may be even more important in decisions regarding the validity of patents. But while most legal systems admit technical experts, they shy away from bringing this expertise into judges’ chambers—in the form of a technical judge. The German experience with technical judges at the Federal Patent Court has been a largely positive one. At the same time, the European Commission is now proposing to introduce a centralized European court system for patent litigation, but to dispense with technical judges. That may not be a prudent step; its implications need to be considered before turning such a proposal into law. Do Not Reduce the Costs of Patenting Further Without Solving the Fundamental Problems of the Patent System First Policy-makers in industrialized countries appear to have followed the leitmotiv to make patenting available to any firm at the lowest possible price. Paradoxically, in the presence of constant grant rates, relatively high costs of patenting and patent enforcement may have saved the EPO system from some of the problems besetting the U.S. system. In the current situation, a uniform reduction of patent office fees or subsidization of patenting would lead to a further influx of applications. Financing constraints for SMEs may have to be addressed by public policy programs, but patents at discount prices will lead to more questionable patents, given current low hurdles of inventive step or nonobviousness. Initiate an Open Discussion Process About the Future of the European Patent System To this observer, the handling of the discussion surrounding the European Commission’s Directive on Computer-Implemented Inventions (CII) serves as an example as to how public policy-making in the patent system should better not be undertaken. Introducing major changes into the complex patent system without first bringing scientific evidence and stakeholder opinions together may result in controversy that is not productive. The European Commission should take the FTC’s hearing process as an example of an open and timely way to collect data and opinions on major issues concerning the patent system. Studies commissioned by the European Commission but essentially performed by one or very few institutions may be productive elements of such a process, but they cannot be a full substitute for a broadly based academic study (such as the one produced by the National Research Council) in which many researchers and practitioners contribute their views and analyses. With the failure of the CII Directive in Europe, there is a new chance for initiating a policy-making process that can result in major improvements and productive

Patent Quantity and Quality

347

reforms in the European patent system. Since the European system is not in an immediate crisis, Europe can afford to take its time for a thorough discussion. Acknowledgments Parts of the paper were presented at the conference Advancing Knowledge and the Knowledge Economy at the National Academy of Sciences, January 10–11, 2005. I would like to thank the conference participants for constructive comments. I also thank Dominique Guellec and Niels Stevnsborg for many helpful suggestions and discussions. Moreover, I want to acknowledge the help and comments received from Stuart Graham, Bronwyn Hall, David Mowery, and Mike Scherer. All remaining errors are my own responsibility. Notes 1. See ‘‘Mark Twain Matched German’s Invention,’’ New York Times (July 26, 1924), and—on the patent filing and interference—‘‘Patent Files Hold Mark Twain Story—Believed to Be Only One He Swore To as Being True It Is Based on Vest Strap,’’ New York Times (March 12, 1939). From http://www.twainquotes.com/nytindex.html (accessed January 10, 2005). 2. See, for example, Section II in the Executive Summary of the report. 3. See, for example, the statements made by representatives of various corporations during the symposium Ideas into Action: Implementing Reform of the Patent System, transcribed in Berkeley Law Technology Journal 19(3) (2004): 1122–1155. 4. For most of the following figures, data on EPO application and grant numbers are used, but additional evidence from the German Patent Office (DPMA) and other national offices supports the overall assessment. Focusing on EPO figures can be justified on the ground that the EPO has become the most important patent-granting institution in Europe and for European countries. If not otherwise stated, the computations are made on the basis of EPO patent data published in www.epoline.org. Data on the number of claims were supplied by the EPO separately. A comprehensive set of statistics (some of which are presented in this chapter) is available from the author upon request. 5. OECD R&D expenditures are chosen as the comparison index since OECD country R&D composition roughly mimics the applicant composition at the EPO. 6. Author’s computations based on PATDPA data. 7. See Webb et al. (2005), Fig. 8, for more details. 8. Lemley and Allison (2002) point to the growing complexity of U.S. patents. On the EPO’s approach to ‘‘complex applications,’’ see Dack and Cohen (2001). These authors also call for a claims fee for PCT applications. 9. In a recent paper, van Zeebroeck et al. (2005) analyze the increase in the voluminosity of EPO applications, showing that applications have grown in number of pages and claims.

348

Dietmar Harhoff

They test various hypotheses regarding the increase in the number of claims and conclude that strategic motivations as well as an increasing technological complexity account for a large share of the observed increase. 10. For details see Michel and Bettels (2001). 11. A-type references merely characterize the state of the art. Two or more Y-type references—when taken together or with other documents—can also be harmful in the assessment of the patent’s novelty or inventive step. 12. In 1993, the German Patent Court received 44 annulment filings pertaining to EPOgranted patents and 62 filings pertaining to DPMA-granted patents; in 2000, the filings had increased to 117 and 90, respectively. The figures refer to cases resolved by the end of 2003 and could be subject to censoring biases. 13. See Harhoff and Reitzig (2004), footnote 25. 14. For a detailed discussion of the U.S. situation see Eisenberg (2004). The EPO’s ‘‘problem and solution approach’’ toward assessing the inventive step of an application is detailed in Szabo (1995). 15. Internally in the EPO, a decision to grant must be supported by a motivation to be signed by all three members of the respective examining division, but the motivation is nonpublic. A refusal is accompanied by a detailed explanation of the reasons leading to the decision. Since the refusal decision is public and can be appealed, considerable effort goes into the writing of such explanations. Current legal tools also limit the possibilities for refusals. The applicant has ‘‘a right to be heard.’’ In contrast, the JPO system allows for a refusal of the application directly after the first written communication in the examination procedure. 16. Merges (1999) states the same recommendation for the USPTO. 17. See Scherer and Harhoff (2000).

References Allison, J. R., M. A. Lemley, K. A. Moore, and R. D. Trunkey (2003). ‘‘Valuable Patents.’’ George Mason Law & Economics Research Paper No. 03-31; University of California Berkeley Public Law Research Paper No. 133. American Intellectual Property Law Association (2004). ‘‘AIPLA Response to the October 2003 Federal Trade Commission Report: ‘To Promote Innovation: The Proper Balance of Competition and Patent Law and Policy.’ ’’ Washington, DC: Author. Dack, S., and B. Cohen (2001). ‘‘Complex Applications—A Return to First Principles.’’ International Review of Industrial Property and Copyright Law 32 (May): 485–606. Eisenberg, R. (2004). ‘‘Obvious to Whom? Evaluating Inventions from the Perspective of PHOSITA.’’ Berkeley Law Technology Journal 19(3): 885–906. Federal Trade Commission (2003). To Promote Innovation: The Proper Balance of Competition and Patent Law and Policy. Washington, DC: Government Printing Office. Graham, S. J. H., and D. Harhoff (2005). ‘‘Would the U.S. Benefit from Patent Post-Grant Reviews? Evidence from a ‘Twinning’ Study.’’ Paper presented at the NBER Summer Institute (July 19).

Patent Quantity and Quality

349

———, B. H. Hall, D. Harhoff, and David C. Mowery (2003). ‘‘Post-Issue Patent Quality Control: A Comparative Study of US Patent Re-Examinations and European Patent Oppositions.’’ In Patents in the Knowledge-Based Economy, W. M. Cohen and S. A. Merrill, editors, pp. 74–119. Washington, DC: National Academies Press. Hall, B. H. (2005). ‘‘Exploring the Patent Explosion.’’ Journal of Technology Transfer 30(1– 2): 35–48. ———, and D. Harhoff (2004). ‘‘Post Grant Review Systems at the U.S. Patent Office— Design Parameters and Expected Impact.’’ Berkeley Law Technology Journal 19(3), 989– 1016. ———, and R. H. Ziedonis (2001). ‘‘The Patent Paradox Revisited: An Empirical Study of Patenting in the U.S. Semiconductor Industry, 1979–1995.’’ Rand Journal of Economics 32: 101–128. ———, J. H. S. Graham, D. Harhoff, and D. C. Mowery (2003). ‘‘Prospects for Improving U.S. Patent Quality via Post-Grant Opposition.’’ Innovation Policy and the Economy 4: 115–143. Harhoff, D., and M. Reitzig (2004). ‘‘Determinants of Opposition against EPO Patent Grants—The Case of Biotechnology and Pharmaceuticals.’’ International Journal of Industrial Organization 22(4), 443–480. ———, and S. Wagner (2005). ‘‘Modeling the Duration of Patent Examination at the European Patent Office.’’ CEPR Discussion Paper No. 5283, Centre for Economic Policy Research, London. Heller, M. A., and R. S. Eisenberg (1998). ‘‘Can Patents Deter Innovation? The Anticommons in Biomedical Research.’’ Science 698 (May 1). Hunt, R. M. (2001). ‘‘You Can Patent That? Are Patents on Computer Programs and Business Methods Good for the New Economy?’’ Philadelphia Federal Reserve Bank Business Review 2001(Q1): 5–15. Kortum, S., and J. Lerner (1999). ‘‘Stronger Patent Protection or Technological Revolution: What Is behind the Recent Surge in Patenting?’’ NBER Working Paper 6204, National Bureau of Economic Research, Cambridge, MA. Lanjouw, J. O., and M. Schankerman (2001). ‘‘Enforcing Intellectual Property Rights.’’ NBER Working Paper 8656, National Bureau of Economic Research, Cambridge, MA. Lemley, M. A. (2001). ‘‘Rational Ignorance at the Patent Office.’’ Northwestern University Law Review 95(4): 1495–1532. ———, and J. R. Allison (2002). ‘‘The Growing Complexity of the United States Patent System.’’ Boston University Law Review 82: 77. Merges, R. P. (1999). ‘‘As Many as Six Impossible Patents Before Breakfast: Property Rights for Business Concepts and Patent System Reform.’’ Berkeley High Technology Law Journal 14: 577–615. Michel, J., and B. Bettels (2001). ‘‘Patent Citation Analysis—A Closer Look at the Basic Input Data from Patent Research Reports.’’ Scientometrics 51: 181–201. National Research Council, Board on Science, Technology, and Economic Policy (2004). A Patent System for the 21st Century. Washington, DC: National Academies Press.

350

Dietmar Harhoff

Quillen, C. D., O. H. Webster, and R. Eichmann (2002). ‘‘Continuing Patent Applications and Performance of the U.S. Patent and Trademark Office—Extended.’’ The Federal Circuit Bar Journal 12(1): 35–55. Scherer, F. M., and D. Harhoff (2000). ‘‘Policy Implications for a World with SkewDistributed Returns to Innovation.’’ Research Policy 29: 559–566. Somaya, D. (2002). ‘‘Patent Strategy Viewed Through the Lens of Patent Litigation.’’ Unpublished Ph.D. Thesis, Haas School of Business, University of California Berkeley. Szabo, G. S. A. (1995). ‘‘The Problem and Solution Approach in the European Patent Office.’’ International Review of Industrial Property and Copyright Law 26(4): 457–487. U.S. Patent and Trademark Office (2003). 21st Century Strategic Plan. Version of February 3, 2003. http://www.uspto.gov/web/offices/com/strat21/index.htm. van Zeebroeck, N., B. v. Pottelsberghe, and D. Guellec (2005). ‘‘US Contamination or a Trend Toward Complexity: What Is Behind the Surge in EPO Patent Voluminosity?’’ Unpublished manuscript, Solvay Business School, Brussels. Webb, C., H. Dernis, D. Harhoff, and K. Hoisl (forthcoming). ‘‘Analysing European and International Patent Citations—A Set of EPO Patent Database Building Blocks.’’ OECD Working Paper.

20 Blurred Boundaries: Tensions Between Open Scientific Resources and Commercial Exploitation of Knowledge in Biomedical Research Iain M. Cockburn

Introduction Biomedical research drives some of the most visible and significant sectors of the ‘‘knowledge economy.’’ High margin, high growth, high wage, knowledge-intensive industries such as pharmaceuticals, diagnostics, and medical devices are supported by a global biomedical research budget that likely now exceeds $100 billion per year. In pharmaceuticals in particular there have been very handsome social and private returns to R&D and knowledge creation—generous returns to investors have been accompanied by substantial declines in mortality and other health indicators across a wide range of diseases and health problems that correlate with the number of new drugs introduced.1 But the breathtaking scale of these investments (which, after all, have opportunity costs) naturally raises questions about the efficiency with which new biomedical knowledge is created and used. And after decades of building on advances in basic science to create a steady stream of new drugs responsible for remarkable economic and medical gains in the treatment of conditions such as heart disease, stomach ulcers, and depression (and equally remarkable gains for their stockholders), pharmaceutical companies now face a ‘‘productivity crisis.’’ Against a backdrop of rapid advances in the industry’s science base (marked by major scientific achievements such as completing the sequencing of the human genome) as well as in supporting technologies such as instrumentation and computing, the pipeline of new products appears to be shrinking. In 2002 the FDA approved only 17 new molecular entities for sale in the United States—a disappointing fraction of the 15-year high of 56 NMEs approved in 1996 and the lowest since 1983.2 In 2003, the FDA approved 21 NMEs, of which only nine were designated as ‘‘significant improvements’’ over existing drugs. Alarmingly, this decline occurred despite a substantial increase in R&D: Between 1995 and 2002 R&D expenditures by U.S.-based pharmaceutical companies roughly doubled to about $32 billion.3 Similar trends can be seen in worldwide statistics,

352

Iain M. Cockburn

where the annual number of New Active Substances approved in major markets fell by 50% during the 1990s while private sector pharmaceutical R&D expenditures tripled to $47 billion.4 Numbers such as these have prompted headlines in the popular press and in trade journals referring to ‘‘dry,’’ ‘‘weak,’’ or ‘‘strangled’’ pipelines and suggestions that the industry’s historically successful business model is ‘‘broken’’—with dire consequences for investors, who can expect ‘‘permanently lower multiples,’’ and the taxpayers, patients, and insurers who will have to foot an ever-higher bill if they want to maintain the pace of technological progress in the industry. These concerns about productivity are almost surely overblown: If past experience is any guide, the recent surge in R&D spending should generate a commensurate increase in new drug approvals during the next three to ten years.5 Underlying trends in ‘‘true’’ research productivity (in the sense of the relationship between current R&D expenditures and the stream of future benefits attributable to them) are very difficult to measure. The long and complex process of drug development and the significant role of unpriced knowledge spillovers makes it remarkably difficult to unambiguously attribute specific outputs to specific inputs. Today’s new drugs are the result of R&D expenditures stretching back decades into the past and undertaken by many different institutions. Conversely, today’s R&D will likely contribute to output far into the future, both directly in the form of new products and indirectly in the form of more efficient research. Simple comparisons of current output with current inputs are therefore uninformative. But skepticism about what can be inferred from easily observable statistics should not distract from the imperative to understand underlying productivity trends and their sensitivity to policy changes. Given the extraordinary level of resources committed to medical research, ‘‘bang for the buck’’ is a serious concern. Notwithstanding impressive advances on many fronts, technological progress has been disappointing in other areas. No new broad-spectrum antibiotics have been marketed in almost 40 years, and many forms of cancer, as well as chronic diseases and disorders such as diabetes, Alzheimer’s, Parkinson’s, and schizophrenia, still lack effective and well-tolerated treatments.6 Continuing growth in R&D spending represents investment in overcoming these scientific challenges, but this upward trajectory will only be sustainable if it can be paid for, and as increased research spending collides with ever-intensifying pressure to contain health care expenditures, the factors driving the efficiency of the drug discovery and development process are being brought into sharp focus. Chief among these are the institutions governing creation and use of biomedical knowledge— intellectual property rights, channels for knowledge transfer, and processes for allocating resources and rewarding effort in the research enterprise. These institutions

Open Scientific Resources and Commercial Exploitation in Biomedical Research

353

have undergone substantial transformation and realignment in recent decades, but the long-term consequences for system performance of these changes, particularly the blurring of distinctions and boundaries between noncommercial and for-profit research, remain poorly understood. System Performance Versus Component Performance Biomedical research is conducted by a variety of organizations—for-profit companies, nonprofit institutes, government labs, universities, and hospitals—linked together in a complex industry. In thinking about the impact of changes in institutions governing knowledge creation and exchange on social returns to investment in biomedical R&D, it can be helpful to draw a distinction between system performance and component performance—that is, between the efficiency or productivity of specific entities and the efficiency of interactions among them. In general, the productivity of any organization (whether it be a university lab or a drug company) will be driven by factors such as the quality of inputs to production and the nature of the production activity it is engaged in, as well as managerial factors such as the types of incentives used to motivate its employees and the processes and organizational structure used to allocate resources. In the case of commercial pharmaceutical research, these factors are reasonably well understood. For drug companies, output of new drugs is a function of ‘‘shots on goal,’’ i.e., the number of lead compounds generated or acquired, and the probability of them making it through preclinical and clinical development phases. Studies have shown that, at least in the 1980s, the efficiency of this process was related to the size and diversity of the company’s research effort, its reward systems, and the nature of internal decision-making and distribution of authority.7 Less is known about the factors driving the productivity of academic or government research.8 For the industry as a whole, however, productivity is a function of both the efficiency of its component institutions and of the industry structure—that is to say, the numbers and types of institutions, the allocation of effort among them, and the nature of relationships between them. During the past 30 years the pharmaceutical industry has seen some profound structural changes that are tightly linked to evolving institutions for creating, managing, and exchanging knowledge. These changes have important implications for system performance. The Changing Structure of the Pharmaceutical Industry The postwar evolution of the pharmaceutical industry can be characterized as a process of progressive vertical dis-integration and growing complexity.9

354

Iain M. Cockburn

In the 1960s and 1970s, the industry could be seen as having a fairly simple binary structure with a clear division of effort between upstream not-for-profit institutions, which did curiosity-driven basic research, and downstream for-profit companies that did market-oriented applied research. In the for-profit sector, almost all firms were large and fully integrated, from drug discovery, through clinical development, regulatory affairs, manufacturing, and marketing. Most commercial drug discovery activity was conducted in-house and at least in the early part of this period was dominated by large scale ‘‘random screening’’ programs with limited requirements for deep knowledge about fundamental physiological processes at the molecular level. Licensing activity was driven largely by downstream concerns: rights to sell drugs that were already approved (or were in the late stages of clinical development) would be acquired in order to maintain efficient levels of utilization of manufacturing or marketing assets or, in the international context, to take advantage of local knowledge and access to regulators and distribution channels. Upstream technology was largely acquired either ‘‘for free’’ by reading journals and attending conferences or by purchasing tangible inputs and services, such as instruments or highly skilled graduates. In this industry structure, pharmaceutical firms appropriated returns from R&D through a combination of extensive patenting of production processes and end products, proprietary know-how, brands, regulatory barriers to entry, and favorable product market conditions. Most of these firms were long lived, mature organizations, tracing their roots back many decades, often to the 19th century chemical industry. Their large and sustained investments in R&D, marketing assets, and human and organizational capital were largely financed from internal cash flow. Competitive advantage was driven by firms’ ability to effectively manage product market interactions with regulators and end users, and to ‘‘fill the pipeline’’ with a steady succession of internally developed blockbuster drugs. The productivity of R&D performed by these firms appears to have been driven to a great extent by economies of scale and scope in conducting research, efficient allocation of resources in internal capital markets, and the ability to capture internally and externally generated knowledge spillovers. In the upstream not-for-profit sector, taxpayers (and to some extent philanthropists) supported curiosity-driven research conducted at cottage industry scale inside government labs, universities, research institutes, and teaching hospitals. Legal constraints and a strong set of social norms limited commercial or contractual contacts between the world of open science and pharmaceutical firms in important ways. Resource allocation in the not-for-profit sector was driven by peer-reviewed competition for grants on the basis of scientific merit and the reputation of individual researchers. The importance of establishing priority and reputation drove early and

Open Scientific Resources and Commercial Exploitation in Biomedical Research

355

extensive publication of results, and social norms (and requirements of granting agencies) promoted routine sharing of research materials. Not-for-profit researchers concentrated largely on fundamental science and filed very few patents. This is, of course, a gross oversimplification. Many drug companies invested significant resources in ‘‘blue sky’’ basic research, and specialist for-profit research boutiques generated and sold technology to large firms. Public sector institutions conducted screening programs for drug candidates, and many academic researchers had close financial and contractual links with drug companies through individual consulting arrangements and institutional research grants and contracts.10 Funding priorities reflected political pressure, intellectual fashions, and the dynamics of the Matthew Effect,11 as well as pure scientific merit. Importantly, the ‘‘waterfall’’ model of vertical knowledge spillovers, with a one-way flow of ideas and information down a gradient running from upstream basic science to downstream applied research and clinical practice, appears to have been only partially true. Nobelwinning work in basic science was done in for-profit labs, and non-profit institutions were an important source of data, techniques, and expertise in late-stage drug development, epidemiology, and postmarketing follow-up. Clear institutional boundaries between academic and commercial science did not prevent significant movement of ideas, candidate molecules, research materials, research results, and individuals back and forth across the for-profit/not-for-profit divide. Notwithstanding these caveats, it is still possible to summarize the vertical structure of the industry in this era as being essentially binary, with a clear distinction drawn between upstream open science and a downstream commercial sector dominated by large, highly integrated firms. Since the early 1980s, industry structure has become considerably more complex. After decades of stability and consolidation, in the late 1970s the for-profit side of the industry began to experience significant entry as an intermediate sector emerged between academic research institutions and Big Pharma. By the mid 1990s several thousand biotechnology ventures had been launched, and several hundred had survived to reach sufficient scale to be an important force in the industry. Existing vertical relationships were disrupted and reformed, with consequences that are still far from clear. These new companies straddled the historical divide between for-profit and not-for-profit research. Though they were, for the most part, overtly profit-oriented, they also had much tighter and more explicit links to non-profit research institutions, with close personal, geographical, cultural, and contractual ties to universities, research institutes, and government labs. Academic scientists played a particularly significant role in the founding of these companies, either moving out of academic employment or participating actively in both worlds.12 Many of the smaller pharmaceutical firms have disappeared as leading players have merged and consolidated, and worldwide research activity has gravitated

356

Iain M. Cockburn

toward a handful of locations.13 Relationships between the nonprofit and forprofit sectors of the industry have changed dramatically, and a new class of competitors—the biotechnology companies—has entered the industry at the interface between academic and commercial research. Some ‘‘product’’ biotechnology companies have entered the industry as direct horizontal competitors to established firms, intending to realize profits by using their command of new techniques and insights from molecular biology to develop products that will be sold to end users. Other ‘‘tool’’ companies have inserted themselves into the industry value chain at the interface between academic research and the downstream for-profit pharmaceutical firms, with a business model based on licensing or selling leading-edge knowledge, research tools, or intellectual property to companies focused on less scienceintensive clinical development, manufacturing, and marketing. By taking over a certain amount of research activity from both upstream and downstream entities, these new entrants have forced some important adjustments in university–industry relations and ushered in a new ‘‘partnering’’ mode of research. Large incumbent firms with marketing, manufacturing, regulatory affairs, and clinical development capabilities now rely heavily on research tools and candidate molecules acquired from upstream sources through complex contracts and collaborative agreements. Between 25% and 40% of Big Pharma’s sales are now reported to come from drugs originated in the biotech sector.14 Factors Driving Structural Change This vertical dis-integration appears to have been driven by a number of interlinked economic and legal forces. Perhaps the most salient of these are the developments in law and administrative practice that have brought much of molecular biology and the life sciences within the ambit of the patent system. Patents are now routinely awarded on fundamental scientific knowledge such as genetic sequence information, cell receptors, and fundamental metabolic pathways. This extension of exclusionbased intellectual property into the domain of basic science means that marketbased competition based on proprietary rights over biomedical knowledge now plays a very significant role in determining the overall rate and direction of technological progress. Pharmaceutical and biotechnology companies have become important participants in basic biomedical research while, in parallel, universities and other non-profit entities have become enthusiastic participants in the patent system. Interestingly, at the same time that exclusionary property rights have become a significant feature of basic research, aspects of classic ‘‘Mertonian’’ rules and norms governing production and exchange of knowledge in ‘‘open science’’ have diffused into commercial research. Many commercial entities increasingly manage internal

Open Scientific Resources and Commercial Exploitation in Biomedical Research

357

and external production and exchange of knowledge in ways that closely resemble academic research, emphasizing collaboration, interaction, peer review, and publication.15 And as biology has become increasingly focused on computational methods and digital data, the anti-exclusionary mechanisms of open source software development are playing an increasingly important role in the development of databases and software tools used in bioinformatics. This ‘‘creeping propertization’’ of basic biomedical research is not the only way in which boundaries between for-profit commercial research and academic science have been breached and blurred. A number of other legal and economic changes have played an important role, particularly the passage of the Bayh–Dole Act, the Stevenson–Wydler Act, and other legislation enabling and encouraging commercialization of publicly funded research,16 together with the rise of a venture capital industry (and ultimately a stock market) that was (periodically) willing to provide substantial amounts of capital to inexperienced science-based companies with limited prospects of short-term profitability and enormous unresolved technology risk. Venture funding of biotechnology is closely associated with general increases in the supply of venture capital as a result of the relaxation in 1979 of the ‘‘prudent man rule’’ governing pension fund investment decisions, although other developments in the capital markets have contributed to the rise of the biotechnology sector. New financial technologies have been developed for pricing and managing risk, and at least in the United States, there appears to have been a significant increase in investors’ tolerance for risk, as evidenced by the falling equity premium imputable from stock market returns. Equally significant, however, are the organizational and managerial impacts of the changes in the technology of pharmaceutical research that arose from the revolution in life sciences. One important factor was the rapid increase in the cost and scale of basic research projects. Another was that drug discovery became progressively more science-intensive, with increased emphasis on understanding and exploiting deep understanding of physiology and disease states at the molecular level. As ‘‘rational drug design’’ took center stage in the late 1980s, changes in the nature of research activity were accompanied by complementary changes in the internal structure and incentives of commercial R&D organizations. Drug companies began to look more like universities and behave more like universities, with increasing emphasis on publication and individual collaboration across institutions.17 These changes in business practice were accompanied by increased willingness to consider acquiring external sources of technology in the form of research projects conducted as joint ventures or strategic partnerships. Thus, an environment was created in which specialist research firms could expect if not to prosper, at least to survive. At the same time the growing costs and complexity of academic research

358

Iain M. Cockburn

projects forced successful scientists to acquire managerial and organizational skills—making them better equipped and more favorably disposed toward business ventures and looking much more like entrepreneurs and managers to outside investors or business partners. As ever-increasing resource requirements and growing societal pressure to justify their budgets pushed universities and other governmentfunded institutions to become more tolerant of ‘‘just-off-campus’’ commercial activity, or even to actively encourage it, this rising cadre of scientist–entrepreneurs were well positioned to take advantage of the opportunities created. Consequences for Industry Research Performance The implications of this new industry structure for long-term research performance are far from clear. Standard economic analysis holds that strong property rights, competition, and the profit motive tend to result in socially optimal allocation of resources. To the extent that vertical dis-integration of pharmaceutical research promotes specialization, competition, and risk-taking, and substitutes market signals for bureaucratic allocation of research funds, there may therefore be very large gains in efficiency. On the other hand, the nature of the research process—and particularly the central role played by creation and exchange of scientific knowledge in the economics of the industry—provides less cause for optimism. Arguments in favor of specialization and market exchange presume a world with perfect information, competitive markets, and no transactions costs. Stepping away from this benchmark and focusing for the moment on commercial knowledge production, it has long been clear that large, vertically integrated firms are an efficient response to a number of real world problems. These include limited ability to diversify risk where capital markets are incomplete or imperfect, the presence of transactions costs when complete contracts cannot be written, problems in capturing spillovers or other externalities, and a variety of familiar difficulties that arise from flaws in markets for knowledge. In fact, there is a strong presumption that vertical integration is the first best solution to economic problems such as those encountered during commercial drug discovery and development, i.e., financing and managing multiple projects that are long-term, risky, complex, costly to monitor, require substantial project-specific unrecoverable investments, and have shared costs and vertically complementary outcomes.18 Here, problems with transactions costs, pricing, and access to information are minimized by internalizing decisions within the firm and allocating resources through an internal capital market. Under the old bipartite industry structure, therefore, research performance reflected a world in which most exchanges of scientific knowledge were not explicitly priced, and patents excluded industry participants only from the final product

Open Scientific Resources and Commercial Exploitation in Biomedical Research

359

market. In sharp contrast, in today’s industry exchange, access, and use of knowledge are governed by an active market for licenses and partnership deals. Prices in this market play an important role in allocation of resources in commercial research, and system performance thus relies critically on the market for upstream research generating the ‘‘right’’ signals for downstream resource allocation and for further investment in upstream knowledge creation. ‘‘Transactional optimists’’ believe that this market works well, arguing that potential distortions arising from informational asymmetries, thin markets, bargaining problems, and other sources of market failure can be minimized by creative use of contractual provisions in license agreements and partnership deals.19 Markets for knowledge are, however, notoriously inefficient due to the unique properties of knowledge as an economic good, and in the context of vertical agreements in biomedical research there are particularly good grounds for skepticism about the ability of these markets to ‘‘get the prices right.’’ Consider the stylized case of a small biotech company that holds a valid and enforceable patent on a gene coding for a target, whose claims will be infringed by any attempt by a downstream pharmaceutical company to develop a marketable drug. The pharmaceutical company, in turn, blocks the biotech company’s access to the end user with its own product or use patents. The two parties are clearly better off if they can agree on a license or partnership deal that divides profits between them. Bargaining is likely to be easy and efficient when both participants can agree on the payoff, neither has an informational advantage, and both are equally risk averse. However, in this context these assumptions are surely violated, and it is quite likely that the two firms will find it hard to agree. Experience suggests that the biotech company will tend to have overinflated expectations of the value it brings to the table, while the pharmaceutical company will be in a stronger bargaining position given its greater size, wider range of other opportunities, and potentially a credible threat to invent around the biotech company’s patent—or litigate it to death. Both sides will likely have plenty of private information (the pharmaceutical company will be better informed about market prospects and product development risks, while the tool company will be better informed about its technology) and incentives to act opportunistically on that information, raising the costs of drawing up a contract or inducing the parties to make defensive investments. To cap it all, imperfect capital markets mean that biotech company will not infrequently be facing a very real threat of bankruptcy. Outside investors’ interest in biotechnology periodically waxes and wanes, and when the ‘‘funding window’’ is closed, cash-poor companies are easily pressured into entering an agreement on adverse terms: a low fixed fee rather than a high reach-through royalty rate, plus exclusivity provisions that limit its ability to sell its technology elsewhere or exploit it

360

Iain M. Cockburn

through internal development. Add a little more realism to this picture by introducing the costs of coordinating contracts with multiple upstream technology vendors, potential anti-commons problems created by overlapping rights, and uncertainty about the ultimate validity and enforceability of broadly written patents and it becomes increasingly difficult to be optimistic about efficient outcomes being reached in licensing negotiations. There are, of course, a number of arguments in favor of vertical specialization supported by strong, broad patents on upstream basic technology. First, basic technologies tend to have broad applicability, often in ways that are very difficult to anticipate. To the extent that markets for upstream stimulate development of commercially relevant tools, and competition forces down their prices, there may be faster and more widespread impacts on downstream product development. Development of tool technologies in secret is undoubtedly socially costly, and therefore the prompt disclosure of early stage tools or platform technologies in patent applications may also promote knowledge spillovers and raise social returns. Second, relying on incumbent firms to develop tools may result in delayed development. Incumbents may have incentives to slow down technology development to avoid cannibalizing existing products. They may also shelve or abandon new technologies that threaten other sources of quasi-rents. Limiting proprietary rights in early stage technologies can reinforce the competitive position of incumbents. The ‘‘Strategy of the Commons’’ argument suggests, for example, that incumbent firms can deter entry into their markets by putting new technology in the public domain.20 Entrants are thus denied the opportunity to establish patent rights, sharply limiting their ability to raise capital and establish a proprietary market position. The SNP Consortium has been suggested as an example of this dynamic in action. (An interesting variant of this strategy is to sponsor university research, but only on condition that it be licensed nonexclusively.) Third, while large, vertically integrated firms minimize some costs, they may also raise others. Gains from integration come at the cost of creating internal bureaucracies to coordinate and control activity. These systems are costly to maintain and may cause rigidity, organizational ‘‘slack,’’ and a bias toward conservative decisions—limiting the ability of these firms to respond to new technological opportunities. It is widely believed that new enterprises are faster at recognizing and developing new technologies, and they may also enjoy cost advantages in doing research arising from specialization, flexibility, and ‘‘focus.’’ Fourth, the prospect of obtaining broad patent rights in early stage technologies may stimulate socially valuable investment in R&D—and further rapid innovation as second movers invent around the first round of patents on a new technology. Models of sequential innovation highlight the importance of balancing the division

Open Scientific Resources and Commercial Exploitation in Biomedical Research

361

of rents between first movers and second movers for equilibrium levels of R&D, and reluctance to grant patent rights to early innovators may therefore have deleterious effects. Last, though the ‘‘gold rush’’ and ‘‘land grab’’ metaphors commonly employed to describe upstream patenting raise the specter of socially wasteful rent dissipation, such ‘‘racing’’ behavior may also have beneficial effects. Competitive races finish faster. Falling behind in a protracted race may cause weak competitors to drop out, weeding out bad ideas or poorly conceived enterprises. Indeed, some game theoretic modeling of technology races suggests that in some circumstances social surplus can be raised by awarding patents early rather than late in the development of a technology.21 To summarize, vertically dis-aggregated industries are not necessarily inefficient, and specialized research firms can play an important role in the right circumstances.22 One can be optimistic about efficiency being raised by increased vertical specialization in industries where horizontal intrasegment competition is high, where specialization reduces costs, where vertical coordination is relatively unimportant, where prices reached in the market for the upstream technology accurately reflect marginal opportunity costs, and where bargaining and contracting are easy and effective. Unfortunately, it is far from clear that these conditions prevail in biomedical research. High levels of uncertainty and high transactions costs imply serious contracting problems. Horizontal competition in specific areas of technology is often limited, and price signals from end users are muted at best. The considerations all suggest limited economic gains from vertically dis-integrating the industry, and if this is indeed the case then further vertical restructuring induced by regulatory or technological change may have adverse effects on social welfare. ‘‘More and stronger patents’’ could make things worse if they induce excess entry upstream, exacerbate contracting problems, or strike the wrong balance between incentives for pioneers and subsequent innovators.23 Anecdotal evidence and the relatively low stock market returns to biotech tool companies support this pessimistic view. For example, the apparently broad claims of patents on DNA sequences have not yet translated into the ability to extract a significant share of the rents accruing to downstream incumbents.24 In part this reflects the superior bargaining position of the downstream firms, which have largely been able to dictate contractual terms to tool companies. But it also reflects what Richard Nelson called ‘‘the simple economics of basic scientific research’’— patents or no patents, capturing the value that ultimately derives from fundamental early stage research is extraordinarily difficult for profit-oriented organizations. Those firms that succeeded in doing this have, historically, been large, stable, highly

362

Iain M. Cockburn

integrated firms, sufficiently diversified in product markets to capture spillovers and financially strong enough to be able to effectively manage risk internally. The ‘‘pure play’’ biotech tool companies seem unlikely to replicate the success of product winners such as Amgen. Falling stock market valuations may reflect a realization by investors that large portfolios of gene patents are unlikely to confer significant access to blockbuster downstream revenues. In fact, licensing revenues may for the most part be confined to one-time payments or periodic user fees, with any royalties eventually realized from sales of downstream products shared with other tool providers. Many tool companies have therefore changed their business strategies. Some have switched to emphasizing product development, while others have moved toward much closer relationships with downstream firms, emphasizing long-term mutual interests, proprietary nondisclosed information, and close coordination, i.e., a ‘‘quasi-integration solution.’’ Tools (and associated patents) that the passage of time reveals to be truly valuable are likely to be acquired by downstream firms— potentially raising fresh issues in the antitrust area about vertical foreclosure. One thing that upstream patents on basic research do seem to have done effectively is the creation of powerful incentives for new entrepreneurial companies to enter the pharmaceutical industry as vertical competitors against the established firms. But it is far from clear that these new entrants have, on net, increased value creation in the industry. In one area—gene sequencing and genomics—the new entrants do appear to have dramatically reduced the costs of finding (and then using) biologically significant sequence information. Competitive pressure appears to have rapidly pushed down the cost of gene sequencing and to have brought the global effort to sequence the human genome to completion much faster. The effort induced by incentives to search for patentable DNA sequences may also have had the benefit of generating spillovers to other technologies. But these achievements must be set against the costs of racing behavior, whether they be socially wasteful duplicative effort or simply the opportunity cost of employing extra resources to finish faster. Other than inducing potentially inefficient levels of entry and investment into the tool sector, the impact of gene patents, at least in the medium term, may be quite small. On the positive side they prompted voluminous disclosure of fundamentally important information—though to some extent this information was being created and published elsewhere. On the negative side, in some highly publicized cases gene patents are apparently being used in ways that limit nonprofit research activity or otherwise raise the costs of doing research.25 Relatively low marginal costs of generating some types of applications for gene patents also appears to have had adverse consequences: Early in the gene patent ‘‘gold rush’’ the Patent Office was flooded with ultimately fruitless applications on ESTs, straining its resources and likely low-

Open Scientific Resources and Commercial Exploitation in Biomedical Research

363

ering the quality of examination. Anecdotal reports suggesting that some genomics companies have had ‘‘more than 60,000’’ applications pending do nothing to assuage these concerns: Though increased stringency of examination may have resulted in some of these applications being abandoned or consolidated, given the very long pendency period for complex molecular biology patents many of these may still be in the pipeline. Impact on Academic Science Aside from any impact on the productivity of commercial biomedical science, the extension of exclusionary property rights into basic biomedical research also has the potential to weaken academic research, a vital but fragile component of the biomedical innovation system. Historically, academic research has been driven by social norms and resource allocation procedures that largely ignored market signals and commercial concerns. Patents and the profit motive are largely antithetical to the governance mechanisms of publicly funded science, and their steady perfusion into academic institutions has generated considerable alarm.26 Open science has relied heavily on priority and reputation-based incentives, investigator-initiated research, peer review, and a ‘‘gift economy’’ of prompt reciprocal sharing of data, materials, and results. These mechanisms may be very difficult to sustain in the face of increasing competition from commercial entities for resources and talented scientists, and the proliferation of patents and proprietary data. Decreased information sharing, increased emphasis on product market potential over scientific merit in funding decisions and agenda-setting, and the corruption of the ‘‘truth-finding’’ mechanisms of scientific communities surely have serious consequences for the future vitality and productivity of fundamental science and for the academic community’s contributions to nonmarket social goals. Evidence on these issues is mixed. Universities have become active participants in patenting discoveries in the life sciences but have begun to experience a growing ‘‘push back’’ from industry in the form of challenges to patents asserted by universities.27 Major funders of biomedical research have become more insistent that licensing deals made by universities be unrestrictive and focus on public benefits. Some surveys suggest important changes in the behavior of individual academic researchers in biomedical disciplines.28 However, studies of patenting and publishing behavior have typically found that participation in patenting or in startup companies is a complement rather than a substitute for publication,29 and compelling evidence for a large ‘‘choking’’ effect of patents on academic research or of any significant swing away from basic science toward commercial applications has yet to emerge.30

364

Iain M. Cockburn

Though there is little quantitative evidence thus far of a negative impact of patents on scientific research activity, their qualitative impact on the norms of scientific inquiry and on institutional culture may ultimately prove to be very significant. Unfortunately, these are particularly difficult to observe directly, and drawing conclusions about the incidence of scientific fraud or the influence of commercial considerations in promotion decisions from the few cases reported in the media is obviously very hazardous. Nonetheless, many observers remain deeply concerned about the impact of expanding exclusionary intellectual property rights into the domain of academic research.31 But in at least one important area of biomedical research, the burgeoning new discipline of computational biology, or bioinformatics, open science appears to be alive and well. Here, academic researchers appear (thus far) to have effectively limited the incursion of exclusionary IP through aggressive use of the public domain and open source licensing.32 In silico biology relies on software algorithms, huge collections of digital data on genetic sequences, molecular structures and disease epidemiology, and interfaces and linkages among them. Situated at the interface of molecular biology and software—two of the most troublesome and controversial areas of IP law and practice—the potential for poor outcomes from widespread acquisition and assertion of exclusionary rights in these types of knowledge would appear to be very high. Yet, with the conspicuous exception of DNA arrays and other hardware technologies, there has been relatively little patenting of bioinformatics. Limited patenting has been accompanied by very few legal disputes and a conspicuous absence of outrage in the trade press over IP issues. One reason for this may be that there are large costs to all participants in bioinformatics from fragmentation of data sources and restrictions on access—here, the value of the whole is clearly much greater than the sum of its parts. But it also seems clear that lessons learned from the struggle over the human genome sequence have been effectively applied by public sector researchers. Important software tools such as ENSEMBL or BLAST are either public domain or ‘‘copylefted’’ and constitute an important source of prior art against attempts to obtain patents on fundamental algorithms and data structures. New large-scale data gathering initiatives such as the International HapMap Project have also at times used ‘‘click-wrap’’ licenses to enforce open access policies or even GPL-type requirements for users to make their improvements or additions to the database available to the community of users. These developments suggest an expanded role for collaborative, open, and inclusive structures governing creation and access to biomedical knowledge in the future. But the long-term viability of such structures is questionable, and much work needs to be done to develop robust legal frameworks and business models that can sup-

Open Scientific Resources and Commercial Exploitation in Biomedical Research

365

port the investments required to bring the results of this research to market. It is also important to recognize that limiting patenting may encourage data sharing and collaboration, but at a cost. Patents force disclosure, and in bioinformatics, for example, vigorous extension of the public domain may have shifted some commercial actors toward greater use of trade secrets, with important tools and data hidden from sight and priced beyond the reach of most academic users. Agreements to act collectively are also very vulnerable to defection and opportunistic behavior, as has been seen in software and communications industries, where consortia convened to facilitate coordination through common standards are under constant threat of being ‘‘hijacked’’ by unanticipated patents. Conclusions The promise of biomedical research to relieve human suffering and create wealth has never been higher. But the ability of the system to deliver on this promise depends critically on its ability to efficiently create, manage, and exchange knowledge. The patent system is perhaps the most important piece of institutional infrastructure that enables these activities, and the evolution of patent law has played a very significant role in restructuring the pharmaceutical industry. The extension of exclusionary intellectual property rights into basic research has unleashed a surge of entrepreneurial energy and risk taking in commercial science, with potentially very significant benefits to society once the technology reaches end users. But these benefits carry with them substantial costs: The patent-driven vertical struggle for rents within the biomedical innovation system may have generated important inefficiencies, waste, and misallocation of resources, and drawing universities more deeply into the patent system may prove costly in the long run. Arguably, some reassessment of the appropriate domain of patents is in order. Restrictions on access to research tools and data are likely to prove very costly in the long run, and stronger protection of the public domain may be a prerequisite for the future health of basic biomedical science. Reforms to patent law and practice suggested by the FTC and the National Academies may go some way toward limiting, if not reversing, decades of patent ‘‘creep’’ into the process of scientific discovery. Statutory protection for research may be necessary, along with more weight given to the implications of university patenting for the conduct of science. The experience of other industries suggests a larger role in biomedical research for collaborative precompetitive research, as well as new mechanisms such as open source development for coordinating and rewarding effort within large-scale research projects. Developing such institutions for the unique technological and economic environment of biomedical research presents an interesting challenge.

366

Iain M. Cockburn

Acknowledgments This essay relies heavily on two earlier papers, ‘‘The Changing Structure of the Pharmaceutical Industry’’ and ‘‘State Street Meets the Human Genome Project,’’ and notes for a lecture given at the 4th EPIP Conference. I am grateful to conference participants and referees of these earlier papers for helpful comments. Notes 1. F. Lichtenberg, ‘‘The Impact of New Drug Launches on Longevity: Evidence from Longitudinal, Disease-Level Data from 52 Countries, 1982–2001,’’ NBER Working Paper No. 9754, 2003. 2. FDA CDER website, http://www.fda.gov/cder/rdmt/pstable.htm. 3. PhRMA, ‘‘Pharmaceutical Industry Profile, 2002.’’ 4. EFPIA, ‘‘The Pharmaceutical Industry in Figures, 2003 Update.’’ 5. Record numbers of new drug candidates have entered the pipeline in recent years, with more than 3200 in the period 2001–2003 alone. (PJB Publications, ‘‘Pharmaprojects Annual Review,’’ May 2003.) 6. Shamefully, very little research has been directed toward tropical diseases such as malaria whose burden falls almost entirely on the populations of the world’s poorest countries. See J. Lanjouw and I. Cockburn, ‘‘New Pills for Poor People? Empirical Evidence after GATT,’’ World Development 29, no. 2 (2001): 265–289. 7. R. Henderson and I. Cockburn, ‘‘Measuring Competence: Exploring Firm Effects in Pharmaceutical Research,’’ Strategic Management Journal 15 (1994): 63–84; R. Henderson and I. Cockburn, ‘‘Scale, Scope, and Spillovers: Determinants of Research Productivity in the Pharmaceutical Industry,’’ RAND Journal of Economics 27, no. 1 (1996): 32–59. 8. Although see, for example, A. Arora, P. David, and A. Gambardella, ‘‘Returns to Scientific Reputation: Funding of Research Projects by the Italian CNR,’’ Annales des Economie et des Statistiques, No. 49/50 (1998): 164–198; P. D. Allison and J. S. Long, ‘‘Departmental Effects on Scientific Productivity,’’ American Sociological Review 55 (1990): 469–478; A. Geuna, The Economics of Knowledge Production: Funding and the Structure of University Research. Cheltenham: Edward Elgar, 1999; S. F. Breschi, F. Lissoni, and F. Montobbio, ‘‘The Scientific Productivity of Academic Inventors: New Evidence from Italian Data,’’ Economics of Innovation and New Technology (in press 2005). 9. See, for example, A. Gambardella, Science and Innovation: The U.S. Pharmaceutical Industry During the 1980s (Cambridge: Cambridge University Press, 1995); I. Cockburn, R. Henderson, L. Orsenigo, and G. Pisano, ‘‘Pharmaceuticals and Biotechnology,’’ in U.S. Industry in 2000: Studies in Competitive Performance, ed. D. Mowery (Washington, DC: National Research Council, 1999), pp. 363–398. 10. These ties have a long history in the pharmaceutical industry; see MacGarvie, M., and J. Furman, ‘‘Early Academic Science and the Birth of Industrial Research Laboratories in the U.S. Pharmaceutical Industry,’’ mimeo, Boston University School of Management, 2005. 11. ‘‘Unto every one that hath shall be given, and he shall have abundance.’’ Matthew 25: 29.

Open Scientific Resources and Commercial Exploitation in Biomedical Research

367

12. L. Zucker, M. Darby, and M. Brewer, ‘‘Intellectual Human Capital and the Birth of U.S. Biotechnology Enterprises,’’ American Economic Review 88, no. 1 (March 1998): 290– 306. 13. See J. Furman, M. Kyle, I. Cockburn, and R. Henderson, ‘‘Knowledge Spillovers, Geographic Location, and the Productivity of Pharmaceutical Research,’’ Annales d’Economie et de Statistique (in press 2005). 14. Source: CMR International, cited in ‘‘The Pharmaceutical Industry in Figures’’ (Brussels: EFPIA, 2000). 15. I. Cockburn, R. Henderson, and S. Stern, ‘‘Balancing Incentives: The Tension Between Basic and Applied Research,’’ NBER Working Paper No. 6882, January 1999; I. Cockburn, R. Henderson, and S. Stern, ‘‘The Diffusion of Science-Driven Drug Discovery: Organizational Change in Pharmaceutical Research,’’ NBER Working Paper No. 7359, September 1999. 16. D. Mowery, B. Nelson, B. Sampat, and A. Ziedonis, ‘‘The Growth of Patenting and Licensing by U.S. Universities: An Assessment of the Effects of the Bayh–Dole Act,’’ Research Policy 30 (2001): 99–119. 17. I. Cockburn and R. Henderson, ‘‘Absorptive Capacity, Coauthoring Behavior, and the Organization of Research in Drug Discovery,’’ Journal of Industrial Economics 46, no. 2 (1998): 157–182. 18. See P. Milgrom and J. Roberts, Economics, Organization & Management (Englewood Cliffs, NJ: Prentice Hall, 1992). 19. A. Arora, A. Fosfuri, and A. Gambardella, Markets for Technology: Economics of Innovation and Corporate Strategy (Cambridge, MA: MIT Press, 2001); J. Gans and S. Stern, ‘‘Incumbency and R&D Incentives: Licensing the Gale of Creative Destruction,’’ Journal of Economics and Management Strategy 9 (2000): 485–511. 20. A. Agrawal and L. Garlappi, ‘‘Public Sector Science and the Strategy of the Commons (Abridged),’’ Best Paper Proceedings, Academy of Management (2002). 21. J. Judd, K. Schmedders, and S. Yeltekin, ‘‘Optimal Rules for Patent Races,’’ Northwestern University, Center for Mathematical Studies in Economics and Management Science, Discussion Paper No. 1343, April 2002. Empirical evidence for racing behavior is thin. For the case of pharmaceutical R&D see I. Cockburn and R. Henderson, ‘‘Racing to Invest? The Dynamics of Competition in Ethical Drug Discovery,’’ Journal of Economics and Management Strategy 2, no. 3 (1994): 481–519. 22. See, for example, the case of specialist engineering firms in the chemicals industry, as documented in A. Arora, and A. Gambardella, ‘‘Evolution of Industry Structure in the Chemical Industry,’’ in Chemicals and Long-Term Economic Growth, ed. A. Arora, R. Landau, and N. Rosenberg (New York: Wiley, 1998). 23. S. Scotchmer, ‘‘Standing on the Shoulders of Giants: Cumulative Research and the Patent Law,’’ Journal of Economic Perspectives 5, no. 1 (1991): 29–41. 24. T. Gura, ‘‘After the Gold Rush: Genome Firms Reinvent Themselves,’’ Science 383 (2001): 1982–1984. 25. Myriad Genetics’ exclusive licensing of the BRCA1 gene is often claimed to restrict academic research. See A. Schissel, J. Merz, and M. Cho, ‘‘Survey Confirms Fears about Licensing of Genetic Tests,’’ Nature 402 (1999): 118.

368

Iain M. Cockburn

26. See Keeping Science Open: The Effects of Intellectual Property Policy on the Conduct of Science (London: The Royal Society, 2003); S. Krimsky, Science in the Private Interest: Has the Lure of Profits Corrupted Biomedical Research (Lanham, MD: Rowan and Littlefield, 2003); R. Eisenberg, ‘‘Property Rights and the Norms of Science in Biotechnology Research,’’ Yale Law Journal 97 (1987): 177–223; A. Rai, ‘‘Regulating Scientific Research: Intellectual Property Rights and the Norms of Science.’’ Northwestern University Law Review, 77 (1999): 94–129. Interesting counterargument given by S. Kieff, ‘‘Facilitating Scientific Research: Intellectual Property Rights and the Norms of Science—A Response to Rai and Eisenberg,’’ Northwestern University Law Review 95 (2000): 691. 27. ‘‘Depth Charges Aimed at Columbia’s Submarine,’’ Science 301 (2003): 448. ‘‘Judge Turns Rochester’s Golden Patent into Lead,’’ Science 299 (2003): 39. 28. D. Blumenthal et al., ‘‘Participation of Life-Science Faculty in Research Relationships with Industry,’’ New England Journal of Medicine 335 (1996): 1734–1739, E. Campbell et al., ‘‘Data Withholding in Academic Genetics: Evidence from a National Survey,’’ JAMA 287 (2002): 473–480. 29. See P. Stephan et al., ‘‘Who’s Patenting in the University? Evidence from a Survey of Doctorate Recipients,’’ mimeo, Georgia State University, 2004; G. Thursby and M. Thursby, ‘‘Patterns of Research and Licensing Activity of Science and Engineering Faculty,’’ mimeo, Georgia Tech, 2003; P. Azoulay, W. Ding, and T. Stuart, ‘‘The Determinants of Faculty Patenting Behavior: Demographics or Opportunities?,’’ mimeo, Columbia University 2005; K. Markiewicz and A. DiMinin, ‘‘Commercializing the Laboratory,’’ mimeo, Boston University, 2004. 30. A recent survey of life scientists found little evidence that patents on research tools were hindering academic research: see J. Walsh, A. Arora, and W. Cohen, ‘‘Research Tool Patenting and Licensing and Biomedical Innovation,’’ in Patents in the Knowledge-Based Economy, ed. W. Cohen and S. Merrill (Washington, DC: National Academies Press, 2004). On the other hand, in one interesting study patents have been shown to negatively affect access to knowledge, as measured by citations. See Murray, F., and S. Stern, ‘‘Do Formal Intellectual Property Rights Hinder the Free Flow of Scientific Knowledge? An Empirical Test of the Anti-Commons Hypothesis.’’ Mimeo, MIT, March 2005. 31. David, P., ‘‘Can ‘Open Science’ Be Protected from the Evolving Regime of IPR Protections?’’ Stanford Department of Economics, Working Paper #03-011, 2003. ‘‘Is the University–Industry Complex Out of Control?’’ [editorial], Nature 409(6817) (Jan. 11, 2001): 119. Shorett, P., et al. ‘‘The Changing Norms of the Life Sciences.’’ Science 21 (2003): 123. Angell, M., ‘‘Is Academic Medicine for Sale?’’ NEJM 342(20) (2000): 1516– 18. Lexchin, J., et al., ‘‘Pharmaceutical Industry Sponsorship and Research Outcome and Quality: A Systematic Review.’’ British Medical Journal 326(7400) (2003): 1167–1170. Shulman, S., ‘‘Trouble on the ‘Endless Frontier’: Science, Invention and the Erosion of the Research Commons.’’ Washington, DC: New America Foundation, 2002. Eisenberg, R., and R. Nelson, ‘‘Public vs. Proprietary Science: A Fruitful Tension?’’ Daedalus 131(2) (2002): 89–101. 32. I. Cockburn, ‘‘State Street Meets the Human Genome Project: Intellectual Property and Bioinformatics,’’ in Intellectual Property Rights in Frontier Industries: Biotechnology and Software, ed. R. Hahn (Washington, DC: AEI-Brookings Press, 2005).

21 The Economics of Technology Sharing: Open Source and Beyond Josh Lerner and Jean Tirole

The open source process of production and innovation seems very unlike what most economists might expect. Private firms usually pay their workers, direct and manage their efforts, and control the output and intellectual property thus created. In an open source project, however, material is made publicly available for others to use, often under certain conditions (including requirements for openness and sometimes limitations on future uses). In many cases, anyone who distributes the material must agree to make all enhancements to the original material available under these same conditions. This rule distinguishes open source production from material in the public domain. Many of the contributors to open source projects are unpaid. Indeed, contributions are made under licenses that often restrict the ability of contributors to make money on their own contributions. Open source projects are often loosely structured, with contributors free to pursue whatever area they feel most interesting. The most prominent example of open source production is software, which involves developers at many different locations and organizations sharing code to develop and refine computer programs. The importance of open source software can be illustrated by considering a few examples. The market for server software, which is used by the computers that make web pages available to users through the Internet, has been dominated by the open source Apache project since the inception of systematic tracking by Netcraft in 1995. As of March 2004, more than twothirds of servers employed this or other open source products, rather than commercial alternatives from Microsoft, Sun, and other firms. The open source operating system called Linux accounts for 23% of the operating systems of all servers; moreover, Linux has rapidly outstripped Microsoft’s Windows program as the operating system most frequently embedded into products ranging from mobile phones to video recording devices.1 Open source software is dominant in a number of other areas as well; for example, PERL and PHP are the dominant scripting languages. Recent years have also seen a rise of major corporate investments into open source projects; for instance, IBM is reported to have spent more than $1 billion in 2001 alone on such projects.2

370

Josh Lerner and Jean Tirole

Open source software seems poised for rapid growth in the future. A recent survey of chief information officers suggests that Linux will play an increasingly important role as the operating system for web servers. Linux also has plenty of room to grow in the market for desktop operating systems; at the end of 2003, only 1.4% of the queries to Google came from machines running Linux, although that share was rising.3 The dissemination of open source databases remains in its infancy, but these are projected to become by 2006 significant challengers to commercial systems sold by firms such as IBM and Oracle. The challenge is expected to be led by MySQL, which received a $16 million financing from the venture capital organizations Accel and Benchmark in 2003. MySQL provides its program for free under an open source license and for a substantial fee under a commercial license.4 As of March 2004, the website SourceForge.net, which provides free services to open source software developers, listed more than 78,000 open source projects. A Brief History of Open Source Software Software development has a tradition of sharing and cooperation. But in recent years, both the scale and formalization of the activity have expanded dramatically with the widespread diffusion of the Internet. We highlight three distinct eras of cooperative software development.5 During the first era, the 1960s and 1970s, many of the key features of computer operating systems and the Internet were developed in academic settings such as Berkeley and MIT, as well as in central corporate research facilities where researchers had a great deal of autonomy, such as Bell Labs and Xerox’s Palo Alto Research Center. Software can be transmitted in either ‘‘source code’’ or ‘‘object (or binary) code.’’ Source code is the code using languages such as Basic, C, and Java. Object, or binary, code is the sequence of 0s and 1s that directly communicates with the computer but that is difficult for programmers to interpret or modify. Most commercial software vendors today provide users only with object, or binary, code; when the source code is made available to other firms by commercial developers, it is typically licensed under very restrictive conditions. However, in this first era, the sharing by programmers in different organizations of the source code for computer operating systems and for widely used transmission protocols was commonplace. These cooperative software development projects were undertaken on a highly informal basis. Typically, no efforts to delineate property rights or to restrict reuse of the software were made. This informality proved to be problematic in the early 1980s, when AT&T began enforcing its (purported) intellectual property rights related to the operating system software UNIX, to which many academics and corporate researchers at other firms had made contributions.

The Economics of Technology Sharing

371

In response to the threats of litigation over UNIX, efforts to formalize the ground rules behind the cooperative software development process emerged, which ushered in the second era. The critical institution during this period was the Free Software Foundation, begun by Richard Stallman of the MIT Artificial Intelligence Laboratory in 1983. The foundation sought to develop and disseminate a wide variety of software without cost. The Free Software Foundation introduced a formal licensing procedure under the General Public License for the software produced under the GNU Project. (The name GNU is a recursive acronym which stands for ‘‘GNU’s Not UNIX.’’) In keeping with the philosophy of the organization that this software should be free to use, free to modify, and free to redistribute, the license aimed to preclude the private assertion of copyright or patents that might impede use of cooperatively developed software. In exchange for being able to modify and distribute the GNU software, software developers had to agree to (1) make the source code freely available (or at a nominal cost) to whomever the program is distributed and (2) insist that others who use the software agree to do likewise. All enhancements to the code—and even in many cases code that intermingled the cooperatively developed software with that developed separately—had to be licensed on the same terms. This kind of license is sometimes called ‘‘copyleft,’’ because if copyright seeks to keep intellectual property private, copyleft seeks to keep intellectual property free and available. These contractual terms are distinct from ‘‘shareware,’’ where the binary files, but not necessarily the underlying source code, are made freely available, possibly for a trial period only. The terms are also distinct from public-domain software, where no restrictions are placed on subsequent users of the source code: Those who add to material in the public domain do not commit to put the new product in the public domain. Some open source projects, such as the Berkeley Software Distribution (BSD) effort, took less radical approaches. The BSD license allows anyone to freely copy and modify the source code, but it is much less constraining than the General Public License: Anyone can modify the program and redistribute enhanced versions as proprietary software as long as they acknowledge the original source. The widespread diffusion of Internet access in the early 1990s led to the third era, which saw a dramatic acceleration of open source activity. The volume of contributions and diversity of contributors expanded sharply, and numerous new open source projects emerged, most notably Linux, an operating system related to UNIX, developed by Linus Torvalds in 1991. Another innovation during this period was the proliferation of alternative approaches to licensing cooperatively developed software. In 1997, a number of individuals involved in cooperative software development adopted the ‘‘Open Source Definition,’’ which encompassed BSD-style licensing, as well as the General Public License.

372

Josh Lerner and Jean Tirole

Key Questions on Open Source What Motivates Open Source Contributors? The decision to contribute without pay to freely available software may seem mysterious to economists. However, the standard framework of labor economics can be adapted to capture activity in the open source environment (Lerner and Tirole 2002). The unpaid programmer working on an open source software development project faces a variety of benefits and costs. An independent programmer forgoes the monetary compensation that could otherwise be earned by working for a commercial firm or a university. For a programmer with a commercial company, university, or research lab affiliation, working on open source software bears the opportunity cost of not focusing on other tasks, such as research or the development of proprietary software. Several short- or long-run benefits may counter these costs. First, open source programmers may improve rather than reduce their performance in paid work. This outcome is particularly relevant for system administrators looking for specific solutions for their companies. Second, the programmer may find intrinsic pleasure if choosing a ‘‘cool’’ open source project is more fun than a routine task set by an employer. Third, in the long run, open source contributions may lead to future job offers, shares in commercial open source–based companies or future access to the venture capital market, and last (but not least) ego gratification from peer recognition. Of course, different programmers may put different values on monetary or personal payoffs and on short-term or long-term payoffs. Economic theory suggests that long-term incentives are stronger under three conditions: (1) the more visible the performance to the relevant audience (peers, labor market, and venture capital community); (2) the higher the impact of effort on performance; (3) the more informative the performance about talent (for example, Holmstro¨m 1999).6 The first condition gives rise to ‘‘strategic complementarities.’’ To have an ‘‘audience,’’ programmers will want to work on software projects that will attract a large number of other programmers. This argument suggests the possibility of multiple equilibria. The same project may attract few programmers because programmers expect that other programmers will not be interested; or it may flourish as programmers gain faith in the project. From the standpoint of the individual, commercial projects typically offer better current compensation than open source projects, because employers are willing to offer salaries to software programmers in the expectation that they will capture a return from a proprietary project. Yet, even commercial firms that compensate programmers may want their employees to work on open source projects. Besides the

The Economics of Technology Sharing

373

strategic reasons described below, we already noted that the impossibility of appropriating one’s contribution to an open source project can be offset if the activity brings private benefits like the ability to fix bugs and customize the product to one’s own ends. (Commercial software vendors—like Microsoft in its shared source initiative—have sometimes tried to emulate this benefit by opening their code to selected users under a confidentiality arrangement.) Also, open source code may already be familiar to programmers: Because it is freely available to all, it can be used in schools and universities for learning purposes, thus creating an ‘‘alumni effect.’’ (Again, commercial software vendors are trying to emulate this benefit through university licenses to, say, Windows code.) When we consider the delayed rewards of working on an open source project, the ability to signal a high level of competence may be stronger in the open source mode for three reasons. First, in an open source project, outsiders can see the contribution of each individual, whether that component ‘‘worked,’’ whether the task was hard, whether the problem was addressed in a clever way, or whether the code can be useful for other programming tasks in the future. Second, the open source programmer takes full responsibility for the success of a subproject, with little interference from a superior, which generates information about ability to follow through with a task. Finally, since many elements of the source code are shared across open source projects, more of the knowledge they have accumulated can be transferred to new environments, which makes programmers more valuable to future employers. These incentives are likely to be stronger and the project more successful if there is an effective leader. While the leader of an open source project has no formal authority—that is, he cannot direct any one to do anything—the leadership often has considerable ‘‘real authority.’’7 Leaders play a key role in formulating the initial agenda, setting goals as the project evolves, and resolving disputes that might lead to the splintering or outright cessation of the project. The empirical evidence is largely consistent with the belief that individual contributors to open source projects do benefit directly. The sole nonsurvey study we are aware of, by Hann et al. (2004), examines contributors to the Apache project, drawing on a wide variety of project records. The results suggest that sheer volume of contributions to the Apache project has little impact on salary. But individuals who attain high rank in the Apache organization enjoy wages that are 14% to 29% higher, whether or not their work directly involves the Apache program. Academics have often attempted to understand motivations of those who work on open source projects through surveys. Given the inherent subjectivity of these assessments and the self-serving biases in reporting, the low response rates that many of these surveys have obtained, and the sensitivity of some of the questions,

374

Josh Lerner and Jean Tirole

it is perhaps not surprising that self-reported motivations vary considerably across studies. For instance, Haruvy et al. (2003) find that commercial objectives— particularly the promise of higher future earnings—are an important driver of contributions to open source projects. However, Lakhani and von Hippel (2003) suggest that the overwhelming driver of open source contributors is the need to solve their own specific programming needs, while a Boston Consulting Group (2003) survey implies that intellectual curiosity is the most important determinant. How Do Commercial Firms Work and Compete with Open Source? Commercial companies may interact with an open source project in a number of ways. While improvements in the open source software may not be not appropriable, commercial companies can benefit if they also offer complementary expertise in some proprietary segment of the market. Firms may temporarily encourage their programmers to participate in an open source project to learn about the strengths and weaknesses of the project’s technical approach. For-profit firms may compete directly with open source providers in the same market. Finally, commercial companies may interface with the open source world because it generates good public relations with programmers and customers. A for-profit firm that seeks to provide services and products that are complementary to the open source product but not supplied efficiently by the open source community can be described as ‘‘living symbiotically.’’ IBM, which has made open source software a major resource for its systems integration and consulting work, exemplifies this approach. A company in this situation will want to have extensive knowledge about the open source movement and may even want to encourage and subsidize open source contributions, both of which may cause it to allocate some programmers to the open source project. Because firms do not capture all of the benefits of the investments in the open source project, however, the free-rider problem often discussed in the economics of innovation should apply here as a limiting factor. The code release strategy arises when companies release some existing proprietary code and then create a governance structure for the resulting open source development process. For example, IBM released half a million lines of its Cloudscape program, a simple database that resides inside a software application instead of as a full-fledged database program, to the Apache Software Foundation. HewlettPackard released its Spectrum Object Model-Linker to the open source community to help the Linux community write software to connect Linux with Hewlett Packard’s RISC computer architecture. This strategy is akin to giving away the razor (the code) to sell more razor blades (the related consulting services that IBM and HP hope to provide).8

The Economics of Technology Sharing

375

When can it be advantageous for a commercial company to release proprietary code under an open source license? In general, it will make sense if the increase in profit in the proprietary complementary segment offsets any profit that would have been made in the primary segment had it not been converted to open source. Thus, the temptation to go open source is particularly strong when the product is lagging behind the market leader but the firm sees a possibility that widespread use and further development will increase the profitability of the complementary product or service. If network effects and switching costs are very strong, the second-best commercial package might have a small and diminishing market share. In these cases, the cost to corporations of releasing code may be very small. Moreover, such a strategy may reassure present and potential users that the released software will never be withdrawn (i.e., they will always be able to maintain the product themselves). This motivation can also depend on the evolution of vertical relationships between small and large firms in the software industry in commercial software environments, a subject that merits further study. Indeed, many small developers are uncomfortable doing business with leading software firms. They fear that the commercial platform owner has an incentive to introduce substitutes in the developers’ segment in order to force prices down in that segment, and to raise the demand for licenses to the broad software platform (Farrell and Katz 2000). By contrast, when a large firm makes its platform available on an open source basis through a restrictive license, such as the GPL, the small firm need no longer fear being squeezed in this way. Numerous challenges appear, though, when a for-profit firm seeks to become the center of an open source development project. Leadership by a commercial entity may not internalize enough of the objectives of the open source community. In particular, a corporation may not be able to credibly commit to keeping all source code in the public domain and to highlighting important contributions adequately. These difficulties help to explain why Hewlett-Packard released its code through Collab.Net, a venture by leading open source programmers, which organizes open source projects for corporations that wish to open up part of their software. In effect, Collab.Net offers a kind of certification that the firm is committed to the open source project. (The Apache Software Foundation plays a similar role in the Cloudscape case mentioned above.) In a theoretical model, Dessein (2002) shows that a principal with formal control rights over an agent’s activity in general gains by delegating control rights to an intermediary with preferences or incentives that are intermediate between the principal’s and the agent’s. The partial alignment of the intermediary’s preferences with the agent’s fosters trust and boosts the agent’s initiative, ultimately offsetting the partial loss of control for the principal. In the case of Collab.Net, the congruence with the open source developers is obtained through the

376

Josh Lerner and Jean Tirole

employment of visible open source developers and the involvement of O’Reilly, a technical book publisher with strong ties to the open source community. While the relative merits of open source and proprietary software are discussed in a number of contributions, direct competition between the two paradigms has received little attention. An exception is Gaudeul (2004),9 who builds a duopoly model with one open source and one proprietary software project. In his model, open source software has both costs and benefits relative to proprietary software. Open source software suffers from some lack of coordination: The same code may be written twice or not at all. Another cost of open source software in Gaudeul’s model is that its designers, the developers, may not bother developing interfaces that appeal to unsophisticated users. By contrast, the profit-maximizing proprietary software firm in his model is keener to develop such an interface. However, the proprietary model must pay its developers and, despite good project coordination, may choose to develop a limited set of features. In this model, the proprietary software is sold to users at a positive price that excludes some possible users. In equilibrium, the open source software, if it survives, is used by either low-demand or low-income consumers, who cannot afford buying the proprietary software, or by developers who like the potentially larger set of features and do not care about the missing or insufficient user interface. Furthermore, the presence of open source software raises welfare, at least if it does not discourage the development of proprietary software with a good interface. How Does the Legal System Affect Open Source? Open source software is shaped by the legal rules under which it operates. In each case, the product originator gives users the right to employ the copyrighted code through a license. But the licenses differ tremendously in the extent to which they enable licensors and contributors to profit from the code that is contributed. In Lerner and Tirole (2005), we explore what drives firms to choose particular licenses. We begin with a model of license choice. We suppose that an entity, either an individual or a firm, (1) is deciding whether to make some software available under an open source license and (2) if so, what type of license to employ. We depict the interactions between the licensor and the community of programmers. The programmers’ benefits from working on the project may depend on the choice of license. The licensor must assess how its choice of license, together with project characteristics—such as the environment, the nature of the project, and the intended audience—impacts the project’s likely success. The model suggests that permissive licenses such as the BSD, where the user retains the ability to use the code as he sees fit, will be more common in cases where projects have strong appeal to the community of open source contributors—for instance, when contributors stand to ben-

The Economics of Technology Sharing

377

efit considerably from signaling incentives or when the licensors are well trusted. Conversely, restrictive licenses such as the GPL will be commonplace when such appeals are more fragile. Examples of cases where we would expect a restrictive license are projects geared for end users who are unlikely to appreciate the coding, such as computer games, or those sponsored by corporations, which potential contributors might fear would ‘‘hijack’’ the project. One of the most visible of the disputes over licensing was the Mozilla case alluded to above. Netscape initially proposed the ‘‘Netscape Public License,’’ which would have allowed Netscape to take pieces of the open source code and turn them back into a proprietary project again (Hamerly et al. 1999). Ultimately, the firm announced the ‘‘Mozilla Public License,’’ under which Netscape cannot regain proprietary rights to modifications of the code: In fact, the terms of the final license are even stricter than those of the General Public License. In Lerner and Tirole (2005), we also present an empirical analysis of the prevalence of different types of open source licenses. The analysis employs nearly 40,000 open source projects in the SourceForge database. Since all of the projects in this database are open source, we focus on whether the license requires that when modified versions of the program are distributed, the source code must be made generally available and/or whether the license restricts modified versions of the program from mingling their source code with other software that does not employ such a license. We term such licenses ‘‘restrictive.’’ We find that restrictive licenses are more common for applications geared toward end users and system administrators—like desktop tools and games. Restrictive licenses are significantly less common for those applications aimed toward software developers. Restrictive licenses are also less common for projects operating in commercial environments or that run on proprietary operating systems. Projects whose natural language is not English, whose community appeal may be presumed to be much smaller, are more likely to employ restrictive licenses. Projects with less restrictive licenses tend to attract more contributors. What Is the Relative Quality of Open Source Software? One of the most contentious issues in the literature has been the relative virtues of the open source and proprietary development process. Advocates of open source software have long claimed that the open source development process leads to superior software (for example, Raymond 1999). A number of studies have sought to explore these claims, but consensus remains elusive. Kuan (2001) was the first to offer a formal model of some of the advantages of open source software for users. She focused on the consumer’s choice between employing off-the-shelf commercial software and adapting open source. While the

378

Josh Lerner and Jean Tirole

proprietary software can (and indeed must) be used ‘‘as is,’’ open source code can be enhanced in quality through the user’s efforts. Kuan shows that under certain circumstances, some consumers will prefer the open source option and invest in producing software that is of superior quality to commercial alternatives. The paper tests this model by comparing dates at which program errors or ‘‘bugs’’ were reported and fixed in three open source programs—Apache, FreeBSD, and Gnome—with three commercial projects matched by subject matter and age. For two of the three pairs that she examines, the rate at which bugs are fixed is significantly faster in the open source project, and there is little difference in the third case. Bessen (2002), in a highly related paper, examines the ability of heterogeneous users to customize open source software to meet their own particular needs. Proprietary software manufacturers cannot anticipate and offer every conceivable variation that consumers might desire. Again, consumers face a ‘‘make versus buy’’ choice, where the complexity and idiosyncrasy of the project, as well as the cost of modifications, will drive the choice. Franke and von Hippel (2003) find that onefifth of Apache users adapted security features to meet their particular needs, as consistent with Bessen’s model. While Kuan and Bessen attribute the superiority of open source projects to the ability of end users to adapt an initial code base, Johnson (2004) suggests that open source programs may avoid pathologies that affect commercial projects. He argues that workers in commercial firms may collude not to report programming errors of fellow employees lest their own reputation and future earnings be damaged. He hypothesizes that because programmers do not receive wages in open source projects, they will have fewer incentives to engage in such collusion. Although the ego gratification and career incentives may motivate collusion, perhaps the large number of potential eyeballs in open source software makes collusion difficult to sustain. Johnson argues that reduced collusion will lead to more peer review and higher quality. Open source advocates have argued that when source code is open and freely visible, programmers can readily identify security flaws and other problems: As Eric Raymond (1999) has argued, ‘‘Given enough eyeballs, all bugs are shallow.’’ Proponents of proprietary software, on the other hand, argue that the openness of the source code allows malicious hackers to figure out its weaknesses. Anderson (2002) argues that under certain plausible assumptions, the openness of the system should have no impact on its security. Making bugs harder for hackers to find by keeping the source code hidden will also mean that software companies have a more difficult time identifying errors through ‘‘beta’’ testing, where lead users experiment with the product, also without access to the underlying source code. (While software firms will also do internal testing by employees with access to the source code, the effort

The Economics of Technology Sharing

379

devoted to these ‘‘alpha’’ tests is usually many times smaller than that in later-stage tests.) Thus, he concludes, ‘‘other things being equal, we expect that open and closed systems will exhibit similar growth in reliability and in security assurance.’’ However, Anderson does not attempt to assess this claim empirically. Any such effort is difficult because hackers may attack a software program for reasons unrelated to the intrinsic security of the program; for instance, some hackers may derive more gratification from an attack on a leading public company, even though hackers have targeted both commercial and open source programs on various occasions.10 What Are Appropriate Public Policies Toward Open Source? Government commissions and agencies have proposed—and in some cases implemented—a variety of measures to encourage open source developers. For example, in the United States, the President’s Information Technology Advisory Committee (2000) recommended direct federal subsidies for open source projects to advance high-end computing. Many European governments have policies to encourage the use and purchase of open source software for government use (‘‘Microsoft at the Power Point,’’ 2003). Governments may even mandate the development of localized open source projects, as has occurred in China (Open Source Development Labs, 2004). Economists have sought to understand the consequences of a vibrant open source sector for social welfare. Perhaps not surprisingly, definitive or sweeping answers have been difficult to come by; instead, the policy conclusions focus on specific instruments in the specific contexts. Most analyses have suggested that government support for open source projects is likely to have an ambiguous effect on social welfare. For example, Johnson (2002) presents a model where programmers decide whether to devote effort to a project, in which their contributions become a public good once they are developed. Users thus face a decision whether to enhance an existing open source program or to wait in the hope that another programmer will undertake the development process. Johnson then compares this process to a stylized depiction of the development of proprietary software in a corporate setting. Open source projects have the advantage of being able to access the entire pool of developer talent, not just employees in a single firm. Given the larger talent pool, they can aggregate and exploit more private information. But because of the free-riding problem, some potentially valuable projects will not be developed under an open source system. Johnson concludes that a comparison of the social welfare consequences of these two systems is ambiguous. Casadesus-Masanell and Ghemawat (2003) depict competition between an open source operating system available at no cost and a proprietary commercial product. The crucial feature of their model is on the demand side: The larger the market

380

Josh Lerner and Jean Tirole

share of a given operating system, the more valuable that system to users. This effect could be due to better learning about the program’s features (if users contribute comments and suggestions to improve the product) or to the presence of complementary software developed by other firms. In this setting, the presence of an open source operating system leads the commercial firm to set lower prices, which in turn means that the overall use of operating systems is higher. However, the value of the commercial system for users is lower: For instance, the presence of a competing product may lead third-party developers to develop fewer complementary products for the commercial operating system. Thus, the presence of open source projects may make society either better or worse off. This model also suggests that in some cases, the proprietary operating system may be able to drive the market share of the open source alternative to zero, which may well be socially desirable. Schmidt and Schnitzer (2003) highlight similarly that open source software has social costs and benefits. Building on a line of economic reasoning that extends back to Arrow (1962) and even earlier, they highlight two countervailing effects. From a static point of view, free or nearly free open source will ensure greater social welfare, since virtually any potential user will be able to access software. But from a dynamic perspective, with so few profits to be gleaned, developers may lack incentives to introduce new products. While career concerns and other incentives may motivate developers to identify bugs in open source programs and undertake certain modest adaptations to meet their own needs, they are unlikely to be sufficient to encourage major breakthroughs. The authors argue that while open source programs will enhance social welfare in some settings, this will be far from universal. They caution against subsidies that may lead to an undesirably high level of open source activity. Saint-Paul (2003) reaches an even bleaker conclusion about the open source phenomenon. He employs a Romer-style endogenous growth model, in which both commercial firms and ‘‘philanthropists’’—individuals who are willing to give their contributions away for free—innovate. He shows that the free contributions will lead to economic growth but also reduce the profits, and hence the incentives to innovate, among commercial firms. Unless the proprietary sector is quite profitable, then the second effect will dominate, and innovation and growth will be harmed by the presence of open source software. He argues that the negative effect is likely to be even stronger than his model shows, because he neglects, for instance, the possibility that philanthropic products do not meet users’ needs as well as commercial products (though see the previous section for a counterargument) and can also divert programming talent that could have been devoted to commercial products. In a more informal piece, Shapiro and Varian (2004) suggest another consideration that formal models have not so far discussed: the impact on human capital

The Economics of Technology Sharing

381

and entrepreneurship. They suggest that an open system will facilitate learning by students as to how to program and will provide opportunities for third-party developers to introduce complementary products. They argue that all else being equal, these considerations should lead public policy makers in nations that seek to encourage the development of their software industries to boost the development of open source activity. How Will Software Patents Affect Open Source? Software patents will interact with open source activity.11 This issue is clearly a timely one, in light of the litigation launched by the SCO Group, which holds (at least partial) rights to UNIX (acquired from Novell, who in turn had purchased them from AT&T). Beginning in 2003, the firm initiated a series of lawsuits against, among others, AutoZone, DaimlerChrysler, IBM, and Novell, alleging that they were violating SCO’s intellectual property by contributing to or using Linux.12 The allegedly detrimental impact of software patents on open source was also a frequently invoked reason for opposing software patents in the ongoing debate in the European Parliament on this question.13 Software patents create the possibility of holding up software producers. In the case of commercial software, individuals and companies that do not produce software themselves (e.g., hardware manufacturers and software users) but hold a software patent can try to obtain royalty payments from software vendors. (Large software vendors are less likely to engage in such behaviors against each other, since they have accumulated patent portfolios that they can use for retaliatory purposes.) Open source software is vulnerable in a different way. After all, the code is free of charge and the contributors hardly solvent for the most part, so attempting to collect royalties is not a powerful incentive. However, firms with software patents may seek damages from firms that service or use open source software. Software firms facing competition from open source alternatives may sue to enjoin utilization or distribution of open source code. It remains to be seen whether the open source movement will itself enter into defensive patenting, as large commercial vendors already do, or at least make a more concerted effort to forestall patenting by others by aggressively publishing. One intriguing initiative is the Red Hat Assurance Plan, in which the Linux distributor is offering partial protection against intellectual property litigation.14 Another interesting area of study concerns the consequences of users paying royalties for an open source program that also included some commercially patented material. Such royalty demands might trigger ‘‘sweetheart’’ deals between firms, the splitting of open source projects into different branches (often termed ‘‘forking’’), and the privatization of blocks of code. The General Public License seeks to

382

Josh Lerner and Jean Tirole

address these problems by providing, ‘‘[ I]f a patent license would not permit royaltyfree redistribution of the Program by all those who receive copies directly or indirectly through you, then the only way you could satisfy both it and this License would be to refrain entirely from distribution of the Program.’’ Many other types of open source licenses, however, do not address this issue.15 Another, less prominent, question relates to the impact of patents on the dynamics of information sharing and collaboration among open source contributors. To what extent will the ability of programmers to protect their discoveries with strong patent rights reduce their incentives to participate in open source projects? To date, very little systematic analysis has examined the implications of patents for open source. However, a broader literature has scrutinized the impact of patenting on the generation and diffusion of scientific knowledge more generally, especially the augmented ability of academic institutions, government laboratories, and nonprofit institutions to patent the results of publicly funded research. This literature has sought to understand the pervasiveness and consequences of the ‘‘anticommons’’ problem (Heller and Eisenberg 1998): the concern that the patenting of scientific knowledge will lead to lower research productivity and hence eventually to reduced economic growth. Much of the discussion of these questions to date has featured broad assertions and anecdotal examples (as in Bok 2003). It is clear from these studies that institutions and researchers have responded to the increased incentives to commercialize products by engaging in more patenting and commercialization activities (for instance, Jaffe and Lerner 2001; Lach and Schankerman 2003). Whether these commercial activities have detrimental effects on research and social welfare is much more ambiguous.16 Given this preliminary and somewhat contradictory evidence, our ability to draw conclusions for consequences of formal intellectual property rights for open source software is quite limited. Can Open Source Work Beyond Software? An interesting question is whether the open source model can be transposed to other industries. Many industries involve forms of cooperation between commercial entities in the form of for-profit or not-for-profit joint ventures. Others exhibit userdriven innovation or open science cultures. Although some aspects of open source software collaboration (such as electronic information exchange across the world) could easily be duplicated, other aspects would be harder to emulate. Consider, for example, the case of biotechnology. It may be impossible to break up large projects into small manageable and independent modules, and there may not be sufficient sophisticated users who can customize the molecules to their own needs. The tasks that are involved in making the product available to the end user involve larger expenditures than simply providing consumer support and friendlier

The Economics of Technology Sharing

383

user interfaces as in software. The costs of designing, testing, and seeking regulatory approval for a new drug are enormous. More generally, in most industries the development, production, and distribution of individual components requires substantial capital costs as opposed to (for some software programs) individual contributions and no capital investment (besides the computer the programmer already has). Another obstacle is that in mass-market industries users are numerous and rather unsophisticated and so deliver little peer recognition and ego gratification. How Can Firms Realize the Benefits of Open Source? As the earlier discussion pointed out, corporations may emulate some of the benefits attached to open source production either by getting involved in open source themselves or by adopting institutional arrangements that deliver some of these benefits. First, using open source technology encourages users that they will not be ‘‘held up’’ by a future price increase after adopting a technology and that they will always be able to tailor their technology to their own particular needs without depending on the good will or health of a vendor. Second, open source limits the problems of ‘‘patent thickets,’’ when multiple firms have overlapping intellectual property rights and at least one party attempts to extract a high fee for its particular contribution. Third, a firm might make a technology open source as a way of trying to certify a technological standard, in which case firms may contribute software to open source to benefit from the endorsement of such a standard, as the HP case discussed above illustrates. Firms can also address these problems in non–open source ways, such as patent pools, standard-setting organizations, and self-imposed commitments. In a patent pool, firms combine their patents with those of other firms. These pools allow users to access a number of firms’ patents simultaneously, thereby avoiding the ‘‘patent thicket.’’ In many cases, the pricing schedule is also specified in the agreement that establishes the pool, ensuring that no party attempts to extract very high fees or to increase its fees after users are locked in (Lerner and Tirole 2004a; Lerner et al. 2003). Standard-setting organizations offer an alternative path for the certification of new technologies. Often firms can choose between standard-setting organizations, and they can seek an endorsement for an emerging technology from an independent and prestigious organization or use a more complacent one (Lerner and Tirole 2004b). These bodies also help address the other concerns, frequently asking contributors of the key technologies to commit to license the technology on ‘‘reasonable and nondiscriminatory’’ terms or to make various other concessions. Self-imposed commitments can serve much the same role. For instance, firms can commit to license technologies at a given price schedule, or they can commit to

384

Josh Lerner and Jean Tirole

provide sufficient information so that users can tailor the technology, such as in Microsoft’s Shared Source Initiative. One open question about many of these selfimposed programs is the extent to which the commitments can be enforced if the firm subsequently changes its design.17 Open Source and Academia Open source and academia have many parallels. The most obvious parallel relates to motivation. As in open source, the direct financial returns from writing academic articles are typically nonexistent, but career concerns and the desire for peer recognition provide powerful inducements. Other similar dynamics are also at work. Consider, for instance, the discussion of motivation for programmers when choosing an open source project to contribute to. As we highlight above, a critical goal is the selection of a project that is likely to continue to be successful, so that the programmers’ contributions are widely recognized, yet which at the same time has interesting and challenging programming challenges to be addressed. These criteria should be familiar to anyone who has advised a doctoral student on the choice of a thesis topic! At the same time, however, there are some substantial differences between the two realms. Here, we highlight two areas where academic economists could learn from the open source realm. The first of these relates to the incentives to create public goods. Open source contributors often create substantial bodies of code, which are made widely available when completed. Similarly, while we can cite some examples of efforts to create shared resources that can be widely used by the economics community—the NBER Patent Citations Database created by Bronwyn Hall, Adam Jaffe, and Manuel Trajtenberg is a recent important example—far too often these efforts are neglected because the returns to the project leaders are low. Why it is not commonplace to see economists frequently seeking to establish their reputation by creating original, widely accessible datasets is an interesting question. (Akin to open source, we might anticipate that this strategy would be especially effective for those at smaller and less centrally located institutions.) One explanation might be that data collection is often inspired by what analyses one wants to perform, so it is harder to separate data collection and analysis. In any case, the design of mechanisms that successfully encourage such investments is an important challenge for academic economists. A second area relates to access to published work. As we have highlighted, contributors to open source projects seem to be powerfully spurred by the provisions of these licenses. The assurance that contributions—and subsequent contributions that build on them—will remain publicly accessible incentivizes programmers to write code. By way of contrast, in academic economics, it is standard to assign the copy-

The Economics of Technology Sharing

385

right to one’s work to a commercial publisher (although the ideas remain free for others to build on). In other areas of academia, this approach is under increasing attack. For instance, recent years have seen the rise of ‘‘open access’’ journals, such as the Public Library of Science, which make all articles freely accessible and distributable. In response to this challenge, a number of established science journals, such as the Proceedings of the National Academies of Sciences, have not only begun providing free access to older issues but even allowing authors to opt to have their articles immediately publicly accessible with the payment of an additional fee.18 It is an interesting question as to whether open access will have the same appeal for the economics community. Final Thoughts This paper has reviewed our understanding of the growing open source movement. We have highlighted how many aspects of open source software appear initially puzzling to an economist. As we have acknowledged, our ability to answer confidently many of the issues raised here is likely to increase as the open source movement itself grows and evolves. At the same time, it is heartening to us how much of open source activities can be understood within existing economic frameworks, despite the presence of claims to the contrary. The labor and industrial organization literature provides lenses through which the structure of open source projects, the role of contributors, and the movement’s ongoing evolution can be viewed. Acknowledgments We thank the National Science Foundation and Harvard Business School’s Division of Research for financial support. The Institut D’Economie Industrielle receives research grants from a number of corporate sponsors, including France Telecom and the Microsoft Corporation. We thank Christophe Bisiere, Alexandre Gaudeul, Jacques Cremer, Justin Johnson, Hal Varian, and Brian Kahin for helpful comments. All errors are our own. Notes 1. On web server software and Apache, see http://news.netcraft.com/archives/web_server_ survey.html (accessed March 21, 2004). On the use of Linux in web server operating systems, see http://www.pcworld.com/news/article/0,aid,112840,00.asp (accessed March 31, 2004). On the use of Linux for embedded software, see http://www.linuxdevices.com/articles/ AT8693703925.html (accessed March 21, 2004).

386

Josh Lerner and Jean Tirole

2. See http://news.com.com/2100-1001-825723.html (accessed March 21, 2004). 3. For the survey of chief information officers, see http://www.morganstanley.com/ institutional/techresearch/pdfs/ciosurvey1203.pdf (accessed March, 21, 2004). On Linux software used for Google searches, see http://www.internetnews.com/dev-news/article.php/ 3302941 (accessed March 31, 2004). 4. See http://www.informationweek.com/story/showArticle.jhtml?articleID=18312009 (accessed August 8, 2004). 5. This history is highly abbreviated. See Lerner and Tirole (2002) and the sources cited therein for a longer account. 6. For a discussion as to how firms might otherwise have superior information about employees and how this might deter job offers from outsiders—a problem that open source programming can address—see Greenwald (1986) and Waldman (1984). 7. To use the terminology of Aghion and Tirole (1997). 8. For more details, see http://www.infoworld.com/article/04/08/03/HNclouscape_1.html (accessed August 3, 2004), http://www.collab.net/customers/cdp_solutions_at_work.html (accessed March 31, 2004), and the associated links. 9. See also the discussion below of Casadesus-Masanell and Ghemawat (2003). 10. For a discussion of a hacker attack on Apache, see http://thewhir.com/marketwatch/ hac062102.cfm (accessed March 31, 2004). 11. In this section, we avoid discussing the highly contentious and unsettled question of the economic impact of software patents more generally; for more on this topic see, for instance, Bessen and Hunt (2003), Caillaud (2003), Graham and Mowery (2003), and Hahn and Wallsten (2003). 12. Patent concerns have also slowed the adoption of Linux in the public sector; see, for instance, http://www.informationweek.com/story/showArticle.jhtml?articleID=26806464 (accessed August 25, 2004) for a discussion of the impact of these concerns on the city of Munich’s open source effort. 13. See, for instance, http://news.zdnet.co.uk/business/legal/0,39020651,39116053,00.htm (accessed March 26, 2004). 14. See http://www.redhat.com/about/presscenter/2004/press_blackduck.html (accessed August 24, 2004). One challenge is the extent and dispersion of the patent holdings that may impact open source projects: The insurer Open Source Risk Management estimates that there are 283 patents that might be used in claims against the Linux kernel alone (http://www .eweek.com/article2/0,1759,1631336,00.asp, accessed August 24, 2004). 15. A related danger is that programs will inadvertently infringe patents. Programmers may lack the incentives and skills needed to check whether their contribution infringes awards. As an effort to limit this problem, beginning in May 2004, Linux contributors were required to attest that they have the right to make that contribution. For a discussion, see http://www .computerworld.com/softwaretopics/os/linux/story/0,10801,93395,00.html (accessed August 8, 2004). 16. For instance, Thursby and Thursby’s (2003) study of six major research universities suggests that while the probability that a faculty member will indicate to his university’s technol-

The Economics of Technology Sharing

387

ogy transfer office that he has made a new discovery has increased tenfold over the past decade, research productivity in basic research journals has remained constant. On the other hand, Murray and Stern (2003) have shown that papers published in the journal Nature Biotechnology are somewhat less likely to be cited in other articles once the corresponding patent application issues. They find that the papers with corresponding patents are initially more heavily cited than those without, but then their citation rate declines more sharply over time. 17. This is also a question for other commitments as well. For one illustration in a standard setting context, see Rambus Inc. v. Infineon Techs. AG, 318 F.3d 1081 (Fed. Cir. 2003). 18. See, for instance, http://www.plos.org/about/openaccess.html and http://www.pnas.org/ cgi/content/full/101/23/8509 (accessed August 10, 2004).

References Aghion, Philippe, and Jean Tirole (1997). ‘‘Formal and Real Authority in Organizations.’’ Journal of Political Economy 105: 1–29. Anderson, Ross (2002). ‘‘Security in Open versus Closed Systems—The Dance of Boltzman, Coase and Moore.’’ Unpublished working paper, Cambridge University. Arrow, Kenneth J. (1962). ‘‘Economic Welfare and the Allocation of Resources for Invention.’’ In The Rate and Direction of Inventive Activity: Economic and Social Factors, Richard R. Nelson, editor, pp. 609–626. Princeton, NJ: Princeton University Press. Bessen, James (2002). ‘‘Open Source Software: Free Provision of Complex Public Goods.’’ Unpublished working paper, Research on Innovation. ———, and Robert M. Hunt (2003). ‘‘An Empirical Look at Software Patents.’’ Working Paper 03-17, Federal Reserve Bank of Philadelphia. Bok, Derek (2003). Universities in the Marketplace: The Commercialization of Higher Education. Princeton, NJ: Princeton University Press. Boston Consulting Group (2003). ‘‘Boston Consulting Group/OSDN Hacker Survey.’’ Caillaud, Bernard (2003). ‘‘La Proprie´te´ Intellectuelle sur les Logiciels.’’ Proprie´te´ Intellectuelle, Conseil D’Analyse Economique, Rapport 41: 113–171. Casadesus-Masanell, Ramon, and Pankaj Ghemawat (2003). ‘‘Dynamic Mixed Duopoly: A Model Motivated by Linux vs. Windows.’’ Strategy Unit Working Paper 04-012, Graduate School of Business Administration, Harvard University. Dessein, Wouter (2002). ‘‘Authority and Communication in Organizations.’’ Review of Economic Studies 69: 811–838. European Commission, Interchange of Data between Administrations (2001). ‘‘Study into the Use of Open Source Software in the Public Sector.’’ June. http://europa.eu.int/ISPO/ida/jsps/ index.jsp?fuseAction=showDocument&documentID=333&parent=chapter&preChapterID= 0-17-134. Farrell, Joseph, and Michael L. Katz (2000). ‘‘Innovation, Rent Extraction, and Integration in Systems Markets.’’ Journal of Industrial Economics 48: 413–432. Franke, Nikolaus, and Eric von Hippel (2003). ‘‘Satisfying Heterogeneous User Needs via Innovation Tool Kits: The Case of Apache Security Software.’’ Research Policy 32: 1199–1215.

388

Josh Lerner and Jean Tirole

Gaudeul, Alexandre (2004). ‘‘Competition between Open-Source and Proprietary Software: The (LA )TE X Case Study.’’ Unpublished working paper, Universities of Toulouse and Southampton. Graham, Stuart, and David C. Mowery (2003). ‘‘Intellectual Property Protection in the Software Industry.’’ In Patents in the Knowledge-Based Economy: Proceedings of the Science, Technology and Economic Policy Board, Wesley Cohen and Steven Merrill, editors. Washington, DC: National Academies Press. Greenwald, Bruce C. (1986). ‘‘Adverse Selection in the Labour Market.’’ Review of Economic Studies 53: 325–347. Hahn, Robert W., and Scott J. Wallsten (2003). ‘‘A Review of Bessen and Hunt’s Analysis of Software Patents.’’ Unpublished working paper, American Enterprise Institute–Brookings Joint Center for Regulatory Studies. Hamerly, Jim, Tom Paquin, and Susan Walton (1999). ‘‘Freeing the Source: The Story of Mozilla.’’ In Open Sources: Voices from the Open Source Revolution, Chris DiBona, Sam Ockman, and Mark Stone, editors, pp. 197–206. Cambridge, MA: O’Reilly. Hann, Il-Horn, Jeff Roberts, Sandra Slaughter, and Roy Fielding (2004). ‘‘An Empirical Analysis of Economic Returns to Open Source Participation.’’ Unpublished working paper, Carnegie-Mellon University. Haruvy, Ernan E., Fang Wu, and Sujoy Chakravarty (2003). ‘‘Incentives for Developers’ Contributions and Product Performance Metrics in Open Source Development: An Empirical Investigation.’’ Unpublished working paper, University of Texas at Dallas. Heller, Michael, and Rebecca Eisenberg (1998). ‘‘Can Patents Deter Innovation? The Anticommons in Biomedical Research.’’ Science 280: 698–701. Holmstro¨m, Bengt (1999). ‘‘Managerial Incentive Problems: A Dynamic Perspective.’’ Review of Economic Studies 66: 169–182. Jaffe, Adam B., and Josh Lerner (2001). ‘‘Reinventing Public R&D: Patent Law and Technology Transfer from Federal Laboratories.’’ Rand Journal of Economics 32: 167–198. Johnson, Justin P. (2002). ‘‘Open Source Software: Private Provision of a Public Good.’’ Journal of Economics and Management Strategy 11: 637–662. ——— (2004). ‘‘Collaboration, Peer Review and Open Source Software.’’ Unpublished working paper, Cornell University. Kuan, Jennifer (2001). ‘‘Open Source Software as Consumer Integration into Production.’’ Unpublished working paper, Stanford University. Lach, Saul, and Mark Schankerman (2003). ‘‘Incentives and Invention in Universities.’’ Discussion Paper No. 3916, Centre for Economic Policy Research. Lakhani, Karim, and Eric von Hippel (2003). ‘‘How Open Source Software Works: ‘Free’ User-to-User Assistance.’’ Research Policy 32: 923–943. Lerner, Josh, and Jean Tirole (2002). ‘‘Some Simple Economics of Open Source.’’ Journal of Industrial Economics 52: 197–234. ———, and Jean Tirole (2004a). ‘‘Efficient Patent Pools.’’ American Economic Review 94: 691–711.

The Economics of Technology Sharing

389

———, and Jean Tirole (2004b). ‘‘A Model of Forum Shopping, with Special Reference to Standard Setting Organizations.’’ Unpublished working paper, Harvard University and University of Toulouse. ———, and Jean Tirole (2005). ‘‘The Scope of Open Source Licensing.’’ Journal of Law, Economics, and Organization 21 (forthcoming). ———, Marcin Strojwas, and Jean Tirole (2003). ‘‘Cooperative Marketing Agreements between Competitors: Evidence from Patent Pools.’’ Working Paper No. 9680, National Bureau of Economic Research. ‘‘Microsoft at the Power Point’’ (2003). Economist (September 11). Murray, Fiona, and Scott Stern (2003). ‘‘Do Formal Intellectual Property Rights Hinder the Flow of Scientific Knowledge? Evidence from Patent-Paper Pairs.’’ Unpublished working paper, Massachusetts Institute of Technology and Northwestern University. Open Source Development Labs (2004). ‘‘OSDL Announces First Chinese Member.’’ January 30. http://www.osdl.org/newsroom/press_releases/2004/2004_01_30_beaverton.html. President’s Information Technology Advisory Committee, Panel on Open Source Software for High End Computing (2000). ‘‘Developing Open Source Software to Advance High End Computing.’’ October. http://www.hpcc.gov/pubs/pitac/pres-oss-11sep00.pdf. Raymond, Eric (1999). The Cathedral and the Bazaar: Musings on Linux and Open Source by an Accidental Revolutionary. Cambridge, MA: O’Reilly. Saint-Paul, Gilles (2003). ‘‘Growth Effects of Non-Proprietary Innovation.’’ Journal of the European Economic Association: Papers and Proceedings 1: 429–439. Schmidt, Klaus, and Monika Schnitzer (2003). ‘‘Public Subsidies for Open Source? Some Economic Policy Issues of the Software Market.’’ Discussion Paper No. 3793, Centre for Economic Policy Research. Shapiro, Carl, and Hal R. Varian (2004). ‘‘Linux Adoption in the Public Sector.’’ Unpublished working paper, University of California. Thursby, Jerry, and Marie Thursby (2003). ‘‘Has Licensing Changed Academic Research? Issues of Productivity, Faculty Incentives, and Public Policy.’’ Unpublished working paper, Emory University and Georgia Institute of Technology. Waldman, Michael (1984). ‘‘Job Assignments, Signaling, and Efficiency.’’ Rand Journal of Economics 15: 255–267.

22 ‘‘Open and Collaborative’’ Biomedical Research: Theory and Evidence Arti K. Rai

Mounting empirical evidence suggests that biomedical research has, during the past 25 years, become increasingly proprietary1 and secretive.2 Given the cumulative nature of research, this trend has raised fears that future progress may be impeded by access and licensing difficulties.3 One important response has involved calls for improving access by requiring publicly funded scientists and research institutions to put data and certain types of research tools into the public domain or, at a minimum, to license them widely and nonexclusively at a reasonable fee.4 This emphasis takes the current organizational structure of research as a given but seeks to reduce the intensity of exclusionary behavior associated with the research. A response that is perhaps more dramatic has begun to emerge, however. Public funding bodies, prominent scientists, and even some pharmaceutical firms have taken steps in the direction of what might be called ‘‘open and collaborative’’ science. Open and collaborative projects not only disavow exclusionary behavior but also move beyond the traditional small lab/small firm–based structure of biomedical research. The rise of arguments for open and collaborative biomedical research has coincided with two prominent phenomena: (1) the increased importance of computation in such research5 and (2) the well-documented emergence of so-called ‘‘open source’’ methods of innovation in computation-heavy areas of research and development, primarily software. In some recent cases, the modeling on open source software has been quite explicit—for example, the federally funded haplotype mapping project, which aims to create a database that catalogs all human genetic variation, has adopted a licensing policy that is self-consciously modeled on the ‘‘copyleft’’ system of open source software licensing.6 Under the copyleft version of open source development, users can access source code freely, but such access is conditioned on the user making his or her improvements to such information available under the same conditions. Although some commentators have discussed the application of open source–type principles to biomedical research, the empirical literature on such research is sparse.

392

Arti K. Rai

In this chapter, I give some preliminary results from an ongoing empirical inquiry into such research. The first part of the paper gives economic and institutional background on biopharmaceutical innovation, with an eye toward highlighting the innovation challenges to which the open and collaborative model could respond. The next part discusses how open collaboration has operated in the context of software and of some other Web-based information projects and reviews results from empirical investigation of some prominent open and collaborative biomedical research projects. I then use these results as well as the theoretical literature to elucida