Você está na página 1de 313

Maastricht University School of Business and Economics

Academic Year 2011/2012

Course Book EBC2148
Innovation in Business and Economic Growth

This version of 11 April 2012

2009, 2010, 2011, 2012 F. Guadagno, B. Verspagen & H. Meijers
This work is licensed under the Creative Commons Attribution-Non-Commercial-Share Alike 3.0 Nether-
lands License. To view a copy of this license, visit http://creativecommons.org/licenses/by-nc-sa/3.0/nl/
or send a letter to Creative Commons, 171 Second Street, Suite 300, San Francisco, California 94105,
1 | P a g e

Table of Contents
1. Introduction .......................................................................................................................................... 2
2. Organization of the course.................................................................................................................... 3
3. Examination and grading ...................................................................................................................... 4
4. Guidelines for the hosted sessions ....................................................................................................... 4
5. Guidelines for the business case ........................................................................................................... 6
6. Required literature ................................................................................................................................ 8
7. Planning group ...................................................................................................................................... 9
8. Meetings ............................................................................................................................................... 9
Meeting 1 | Innovation, its knowledge context and the innovative firm ............................................... 10
Meeting 2 | General Purpose Technologies ........................................................................................... 11
Meeting 3 | Networks ............................................................................................................................. 12
Meeting 4 | Systems of Innovation, Importance of Learning and Interactions ..................................... 13
Meeting 5 | Spillovers, Agglomeration and Proximity ............................................................................ 14
Meeting 6 | Global value chains ............................................................................................................. 15
Meeting 7 | Dispersed R&D .................................................................................................................... 16
Meeting 8 | Markets for technology ...................................................................................................... 17
Meeting 9 | Globalization and Emerging Economies ............................................................................. 18
Meeting 10 | Business case presentation ............................................................................................... 19

2 | P a g e

1. Introduction
This is a course about innovation and about how firms, in interaction with their technological, economic
and social context, create, use and take advantage from it. In order to tackle these issues, the course
combines a theoretical perspective with highly practical elements on innovation processes in firms.
What distinguishes this course from other courses on innovation in your study program is its explicit fo-
cus on the broader economic context of innovation. In other words, this course does not just look at
what goes on within the firm or its immediate environment, but also considers what the impact of inno-
vation on the broad economy is.

This broad focus is evident from the main background document for the course: The Netherlands of
2040, released by the CPB Netherlands Bureau for Economic Policy Analysis. This report addresses the
question of how the Dutch will earn money in 2040. It focuses on who will produce and where, and de-
velops 4 possible scenarios of the future of the Dutch economy. By combining different degrees of la-
bour specialization (who produces) and geographical concentration (where), 4 scenarios are derived:
1. Talent Towns are characterized by geographical scattering and labour specialization. The econ-
omy is centered on the service sector and scattering is made possible by ICTs.
2. Cosmopolitan Centers are large cities populated by specialized workers and firms. The concen-
tration of different fields of expertise facilitates the development of new radical innovations.
3. Egalitarian Ecologies are medium sized cities hosting generalist workers. They are characterized
by high-quality but small size activities which result in moderate growth and low inequality.
4. Metropolitan Markets are large cities where economic activities are concentrated. Proximity al-
lows for high degrees of interaction among generalists in charge of exploiting the rise of a new break-
through innovation.
3 | P a g e

In this course, students will work in the tutorial group, and smaller subgroups of these. Each small sub-
group chooses one of the 4 scenarios, puts themselves in the shoes of an existent or imaginary firm fac-
ing the selected scenario, and discusses how the firm will make money in 2040 and how their manageri-
al decisions will impact on the society and economy at large. In order to prepare this business case, the
tutorial group will explore the innovation literature, whose insights and concepts will be ultimately ap-
plied in the case (guidelines for the business case are in section 5).

2. Organization of the course
The course starts with a lecture that sketches the broad relation between innovation and economic
growth. After this, all work is done in the tutorial group meetings (and the preparation that they re-
quire), and the smaller subgroups that will prepare the business case (these subgroups must organize
their own work).
Tutorial group meetings are composed of two parts (of roughly one hour each): the first is focused on
the assigned literature and the second on the application of the literature to the business cases. The first
part is a Seminar Session where 1 or 2 students host a seminar on the theories object of the meeting
(details on hosted sections are in section 4).
In the second part of the tutorial group meeting, the subgroups will first (about 20 minutes) brainstorm
about the application of the theoretical concepts developed in the first part of the meeting to their spe-
cific business case. In the remaining time, each subgroup (informally) discusses its ideas with the rest of
the tutorial group, and gives and receives feedbacks. Each subgroup can take advantage of the set of
questions that is found under each meeting (the questions for the business case). These questions are
also meant to guide the subgroups in the preparation of their business case.
The course is divided in 3 thematic parts:
Part 1. Meetings 1- 4 The knowledge context
Part 2. Meeting 5 - 7 Agglomeration/Dispersion
Part 3. Meeting 7 -10 Strategies to cope with the scenarios

This division will guide you to explore the main concepts of the course, and help you prepare the busi-
ness case. Subtopics in each of the three parts are highly interconnected. You should try to explore the
relations between the different theoretical concepts of each part, and propose interpretations on how
to connect them during the discussion.
The first part of the course deals with the knowledge context in which firms operate. This part sketches
the broad context of the innovating firm in the economy, i.e. the x-axis of the graph in the introduction.
The theoretical and practical issues that are reviewed here will help you to obtain a framework for think-
ing about your business case. The second part treats the geography of innovation and R&D. Since geog-
raphy is a crucial issue in the scenarios, this part is essentially grasping the specific issues addressed by
each scenario. In the final part, issues related to the organization of innovation within the firm are ex-
4 | P a g e

plored. This is where the link between the broad context (in terms of type of knowledge and geography)
and the specific business case is realized. As will be explained in more detail below, these 3 sections
should be clearly identifiable in the business case that you will present (although you are not expected
to use each and every single concept explored during the course).
It is essential that you study the reading material before each class, and the indicated sections of the
CPB report before the beginning of the course (more information on this is given in the section with
guidelines for the business case). Failure to do so will make it impossible to have an effective meeting,
will slow down your drafting of the business case and your preparation to the final exam, and will lower
your participation grade. In summary, you are expected to be well prepared for every meeting. This also
means that you should be ready to raise questions and critical remarks during the group meetings.

3. Examination and grading
The course examination consists of 4 elements:
A final written exam covering all required readings (but not the additional readings, additional
material presented during students presentations or the CPB report). The exam is based on
open questions. This part contributes 50% of your final grade.
The Business Case (written document and presentation). This part contributes 30% of your final
Hosted session (incl. presentation). This part contributes 10% of your final grade.
Participation (in the meetings). This part contributes 10% of your final grade.
Please note that the evaluation of your participation will be based on the quality of your comments. A
high quality comment is grounded in the theory and helpful for you and the other group members.

4. Guidelines for the hosted sessions
Meetings will be hosted by 1 or 2 students (each student will host once during the course). The number
of hosts per meeting is indicated under each meeting. You choose a topic and register as a host for one
meeting during the course opening lecture. The host prepares the meeting and chairs it. The host is en-
tirely responsible for the meeting. In this way, we take bachelor students seriously as self-guiding intel-
The hosted sessions form the first part of the meeting (as described above in Section 2), and follow a
standard format:
1. Opening by one of the host (this same student acts as the primary chair of the seminar).
2. A critical presentation of the literature by the seminar host (appr. 30 minutes).
3. Questions and clarifications by the other group members.
4. Discussion of additional points raised by the host or the seminar members.
5 | P a g e

5. Evaluation of seminar and host.
6. Closing.
The presentation by the host must include the required and additional literature assigned to the session
(see below), plus some additional material of your choosing. This can be in the form of pieces of litera-
ture (theoretical or empirical), business cases, newspapers or your own experience. Moreover, hosts are
also highly encouraged to come up with questions and inputs for class discussions.
The presentation will be evaluated in the following way:
1. 50% of the mark will be based on the content of your presentation. Consistency of the ar-
guments and validity of the additional material are the most important criteria for the eval-
uation of the content of the presentation.
2. 50% of the mark will depend on how you managed your seminar.
Even though during the hosted session discussions are primarily the responsibility of the host(s), the
other seminar members are also welcome to come up with questions and input for fruitful class discus-
sion. Dont forget that you will be asked to put the theoretical concepts in good practice in your business
case: having a clear vision of all implications of the theoretical concepts will be crucial at the stage of de-
veloping and writing your business case!
Tips for an interesting presentation:
With your presentation, you are responsible for your classmates learning proceeds. The best
way to make the most of it is to offer many inputs: the more you put it, the more you are likely
to get out of it.
Try to make it intellectually stimulating and original. Do not list or sum up the literature. Try to
evaluate the concepts you are presenting: plausibility, limitations...
Try to expand the theoretical insights of the literature with examples (e.g., cases, newspapers,
magazines, short videos, your experience...).
Find the link between the required and additional readings. This will help you organize your
Try to connect the theoretical concepts of your presentation with the other concepts explored
during the course.
Think of and show the theoretical and practical implications of the concepts on firms, society
(government, practitioners) and the economy at large.
Give a presentation that you yourself would find interesting and helpful in the understanding of
the literature and in its application to real life situations. Keep in mind that your audience is fa-
miliar with the pieces of literature proposed in the course manual so a presentation focused
mainly on them will probably be uninteresting.
In reading the proposed material, ask yourself what is the main point that the authors are trying
to make and how it is related to the issue at hand.
Tips on how to proceed with the drafting of your presentation:
6 | P a g e

Each of you reads carefully the proposed material and tries to devise an overall picture of what
the literature is telling you in a few sentences.
Confront your picture with the one of your fellow-host, and discuss it. Come up with an idea on
how to tell the story in your own words, identify the main questions and points that you should
address in your presentation.
Expand these points by searching additional material.
Raise questions on things that are unclear to you or you do not agree on. Include these points in
your presentation. Questions can be on theoretical, methodological or practical grounds.
After the presentation, the hosts chair the rest of the seminar. This part of the seminar can be com-
posed of clarification questions on the content of the literature by the other seminar members and of
the questions raised by the host.

5. Guidelines for the business case
The tutorial group is divided into subgroups of 3 or 4 students, leading to 4 subgroups in each tutorial
group. Each subgroup prepares a business case and hands it in on the day of the 10
group meeting
(01/06). A paper copy and an electronic version through SafeAssign (deadline 31/05 at 10.00) are both
required. The business case should be around 5,000 words, be properly structured (more details later on
in this section) and contain literature references.
These might be the initial steps to follow in order to prepare the business case:
1. Read these sections of the CPB report:
The first part of the summary: Summary, pp. 9-19
Chapter 8: Unfolding the scenarios, pp. 141- 163
2. Make sure you understand the characteristics and implications of all 4 scenarios.
3. Pick one of the 4 scenarios. Coordinate with the other subgroups in your tutorial group, so
that all scenarios are covered. Each group has to report on the choice of the scenario during
the first group meeting (24 April). Changing scenario will not be possible after the first meet-
4. Choose an existing or imaginary firm (in this second case, you can imagine to be entrepre-
neurs of a new start up). Each group has to report on the choice of the firm during the se-
cond group meeting (27 April).
As explained above, during each meeting, the subgroups have some time to discuss about their case and
also have the chance to get feedback from other students and the tutor. The quality of this discussion is
also considered when evaluating participation.
In the business case, you describe how your firm can take optimal advantage of the circumstances and
characteristics of the chosen scenario. You take the circumstances of the scenario as given, and describe
which strategic choices the firm makes to take advantage of these, and what role innovation (in all its
forms) takes in this strategic perspective of your firm.
7 | P a g e

In the description of each tutorial group meeting, you will find a set of questions that are aimed at guid-
ing you through the preparation of the case. You might find it useful to follow these questions in order
to check that you are not missing any important issue, but you dont need to specifically address each of
the questions, or discuss them in class and include them in your business case.
The business case should contain references to the theoretical concepts discussed in class. You can
choose how to structure your report but try to equilibrate theoretical and applied contents in such a
way that you dont just repeat the content of the papers but also dont lose the focus on the theoretical
underpinning of your company strategy.
In case your firm is an existent one, you can look at the company website and other public documents in
order to discuss the changes that are needed to take advantage of the opportunities of your scenario. In
case your firm is imaginary, make use of your creativity but be also practical and realistic.
Finally, a few specific guidelines for the business case:
1. Give an introductory description of the company you selected.
2. In case your firm operates in many sectors, you might select a particular sector/technology/
product of your interest or you might rearrange the sectors and invest in new sectors/ tech-
nologies if this fits with your scenario about the Netherlands in 2040.
3. Imagine and describe your new product/technology/set of activities in 2040. Describe also
the gradual process that over time will lead you to 2040. Embed them in one of the 4 sce-
narios you have chosen. Justify your choices.
4. Discuss the challenges and opportunities that this new product/technology/ set of activities
entails in terms of the concepts discussed in class and of the scenario that the firm faces.

8 | P a g e

6. Required literature
The following is a summary of the required literature for the written exam:

Lazonick, W. (2004), The innovative firm, In J. Fagerberg, D.C. Mowery and R. R. Nelson (eds), The Ox-
ford Handbook of Innovation, Oxford University Press: New York, p. 29-55.
Jovanovic, B., and Rousseau, P.L. (2005), General Purpose Technologies, In P. Aghion and S. N. Durlauf
(eds), Handbook of Economic Growth, Elsevier, Volume 1, pp. 1181-1224.
Powell, W.W., and Grodal, S. (2004), Networks of Innovators, In J. Fagerberg, D.C. Mowery and R. R.
Nelson (eds), The Oxford Handbook of Innovation. Oxford University Press: New York, p. 56-85.
Edquist, C. (1997), Systems of Innovation: Perspectives and Challenges, In J. Fagerberg, D. C. Mowery, R.
R. Nelson (eds), The Oxford Handbook of Innovation. Oxford University Press: New York, p. 181-208.
Audretsch, D. B., & Feldman, M. P. (1996), R & D Spillovers and the Geography of Innovation and Pro-
duction, The American Economic Review, 86(3): 630-640.
Ernst, D and Kim, L (2002), Global production networks, knowledge diffusion, and local capability for-
mation, Research Policy, 31: 14171429.
Von Zedwitz and Gassmann (2002), Market versus technology drive in R&D internationalization: four dif-
ferent patterns of managing research and development, Research Policy, 31(4): 569-588.
Arora, A., Fosfuri, A., and Gambardella, A., 2001, Markets for Technology and their Implications for Cor-
porate Strategy, Industrial and Corporate Change, 10(2): 417-449.
Lee, K. & C. Lim, 2001, Technological regimes, catching-up and leapfrogging: findings from the Korean
industries, Research Policy, 30: 459-483.

9 | P a g e

7. Planning group
The planning group of this course consists of:
Prof Bart Verspagen, E-mail: b.verspagen@maastrichtuniversity.nl
Francesca Guadagno, E-mail: f.guadagno@maastrichtuniversity.nl

8. Meetings
This is the overview of the 10 groups meeting.
Meeting Date Topic Agenda
1 24/4 Knowledge and Innovation Students presentations
27/4 General Purpose Technologies Students presentations
1/5 Network Students presentations
4 4/5 Systems of Innovation Students presentations
8/5 Agglomeration Students presentations
11/5 Global Value Chains Students presentations
15/5 Dispersed R&D Students presentations
8 22/5 Markets for Technology Students presentations
25/5 Globalization and Emerging Economies Students presentations
10 1/6 - Business Case Presentations

Under each meeting, you will also find details on the literature, questions for discussion, food for
thought about your business case, and the structure of the meeting.

10 | P a g e

Meeting 1 | Innovation, its knowledge context and the innovative firm
In the first half of this meeting, two students host a session around innovation and the innovative firm.
The required literature deals with the characteristics and contexts of innovating firms in a comparative-
historical perspective. The additional reading, instead, focuses on the knowledge- and industrial context
in which innovation takes place.
Required reading:
Lazonick, W. (2004), The innovative firm, In J. Fagerberg, D.C. Mowery and R. R. Nelson (eds), The Ox-
ford Handbook of Innovation, Oxford University Press: New York, p. 29-55.
Additional reading:
Malerba, F. and L. Orsenigo (1997), Technological Regimes and Sectoral Patterns of Innovative Activities,
Industrial and Corporate Change, 6(1): 83- 117.
Questions for the business case:
1. Strategize, finance and organize your resources.
2. Think of the resources you have in your firm in comparison to those you need for your fu-
ture success.
3. Discuss the status and role of your capabilities.
4. Think of the social characteristics of your firm in relation with your predictions on how the
world will be in 2040. How will the social conditions of 2040 impact on your firms innova-
tiveness and success?
5. In which sector is your firm? How would you classify it (in terms of the Malerba and
Orsenigo Schumpeterian patterns)?
6. Which are the knowledge characteristics of this sector?
7. How do you benefit from this setting?
8. Which firms characteristics are better suited for this type of knowledge? Do you think that
your firm possesses these characteristics? If not, how do you plan to acquire them?

11 | P a g e

Meeting 2 | General Purpose Technologies
In the first half of this meeting, one student hosts the session around the topic of GPT. The CPB report
refers to the arrival of a new GPT and discusses its opportunities and implications for business, the
economy and the society. In this meeting, we will therefore try to understand in dept what general pur-
pose technologies are.
Required reading:
Jovanovic, B., and Rousseau, P.L. (2005), General Purpose Technologies, In P. Aghion and S. N. Durlauf
(eds), Handbook of Economic Growth, Elsevier, Volume 1, pp. 1181-1224.
Additional reading:
Youtie, J., M. Iacopetta and S. Graham, 2008, Assessing the nature of nanotechnology: can we uncover
an emerging general purpose technology?, Journal of Technology Transfer, 33(3): 315-329.
Questions for the business case:
1. Given your scenario, do you forecast the arrival of a new GPT before 2040 or around that peri-
2. Would your company be involved in/affected by it?
3. If you think that a GPT will disrupt your industry, do you think that your firm is technologically
and organizationally prepared for that?
4. How will your firm benefit from it (if it can)?
5. What strategies, competencies and firm organizational practices do you intend to use to deal
with the new (or old) GPT? Do you possess them already? If not, how do you plan to build them?

12 | P a g e

Meeting 3 | Networks
In the first half of this meeting, two students host the session around the topic of networks. Networks
are a means to pool and exchange knowledge and to jointly create new knowledge and skills. In scien-
tific and technological fields where technological change is fast and sources of knowledge dispersed,
network become pivotal for firms, society and the economy.
Required reading:
Powell, W.W., and Grodal, S. (2004), Networks of Innovators, In J. Fagerberg, D.C. Mowery and R. R.
Nelson (eds), The Oxford Handbook of Innovation. Oxford University Press: New York, p. 56-85.
Additional reading:
Verspagen, B., and Duysters, G. (2004), The small worlds of strategic technology alliances, Technovation,
24(7): 563-571.
Questions for the business case:
1. How would you describe the network in which your firm is embedded?
2. Which is your position in the network and how do you benefit/suffer from it? How would you
improve your position in case you want to?
3. Given your sector and technology, are this structure and its properties beneficial to your firm?
4. If yes, why? If not, which kind of network would be ideal for your firms success?

13 | P a g e

Meeting 4 | Systems of Innovation, Importance of Learning and Interactions
In the first part of the meeting, two students host the session on the importance of learning and interac-
tions for firms, and explore the concept of system of innovation as the locus where learning and interac-
tion take place. Because this is the last meeting on the first part of the course, we will also spend some
time in putting all pieces of literature together and depict an overall picture of the innovation and
knowledge context for firms.
Required reading:
Edquist, C. (1997), Systems of Innovation: Perspectives and Challenges, In J. Fagerberg, D. C. Mowery, R.
R. Nelson (eds), The Oxford Handbook of Innovation. Oxford University Press: New York, p. 181-208.
Additional reading:
Mowery, D. C., and Rosenberg, N. (1992), The U.S. National Innovation System, in R. R. Nelson (ed),
National Innovation Systems: A Comparative Analysis, Oxford University Press: New York, p. 29-75.
Questions for the business case:
Describe the innovation system in which your firm is embedded:
1. Describe your system of innovation and its boundaries.
2. Given your sector and GPT, is the innovation system in which your firm is embedded favorable
to your business? If yes, why? If not, how should it improve?
3. Do you interact with some universities? Which one? What is its role and contribution for your
business? How do you manage this relationship?
4. Which policies are there to favour such interactions?
14 | P a g e

Meeting 5 | Spillovers, Agglomeration and Proximity
In the first half of this meeting, two students host the session on geography of innovation. Notwith-
standing the long tradition in economic literature, this topic is still very relevant. In fact, despite the
economy is becoming more and more global, empirical evidence shows that innovative activity is still
geographically concentrated. This has evident implications for people, governments and firms.
Required reading:
Audretsch, D. B., & Feldman, M. P. (1996), R & D Spillovers and the Geography of Innovation and Pro-
duction, The American Economic Review, 86(3): 630-640.
Additional reading:
Bathelt, H., Malmberg, A., and Maskell, P. (2004), Clusters and knowledge: local buzz, global pipelines
and the process of knowledge creation, Progress in Human Geography, 28: 31-56.
Questions for the business case:
1. Where is your firm located? Describe the main features (pros and cons) of the region. Refer
to the strength of the agglomeration and dispersion forces mentioned in the CPB report and
to how your location impacts on living conditions of your employees.
2. How important are knowledge spillovers given your sector/technology and its relative stage
of development?
3. Given your technology, innovation system and network, do you think that geographical
proximity to your network partners is essential? Why?
4. If possible, would you relocate your firm somewhere else? If yes, where? If not, why?
5. If you are in a cluster, would you define your cluster more as a centre of diversity or as a
cluster of homogeneity? Which of the two types of cluster would be more useful to your
business? Why?
6. How do the global and local dimensions of knowledge creation and innovation intersect in
your business?
15 | P a g e

Meeting 6 | Global value chains
In the first half of this meeting, one student hosts the session around the topic of Global Value Chains.
The increased global competition has recently resulted in global production networks. These new inter-
national structure of production and innovation has clear implications for knowledge diffusion and local
capabilities in both developed and developing countries.
Required reading:
Ernst, D and Kim, L (2002), Global production networks, knowledge diffusion, and local capability for-
mation, Research Policy, 31: 14171429.
Additional reading:
Humphrey, J., and Schmitz, H. (2002), How does insertion in global value chains affect upgrading in in-
dustrial clusters?, Regional studies, 36(9): 1017-1027.
Questions for the business case:
1. Is your firm part of any global production network? If yes, which is its role/position?
2. How does your firm benefit from it?
3. If you think that your firm is not benefiting from it, how would you create and manage your
global production network and which benefits do you expect from it?

16 | P a g e

Meeting 7 | Dispersed R&D
In the first half of this meeting, one students hosts the session around the topic of geographically dis-
persed R&D. Along the same line of the previous meeting, this topic focuses on the new trends in the
organization of R&D activities. Many studies, in fact, have recently pointed out that global R&D is in-
creasingly more organized in flat independent networks of mutually supportive and globally spread facil-
ities. Opportunities and challenges arise for firms and governments.
Required reading:
Von Zedwitz and Gassmann (2002), Market versus technology drive in R&D internationalization: four dif-
ferent patterns of managing research and development, Research Policy, 31(4): 569-588.
Additional reading:
Criscuolo, P. (2005), On the road again: Researcher mobility inside the R&D network, Research Policy,
34(9): 1350- 1365.
Questions for the business case:
1. Where is your R&D located? How would classify your firm using the Von Zedwitz and Gassmann
2. Is this the optimal organization given your technological status?
3. How do you manage your R&D organizational structure?
4. Do you forecast any trend in your R&D internationalization strategy? Why?
17 | P a g e

Meeting 8 | Markets for technology
In the first part of the meeting one student hosts a session about the importance of learning and inter-
actions for firms and explore the concept of system of innovation as the locus where learning and inter-
action take place.
Required reading:
Arora, A., Fosfuri, A., and Gambardella, A., 2001, Markets for Technology and their Implications for Cor-
porate Strategy, Industrial and Corporate Change, 10(2): 417-449.
Additional reading:
Gambardella, A., Harhoff, Q., and Verspagen, B. (2008), The Value of European Patents, European Man-
agement Review, 5: 69-84.
Questions for the business case:
1. Do you see a market for technology in your sector?
2. If you think there is, is your company involved in/affected by it?
3. How does it benefit from it (if it can)?
4. Which role does your firm play in the market?
5. Which competencies and firm organizational practices are needed in a market for technology?

18 | P a g e

Meeting 9 | Globalization and Emerging Economies
In the first part of the meeting one student hosts a session on the opportunities opened up by globaliza-
tion and the rise of emerging economies. On the one hand, by acquiring capabilities, emerging econo-
mies are entering the global market and competing with advanced economies firms. On the other hand,
emerging and developing countries are potential new markets to both domestic and international pro-
Required reading:
Lee, K. & C. Lim, 2001, Technological regimes, catching-up and leapfrogging: findings from the Korean
industries, Research Policy, 30: 459-483.
Additional reading:
Prahalad, C. K. (2005), The Market at the Bottom of the Pyramid, in C. K. Prahalad, The Fortune at the
Bottom of the Pyramid: Eradicating Poverty through Profits. Wharton School Publishing: Upper Saddle
River, NJ, p. 3-22.
Questions for the business case:
1. How can your firm benefit from the opportunities offered by globalization?
2. Is your firm facing competition from emerging markets?
3. How does your firm react to this new source of competition?
4. Are there possibilities for cooperation with emerging economies firms?
5. Given your scenario and sector, do you see fortunes at the Bottom of the Pyramid? If yes, how
would you profit from this new market?

19 | P a g e

Meeting 10 | Business case presentation
In this meeting the 4 groups present their business case. Each presentation should take around 20
minutes, including Q&A. Hand the electronic copy of the business case through SafeAssign (deadline
31/05 at 10 am) and the hard copy at the beginning of the 10
tutorial. Be aware that two versions are
the same because only the electronic version will be evaluated!

p a r t i
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 27
Introduction to Part I
Most innovations occur in Wrms or other types of organizations.
The contributions in this section survey our current knowledge on
the organizational structure and context of the process of innov-
ation. Chapter 2, by Lazonick, provides a historical perspective on
the development of innovative Wrms, fromthe small and medium-
sized Wrms of the First Industrial Revolution through the multi-
divisional diversiWed industrial Wrms of the US and Japan in the
twentieth century to the current debate on the New Economy
and network-based business models. Powell and Grodal deal more
extensively with the role of networks in innovation in the subse-
quent chapter. Chapter 4, by Pavitt, discusses innovation processes
within Wrms, and uses an extensive survey of the relevant literature
to provide an analytical perspective on the factors aVecting the
performance and management of innovationwithin the large Wrm.
A complementary chapter by Lam (Chapter 5) focuses on Wrms
experiences with organizational innovation. Finally, Chapter 6 by
Smith deals with an indispensable prerequisite for the study of
innovation, the measurement of innovation-related activities, par-
ticularly in Wrms.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 28
c h a p t e r 2
wi lli am lazoni ck
2. 1 Introduction
What makes a Wrm innovative? How have the characteristics of innovative Wrms
changed over time? To address these questions, one requires a conceptual framework
for analyzing how a Wrm transforms productive resources into goods and services
that customers want at prices they can aVord. To make this productive transform-
ation, a Wrm must engage in three generic activities: strategizing, Wnancing, and
organizing. The types of strategy, Wnance, and organization that support the innov-
ationprocess change over time andcanvary markedly across industrial activities and
institutional environments at any point in time. The innovative Wrm must, there-
fore, be analyzed in comparativehistorical perspective. This chapter presents and
illustrates a framework for analyzing the social conditions of innovative enterprise
in the comparativehistorical experiences of the advanced economies.
Section 2.2 builds upon prominent theories of the innovative Wrm to derive the
social conditions of innovative enterprise framework. Section 2.3 focuses on the
regional agglomerations of capabilities, now known as Marshallian industrial
districts, that, by the late nineteenth century, had enabled Britain to emerge as
the worlds Wrst industrial nation. Section 2.4 provides a perspective on the emer-
gence and growth of the US managerial corporation that propelled the US economy
to international industrial leadership during the Wrst half of the twentieth century.1
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 29
Over the past fewdecades, the greatest challenges to the US managerial corporation
have come from Japan. Section 2.5 identiWes the social conditions of innovative
enterprise that have characterized the Japanese model, while Section 2.6 outlines the
distinctive characteristics of the US NewEconomy Wrmthat has gained competitive
advantage in a number of critical product markets in the information and commu-
nication technology (ICT) industries. Section 2.7 draws some general conclusions
from this essays comparativehistorical perspective concerning strategy, Wnance,
and organization in the innovative Wrm, and the methodology for studying these
2. 2 Social Conditions of
Innovative Enterprise
Firms strategize when they choose the product markets in which they want to
compete andthe technologies withwhichthey hope tobe competitive. Firms Wnance
when they make investments to transform technologies and access markets that can
only be expected to generate revenues sometime in the future. Firms organize when
they combine resources in the attempt to transform them into saleable products.
To strategize, Wnance, and organize is not necessarily to innovate. By deWnition,
innovation requires learning about how to transform technologies and access
markets in ways that generate higher quality, lower cost products. Learning is a
social activity that renders the innovation process uncertain, cumulative, and
collective (OSullivan 2000b). The innovation process is uncertain because, by
deWnition, what needs to be learned about transforming technologies and accessing
markets can only become known through the process itself. By investing in learning,
an innovative strategy confronts the uncertain character of the innovation process.
The innovationprocess is cumulative whenlearning cannot be done all at once; what
is learned today provides a foundation for what can be learned tomorrow. Invest-
ments in cumulative learning, therefore, require sustained, committed Wnance. The
innovation process is collective when learning cannot be done alone; learning
requires the collaboration of diVerent people withdiVerent capabilities. Investments
in collective learning, therefore, require the integration of the work of these people
into an organization.
What is the theory of the Wrm that can comprehend how strategizing, Wnancing,
and organizing can support the innovation process? Over the past century, the
theoretical eVorts of economists have focused mainly on the optimizing Wrm
rather than the innovating Wrm. The optimizing Wrm takes as given technological
30 william lazonick
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 30
capabilities and market prices (for inputs as well as outputs), and seeks to maximize
proWts on the basis of these technological and market constraints. In sharp contrast,
in the attempt to generate higher quality, lower cost products than had previously
been available, and thus diVerentiate itself from competitors in its industry, the
innovating Wrmseeks to transformthe technological and market conditions that the
optimizing Wrm takes as given constraints. Hence, rather than constrained opti-
mization, the innovating Wrm engages in what I call historical transformation, a
mode of resource allocation that requires a theoretical perspective on the processes
of industrial and organizational change (Lazonick 2002a).
The distinction between the innovating and optimizing Wrm is implicit in the
work of Alfred Marshall, whose Principles of Economics, published in eight editions
between 1890 and 1920, placed the theory of the Wrm at the center of economic
analysis. Although Marshalls followers used his arguments to construct the theory
of the optimizing Wrm that remains entrenched in economics textbooks, Marshall
(1961: 315) himself displayed considerable insight into the dynamics of the innovat-
ing Wrm, as revealed in the following passage:
An able man, assisted by some strokes of good fortune, gets a Wrm footing in the trade, he
works hard and lives sparely, his own capital grows fast, and the credit that enables him to
borrow more capital grows still faster; he collects around him subordinates of more than
ordinary zeal and ability; as his business increases they rise with him, they trust him and he
trusts them, each of them devotes himself with energy to just that work for which he is
specially Wtted, so that no high ability is wasted on easy work, and no diYcult work is
entrusted to unskillful hands. Corresponding to this steadily increasing economy of skill, the
growth of his Wrm brings with it similar economies of specialized machines and plants of all
kinds; every improved process is quickly adopted and made the basis of further improve-
ments; success brings credit and credit brings success; success and credit help to retain old
customers and to bring new ones; the increase of his trade gives him great advantages in
buying; his goods advertise one another and thus diminish his diYculty in Wnding a vent for
them. The increase of the scale of his business increases rapidly the advantages which he has
over his competitors, and lowers the price at which he can aVord to sell.
What then constrains the growth of such a Wrm? In Industry and Trade, published
in 1919, Alfred Marshall acknowledged that over the previous decades the large-scale
enterprise had become dominant in advanced nations such as the United States and
Germany. He invoked, however, the aphorism, shirtsleeves to shirtsleeves in three
generations (Marshall 1961: 621) to explain the limit to the growth of the Wrm that
would prevent a small number of large Wrms from dominating an industry. An
owner-entrepreneur of exceptional ability would found and build a successful Wrm.
In the second generation, control would pass to descendants who could not be
expected to have the capabilities or drive of the founder, and as a result the Wrm
would grow more slowly or even stagnate. The third generation would lose touch
with the innovative legacy of the Wrst generation, and the Wrmwould wither away in
the face of new entrepreneurial competition.
the innovative firm 31
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 31
Writing in the Wrst decades of the twentieth century, Joseph Schumpeter (1934)
also focused on the innovative entrepreneur who, by creating new combinations
of productive resources, could disrupt the circular Xow of economic life as
conditioned by given circumstances. In eVect, Schumpeter was arguing that,
through entrepreneurship, which he called the fundamental phenomenon of
economic development, innovating Wrms could challenge optimizing Wrms, and
thereby drive the development of the economy. In 1911, when he Wrst published
The Theory of Economic Development (in German), Schumpeter, like Marshall,
viewed the innovative Wrm as the result of the entrepreneurial work of an extra-
ordinary individual. Over the subsequent decades, however, as Schumpeter ob-
served the actual development of the leading economies, he came to see the large
corporation as the innovating Wrm, engaged in what he called a process of creative
destruction; the creation of new modes of productive transformation destroyed
existing modes that had themselves been the result of innovative enterprise in
the past.
In Capitalism, Socialism, and Democracy, Wrst published in 1942, Schumpeter
(1950: 118, 132) argued that technological progress tends, through systemization
and rationalization of research and management, to become more eVective and
sure-footed as it is undertaken as the business of teams of trained specialists who
turn out what is required and make it work in predictable ways. In a series of major
works, Alfred Chandler (1962, 1977, 1990) documented the rise of the managerial
corporation in the United States fromthe last decades of the nineteenth century, the
evolution of its multidivisional structure from the 1920s, and the emergence of
managerial enterprise in Britain and Germany. In The Theory of the Growth of the
Firm, Wrst published in 1959, Edith Penrose (1995) conceptualized the modern
corporate enterprise as an organization that administers a collection of human
and physical resources. People contribute labor services to the Wrm, not merely as
individuals, but as members of teams who engage in learning about how to make
best use of the Wrms productive resourcesincluding their own.
At any point in time, this learning endows the Wrm with experience that gives it
productive opportunities unavailable to other Wrms, even in the same industry, that
have not accumulated the same experience. The accumulation of innovative experi-
ence enables the Wrm to overcome the managerial limit that in the theory of the
optimizing Wrm causes the onset of increasing costs and constrains the growth of
the Wrm(Penrose 1995: chs. 5, 7, and 8). The innovating Wrmcantransfer and reshape
its existing productive resources to take advantage of new market opportunities.
Each move into a new product market enables the Wrmto utilize unused productive
services accumulated through the process of organizational learning. These unused
productive services can provide a foundation for the growth of the Wrm, through
both in-house complementary investments in new product development and the
acquisition of other Wrms that have already developed complementary productive
32 william lazonick
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 32
From the 1980s many business school academics, working in the strategy area,
cited Penroses 1959 book as an intellectual foundation for a resource-based view
of the Wrm. Resource-based theory focused on the characteristics of valuable re-
sources that one Wrm possessed and that competitor Wrms found it diYcult to
imitate. Resource-based theory, however, provided no perspective on why and
how some Wrms rather than others accumulated valuable and inimitable resources,
or indeed what made these resources valuable and inimitable (see Lazonick 2002a).
Independently of the resource-based perspective, however, Richard Nelson and
Sidney Winter (1982) fashioned a theory of the persistence of the large industrial
corporation based on organizational capabilities, characterized by tacit knowledge
and embedded in organizational routines, thus adding a cumulative dimension to
the theory of the Wrm. Drawing on a highly eclectic set of sources from a number of
disciplines, Bruce Kogut and Udo Zander (1996: 502) argued that [f]irms are
organizations that represent social knowledge of coordination and learning, thus
emphasizing the collective dimension in the theory of the Wrm.
In Why Do Firms DiVer, and HowDoes It Matter? Nelson (1991: 72) argued that
it is organizational diVerences, especially diVerences in abilities to generate and
gain frominnovation, rather thandiVerences incommandover particular technolo-
gies, that are the source of durable, not easily imitable, diVerences among Wrms.
Particular technologies are much easier to understand, and imitate, than broader
Wrm dynamic capabilities. David Teece, Gary Pisano, and Amy Shuen (1997: 516)
deWneddynamic capabilities as the Wrms ability to integrate, build, andreconWgure
internal and external competences to address rapidly changing environments. They
also argued that the Wrms strategy entails choosing among and committing to long-
term paths or trajectories of competence development (Teece at al. 1997: 524).
Whereas the Wrms asset positions determine its competitive advantage at any
point in time and its evolutionary path constrains the types of industrial activities
in which a Wrm can be competitive, its organizational processes transform the
capabilities of the Wrm over time.
While Teece et al. (1997: 519) stressed the importance of learning processes that are
intrinsically social and collective, their dynamic capabilities perspective lacks
social content. The framework does not ask what types of people are able and willing
to make the strategic investments that can result in innovation, how these strategic
decision makers mobilize the necessary Wnancial resources, and how they create
incentives for those people within the Wrms hierarchical and functional division of
labor to cooperate in the implementation of the innovative strategy. These questions
about the roles of strategizing, Wnancing, and organizing in the innovating Wrm are
at the center of what Mary OSullivan and I have called the social conditions of
innovative enterprise perspective (Lazonick and OSullivan 2000; OSullivan
2000b; Lazonick 2002b).
This perspective asks how and under what conditions the exercise of strategic
control ensures that the enterprise seeks to grow using the collective processes and
the innovative firm 33
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 33
along the cumulative paths that are the foundations of its distinctive competitive
success. The perspective emphasizes the role of human agency in determining
whether and how the enterprise accumulates innovative capability, and thus adds
an explicitly social dimension to work on dynamic capabilities. SpeciWcally,
strategic control determines how strategic decision makers choose to build on
asset positions; Wnancial commitment determines whether the enterprise will
have the resources available to it to persist along an evolutionary path to the
point where its accumulation of innovative capability can generate Wnancial returns;
and organizational integration determines the structure of incentives that charac-
terize organizational processes that can transformindividual actions and individ-
ual capabilities (including those of strategic managers) into collective learning.
Of central importance to the accumulation and transformation of capabilities in
knowledge-intensive industries is the skill base in which the Wrminvests in pursuing
its innovative strategy. Within the Wrm, the division of labor consists of diVerent
functional specialties and hierarchical responsibilities. At any point in time a
Wrms functional and hierarchical division of labor deWnes its skill base. In the
eVort to generate collective and cumulative learning, those who exercise strategic
control can choose how to structure the skill base, including how employees move
around and up the functional and hierarchical division of labor over the course of
their careers. At the same time, however, the organization of the skill base will be
constrained by both the particular learning requirements of the industrial activities
in which the Wrm has chosen to compete and the alternative employment oppor-
tunities of the personnel for whom the Wrm must compete.
In cross-national comparative perspective, the skill base that enterprises employ
to transform technologies and access markets can vary markedly even in the same
industrial activity during the same historical era, with diVerent innovative out-
comes. Precisely because innovative enterprise depends on social conditions, the
development and utilization of skill bases that occur in one institutional environ-
ment may not, at a point in time at least, be possible in another institutional
environment. Moreover, even within the same industry and same nation, dynamic
capabilities that yielded innovative outcomes inone historical era may become static
capabilities that inhibit innovative responses in a subsequent historical era.
The innovative Wrm requires that those who exercise strategic control be able
to recognize the competitive strengths and weaknesses of their Wrms existing
skill base and, hence, the changes in that skill base that will be necessary for an
innovative response to competitive challenges. These strategic decision makers must
also be able to mobilize committed Wnance to sustain investment in the skill base
until it can generate higher quality, lower cost products than were previously
available. As the following comparativehistorical syntheses illustrate, given stra-
tegic control and Wnancial commitment, the essence of the innovative Wrm is the
organizational integration of a skill base that can engage incollective and cumulative
34 william lazonick
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 34
2. 3 The British Industrial District
In last half of the nineteenth century, Britain became known as the workshop of the
world. Britains position in the world economy owed much to its mercantile power,
developed through global commerce and related wars with other leading nations
over the previous centuries. Mercantilism gave British industry access to world
product markets and sources of raw materials, but it was the transformation of
production from the late eighteenth century that enabled Britain to emerge as the
worlds leading (and indeed Wrst) industrial nation.
In the late nineteenth century, Britains productive power resided in industrial
districts that, for building machines and using them to manufacture products as
varied as cloth and ships, possessed an immense accumulation of capabilities.
Beyond evening courses at local mechanics institutes, formal vocational or pro-
fessional education played no role in the development of Britains skilled labor force.
Nor did British industry make use of corporate, university, or government research
labs to develop new technology. Regionally based on-the-job apprenticeship ar-
rangements, through which craft workers passed on their skills to the next gener-
ation, constituted in eVect the national innovation system of the worlds Wrst
industrial economy.
What accounts for the importance of the craft worker for Britains industrial
leadership? While the mechanization of the factory was a central feature of the
British industrial revolutionand in its time a wonder of the worldthe standard-
ization of materials and the automation of machinery that British industry achieved
during its industrial revolution were, in historical retrospect, incipient. Skilled craft
workers maintained critical roles in keeping imperfect machinery in motion and
ensuring high levels of throughput of work-in-progress made from imperfect
materials. Within the Wrm, experienced workers typically were responsible for
training younger workers in the craft, supervising their work, and coordinating
the Xow of work through the production process. In some industries, the central
employment relation took the form of an internal subcontract system; for example,
in the cotton spinning industry, employers paid piece-rates to senior workers,
known as self-acting minders, who in turn trained, supervised, and paid time
wages to junior workers known as piecers and doVers. In the metalworking
industries, specialized workers such as turners and Wtters were generally clas-
siWed as engineers, an appellation that in the British context signiWed membership
in the labor aristocracy of skilled production workers (Lazonick 1990: chs. 16).
The localized, on-the-job character of skill formation was the major factor
underlying the growth of industrial districts that made use of particular specialized
craft skills. As Alfred Marshall (1961: 271) famously put it, in the British industrial
districts mysteries of the trade become no mysteries; but are as it were in the air. In
periods of strong product-market demand, the ready availability of specialized craft
the innovative firm 35
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 35
labor induced newspecialized manufacturing Wrms, often founded by craft workers
themselves, toset upinthese districts. The growthof a district inducedother Wrms to
invest in regionally speciWc communication and distribution facilities for the supply
of materials, the transfer of work-in-progress across vertically specialized Wrms, and
the marketing of output.
Regional concentration encouraged vertical specialization, which in turn eased
Wrm entry into a particular speciality, thus resulting in high levels of horizontal
competition. Firms could be owned and managed by the same people; there was no
need to invest in the types of managerial organization that by the late nineteenth
century were becoming central to the growth of Wrms in the United States, Germany,
and Japan. In the industrial districts, economies of scale were, as Marshall argued,
external, rather than internal, to the Wrm.
As producers and users of machinery, craft workers constituted the prime source
of innovation in a particular region. Over time they devised incremental techno-
logical and organizational improvements that, through the local trade press (includ-
ing workers newspapers) as well as the movement of workers (especially trained
apprentices) to newemployers, diVusedacross Wrms inthe district. Some specialized
engineering Wrms distinguished themselves through in-house learning. But even the
strongest of these Wrmsfor example, the textile machinery Wrm of Platt Brothers
based in Oldhamdid no in-house R&D, and from the last half of the nineteenth
century generated no signiWcant technological innovations. Their strength resided
in their employment of craft labor that could Xexibly produce customized machines
for many diVerent types of users (Farnie 1990).
The importance of localized craft labor to the innovative capabilities of local Wrms
meant that it was the industrial district, and often a particular townwithin a district,
not the individual Wrm, that constituted the learning entity. At the Wrm level, craft
workers made countless strategic decisions to improve products and processes.
For both individual Wrms and the district as a whole, the Wxed costs of developing
this source of innovationwere, in historical and comparative perspective, low. At the
same time, craft-orientedemployment systems encourageda highlevel of utilization
of the plant and equipment in place. Union bargains protected the tenure and
remuneration of senior workers who, paid by the piece, were willing to work long,
hard, and steady. The inducement for junior workers, typically paid time wages, was
that they could eventually join the aristocracy of labor. There is evidence that, within
an industrial district, those localities in which negotiated piece-rate bargains shared
productivity gains between employees and employers on a stable and equitable basis
saw the fastest growth in productivity and market share (Lazonick 1990: chs. 35;
Huberman 1996).
Basedoncraft organization, Britishindustrial districts were highly innovative (see
also Bruland and Mowery, this volume). The fact that it was the industrial district as
a whole, rather than the individual enterprise within it, that was the innovating
entity gave rise to the notion that diVerences among Wrms in an industrial activity
36 william lazonick
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 36
were unimportant to economic performance, and indeed that they could all be
characterized by depicting a representative Wrm that optimized subject to given
technological and market constraints. Within the Marshallian perspective, even
innovation at the district level did not require strategic direction, since the industrial
arts were in the air. Indeed, Marshall (1919: 6001) described the organization of
the Lancashire cotton textile industry, with its high degrees of horizontal competi-
tion and vertical specialization, as perhaps the present instance of concentrated
organisation mainly automatic. Yet just as Marshall was writing these words, the
cotton textile industry, which had accounted for one-quarter of British exports on
the eve of World War I, entered into a long-run decline from which it never
recovered, and the other major British industrial districts suVered a similar fate
(Elbaum and Lazonick 1986).
Fromthe late 1970s, however, the notion of the Marshallian industrial district as
a driver of innovative enterprise saw an academic resurgence, based on the rapid
growth during the 1960s and 1970s of many highly specialized and localized districts
in what became known as the Third Italy (Brusco 1982; Sabel 1982; Becattini 1990).
On the basis of this experience, a number of US academics, headed by Charles Sabel,
Michael Piore, andJonathanZeitlin, posited a newmodel of Xexible specialization
as an alternative to mass production on the US corporate model (Piore and
Sabel 1984; Sabel and Zeitlin 1985). The industrial activities of the districts of the
Third Italy focused on, among other things, textiles, footwear, and light machinery,
just as the British districts had done. Large numbers of vertically specialized propri-
etary Wrms in which craft labor was a prime source of competitive advantage
populated each industrial activity, and many entrepreneurs had previously been
craft workers.
There were, however, two important diVerences between the British industrial
districts that Marshall had observed in the late nineteenth century and those that
experienced rapid growth in the Third Italy more recently. The Wrst diVerence was
the extent to which in Italy collective institutions supported the innovative activities
of small Wrms. Sebastiano Brusco (1992) has emphasized the importance of the red
local governments in Emilia-Romagna in promoting policies to support the activ-
ities of small enterprises, and in particular in facilitating cooperatives that provided
these Wrms with real services related to business administration, marketing, and
training. While consumer cooperatives sprung up in the British industrial districts
of the late nineteenth century, producer cooperatives were rare. The second diVer-
ence, which became more evident in the 1990s, was the extent to which, in some
districts and in some industries, leading Wrms could emerge, drawing on the
resources of the industrial districts while, through their own internal growth,
transforming the innovative capability of the districts (see, for example, Belussi
1999). In contrast, when in the Wrst half of the twentieth century competitive
challenges confronted the British industrial districts, dominant Wrms failed to
emerge to lead a restructuring process.
the innovative firm 37
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 37
2. 4 The US Managerial Corporation
Marshall locatedthe limits to the growthof the Wrminthe problemof succeeding the
original owner-entrepreneur. In The Theory of Economic Development, Schumpeter
(1934: 156) concurred using the same aphorism as Marshall, literally clothed in
diVerent garb and speciWcally identiWed as a US phenomenon: An American
adage expresses it: three generations from overalls to overalls. Critical to this
perspective were two assumptions: Wrst, that the entrepreneur was the essence of
the innovative Wrm, and second, that the integration of ownership and control was a
necessary condition for entrepreneurship. Notwithstanding his own important
study of comparative trends in industrial organization published in Industry and
Trade, Marshall (1919) declined to recognize, as ultimately Schumpeter did, that the
problem of innovative succession could be resolved by the separation of ownership
and control.
Taking place during the same decades in which Marshall wrote his inXuential
books, the separation of share ownership from strategic control was the essence of
what Chandler (among others) would call the managerial revolution in American
business. During this period Germany and Japan also experienced managerial
revolutions (Chandler 1990; Chandler et al. 1997; Morikawa and Kobayashi 1986;
Morikawa 1997). Many British Wrms, especially in the science-based chemical and
electrical industries also made investments in managerial organization, but in such a
constrained manner that it can hardly be said that a managerial revolution occurred
in Britain during the Wrst half of the twentieth century (Hannah 1983; Lazonick 1986;
Chandler 1990; Owen 2000).
In the United States, the managerial revolution began in the 1890s in industries
such as steel, oil reWning, meatpacking, tobacco, agricultural equipment, telecom-
munications, and electric power that owner-entrepreneurs had built up over the
previous decades. Wall Street (and especially the Wrmof J. P. Morgan) organized the
merger of the leading companies, and in the process did what would later become
knownas initial public oVerings (IPOs) inorder to allowthe owner-entrepreneurs
to cash in on their ownership stakes. Many of them then retired from active
management of the company. Taking their places in strategic decision-making
positions were salaried managers, most of whom had themselves been recruited
years or even decades earlier to help build the innovative Wrms that they now
controlled. Hence, Marshalls entrepreneurial limit to the growth of the Wrm was
overcome. By the turn of the century, the separation of ownership and control in
many of the most successful industrial corporations served as a powerful induce-
ment for bright young, and typically White, Anglo-Saxon, Protestant, men to
consider careers as corporate executives (Lazonick 1986; OSullivan 2000a: ch. 3).
Also from the beginning of the twentieth century, a four-year undergraduate
college degree became important for entry into managerial careers, and in 1908
38 william lazonick
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 38
HarvardUniversity launchedthe Wrst graduate school inbusiness administration. In
1900 about 2 per cent of 1824 year olds were enrolled in institutions of higher
education; in 1930 over 7 per cent; and in 1950 over 14 per cent. By the 1920s the top
managers of many large industrial corporations had college degrees. As employers of
university graduates as well as beneWciaries of university research, big business took
an active role in shaping the formand content of higher education to meet its needs
for knowledge assets (Noble 1977; Lazonick 1986).
As they expanded, US industrial corporations tended to diversify into newlines of
business. Capabilities developed for generating goods for one product market could
be used as a basis for gaining entry to newproduct markets. Moreover, as companies
were successful, they could use internally generated revenues to Wnance these new
investments. ProWtable US corporations generally paid ample dividends to share-
holders, but they still generated enough revenues to invest for the future, including
growing expenditures on R&D (Mowery and Rosenberg 1989: ch. 4).
Besides transforming technology, a critical role of the managerial organization
was to gain access to product markets. Without high levels of sales, the high Wxed
costs of developing technology and investing in production facilities would have
simply resulted in high levels of losses. The building of national transportation and
communications infrastructuresthemselves largely put in place by managerial
enterprisescreated the possibility for manufacturing enterprises to sell on mass
markets. To take advantage of this opportunity, however, the industrial corporations
hadtomake complementary investments indistributioncapabilities, including sales
personnel, sales oYces, advertising, and in some cases even customized transporta-
tion facilities. As Chandler (1990) has shown, from the late nineteenth century,
a three-pronged investment in production, distribution, and management was a
necessary condition for the growth of the industrial enterprise.
If the social condition for the growth of the US industrial corporation was an
integrated managerial organization, a distinguishing feature of the same corpor-
ation was a sharp organizational segmentation between salaried managers and what
became known as hourly workers. This segmentation had its roots in the Wrst half
of the nineteenth century when industrial managers faced a skilled labor force that
was highly mobile not only from one Wrm to another but also from one occupation
and one locality to another. In contrast, in Britain the local pools of specialized craft
labor generated by apprenticeship systems meant that employers had access to
ample supplies of skilled labor, even in booms. As a result, there was much less
pressure in Britain than in the United States for managers to invest in the develop-
ment of skill-displacing technologies. In the United States, but not Britain, Wrms
integrated technical specialists into their managerial organizations for precisely that
purpose. Hence the emergence by the mid-nineteenth century of the distinctive
American system of manufactures (Hounshell 1984: chs. 12).
The key to this system was the mass production of standardized, precision-
engineered parts that could be used interchangeably in a product without the
the innovative firm 39
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 39
intervention of a skilled worker to make the parts Wt together. As David Hounshell
(1984) has shown, it took a century of investment in productive capabilities by many
companies in many sectors of US industry before, during the boom of the 1920s,
mass production, so deWned, became a reality. The productivity of the mass-
production enterprise, nevertheless, still relied upon the stable employment of
semi-skilled production workers who tended high-throughput, and very expen-
sive, machinery (Lazonick 1990: chs. 78).
During the Great Depression of the 1930s, such stable employment disappeared,
leading semi-skilled workers at the major mass producers to turn to industrial
unionism(Brody 1980: ch. 3). The major achievement of mass-productionunionism
in the United States was long-term employment security for so-called hourly
workers, with seniority as the governing principle for internal promotion to higher
pay grades and continued employment during company layoVs. In return, these
unionized employees accepted unilateral managerial control over the organization
of work and technological change. During the post-World War II decades, produc-
tion workers enjoyed employment security and rising wages but they were not in
general integrated with managerial personnel into the companys organizational
learning processes.
The result was that going into the second half of the twentieth century US
industrial corporations had powerful managerial organizations for developing
new technology. These corporations also had devised arrangements with their
unionized labor forces to ensure the high level of utilization of these technologies.
In employing thousands and in some cases tens of thousands of production workers
who were not integrated into the companys organizational learning processes,
however, this US model of the innovative Wrm had a fundamental weakness that,
in the 1970s and 1980s, would be exposed in international competition. The Japanese
in particular would demonstrate the innovative capability that could be created by
not only building highly integrated managerial organizations, as the Americans had
done, but also, as a complement, developing the skills of shop-Xoor workers and
integrating their eVorts into the Wrms collective learning processes.
Even the most insightful of the theories of the US managerial corporation could
not, without elaboration, account for the Japanese challenge (Lazonick 2002c). Both
Penrose (1995) and Chandler (1962 and 1977) focused exclusively on the managerial
organization, as did the inXuential perspective of John Kenneth Galbraith (1967)
with its notion of the technostructure as the essence of the modern Wrm. Penrose
did not see that, once confronted by the Japanese challenge, the US managerial
corporationwould have to develop the capabilities of the shop-Xoor worker to make
use of unused managerial resources. Chandler focused on speed or throughput as a
basis for achieving economies of scale and scope, but ignored the role of the shop-
Xoor worker in the process of transforming high Wxed costs into lowunit costs, and
hence did not perceive an important limitation of the US managerial model
(Lazonick 1990).
40 william lazonick
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 40
2. 5 The Japanese Challenge
Within the new structure of cooperative industrial relations that emerged out the
conXicts of the depression years, US industrial corporations were able to take
advantage of the post-World War II boom to re-establish themselves as the worlds
pre-eminent producers of consumer durables such as automobiles and electrical
appliances andrelatedcapital goods suchas steel andmachine tools. Withthe helpof
US government research support and contracts, US companies also became the
leaders in the computer and semiconductor industries.
In the 1970s and 1980s, however, Japanese companies challenged the US industrial
corporations in the very mass-production industriessteel, memory chips,
machine tools, electrical machinery, consumer electronics, and automobilesin
which even as late as the 1960s US corporations seemed to have attained an insur-
mountable competitive advantage. During the 1950s and 1960s many Japanese
companies had developed innovative manufacturing capabilities, often on the
basis of technologies borrowed from abroad to produce mainly for the home
market. As Japanese exports to the United States increased rapidly in the last half
of the 1970s, many observers attributed the challenge to the lower wages and longer
working hours that prevailed inJapan. By the early 1980s, however, withreal wages in
Japan continuing to rise, it became clear that Japanese advantage was based on
superior capabilities for generating higher quality, lower cost products.
The three social institutions that, in combination, formed the foundation for
Japans remarkable success were cross-shareholding, the main bank system, and
lifetime employment. Cross-shareholding provided the managers of Japanese
industrial corporations with the strategic control to allocate resources to invest-
ments that couldgenerate higher quality, lower cost products. The mainbank system
provided these companies with levels of Wnancial commitment that permitted them
to sustain the innovation processes until they could generate returns, Wrst on home
and then on foreign product markets. Given this Wnancial support for strategic
industries, lifetime employment enabled the companies involved to put in place a
new model of hierarchical and functional integration that enabled them to
mobilize broader and deeper skills bases for collective and cumulative learning
(Lazonick 2001). Let us look brieXy at how these institutions became embedded in
the functioning of the Japanese industrial enterprise in the post-World War II
In 1948 the Supreme Commander for the Allied Powers (SCAP)the occupation
authority in Japanbegan the dissolution of the zaibatsu, the giant holding com-
panies that had dominated the Japanese economy from the Meiji era of the late
nineteenth century to World War II. The dissolution process not only dispossessed
the families that owned the zaibatsu but also removed from oYce the top manage-
ment layers of the zaibatsu holding companies and major aYliated Wrms (Morikawa
the innovative firm 41
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 41
1997). Taking control of strategic decision making were third-rank executives,
primarily engineers plucked from the ranks of middle management to take leader-
ship positions of companies whose challenge was to Wnd non-military markets for
their companies accumulated capabilities.
With the reopening of the stock market in 1949, these young and ambitious
executives feared that the new public shareholders might join forces to demand
their traditional rights as owners. To defend themselves against these outside
interests, the community of corporate executives engaged in the practice of cross-
shareholding. Commercial banks and industrial companies took equities oV the
market by holding each others shares. Though not contractual, cross-shareholding
was sustained by the willingness of the entire Japanese business community to accept
that one company would not sell its shareholdings of another company.2 By 1975,
according to its broadest, and most relevant, deWnition as stock in the hands of such
stable shareholders, cross-shareholding represented 60 per cent of outstanding
stocks listed on the Tokyo Stock Exchange. It peaked at 67.4 per cent in 1988, but
by 2000 had declined to 57.1 per cent, mainly because the beleaguered banking sector
had been forced to reduce their shareholdings.
During the era of high-speed growth from the early 1950s to the early 1970s,
most of the Wnancial commitment of Japanese companies came from bank loans,
with the companies debtequity ratios often at 6 : 1 or 7 : 1. Each major industrial
company had a main bank whose job it was to convince other banks to join it in
making loans to the company andto take the lead in restructuring its client company
should it fall into Wnancial distress. Some economists (e.g. Aoki and Patrick 1994)
have accorded the main banks a major role in monitoring the behavior of Japans
corporate managers. In funding the growth of Japanese companies, however, the
Japanese banks were relatively passive agents of government development policy,
with overloans being made by the Bank of Japan to its member banks for provid-
ing highly leveraged Wnance to growing industrial companies. Japanese banks, that
is, playeda critical role in providing Wnancial commitment, but no signiWcant role in
the exercise of strategic control.
Integrated organizations of managers and workers, not Wnancial interests, moni-
tored the behavior of the top executives of Japanese corporations (Lazonick 1999).
The main mode of achieving this organizational integration was the lifetime em-
ployment system, which extended from top executives to male (but not female)
shop-Xoor workers. The origins of the lifetime employment system can be found in
the widespread employment in industry of university graduates as salaried technical
and administrative personnel during the early twentieth century (Yonekawa 1984).
Some companies extended the promise of lifetime employment to shop-Xoor
workers as well when dire economic conditions and democratization initiatives of
the late 1940s had given rise to a militant labor movement. The goal of the new
industrial unions was to implement production control: the takeover of idle
factories so that workers could put them into operation and earn a living (Gordon
42 william lazonick
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 42
1985). Leading companies such as Toyota, Toshiba, and Hitachi Wred militant
workers and created enterprise unions of white-collar (technical and administra-
tive) and blue-collar employees. Foremen and supervisors were members of the
enterprise unions, as were all university-educated personnel, for at least the Wrst ten
years of employment before they made the oYcial transition into management.
The most important achievement of enterprise unionism was the institutional-
ization of lifetime employment, a system that, while not contractually guaranteed,
gave white-collar and blue-collar workers employment security, at Wrst to the
retirement age of 55, then from the 1980s to the age of 60, and currently (in
transition) to the age of 65 (Sako and Sato 1997). This employment security both
won the commitment of the worker to the company and gave the company the
incentive to develop the productive capabilities of the worker. The system did not
diVer in principle from the organizational integration of technical and adminis-
trative employees that was at the heart of the US managerial revolution, except inone
extremely important respect. In the United States there was a sharp segmentation
between salaried managers and shop-Xoor workers, whereas the Japanese com-
panies of the post-World War II decades integrated shop-Xoor workers into a
company-wide process of organizational learning.
Through their engagement in processes of cost reduction, Japanese shop-Xoor
workers were continuously involved in a more general process of improvement of
products and processes that, by the 1970s, enabled Japanese companies to emerge as
world leaders in factory automation (Jaikumar 1989). By the early 1990s the stock of
robots in Japanese factories was over seven times that of the United States. Also of
great importance was the ability of Japanese manufacturers to eliminate waste in
production; by the late 1970s, for example, Japans competitive advantage in televi-
sion sets was not in labor costs or even scale economies but in a savings of materials
costs (Owen 2000: 278; Fagerberg and Godinho in this volume). This productive
transformation became particularly important in international competition in the
1980s as Japanese wages approached the levels of those in North America and
Western Europe and, especially from 1985, as the value of the Japanese yen dramat-
ically strengthened. During the 1980s and 1990s, inXuenced by not only Japans
export performance but also the impact of Japanese direct investment in North
America and Western Europe, many Western companies sought, with varying
degrees of success, to implement Japanese high-quality, low-cost mass-production
During the 1980s most Western analyses of the sources of Japanese competitive
advantage focused on the hierarchical integration of the shop-Xoor worker into the
organizational learning process. By the early 1990s, however, as Japanese companies
captured higher value-added segments of the products markets in which they
competed, the emphasis shifted to the role of cross-functional management,
company-wide quality control, or concurrent engineering in generating not
only lower cost but also higher quality products within highly accelerated product
the innovative firm 43
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 43
development cycles. Much of the discussion of functional integration focused on its
role in new product development in international comparative perspective, with,
as Clark and Fujimoto (1991) showed for the automobile industry, the US manager-
ial corporation performing quite poorly.
Given that the innovative power of the US industrial corporation resided in its
integrated managerial organization, why should it have suVered from functional
segmentation in competition with the Japanese? One reason was that, given the
hierarchical segmentation of shop-Xoor activities from organizational learning
processes in US companies, US engineers were not forced to communicate across
their disciplines to solve real-world manufacturing problems. Another had to do
with the increasing interWrm mobility of US engineers from the 1960smobility
that, as we shall see, was related to the rise of the New Economy high-tech Wrm.
The prospects for interWrm mobility gave scientists and engineers an interest in
developing their reputations among their peers within their particular area of
specialization, even if it detracted fromintegrating their specialist knowledge across
functional areas within the particular Wrmfor which they were working. By contrast,
in the Japanese Wrm both the hierarchical integration of managers and workers and
low levels of interWrm mobility of engineering personnel fostered functional
The evolution of the semiconductor industry provides a vivid example of the
competitive power, but also the limits, of Japanese organizational integration. From
the late 1970s the Japanese mounted a formidable competitive challenge to US
producers in dynamic random access memory (DRAM) chips, forcing most US
companies, including Intel, to withdraw from the market after 1985. Already a
powerhouse in semiconductors before the Japanese challenge, Intel reemerged
even stronger in the 1990s as the leader in microprocessors, a product in which it
was the pioneer in the early 1970s and for which during the 1980s it secured the
franchise for the IBM PC and the subsequent IBM clones (Burgelman 1994).
Organizational integration was critical to the Japanese challenge in DRAMs. As
Daniel Okimoto and Yoshio Nishi (1994) have shown, the most critical interactions
in product and process development in Japanese semiconductor companies were
between personnel in divisional R&D labs and factory engineering labs, with
engineering capability being concentrated in the factory labs. They argue that in
Japan hands-on manufacturing experience . . . is almost a requirement for upward
career and post-career mobility [whereas] [i]n the United States, by contrast,
manufacturing engineers carry the stigma of being second-class citizens (Okimoto
and Nishi 1994: 195).
Value added in microprocessors is in the design that determines the use of the
product, an activity for which US skill bases in semiconductors were more suited.
Value added in memory chips is in process engineering that reduces defects and
increases chip yields, an activity for which Japanese skill bases in semiconductors
were more suited. By the 1980s Japanese companies such as Fujitsu, Hitachi, and
44 william lazonick
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 44
NEC were able to achieve yields in the production of DRAMs that were 40 per cent
higher than the best US companies.
In the 1990s the Japanese economy as a whole has stagnated, to the point where
many Western observers now blame its unique institutional framework, still largely
intact, for its lack of innovation. Yet, in industries such as electronics and automo-
biles, Japanese companies such as Sony and Toyota, among many others, remain
leading innovators in those types of products in which, as during the previous
decades, their integrated skill bases gave them international competitive advantage.
The main microeconomic problems in the Japanese economy are to be found in the
Wnancial system and, relatedly, institutions for creation of new innovative Wrms.
During the boom of the 1980s the leading Japanese manufacturing companies
were able to reduce their reliance on bank debt, just as the banks were awash with
cash to lend. The banks then channeled funds into speculative investments in land
and stocks, thus fuelling the bubble economy of the late 1980s. When the bubble
burst in 1990, the banks were saddled with mountains of bad debt. Although most of
this bad debt has nowbeenwritten oV, the banks remain in fragile condition because
most of their loans are being made to smaller companies that do not have anything
close to the growth potential that was realized by many Japanese companies in the
previous eras of high-speed growth and export expansion (Lazonick 1999). Growth
potential, however, is not exogenous to the social conditions of innovative enter-
prise, as illustrated by the emergence of more powerful modes of strategy, Wnance
and organization in the rise of the New Economy model of the innovative Wrm in
the United States.
2. 6 The New Economy Model
During the 1970s and 1980s while Japanese enterprises were challenging established
US managerial corporations in many industries in which they had been dominant,
there was a resurgence of the US information and communications technology
(ICT) industries, providing the foundation for what by the last half of the 1990s
became knownas the NewEconomy. Historically, underlying the emergence of the
NewEconomy were massive post-World War II investments by the US government,
in collaboration with research universities and industrial corporations, in develop-
ing computer and communications technologies.
By the end of the 1950s, this combined businessgovernment investment eVort
had resulted in not only the Wrst generation of computers, with IBM as the leading
Wrm, but also the capability of imbedding integrated electronic circuits on a silicon
the innovative firm 45
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 45
chip, with Fairchild Semiconductor and Texas Instruments in the forefront of
creating the technology that would become the standard of the semiconductor
industry. Through the early 1960s the US government provided virtually all of the
demand for semiconductors. Fromthe second half of the 1960s, however, a growing
array of commercial opportunities for electronic chips induced the creation of
semiconductor startups. A new breed of venture capitalist, many with prior man-
agerial or technical experience in the semiconductor industry, backed so many
semiconductor startups clustered in the region around Stanford University that by
the early 1970s the district was dubbed Silicon Valley. Innovation in semiconduct-
ors, and especially the development of the microprocessorin eVect a computer on
a chipcreated the basis for the emergence of the microcomputer industry fromthe
late 1970s, which in turn resulted in the enormous growth of an installed base of
powerful hosts in homes and oYces that made possible the Internet revolution of
the 1990s.
As AnnaLee Saxenian (1994) has shown, intense, and often informal, learning
networks that transcended the boundaries of Wrms contributed to the success of
Silicon Valley. Like the Marshallian industrial districts of a century earlier, there is no
doubt that, in Silicon Valley, the mysteries of the trade . . . were in the air. But in its
strategy, Wnance, and organization, the NewEconomy business model that emerged
in Silicon Valley diVered signiWcantly from the Marshallian industrial district. Of
particular importance was the extent to which in Silicon Valley organizational
learning occurred within the Wrm, enabling some particularly innovative Wrms
that grew to employ tens of thousands of employees to drive the development of
the region. In its early stages this organizational learning tended to be backed by
venture capital, a mode of Wnance that through its success in Silicon Valley fromthe
1960s evolved into an industry in its own right. Also of great importance in
supporting the development of technology and the education of personnel available
to Wrms in this high-tech industrial district were state funding and universities,
institutions that for a century had been central to the US managerial model.
The founders of new ICT Wrms were typically engineers who had gained special-
ized experience in existing ICT Wrms, although in some cases they were university
faculty members intent on commercializing their academic knowledge. While some
of these entrepreneurs came from existing Old Economy companies, where it was
often diYcult for their new ideas to get internal backing, New Economy companies
themselves became increasingly important as a source of newentrepreneurs who left
their current employers to start a new Wrm (Gompers et al. 2003). Typically the
founding entrepreneurs of a NewEconomy startup sought committed Wnance from
venture capitalists with whom they shared not only ownership of the company but
also strategic control. Besides sitting on the board of directors of the new company,
the venture capitalists would generally recruit professional managers, who would be
givencompany stock along withstockoptions, to leadthe transformationof the Wrm
from a new venture to a going concern. This stock-based compensation gave these
46 william lazonick
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 46
managers a powerful Wnancial incentive to develop the innovative capabilities of the
company to the point where it could do an IPO or private sale to an established
company. But, both before and after making this transition, their tenure with, and
value to, the company dependedontheir managerial capabilities, not their fractional
ownership stakes.
Key to making this transition from new venture to going concern was the
organizational integration of an expanding body of technical and managerial
talent. Stock options became an important mode of compensation, usually as a
partial substitute for cash salaries, for attracting these highly mobile people to the
startup and retaining their services. The underlying stock would become valuable if
and when they took the form of publicly traded shares. Shortening the expected
period between the launch of a company and its IPO was the practice of most
venture-backed high-tech startups of going public on the NASDAQ exchange
(founded in 1971), with its much less stringent listing requirements than the Old
Economy New York Stock Exchange. If and when the Wrm did an IPO or was
acquired by another publicly listed company, the venture capitalists could sell
their shareholdings on the stock market, thereby exiting from their investments in
the Wrm, while entrepreneurs could also transform some or all of their ownership
stakes into cash. With the companys stock being publicly traded, employees who
exercised their stock options could easily turn their shares into cash.
During the 1980s and 1990s the liberal use of stock as a compensation currency,
not only for top executives as had been the case in Old Economy companies since the
1950s, but also for a broad base of non-executive personnel became a distinctive
feature of NewEconomy Wrms. For example, Cisco Systems, which grewfromabout
200 employees at the time of its IPO in 1990 to 38,000 employees in 2001, awarded
stock options to all of its employees, so that by 2001 stock options outstanding
accounted for over 14 per cent of the companys total stock outstanding. Since Cisco
did hardly any of its ownmanufacturinganother distinctive characteristic of many
New Economy systems integratorsthe people in the skill base to whom these
options were awarded were almost all highly educated employees who were poten-
tially highly mobile on the labor market.
Besides using their own stock as a compensation currency, during the 1990s some
NewEconomy companies grewlarge by using their stock, instead of cash, to acquire
other, smaller and typically younger, New Economy Wrms in order to gain access to
new technologies and markets. Cisco mastered this growth-through-acquisition
strategy; from 1993 through 2002 Cisco made seventy-eight acquisitions (forty-
one of which were during 19992000, the peak years of the New Economy boom),
with stock providing the currency for over 98 per cent of the total value of these
At the same time Cisco conserved cash by paying no dividends, a mode of
Wnancial commitment that also distinguished New Economy from Old Economy
companies. As a result, Ciscos astonishing growth in the 1990s occurred without the
the innovative firm 47
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 47
company taking on any long-term debt. Nevertheless, with the bursting of the New
Economy bubble from mid-2000, Cisco spent billions of dollars repurchasing its
own stock to support its sagging stock price (Carpenter et al. 2003). Even during the
boom, when stock prices were rising, the extent to which New Economy companies
issued stock to make acquisitions and compensate employees meant that some of
them spent billions of dollars on stock repurchases; during 19972000, for example,
Intels stock repurchases totalled $18.8 billion and Microsofts $13.4 billion. By way of
comparison, over these years Intels total expenditures on R&D were $14.2 billion,
while Microsofts were $11.2 billion.
As in the cases of Intel, Microsoft, and Cisco, by the end of the twentieth century a
number of NewEconomy companies had grown to be formidable growing concerns
(Lazonick 2004). In 2002 the top 500 US-based companies by sales included twenty
ICT Wrms founded no earlier than 1965 that had been neither spun-oV from nor
merged with an Old Economy Wrm. These twenty companies had revenues ranging
from $35.4 billion for Dell Computer to $3.0 billion for Computer Associates
International, with an average of $10.4 billion. Their headcounts ranged from
78,700 for Intel to 8,100 for Qualcomm, with an average of 30,084, up from an
average for the same twenty companies of 6,347 in 1993. Nine of these twenty
companies (and seven of the top ten) were based in Silicon Valley, another two in
Southern California, and the other nine in eight states around the country. Compaq
Computer, the forty-sixth largest US company in 2001 with $33.6 billion in sales and
70,950 employees, would have been high up on this list in 2002 had it not been
acquired by Hewlett-Packard.
Many of these large New Economy companies have become important contribu-
tors to the patenting activity of US-based corporations. Samuel Kortum and Josh
Lerner (2000) have shown that in the Wrst half of the 1980s a sharp decline in
patenting by US corporations was counterbalanced by a massive increase in early-
stage venture-capital disbursements. But from the last half of the 1980s patenting
picked up again, in part because it became important to the competitive strategy of
high-growth NewEconomy Wrms. In 2001 Intel was eighteenth in the number of US
patents issued to all companies, and seventh among US-based companies. Ahead of
Intel were not only Old Economy companies such as IBM, Lucent Technologies,
General Electric, and Hewlett-Packard but also two much smaller, but still sizeable,
New Economy semiconductor companies, Micron Technology, founded in 1978
in Idaho, in fourth place, and Advanced Micro Devices (AMD), founded in
Silicon Valley in 1969, in fourteenth place. In 2002 AMD was the 535th largest US
company by sales and had 12,146 employees, while Micron was 554th and employed
Innovative NewEconomy companies have tended to growlarge by upgrading and
expanding their product oVerings within their main lines of business, and thus far at
least have not engaged in the indiscriminate diversiWcation into unrelated technolo-
gies and markets that characterized, and ultimately undermined the performance of,
48 william lazonick
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 48
many leading Old Economy companies in the 1960s and 1970s. At the same time,
NewEconomy companies have become less vertically integrated than Old Economy
companies because equipment manufacturers such as Cisco, Dell, and Sun Micro-
systems have focused their investment strategies on activities that require organiza-
tional learning in their core competencies, while outsourcing activities that, as is the
case with semiconductor fabrication, are too expensive and complex to be done in-
house, or, alternatively, as is the case with printed circuit board assembly, have
become routine. Some of the largest ICT companies in the United States are
upstream electronics components suppliers, most of which are New Economy
Wrms. Among the top 1000 US companies by sales in 2002 were eleven semicon-
ductor companies, with a total employment of 212,354, ranging from Intel with its
78,700 employees to Nvidia (a specialist producer of graphics processors founded
in 1993) with 1,513 employees. The worlds Wve largest contract manufacturers
Flextronics, Solectron, Sanmina-SCI, Celestica, and Jabil Circuitto whom equip-
ment manufacturers outsource the mass production of printed circuit boards
and other components, employed a total of 260,000270,000 people at the begin-
ning of 2003.
The severe downturn in the ICT industries in 2001 and 2002 raised questions
about the sustainability of the New Economy model. A major weakness of the New
Economy model lay in the huge personal gains, often amounting to tens of millions
and even hundreds of millions of dollars, that top executives could reap fromstock-
based rewards in a volatile stock market. When stock prices were rising, executives
had strong personal incentives to allocate resources (or give the appearance of doing
so) in ways that encouraged the speculative market. Many of these allocative deci-
sions undermined the innovative capabilities of the Wrms over which these execu-
tives exercised control (Carpenter et al. 2003). When stock prices began falling, the
same executives had strong personal incentives to cash in quickly by selling stock, so
that they made immense fortunes (in most instances without breaking the law) even
as their companies lost money and, in many cases, struggled to survive (Gimein et al.
A major problem for some of these companies was the way in which the use of
stock as a combination and compensation currency in the New Economy boom
aVected the role of the stock market as a source of cash (OSullivan 2003). Seventy
years earlier, in the stock market boom of the late 1920s, US corporations had sold
stock at speculative prices to pay down debt or bolster their treasuries, thus making
them less Wnancially vulnerable when the boom turned to bust. In the boom of the
late 1990s corporations did not take advantage of the speculative market by selling
stock; if anything, these companies purchased stock to support their already inXated
stock prices. While employees, andparticularly high-level executives, beneWtedfrom
these stock price increases, their companies were weakened Wnancially, as became
painfully evident for many ICT companies from mid-2000 when the stock market
turned down.
the innovative firm 49
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 49
2. 7 Understanding the Innovative
Firm: Implications for Theory
This chapter has illustrated that the social characteristics of the innovative Wrmhave
varied markedly over time and across institutional environments. To study the
innovative Wrm in abstraction from the particular social conditions that enable it
to generate higher quality, lower cost products is to forgo an understanding of why it
became innovative in the Wrst place and how its innovative capabilities may be
rendered obsolete. A comparativehistorical analysis enables us to learn from the
past and provide working hypotheses for ongoing research.
First, the comparativehistorical experience of the innovative Wrm suggests that,
contrary toa commonbelief that has persistedsince the time of Marshall, the formof
Wrmownershipis not the critical issue for understanding the type of strategic control
that supports innovative enterprise. Critical are the abilities and incentives of those
managers who exercise strategic control. Whether they are majority owners of the
Wrm, state employees, or employees of publicly listed companies, we need to know
where and how these strategic managers gained the experience to allocate resources
to the innovation process, and the conditions under which their personal rewards
depend on the Wrms innovative success.
Second, the most fundamental, if by no means the only, source of Wnancial
commitment for the innovative Wrmis to be found in those funds that are generated
by the Wrm itself. When bank Wnance is used to leverage Wnancial commitment, it
requires close relations between Wnancial institutions and innovative Wrms, as for
example in the Japanese model. In certain times and places, the stock market can
provide some well-positioned Wrms with Wnancial commitment. But as a Wnancial
institution, the fundamental role of the stock market is to provide liquidity, not
commitment. It enables owner-entrepreneurs and venture capitalists to cash out of
their investments, and it enables households to diversify their savings portfolios so
that they can (hopefully) tap into the yields of the stock market without having to
devote time and eVort to understanding the innovative capabilities of the companies
that have listed their securities on it.
Third, while strategic control and Wnancial commitment are essential to the
innovative Wrm, it is organizational integration that determines the innovative
capability that the Wrm actually possesses. The types of organizational integration
that result in innovation vary across industries and institutional environments as
well as over time. The hierarchical and functional divisions of labor that, when
integrated into learning organizations, generated innovation in the past cannot
necessarily be expected to do so in the future when faced with changes in technology,
markets, and competitionchanges which to some extent successful innovation in
itself brings about.
50 william lazonick
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 50
When a societys previously innovative Wrms no longer generate innovative
outcomes, there will be pressures on these Wrms to reallocate resources from invest-
ments in existing skill bases to investments in new types of organizations. Such
organizational restructuring does not always occur smoothly or successfully, as the
historical experiences of British industrial districts, US managerial corporations,
and Japanese enterprise groups have shown. Precisely because the innovative Wrmis
a social organization, the reallocation of its resources is a social process in which
diVerent groups of people can have very diVerent interests. An understanding of the
changing organization of the innovative Wrm is important for understanding not
only how a society innovates but also how a society copes with processes of social
disruption in which the gains of some may be the losses of others.
In a theory of innovative enterprise, strategy, Wnance, and organization are
interlinked as a dynamic process with learning as an outcome. To fully comprehend
the innovative Wrm, there is a need to understand the actual learning processes: the
relation between tacit knowledge and codiWed knowledge, between individual
capabilities and collective capabilities, and between what is learned at a point in
time and howthat learning cumulates over time. The prevailing social conditions of
innovative enterprise provide the context for those learning processes, shaping the
types of learning that are attempted, the extent to which these processes are sus-
tained, and the ways in which people interact both cognitively and behaviorally. The
inXuence of the social context is manifested by the functional and hierarchical
integration of skill bases that, as this essay has illustrated, can vary dramatically
across industries and institutional environments as well as over time.
Atheory of innovative enterprise must be based onanunderstanding of compara-
tivehistorical experience that is broad and deep enough to evoke conWdence that
the assumptions and relations forming the substance of the theory capture the
essence of the reality to which the theory purports to be relevant. The development
of relevant theory requires an iterative approach in which theoretical postulates are
derived from the study of the historical record and the resultant theory is used to
analyze history as an ongoing and unfolding process (Lazonick 2002a). The intel-
lectual challenge is to integrate theory and history.
As Edith Penrose (1989: 11) perceptively put it in an article written late in her career:
Theory is, by deWnition, a simpliWcationof reality but simpliWcationis necessary inorder
to comprehend it at all, to make sense of history. If each event, each institution, each fact,
were really unique in all aspects, howcould we understand, or claimto understand, anything
at all about the past, or indeed the present for that matter? If, on the other hand, there are
commoncharacteristics, and if such characteristics are signiWcant inthe determination of the
course of events, then it is necessary to analyse both the characteristics and their signiWcance
and theoretically to isolate them for that purpose.
If we need theory to make sense of history, so we also need history to make sense of
theory. As Penrose concluded: universal truths without reference to time and space
are unlikely to characterise economic aVairs.
the innovative firm 51
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 51
1. For a comparisonof managerial corporations of Europeannations such as Britain, France,
Germany, and Italy with the US model, see Lazonick 2003.
2. When in Wnancial distress, a company might raise cash by selling some of its cross-
shareholdings to other companies at the going market price but with an understanding
that the shares would be repurchased, also at the going market price, if and when its
Wnancial condition improved.
Aoki, M., and Patrick, H. (eds.) (1994), The Japanese Main Bank System: Its Relevance for
Developing and Transforming Economies, Oxford: Oxford University Press.
Becattini, G. (1990), The Marshallian Industrial District as a Socio-Economic Notion, in
F. Pyke, G. Becattini, and W. Sengenberger (eds.), Industrial Districts and Inter-Firm
Cooperation in Italy, International Institute for Labour Studies, 3751.
Belussi, F. (1999), Path-Dependency versus Industrial Dynamics: An Analysis of Two
Heterogeneous Districts, Human Systems Management 18: 16174.
Brody, D. (1980), Workers in Industrial America: Essays on the Twentieth Century Struggle,
Oxford: Oxford University Press.
Brusco, S. (1982), The EmilianModel: Productive Decentralisation and Social Integration,
Cambridge Journal of Economics 6: 16784.
(1992), Small Firms andthe Provisionof Real Services, inF. Pyke andW. Sengenberger
(eds.), Industrial Districts and Local Economic Regeneration, International Institute for
Labour Studies, 17796.
Burgelman, R. (1994), Fading Memories: A Process Theory of Strategic Business Exit in
Dynamic Environments, Administrative Science Quarterly 39(1): 2456.
*Carpenter, M., Lazonick, W., and OSullivan, M. (2003), The Stock Market and
Innovative Capability in the NewEconomy: The Optical Networking Industry, Industrial
and Corporate Change 12(5): 9631034.
Chandler, A. (1962), Strategy and Structure: Chapters in the History of the American Indus-
trial Enterprise, Cambridge, Mass.: MIT Press.
*(1977), The Visible Hand: The Managerial Revolution in American Business, Cam-
bridge, Mass.: Harvard University Press.
(1990), Scale and Scope: The Dynamics of Industrial Capitalism, Cambridge, Mass.:
Harvard University Press.
*Amatori, F., and Hikino, T. (eds.) (1997), Big Business and the Wealth of Nations,
Cambridge: Cambridge University Press.
Clark, K., and Fujimoto, T. (1991), Product Development Performance: Strategy, Organiza-
tion and Management in the World Auto Industries, Cambridge, Mass.: Harvard Business
School Press.
* Asterisked items are suggestions for further reading.
52 william lazonick
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 52
*Elbaum, B., and Lazonick, W. (eds.) (1986), The Decline of the British Economy, Oxford:
Oxford University Press.
Farnie, D. (1990), The Textile Machinery-Making Industry and the World Market,
18701960, Business History 32(4): 15070.
Galbraith, J. K. (1967), The New Industrial State, Boston: Houghton MiVlin.
Gimein, M., Dash, E., Munoz, L., and Sung, J. (2002), You Bought. They Sold. Fortune
146(4): 6472.
Gompers, P., Lerner, J., and Scharfstein, D. (2003), Entrepreneurial Spawning: Public
Corporations and the Genesis of New Ventures, 19861999, NBER Working Paper 9816,
Gordon, A. (1985), The Evolution of Labor Relations in Japan: Heavy Industry, 18531955,
Cambridge, Mass.: Harvard University Press.
*Hannah, L. (1983), The Rise of the Corporate Economy: The British Experience, 2nd edn.,
London: Methuen.
Hounshell, D. (1984), From the American System to Mass Production, 18001932, Baltimore:
Johns Hopkins University Press.
Huberman, M. (1996), Escape from the Market: Negotiating Work in Lancashire, Cambridge:
Cambridge University Press.
Jaikumar, R. (1989), Japanese Flexible Manufacturing Systems: Impact on the United
States, Japan and the World Economy 1(2): 11343.
Kogut, B., and Zander, U. (1996), What Firms Do? Coordination, Identity, and Learning,
Organization Science 7: 50218.
Kortum, S., and Lerner, J. (2000), Assessing the Contribution of Venture Capital to
Innovation, Rand Journal of Economics 31(4): 67492.
Lazonick, W. (1986), Strategy, Structure, and Management Development in the United
States and Britain, in K. Kobayashi and H. Morikawa (eds.), Development of Managerial
Enterprise, Tokyo: University of Tokyo Press, 10146.
(1990), Competitive Advantage on the Shop Floor, Cambridge, Mass.: HarvardUniversity
(1999), The Japanese Economy and Corporate Reform: What Path to Sustainable
Prosperity?, Industrial and Corporate Change 8(4): 60733.
(2001), Organizational Learning and International Competition: The Skill-Base
Hypothesis, in W. Lazonick and M. OSullivan (eds.), Corporate Governance and Sustain-
able Prosperity, Basingstoke: Palgrave, 3777.
(2002a), Innovative Enterprise and Historical Transformation, Enterprise & Society
3(1): 3554.
(2002b), Innovative Enterprise, The Theory of, in M. Warner, (ed.), International
Encyclopedia of Business and Management, 2ndedn., Stamford, Conn.: ThomsonLearning,
(2002c), The US Industrial CorporationandThe Theory of the Growthof the Firm, in
Christos Pitelis (ed.), The Growth of the Firm: The Legacy of Edith Penrose, Oxford: Oxford
University Press, 24977.
(2003), The Social Foundations of Innovative Enterprise, Franco Momigliano Lec-
ture, Instituto per la Cultura e la Storia dImpresa, Terni, Italy, July 10.
(2004), Corporate Restructuring, in S. Ackroyd, R. Batt, P. Thompson, and P. Tolbert
(eds.), The Oxford Handbook of Work and Organization, Oxford: Oxford University Press
the innovative firm 53
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 53
Lazonick, W., and OSullivan, M. (2000), Perspectives on Corporate Governance, Innov-
ation, and Economic Performance, Report prepared for the project on Corporate Gov-
ernance, Innovation, and Economic Performance under the Targeted Socio-Economic
Research Programme of the European Commission, June (www.insead.edu/cgep).
Leslie, S., and Kargon, R. (1996), Selling Silicon Valley: Frederick Termans Model for
Regional Advantage, Business History Review, 70(4): 43572.
Marshall, A. (1919), Industry and Trade, London: Macmillan.
(1961), Principles of Economics, 9th (variorum) edn., London: Macmillan.
Morikawa, H. (1997), Japan: Increasing Organizational Capabilities of Large Industrial
Enterprises, 1880s1980s, in A. Chandler, F. Amatori, and T. Hikino (eds.), Big Business
and the Wealth of Nations, Cambridge: Cambridge University Press: 30735.
and Kobayashi, K. (eds.) (1986), Development of Managerial Enterprise, Tokyo: Uni-
versity of Tokyo Press.
*Mowery, D., and Rosenberg, N. (1989), Technology and the Pursuit of Economic Growth,
Cambridge: Cambridge University Press.
Nelson, R. (1991), Why Do Firms DiVer, and HowDoes It Matter?, Strategic Management
Journal 12, Special Issue: 6174.
and Winter, S. (1982), An Evolutionary Theory of Economic Change, Cambridge, Mass.:
Harvard University Press.
Noble, D. (1977), America by Design: Science, Technology, and the Rise of Corporate Capital-
ism, New York: Knopf.
Okimoto, D., and Nishi, Y. (1994), R&D Organization in Japanese and American Semi-
conductor Firms, in M. Aoki and R. Dore (eds.), The Japanese Firm: The Sources of
Competitive Strength, Oxford: Oxford University Press: 178208.
*OSullivan, M. (2000a), Contests for Corporate Control: Corporate Governance and Eco-
nomic Performance in the United States and Germany, Oxford: Oxford University Press.
(2000b), The Innovative Enterprise and Corporate Governance, Cambridge Journal of
Economics 24(4): 393416.
(2003), The Stock Market as a Source of Cash in the US Corporation, INSEAD
working paper.
*Owen, G. (2000), From Empire to Europe: The Decline and Revival of British Industry since
the Second World War, New York: HarperCollins.
Penrose, E. (1989), History, the Social Sciences and Economic Theory, with Special
Reference to Multinational Enterprise, in A. Teichova, M. Levy-Leboyer, and H. Nuss-
baum(eds.), Historical Studies inInternational Corporate Business, Cambridge: Cambridge
University Press, 713.
*(1995), The Theory of the Growth of the Firm, 3rd edn., Oxford: Oxford University Press
[Wrst published 1959].
Piore, M., and Sabel, C. (1984), The Second Industrial Divide: Possibilities for Prosperity,
Basic Books.
Sabel, C. (1982), Work and Politics, Cambridge: Cambridge University Press.
and Zeitlin, J. (1985), Historical Alternatives to Mass Production: Politics, Markets
and Technology in Nineteenth-Century Industrialization, Past and Present 108: 13376.
*Sako, M., and Sato, H. (eds.) (1997), Japanese Labour and Management in Transition:
Diversity, Flexibility, and Participation, London: Routledge.
54 william lazonick
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 54
Saxenian, A. (1994), Regional Advantage: Culture and Competition inSiliconValley and Route
128. Cambridge, Mass.: Harvard University Press.
Schumpeter, J. (1934), The Theory of Economic Development, Cambridge, Mass.: Harvard
University Press.
(1950), Capitalism, Socialism, and Democracy, 3rd edn., New York: Harper.
Teece, D., Pisano, G., and Shuen, A. (1997), Dynamic Capabilities and Strategic Manage-
ment, Strategic Management Journal 18(7): 50933.
Yonekawa, S. (1984), University Graduates in Japanese Enterprises before the Second
World War, Business History 26(3): 193218.
the innovative firm 55
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 2:30pm page 55
Chapter 18
New York University
Vanderbilt University
Abstract 1182
Keywords 1183
1. Introduction 1184
1.1. What is a GPT? 1185
1.2. Summary of ndings 1186
1.2.1. Similarities between the Electrication and IT eras 1186
1.2.2. Differences between the Electrication and IT eras 1186
2. Measuring the three characteristics of a GPT 1187
2.1. Pervasiveness of the GPT 1187
2.1.1. Pervasiveness in the aggregate 1187
2.1.2. Pervasiveness among sectors 1189
2.1.3. Adoption by households 1193
2.1.4. On dating the endpoints of a GPT era 1194
2.2. Improvement of the GPT 1195
2.3. Ability of the GPT to spawn innovation 1198
2.3.1. Patenting 1198
2.3.2. Investment by new rms vs. investment by incumbents 1200
3. Other symptoms of a GPT 1203
3.1. Productivity slowdown 1204
3.2. The skill premium 1205
3.3. Entry, exit, and mergers should rise 1206
3.4. Stock prices should fall 1207
Handbook of Economic Growth, Volume 1B. Edited by Philippe Aghion and Steven N. Durlauf
2005 Elsevier B.V. All rights reserved
DOI: 10.1016/S1574-0684(05)01018-X
1182 B. Jovanovic and P.L. Rousseau
3.5. Young rms should do better 1208
3.5.1. The age of the leadership 1208
3.5.2. The age of rms at their IPO 1210
3.5.3. The stock market performance of the young vs. old after entry 1215
3.6. Consumption, interest rates, and the trade decit 1217
3.6.1. The trade decit 1218
3.6.2. The consumptionincome ratio 1218
3.6.3. Interest rates 1220
4. Conclusion 1221
Acknowledgements 1221
References 1221
A general purpose technology or GPT is a term coined to describe a new method of
producing and inventing that is important enough to have a protracted aggregate impact.
Electricity and information technology (IT) probably are the two most important GPTs
so far. We analyze howthe U.S. economy reacted to them. We date the Electrication era
from 1894 until 1930, and the IT era from 1971 until the present. While we document
some differences between the two technologies, we follow David [In: Technology and
Productivity: The Challenge for Economic Policy (1991) 315347] and emphasize their
similarities. Our main ndings are:
1. Productivity growth in the two GPT eras tended to be lower than it was in other
periods, with productivity slowdowns taking place at the start of the two eras and
the IT era slowdown stronger than that seen during Electrication.
2. Both GPTs were widely adopted, but electricitys adoption was faster and more
uniform over sectors.
3. Both improved as they were adopted, but measured by its relative price decline,
IT has shown a much faster improvement than Electricity did.
4. Both have spawned innovation, but here, too, IT dominates Electricity in terms of
the number of patents and trademarks issued.
5. Both were accompanied by a rise in creative destruction and turbulence as mea-
sured by the entry and exit of rms, by mergers and takeovers, and by changing
valuations on the stock exchange.
In sum, Electrication spread faster than IT has been spreading, and it did so more
evenly and broadly over sectors. Also, IT comprises a smaller fraction of the physical
capital stock than electried machinery did at its corresponding stage. On the other
hand, IT seems to be technologically more dynamic; the ongoing spread of IT and its
continuing precipitous price decline are reasons for optimism about productivity growth
in the 21st century.
Ch. 18: General Purpose Technologies 1183
electricity, information technology, IT revolution, productivity slowdowns, technology
improvement, creative destruction
JEL classication: O3, N2
1184 B. Jovanovic and P.L. Rousseau
1. Introduction
The term general-purpose technology, or GPT, has seen extensive use in recent treat-
ments of the role of technology in economic growth, and is usually reserved for changes
that transform both household life and the ways in which rms conduct business. Steam,
electricity, internal combustion, and information technology (IT) are often classied as
GPTs for this reason. They affected the whole economy.
As David (1991) has pointed out, however, a GPT does not deliver productivity gains
immediately upon arrival. Figure 1 shows the evolution of the growth in output per man-
hour in the U.S. economy over the past 130 years, with periods of rapid diffusion of the
two major GPTs shaded and the dashed line representing long-term trends as generated
with the HodrickPrescott (HP) lter.
Productivity growth was apparently quite rapid
during the heyday of steam power (c. 1870), but fell as Electrication arrived in the
1890s, with the dening moment in the transition probably being the startup of the rst
hydro-electric facility at Niagara Falls in 1894. It was only in the period after 1915,
which saw the diffusion of machines operated by stand-alone secondary motors and the
widespread establishment of centralized power grids, that Electricity nally pervaded
businesses and households more generally and measures of productivity began to rise.
Figure 1 also shows that the arrival of IT, which we date with Intels invention in 1971
of the 4004 microprocessor (the key component of the personal computer or PC),
did not reverse the decline in productivity growth that had begun more than a decade
earlier. It seems only now that we are nally seeing computers show up in the produc-
tivity gures.
But it is not obvious that the startup of the Niagara Falls dam and the invention
of the 4004 chip should dene the birth of the two GPTs. After all, Thomas Edison
invented the incandescent bulb in 1879 and by 1882 the worlds rst large central power
station had been installed at Pearl Street in New York City, twelve years before we mark
Electricitys arrival. And large mainframe computers predicted the winner of the 1952
U.S. Presidential election, nearly two decades prior to the advent of the microprocessor.
An objective measure is needed, though, and we shall dene the start of a GPT era as
the point in time when the GPT achieves a one-percent diffusion in the median sector.
This is another way to arrive at 1894 and 1971 as the starting points where the shading
begins in Figure 1. Similarly, we would say that the era is over when the diffusion curve
attens out. For Electrication, it takes until about 1929 for net adoption to reach a
plateau, whereas new adoption of IT is still rising today so that, on that criterion, the IT
epoch continues.
Each shaded area in Figure 1 contains a productivity-growth slowdown in its initial
phases. Will the growth slowdown of the current IT era be followed by a rise in growth
Output per man-hour in the business, non-farm sector is from John Kendricks series as published in U.S.
Bureau of the Census (1975, Series D684, p. 162) for 18891947 and from the Bureau of Labor Statistics
for 19482003. For 18741889, we use Kendricks decadal averages for 18691879 and 18791889 and
interpolate between these benchmarks assuming a constant growth rate from 18741884 and 18851889.
Ch. 18: General Purpose Technologies 1185
Figure 1. Annual growth in output per man-hour, 18742004.
in the rst half of the 21st century? If the second shaded area in Figure 1 is in some
fundamental respects like the rst shaded area, then we can expect growth to pick up
over the next several decades. In Jovanovic and Rousseau (2002a) we have argued that
the rst half of the 21st century will have higher growth than, say, the 1950s and 1960s.
Gordon (2000), on the other hand, is pessimistic, arguing that IT does not measure up
to Electricity and that it will not have such positive results. This chapter, while docu-
menting key differences between the diffusion paths of the two technologies, will in the
end conclude that the two GPT eras are strikingly similar in a number of respects. If
anything, our nding that IT is the more revolutionary of the two GPTs suggests that
its full impact is yet to be seen.
This chapter is organized around the presentation of a collection of facts. The facts are
described mainly through graphs and tables which provide evidence on a set of models
that we shall mention as we go along. A primarily analytic survey is Greenwood and
Jovanovic (2001).
1.1. What is a GPT?
So, what are these fundamental features of GPTs that would allow us to compare one
to another? And more generally, what criteria can one use to distinguish a GPT from
other technologies? Bresnahan and Trajtenberg (1996) argue that a GPT should have
the following three characteristics:
1. Pervasiveness The GPT should spread to most sectors.
2. Improvement The GPT should get better over time and, hence, should keep low-
ering the costs of its users.
3. Innovation spawning The GPT should make it easier to invent and produce new
products or processes.
1186 B. Jovanovic and P.L. Rousseau
Most technologies possess each of these characteristics to some degree, and thus a
GPT cannot differ qualitatively from these other technologies. Note, too, that the third
property is, in a sense, a version of the rst property if we phrase the latter to say that
the GPT should also spread to the innovation sector. Moreover, this list can be expanded
to include more subtle features of GPTs, a subject that we consider in Section 3. Yet we
nd these three basic characteristics to be a useful starting point for evaluating and com-
paring the impact of various technologies through history. Investigating how Electricity
and IT measure up on these three dimensions is the focus of Section 2. But rst, we
summarize our overall ndings.
1.2. Summary of ndings
The evidence shows similarities and differences between the Electrication and the IT
eras. Electrication was more pervasive (#1), whereas IT has a clear lead in terms of im-
provement (#2) and innovation spawning (#3). Let us list the similarities and differences
in more detail.
1.2.1. Similarities between the Electrication and IT eras
1. In both eras productivity growth rates are below those attained in the decades
immediately preceding the GPTs arrival.
2. Measures of reallocation and invention the entry and exit of rms to the stock
market, investment by new rms relative to incumbents, and grants of patents and
trademarks are all higher during the GPT eras.
3. Private consumption rises gradually during each GPT era.
4. Real interest rates are about the same during the two GPT eras, and about three
percentage points higher than from 1930 to 1970 the period between the rapid
adoptions of Electricity and IT.
1.2.2. Differences between the Electrication and IT eras
1. Innovation measures are growing much faster for IT than for Electrication
patents and trademarks surge much more strongly during the IT era, and the price
of IT is falling 100 times faster, at least, than did the price of electricity.
2. IT is spreading more slowly than did Electrication, and it comprises a smaller
part of the capital stock. Its net adoption continues to rise in the United States.
3. The productivity slowdown is stronger in the IT era.
4. No comparable sudden collapse of the stock market occurred early on in the Elec-
trication era.
5. The Electrication era saw a surplus in the U.S. trade balance, in part because
Europe had to nance a string of wars, whereas the IT era nds the United States
with consistent trade decits.
Ch. 18: General Purpose Technologies 1187
The differences seem to be quite important. But overall the evidence clearly supports
the view that technological progress is uneven, that it does entail the episodic arrival of
GPTs, and that these GPTs bring on turbulence and lower growth early on and higher
growth and prosperity later. The bottom line is that with a wider body of data and fteen
more years of it than David (1991) had at his disposal, we conrm his hypothesis that
Electrication and IT adoption are manifestations of the same force at work, namely the
introduction of a GPT.
2. Measuring the three characteristics of a GPT
As suggested in Figure 1, we shall choose Electricity and IT as our candidate GPTs, and
the measures that we construct will pertain mostly to these two technologies. In passing,
we shall also touch upon steam and internal combustion. The three subsections below
report, in turn, various measures of each characteristic pervasiveness, improvement,
and innovation for the two GPTs at hand.
2.1. Pervasiveness of the GPT
The rst characteristic is the technologys pervasiveness. We begin by looking at aggre-
gates and proceed to consider individual industrial sectors in more detail.
2.1.1. Pervasiveness in the aggregate
Ideally we would like to track the evolution of various candidate GPTs using continu-
ous time series from about 1850 to the present, but we do not have data that consistently
cover this entire stretch of time, and thus will need to work with two overlapping seg-
ments: 18691954 and 19472003.
Figure 2 shows the shares of total horsepower in manufacturing by power source from
1869 to 1954.
The period covers the decline in usage of water wheels and turbines,
the rise and fall of steam engines and turbines, the rise and gradual attening out of
the internal combustion engines use in industrial applications, and the sharp rise in the
use of primary and secondary electric motors. The symmetry of the plot is striking in
that, with the exception of internal combustion, power-generating technologies seem to
have led for the most part sequential existences. The relative brevity of the entire steam
cycle, which rises and falls within a period of 5060 years, suggests that the technology
We construct the shares of total horsepower in manufacturing as ratios of each power source from DuBoff
(1964, table 14, p. 59) to the total (table 13, p. 58). DuBoff estimates these quantities in 1869, 1879, 1889,
1899, 1904, 1909, 1914, 1919, 1923, 1925, 1927, 1929, 1939 and 1954, and we linearly interpolate between
these years in Figure 2. This source does not include a breakdown of non-electrical capacity (i.e., water,
internal combustion, and steam) after 1939, and so we mark the more broadly-dened non-electrical share
for 1954 with an asterisk.
1188 B. Jovanovic and P.L. Rousseau
Figure 2. Shares of total horsepower generated by the main sources in U.S. manufacturing, 18691954.
which replaced it, Electricity, was important enough to force a rapid transition among
manufacturers. In contrast, the decline of water power was more gradual.
If we could continue Figure 2 to the present, Electricity would surely still command
a very high share of manufacturing power as the next new source (e.g., solar power?)
has not yet emerged to replace it. The persistence of Electricity as the primary power
source, even though its diffusion throughout the manufacturing sector was complete
decades ago, helps to identify it as one of the breakthrough technologies of the modern
Figure 3 shows the diffusion of computers in the U.S. industrial sector as measured
by the share of IT equipment and software in the aggregate capital stock.
Computer and
software purchases appear to have reached the rst inection point in their S-curve
more slowly than Electrication in the early years of its adoption, but it is striking
how much faster the IT share has risen over the past few years. Moreover, while the
diffusion of Electricity had slowed down by 1930, the year which we mark as the end of
the Electrication era, computer and software sales continue their rapid rise to this day.
The vertical axes in Figures 2 and 3 are scaled differently. In Figure 2 the vertical
axis measures the share of total horsepower in manufacturing, whereas in Figure 3 it
is the share of IT equipment and software in the aggregate capital stock. But scaling
aside, a comparison of the shape of the diffusions in the two gures suggests that the
We build the ratio plotted in Figure 3 for 19612001 by summing the capital stocks of 62 SIC industrial
sectors from the detailed nonresidential xed asset tables in constant 1996 dollars made available by the U.S.
Bureau of Economic Analysis (2002, 2004). IT capital includes mainframe and personal computers, storage
devices, printers, terminals, integrated systems, and pre-packaged, custom, and own-account software. The
total capital stock is the sum of all xed asset types.
Ch. 18: General Purpose Technologies 1189
Figure 3. Shares of computer equipment and software in the aggregate capital stock, 19602003.
IT-adoption era will last longer than the 35 years of Electrication. Indeed, the accel-
eration in adoption, which was over by about 1905 for Electrication, did not end until
about 1997 for IT. It also appears that IT forms a smaller part of the physical capital
stock than did electric-powered machinery at the corresponding stages.
Why did Electricity spread faster than IT seems to be doing? Both technologies are
subject to a network externality; Electricity because the connecting of cables and wires
to a neighborhood was more protable when the number of users was larger, and IT
especially so after the Internet was invented. Perhaps electrical technologies were more
protable, or perhaps the rapid price decline of computers and peripherals makes it
optimal to wait and adopt later as Jovanovic and Rousseau (2002a) emphasize.
2.1.2. Pervasiveness among sectors
Cummins and Violante (2002, p. 245) classify a technology as a GPT when the share
of new capital associated with it reaches a critical level, and if adoption is widespread
across industries. Electrication seems to t this description. Figure 4 shows the shares
of total horsepower electried in manufacturing sectors at ten-year intervals from 1889
to 1954.
Electrical adoption was very rapid between 1899 and 1919 but slowed con-
siderably thereafter, with the dispersion in the adoption rates largest around 1919.
The striking feature of Figure 4 is how uniformly electrical technology affected indi-
vidual manufacturing sectors. Table 1, which shows the rank correlations of Electricity
shares across sectors and time, indicates that there was little change in the relative order-
ing of the sectors. This means that the sectors that were the heaviest users of Electricity
The shares of electried horsepower include primary and secondary electric motors, and are computed
using data from DuBoff (1964, tables E-11 and E-12aE-12e, pp. 228235).
1190 B. Jovanovic and P.L. Rousseau
Figure 4. Shares of electried horsepower by manufacturing sector, 18901954.
Table 1
Rank correlations of Electricity shares in total horsepower by manufacturing sector, 18891954
1889 1899 1909 1919 1929 1939 1954
1889 1.000
1899 0.707 1.000
1909 0.643 0.918 1.000
1919 0.686 0.746 0.893 1.000
1929 0.639 0.718 0.739 0.871 1.000
1939 0.486 0.507 0.571 0.750 0.807 1.000
1954 0.804 0.696 0.650 0.789 0.893 0.729 1.000
in 1890 remained among the leaders as adoption slowed down in the 1930s. In sum, the
adoption of Electricity was sweeping and widespread.
Why did that adoption take as long as it did? One answer is that it was costly to set up
the wiring required to electrify households early on. This is apparent from the peculiar
two-stage adoption process that many factories chose in adopting Electricity. Located
to a large extent in New England factory towns, textile rms around the start of the 20th
century readily adapted the new technology by using an electric motor rather than steam
to drive the shafts which powered looms, spinning machines and other equipment [see
Devine (1983)]. Moreover, delays in the distribution of electricity made it more costly
to electrify a new industrial plant fully.
Figure 5 shows the same data as Figure 4, but now in percentile form. We build it by
sorting the Electricity shares in each year and, given that only 15 sectors are represented,
plotting the 2nd, 5th, 8th, 11th and 14th largest shares in each year. The percentile
diffusion curves will be useful when drawing comparisons with the IT era. They also
Ch. 18: General Purpose Technologies 1191
Figure 5. Shares of electried horsepower by manufacturing sector in percentiles, 18901954.
help us in dating Electricity as a GPT. Linear extrapolation between the years 1890 and
1900 suggests that in 1894, about one percent of horsepower in the median industry
was provided by Electricity. Whether or not this is actually the right percentage for
dating the start of the Electrication era, we shall use a one percent share for the median
industry to date the beginning of the IT era as well. This provides a common standard
for choosing the left-end points of the two shaded areas.
In the century before the Electricity revolution, the technology that primarily drove
manufacturing was steam. Figure 6 shows just how slowly steam was replaced between
1899 and 1939.
It is natural that industries such as rubber, primary metals, non-electric
machinery, and stone, clay, and glass, which saw such rapid increases in electricity use
over the same period, would withdraw from steam most rapidly. Indeed, most of the
industries that quickly switched over to electricity had been heavy users of steam. This
is clear from Figures 4 and 6, taken together, and from the rank correlations of steam
shares in total horsepower in Table 2, which decay quickly and suggest a non-uniformity
in the destruction of steam technology across the manufacturing sectors.
The spread of IT was also rapid, but does not appear to have been as widespread as
Electricity. Figure 7 shows the share of IT equipment and software in the net capital
stocks of 62 sectors from 1960 to 2001 plotted as annual percentiles.
Some sectors
The sectoral shares of manufacturing horsepower driven by steam were computed from DuBoff (1964,
tables E-12aE-12e, pp. 229233), and include steam engines and turbines. These shares are available on a
decade basis from 1899 until 1939 only, which is why the time coverage in Figure 6 is shorter than that in
Figure 4.
The sectoral capital stocks are from the detailed non-residential xed asset tables in constant 1996 dollars
made available by the U.S. Bureau of Economic Analysis (2002). We present the sectoral shares for the IT era
in percentile form because the number of sectors covered is much larger than was possible for electrication
1192 B. Jovanovic and P.L. Rousseau
Figure 6. Shares of steam-driven horsepower by manufacturing sector, 18991939.
Table 2
Rank correlations of steam shares in total horsepower by manufacturing sector, 18891939
1899 1909 1919 1929 1939
1899 1.000
1909 0.825 1.000
1919 0.604 0.800 1.000
1929 0.525 0.604 0.832 1.000
1939 0.261 0.282 0.496 0.775 1.000
adopted IT very rapidly, and by 1975 six of them (the 90th percentile) had already
achieved IT equipment and software shares of more than 5 percent. Other sectors lagged
behind, and some did not adopt IT in a substantive way until after 1985.
On the other hand, the rank correlations of the IT shares across sectors, shown in
Table 3, are even higher than those obtained for Electrication. On the face of it then,
Electrication would appear to have been the more sweeping GPT-type event because
it diffused more rapidly in the U.S. economy and all sectors adopted it pretty much at
the same time, whereas IT diffused rapidly in some sectors and not-so-rapidly in others.
Nonetheless, the recent gains in IT shares show that the diffusion of this GPT has yet to
slow down in the way that Electrication did after 1929.
So far we have discussed adoption by rms, and have used this concept to determine
the dating of the two GPT-epochs. We turn to households next.
and steam. Changes in the industrial classications and the level of detail provided in the BEAs publicly-
available xed asset tables require us to end the series in 2001.
Ch. 18: General Purpose Technologies 1193
Figure 7. Shares of IT equipment and software in the capital stock by sector in percentiles, 19602001.
Table 3
Rank correlations of IT shares in capital stocks by sector, 19612001
1961 1971 1981 1991 2001
1961 1.000
1971 0.650 1.000
1981 0.531 0.806 1.000
1991 0.576 0.746 0.847 1.000
2001 0.559 0.682 0.734 0.909 1.000
2.1.3. Adoption by households
Households also underwent Electrication and the purchase of PCs for home use dur-
ing the respective GPT eras. Figure 8 shows the cumulative percentage of households
that had obtained electric service and that owned a PC in each year following the ar-
rival of the GPT.
If we continue to date Electricity as arriving in 1894 and the PC
in 1971, Figure 8 shows that households adopted Electricity about as rapidly as they
are adopting the PC. By the time the technology was ofcially 35 years old (in 1929),
nearly 70 percent of households had electrical connections. A comparison with Figure 5
shows that this is just a little higher than the 1929 penetration of electried horsepower
Data on the spread of electricity use by consumers are approximations derived from U.S. Bureau of the
Census (1975) Historical Statistics of the United States (series S108 and S120). Statistics on computer own-
ership for 19751998 are from Gates (1999, p. 118), and from the U.S. Bureau of the Census (20002004)
Current Population Survey thereafter.
1194 B. Jovanovic and P.L. Rousseau
Figure 8. Percent of households with electric service and PCs during the two GPT eras.
as measured by its share in the median manufacturing sector. As in the case of rms, the
Electrication of households reaches a plateau in 1929, although it resumed its rise a
few years later. On the other hand, there is no sign yet that the diffusion of the computer
among either households or rms is slowing down.
With households, as with rms, diffusion lags seem to arise for different reasons for
the two technologies. Rural areas were difcult for Electricity to reach, but this is not
the case for the PC, where the main barrier is probably the cost of learning how to use
it. This barrier seems to have more to do with human capital than was the case with
In some ways it is puzzling that the diffusion of the PC has not been much faster
than that of Electricity. The price of computing capacity is falling much faster than the
price of Electricity did. Affordable PCs came out in the 1980s, when the technology
was some 15 years old. On the other hand, households had to wait longer for afford-
able electrical appliances. Only after 1915, when secondary motors began to diffuse
widely and electrical appliances began to be invented, did the benets of Electrication
outweigh the costs for a majority of households. Greenwood, Seshadri and Yorukoglu
(2002) document the spread of electrically-powered household appliances and argue
that their diffusion helped to raise female labor-force participation by freeing up their
time from housework.
2.1.4. On dating the endpoints of a GPT era
Our dating procedure reects net adoption rates by rms, but the dates would not change
much if we had instead used net adoption by households. The shaded areas in our gures
are periods when the S-shaped adoption curves are, for the most part, rising. Whether
or not they start to fall later should not affect the designated adoption eras. For instance,
Ch. 18: General Purpose Technologies 1195
Electricity has not yet been replaced in the same way that steam was phased out in
the rst half of the 20th century, but the Electrication era still ends in 1930 because
adoption as measured in Figures 2, 4, and 5 attens out. Figures 2 and 6 show that the
steam era must have ended sometime around 1899 because net adoption had already
become negative.
Net adoption is endogenous and should reect the protability of the technology at
hand compared to that of other technologies. The Niagara Falls dam in 1894 and the
development of alternating current made it possible to produce and distribute electricity
more cheaply at greater distances. Figures 4 and 5 show that at the outset, some sectors
(like printing) raced ahead of others in terms of how quickly they adopted. Later on,
as the technology matures, its adoption becomes more universal. Eventually, the lag-
ging sectors tend to catch up a bit, in relative terms, but not completely. Inequality of
adoption is highest in the middle of the adoption era. We also see such a temporary rise
in inequality among the declining steam shares (Figure 6) at about the same time that
inequality was greatest in Electricity adoption.
2.2. Improvement of the GPT
The second characteristic that Bresnahan and Trajtenberg suggested is improvement
in the efciency of the GPT as it ages. Presumably this would show up in a decline
in prices, an increase in quality, or both. How much a GPT improves can therefore
be measured by how much cheaper a unit of quality gets over time. If technology is
embodied in capital, then presumably capital as a whole should be getting cheaper faster
during a GPT era, but especially capital that is tied to the new technology.
To investigate these implications, we rst look at the price of capital goods generally
and then at the prices of capitals components. Figure 9 is a quality-adjusted series
Figure 9. The price of equipment relative to consumption goods.
1196 B. Jovanovic and P.L. Rousseau
for the relative price of equipment as a whole, p
(i.e., relative to the consumption
price index) since 1885, constructed from a number of sources with a linear time trend
The gure shows that equipment prices declined most sharply between 1905
and 1920, and again after 1975. The 19051920 period is also the one that showed the
most rapid growth of Electricity in manufacturing (see Figure 4) and in the home (see
Figure 8). The post-1975 period follows the introduction of the PC.
Figure 10 considers the prices of components of the capital stock that are tied to
our candidate GPTs (as well as to internal combustion), with all prices relative to the
aggregate CPI. We use the price of electricity itself because deators for electrically-
powered capital are not available for the rst half of the 20th century.
Declines in motor
vehicle prices should capture the improvement in internal combustion as a possible
The use of the left-hand scale for electricity and motor vehicles and the right-
hand scale for computers underscores the extraordinary decline in computer prices since
1960 compared to the earlier technologies.
While the relative prices of electricity and
motor vehicles fall by a factor of 10, the index of relative computer prices falls by a
factor of 10,000.
Krusell et al. (2000) build such a series from 1963 using the consumer price index to deate quality-
adjusted estimates of producer equipment prices from Gordon (1990, table 12.4, col. 2, p. 541). Since
Gordons series ends in 1983, they use VAR forecasts to extend it through 1992. We start with Krusell et al.
and work backward, deating Gordons remaining estimates (19471962) with an index for non-durable
consumption goods prices that we derive from the National Income Accounts. Since we are not aware of a
quality-adjusted series for equipment prices prior to 1947, we use the average price of electricity as a proxy
for 19021946, and an average of Bradys (1966) deators for the main classes of equipment for 18851902.
We deate the pre-1947 composite using the Bureau of Labor Statistics (BLS) consumer price index of all
items [U.S. Bureau of the Census (1975, series E135)] for 19131946 and the Burgess cost of living index
[U.S. Bureau of the Census (1975, series E184)] for 18851912.
Electricity prices are averages of all electric services in cents per kilowatt hour from U.S. Bureau of the
Census (1975, series S119, p. 827) for 1903, 1907, 1917, 1922 and 19261970, and from various issues of the
Statistical Abstract of the United States for 19711989. We interpolate under a constant growth assumption
between the missing years in the early part of the sample. For 19902000, prices are U.S. city averages (June
gures) from the Bureau of Labor Statistics (http://www.bls.gov). We then set the index to equal 1000 in the
rst year of the sample (i.e., 1903).
Motor vehicle prices for 19131940 are annual averages of monthly wholesale prices of passenger vehicles
from the National Bureau of Economic Research (Macrohistory Database, series m04180a for 19131927,
series m04180b for 19281940, http://www.nber.org). From 19411947, they are wholesale prices of motor
vehicles and equipment from U.S. Bureau of the Census (1975, series E38, p. 199), and from 19482000 they
are producer prices of motor vehicles from the Bureau of Labor Statistics (http://www.bls.gov). To approx-
imate prices from 19011913, we extrapolate backward assuming constant growth and the average annual
growth rate observed from 19131924. We then join the various components to form an overall price index
and set it to equal 1000 in the rst year of the sample (i.e., 1901).
To construct a quality-adjusted price index, we join the nal price index for computer systems from
Gordon (1990, table 6.10, col. 5, p. 226) for 19601978 with the pooled index developed for desktop and
mobile PCs by Berndt, Dulberger and Rappaport (2000, table 2, col. 1, p. 22) for 19791999. Since Gordons
index includes mainframe computers, minicomputers, and PCs while the Berndt et al. index includes only
PCs, the two segments used to build our price measure are themselves not directly comparable, but a joining
of them should still reect quality-adjusted price trends in the computer industry reasonably well. We set the
index to 1000 in the rst year of the sample (i.e., 1960).
Ch. 18: General Purpose Technologies 1197
Figure 10. Price indices for products of the two GPT eras.
Figure 11. Comparison of the decline in general and GPT-specic equipment prices.
The more interesting question, however, is how the general decline in equipment
prices relates to the declines associated more directly with the GPTs of each epoch.
Figure 11 makes this comparison by plotting the relative prices of all three GPTs along
with the general equipment price index on the same logarithmic scale, with the starting
point for each of the GPTs normalized to the level of the general equipment index in
that year. By this measure, it is clear that electricity and motor vehicle prices declined
at about the same pace as that of equipment generally until the start of the IT price data,
though it is also interesting that motor vehicle prices appear to have declined faster than
electricity prices. After 1960, declining computer prices and rising shares of computers
1198 B. Jovanovic and P.L. Rousseau
in equipment stocks seem to have drawn the general index downward, while computing
prices fell thousands of times faster than the general index.
It can be said that the Electricity index, being the price of a kilowatt-hour, understates
the accompanying technological change because it does not account for improvements
in electrical equipment, and especially improvements in the efciency of electrical mo-
tors. Such improvements may be contained in the price series for capital generally.
But based on the price evidence in Figures 10 and 11, electricity, motor vehicles, and
computers might all qualify as GPTs. Computers, however, are clearly the most revolu-
tionary of the three.
2.3. Ability of the GPT to spawn innovation
The third characteristic that Bresnahan and Trajtenberg suggested was the technologys
ability to generate innovation. Any GPT will affect all sorts of production processes,
including those for invention and innovation. Some GPTs will be biased towards help-
ing to produce existing products, others towards inventing and implementing new ones.
An example of a more specic technology that was heavily skewed towards future prod-
ucts was hybrid corn. Griliches (1957, p. 502) explains why hybrid corn was not an
invention immediately adaptable everywhere, but was rather an invention of a method
of inventing, a method of breeding superior corn for specic localities.
Electricity and IT have both helped reduce costs of making existing products, and
they both spawn innovation, but IT seems to have more of a skew towards the latter.
There is no doubt that the 1920s, especially, also saw a wave a new products pow-
ered by electricity, but as the patenting evidence will bear out, the IT era has seen an
unprecedented increase in inventive activity. For example, the role of the computer in
simulation should be known to many of us writing research papers. Feder (1988) de-
scribes how computers play a similar role in the invention of new drugs.
2.3.1. Patenting
Patenting should be more intense after a GPT arrives and while it is spreading due
to the introduction of related new products. Figure 12, which shows the numbers of
patents issued per capita on inventions annually from 1790 to 2002 and trademarks reg-
istered from 1870 to 2002, shows two surges in activity between 1900 and 1930, and
again after 1977.
Is it mere chance that patenting activity was most intense during our
We use the total number utility (i.e., invention) patents from the U.S. Patent and Trademark Ofce for
19632002, and from the U.S. Bureau of the Census (1975, series W-96, pp. 957959) for 17901962. The
number of registered trademarks are from the U.S. Bureau of the Census (1975, series W-107, p. 959) for
18701969, and from the Statistical Abstract of the United States [U.S. Bureau of the Census (1980, 1992,
2003)] for later years. Population gures, which are for the total resident population and measured at mid-year,
are from U.S. Bureau of the Census (1975, series A-7, p. 8) for 17901970, and from the Census Bureaus
web site thereafter.
Ch. 18: General Purpose Technologies 1199
Figure 12. Patents issued on inventions and trademarks registered in the United States per million persons,
GPT eras? Moreover, it appears that patenting activity picks up after the end of the U.S.
Civil War in 1865, and again at the conclusion of World War II in 1945. The slowdown
in patenting during the wars and the acceleration immediately thereafter suggest that
there may be some degree of intertemporal substitution in the release of new ideas away
from times when it might be more difcult to popularize them and towards times better
suited for the entry of new products.
Does the surge in patenting reect a rise in the number of actual inventions, or was
the surge prompted by changes in the law that raised the propensity to patent? This
question is important because, over longer periods of time, patents may reect policy
rather than invention. Figure 13 analyzes data described in Lerner (2000) and shows that
worldwide, changes in patent policy are correlated with the patent series in Figure 12.
It is possible, therefore, that the U.S. series reects the stance of the courts regarding
enforcement. Kortum and Lerner (1998) analyze this question and found that the surge
of the 1990s was worldwide, but not systematically related to country-specic policy
changes. They conclude that technology was the cause for the surge.
Further support for this viewcomes fromthe behavior of trademarks per capita, which
we also plot in Figure 12. Trademarks behave more or less the same as patents do, except
for their more sharply rising trend. Trademarks are easier to obtain than patents and are
not governed by legal developments concerning patents. But with trademarks we have
a different concern: Do trademarks proxy for the number of products, or do they just
measure duplicative activity and the amount of competition? The answer may depend on
what market one looks at. In the market for bananas, for example, Wiggins and Raboy
(1996) nd that brand names are correlated with measures of quality that do explain
price variation, suggesting that brand names do signify product differentiation.
1200 B. Jovanovic and P.L. Rousseau
Figure 13. Indices of worldwide changes in patent laws.
2.3.2. Investment by new rms vs. investment by incumbents
New rms do not have costs sunk in old technologies and they are more exible or-
ganizationally than existing rms. One should therefore expect to see job-reallocation
waves and waves of entry and exit during the GPT eras. One measure of entry is the ex-
tent of new listings on the stock exchange. Figure 14 shows the value of rms entering
the New York Stock Exchange (NYSE), the American Stock Exchange (AMEX), and
NASDAQ in each year from 1885 through 2003 as percentages of total stock market
As predicted by Helpman and Trajtenberg (1998a, 1998b), initial public offer-
ings (IPOs) surge between 1895 and 1929, and then after 1977, which again closely
matches the dating of our two GPT eras.
The dashed line in Figure 14 is private investment since 1870 as a percent of the net
stock of private capital in the U.S. economy as a whole, and as such is the aggregate
analog of the solid line that covers only the stock market.
The solid line in Figure 15
The data used to construct Figure 14 and others in this chapter that use stock market valuations are from the
University of Chicagos Center for Research in Securities Prices (2004) (CRSP) les for 19252003. NYSE
rms are available in CRSP continuously, AMEX rms after 1961, and NASDAQ rms after 1971. We ex-
tended the CRSP stock les backward from their 1925 starting year by collecting year-end observations from
1885 to 1925 for all common stocks traded on the NYSE. Prices and par values are from the The Commercial
and Financial Chronicle (18851925), which is also the source of rm-level data for the price indices reported
in the inuential study by Cowles and Associates (1939). We obtained rm-level book capitalizations from
Bradstreets [Bradstreet Co. (18851925)], The New York Times and The Annalist [The New York Times Co.
(18971928 and 19131925)]. The resulting dataset, which includes 25,319 rms, complements others that
have begun to build a more complete view of securities prices for the pre-CRSP period [see, for example,
Rousseau (1999) on Bostons 19th century equity market].
To build the investment rate series, we start with gross private domestic investment in current dollars from
the U.S. Bureau of Economic Analysis (2004) for 19292003 and then join it with the gross capital formation
Ch. 18: General Purpose Technologies 1201
Figure 14. IPOs as a percent of stock market value, and private domestic investment as a percent of the net
capital stock, 18702003.
shows the ratio of the solid and dashed lines in Figure 14. In both gures it is clear that,
during the Electrication epoch, investment by stock market entrants accounted for a
larger portion of stock market value than overall new investment in the U.S. economy
contributed to the aggregate capital stock. This is consistent with the adoption of Elec-
tricity favoring the unencumbered entrant over the incumbent, who may have incurred
substantial adjustment costs in using the new technology. We say this because aggregate
investment, while indeed including new rms, has an even larger component attribut-
able to incumbents. Moreover, the solid line in Figure 15 was highest in the early years
of the Electrication period, which is when these adjustment costs would have been
Although the solid line in Figure 15 has so far stayed below unity for most of the
IT era, it has rapidly risen to a higher level in recent years. This could be because IT
adoption involved very large adjustment costs for both incumbents and entrants in the
early years until the price of equipment and software fell enough to generate a wave of
adoptions by new rms.
series in current dollars, excluding military expenditures, from Kuznets (1961b, Tables T-8 and T-8a) for
18701929. We construct the net capital stock using the private xed assets tables of the U.S. Bureau of
Economic Analysis (2004) for 19252003. Then, using the estimates of the net stock of nonmilitary capital
from Kuznets (1961a, Table 3, pp. 6465) in 1869, 1879, 1889, 1909, 1919 and 1929 as benchmarks, we
use the percent changes in a synthetic series for the capital stock formed by starting with the 1869 Kuznets
(1961a) estimate of $27 billion and adding net capital formation in each year through 1929 from Kuznets
(1961b) to create an annual series that runs through the benchmark points. Finally, we join the resulting series
for 18701925 to the later BEA series. The investment rate that appears in Figure 14 is the ratio of our nal
investment series to the capital stock series, expressed as a percentage.
1202 B. Jovanovic and P.L. Rousseau
Figure 15. Other investment ratios, 18852003.
The solid line in Figure 15 shows a downward trend mainly because the stock market
became more important as a vehicle for corporate nancing among industrial rms in
the early part of the 20th century. IPOs are normalized by total stock market value,
which was small early on, and has since become larger. The dotted line in Figure 15
shows the ratio of the dollar values of IPOs and aggregate investment. It is upward
sloping for the same reason: IPOs were not that important early on because the stock
market was small. After 1970, IPOs capture a much larger share of investment by new
entrants than they did before World War I, for example, and even a larger fraction than
in the 1920s. When we consider both lines together, we do get the impression that new
rms invest more during the GPT eras than at other times.
Does the distribution of entries across sectors shed light on the role of technological
factors in the entry waves? Perhaps so. Figure 16 is a scatterplot of the share of IPOs in
the market capitalizations of 15 manufacturing sectors between 1890 and 1930 vs. their
respective shares of horsepower driven by electricity in 1929.
In other words, we ask
whether sectors with more IPOs ended up adopting the new technology more vigorously
than sectors with less entry. The regression line plotted in Figure 16 has a positive slope
coefcient, though with only 15 observations it is not statistically signicant.
In Figure 17, we regress IPOs over the 19712001 period on shares of computers and
peripherals in equipment investment in 2000, and once again obtain a sectoral scatter
with a positive slope coefcient, though like our result for the Electrication era, it is
not statistically signicant.
We compute the IPO shares by summing year-end IPO values by sector for 18901930, converting the
annual totals for each sector into real terms using the implicit price deator for GDP, and then summing
across years. We do the same for all listed rms by sector, and use the ratio of sectoral IPO values to total
sector capitalization to compute the shares shown in Figure 16.
Ch. 18: General Purpose Technologies 1203
Figure 16. Scatterplot of IPOs as shares of sectoral market values, 18901930 vs. shares of horsepower
electried in 1929.
Figure 17. Scatterplot of IPOs as shares of sectoral market values, 19712001 vs. shares of IT in equipment
investment in 2000.
3. Other symptoms of a GPT
So far we have provided some measures of the three qualities of a GPT its pervasive-
ness, its rate of improvement, and its innovation-spawning tendency. Now we turn to
less direct measures as suggested by various theoretical models that deal with GPTs.
These models predict the following symptoms:
1. Productivity should slow down The new technology may not be user-friendly at
rst, and output may fall for a while as the economy adjusts.
1204 B. Jovanovic and P.L. Rousseau
2. The skill premium should rise If the GPT is not user-friendly at rst, skilled peo-
ple will be in greater demand when the new technology arrives and their earnings
should rise compared to those of the unskilled.
3. Entry, exit and mergers should rise These are alternative modes for the realloca-
tion of assets.
4. Stock prices should initially fall The value of old capital should fall. How fast it
falls depends on the way that the market learns of the GPTs arrival.
5. Young and small rms should do better The ideas and products associated with
the GPT will often be brought to market by new rms. The market share and
market value of young rms should therefore rise relative to old rms.
6. Interest rates and the trade decit The rise in desired consumption relative to
output should cause interest rates to rise or the trade balance to worsen.
These, roughly speaking, are the hypotheses that emerge from the theoretical work
on GPTs. Now we examine each empirically in turn, and as we go through the facts, we
shall mention some of the relevant theories.
3.1. Productivity slowdown
As Bahk and Gort (1993) show, even in routine activities, learning seems to cause de-
lays of several years before plant productivity peaks. It is far from settled, however,
whether IT is the reason for the productivity slowdown Bessen (2002) nds that IT
did cause a big part of the slowdown, whereas Comin (2002) argues the opposite. It is
also not yet denitely known from the work of Caballero and Hammour (1994) and
others whether recessions at business-cycle frequencies are episodes of heightened re-
allocation. At any rate, the theoretical models of Atkeson and Kehoe (1993), Hornstein
and Krusell (1996), Jovanovic and Nyarko (1996), Greenwood and Yorukoglu (1997)
and Jovanovic and Rousseau (2002a) emphasize various adjustment costs and learning
delays that may cause output to fall at rst when a GPT arrives. David (1991) argues
that the speed with which a new technology diffuses depends on the pool of invest-
ment opportunities that are available when it arrives, and remarks that the quality of this
pool in the late 1960s was low because a large backlog from the post-war period had
just and nally been eliminated. He also points out that there can often be slippage
between the technological frontier and implementation due to high input costs and the
slow introduction of complementary products.
Figure 1 shows that productivity did not rise quickly in the early phases of the two
GPTs, though there is some evidence of greater productivity between 1918 and 1929 and
after 1997 or so. Productivity was high in the early years of the Electrication period but
fell rapidly as the technology matured. It stayed low through the Depression and 1940s,
and then rose rapidly before the IT-age arrived. This pattern is consistent with Davids
view of exhausted investment opportunities. And while it is interesting to consider the
productivity slowdown after 1971, it is also important to recognize that productivity is
considerably higher today than it was before ITs arrival.
Ch. 18: General Purpose Technologies 1205
3.2. The skill premium
As Nelson and Phelps (1966) and Griliches (1969) argued, and Bartel and Lichtenberg
(1987) and Krusell et al. (2000) have conrmed, new technology should raise the rela-
tive earnings of the skilled. Figure 18 presents a series for the earnings of skilled relative
to unskilled labor. We construct the series by combining estimates of the wage ratio for
urban skilled and unskilled workers for 18701894 from Williamson and Lindert (1980,
p. 307) with estimates of the ratio of clerical to manufacturing production wages for
18951938 and the returns to 16 versus 12 years of schooling for men for 19391995
from Goldin and Katz (1999b).
Although interpreting time series patterns in a continuous series formed from such
disparate sources must be done with caution, we note that the series does have a
U-shape, with the skill premium high in the early stages of Electrication (i.e., 1890
to 1918) and then rising rapidly during the post-1978 part of the IT epoch. We suspect
that the decline in the skill premium from 19181924 would have been less deep, and
thus the overall U-shape of Figure 18 more apparent, had it not been for the rapid rise of
the public higher-education system after the end of World War I [see Goldin and Katz
(1999a, p. 10)].
Figure 18. The skill premium.
Combining several very different series into a continuous skill premium is necessary due to sectoral
shifts in the skilled and unskilled labor forces that render some measures of skill more applicable to certain
periods than others. For example, a college education appears to have become a more important determinant
of income in the postwar period than it was in earlier years. Since the observations from Goldin and Katz
(1999b) are generally decadal, we interpolate between them to obtain an annual series for 18951995. The
vertical dotted lines in Figure 18 mark the points where we need to change data sources.
1206 B. Jovanovic and P.L. Rousseau
3.3. Entry, exit, and mergers should rise
Gort (1969) argued that technological change will generate merger waves. Evidence
since then has shown that mergers and takeovers play a re-allocative role for an econ-
omys stock of human and physical capital. Lichtenberg and Siegel (1987), McGuckin
and Ngyen (1995) and Schoar (2002) nd that the productivity of a target rm rises fol-
lowing a takeover. Jovanovic and Rousseau (2002b, 2002c) study the trade-off between
exits and acquisitions at the margin for an economy that needs to update its capital stock.
This last pair of papers shows that, at times when the value of organization capital is
high, rms are more likely to place themselves on the merger market than to disassem-
ble and sell their assets. Further, reallocation of assets among rms in general (i.e., by
merger, consolidation, or purchases of unbundled used capital) is more likely to occur
than purchases of new capital when rms need to make large adjustments to their capi-
tal stocks because of xed costs associated with entering the merger market. We believe
that both of these conditions are likely to hold during times of sweeping technological
The U-shaped top line of Figure 19 is our estimate of the total amount of capital that
has been reallocated on the U.S. stock market from 1890 to 2003. Its components are
the stock market capitalization of entering and exiting rms divided by two, and the
value of merger targets.
Entries and exits divided by two, given by the center line,
is a rough measure of how much capital exits from the stock market and comes back
in under different ownership, or at least under a different name.
The lower line is the
stock-market value of merger targets. Regardless of whether reallocation occurs through
mergers or through entry and exit, it is much more prevalent during the periods that we
associate with Electrication and IT.
We identify targets for 19262003 using the CRSP stock les and various supplementary sources. CRSP
itself identies 9,758 rms that exited the database by merger between 1926 and 2003. We collected infor-
mation on all mergers for 18951930 in the manufacturing and mining sectors from the original worksheets
underlying Nelson (1959), and identied mergers from 1885 to 1894 from the nancial news section of
weekly issues of The Commercial and Financial Chronicle (18851925). The resulting target series includes
the market values of 10,788 exchange-listed rms in the year prior to their acquisition. Stock market capital-
izations are from our extension of CRSP backward to 1885 (see footnote 13). Before assigning a rm that no
longer carries a price in our database as an exit, we check the list of hostile takeovers from Schwert (2000)
for 19751996 and individual issues of the Wall Street Journal [Dow Jones and Company, Inc. (19972004)]
from 19972003 to ensure that we record rms taken private under a hostile tender offers as mergers.
For this to be exactly true, of course, would require that the assets of all rms exiting the stock market be
purchased by new rms that ended up listing on one of the major exchanges, and that the capital stocks of
these new rms consist of only these used assets, assumptions that we know to be violated in practice. If the
quantity of assets that do not return to the stock market through entry is roughly the same as the quantity of
assets brought into the stock market by new entrants that are not associated with exiting rms, however, our
measure would be roughly correct.
Ch. 18: General Purpose Technologies 1207
Figure 19. Reallocated capital and its components as percentages of stock market value, 18902003.
3.4. Stock prices should fall
The value of old capital should fall suddenly if the arrival of the GPT is a surprise,
as in Greenwood and Jovanovic (1999), Hobijn and Jovanovic (2001), Jovanovic and
Rousseau (2002a) and Laitner and Stolyarov (2003), or more gradually as in Help-
man and Trajtenberg (1998a, 1998b). Figure 20 shows that the stock market declined
in 19731974.
No such sudden drop is visible for stock prices in the early 1890s.
Why not? Maybe because the market was thin and unrepresentative in those days, with
railway stocks absorbing a large share of market capitalization. More likely, the realiza-
tion that the new technology would work well was more gradual and not prompted by
any single event such as the activation of the Pearl Street power station in 1882 or the
completion of the Niagara Falls dam in 1894.
In other words, perhaps a decline in the stock market did not occur early in the Elec-
trication period because the events of the early 1890s were foreseen, as would be the
case in Helpman and Trajtenberg (1998a, 1998b). It also could be, as in Boldrin and
Levine (2001), that old capital is essential to the production of new capital and that its
value may not fall in quite the way that it would when capital can be produced from
consumption goods alone, as is the case in many growth models including Jovanovic
and Rousseau (2002a).
If stock price declines were caused by the threat of IT to incumbents, this should
relate especially to those sectors that later invested heavily in IT. Hobijn and Jovanovic
(2001, p. 1218) conrm this using regression analysis.
We obtain the composite stock price index at the end of each year from Wilson and Jones (2002), updating
through November 2004 using various issues of the Wall Street Journal. We deate using the CPI.
1208 B. Jovanovic and P.L. Rousseau
Figure 20. The real Cowles/S&P stock price index across the two GPT eras.
3.5. Young rms should do better
If new technologies are brought to market most effectively by new rms, we would
expect younger rms in general to performbetter than older rms during the eras of GPT
adoption. The evidence on this hypothesis turns out to be mixed, but positive overall.
3.5.1. The age of the leadership
As a GPT takes hold, we should not only expect to see rms coming to market more
quickly, but the market leaders getting younger as well. In other words, every stage in
the lifetime of the rm should be shorter. This stands in contrast to Hopenhayn (1992),
in which the age distribution of an industrys leadership is invariant when an industry is
in a long-run stochastic equilibrium. That is, the average age of, say, the top 5 percent
or top 10 percent of rms is xed. Some leaders hold on to their positions and this tends
to make the leading group older, but others are replaced by younger rms, and this has
the opposite effect. In equilibrium the two forces offset one another and the age of the
leadership stays the same. Keeping the age of the leaders at requires, in other words,
constant replacement.
Figures 21 and 22 plot the value-weighted average age of the largest rms whose
market values sum to 5 and 10 percent of GDP, respectively. A rms age is measured
as the number of years since incorporation and since being listed on a major stock
exchange. We label some important entries and exits from this group in Figure 21 (with
exits denoted by X). The two gures show that, overall, the age of the leaders is
anything but at. It sometimes rises faster than the 45

line, indicating that the age of

the leaders is rising faster than the passage of time. At other times it is at or falling,
indicating replacement.
Ch. 18: General Purpose Technologies 1209
Figure 21. Average age (in years) of the largest rms whose market values sum to 5 percent of GDP,
Figure 22. Average age (in years) of the largest rms whose market values sum to 10 percent of GDP,
Based upon years from incorporation, for example, the leading rms were being re-
placed by older rms over the rst 30 years of our sample, because the solid line is
then steeper than the 45

line. In the two decades after the Great Depression the leaders
held their relative positions as the 45

slopes of the average age lines show. The leaders

got younger in the 1990s, and their average ages now lie well below the 45

The volatility in these series derives not from aggregate stock-market volatility, but from the volatility of
individual rm valuations. The large dip and subsequent recovery in both series in 19992001, for example,
1210 B. Jovanovic and P.L. Rousseau
Both gures show, however, that the lines are at or falling during the Electricity and
IT periods, so that replacement at these times was high. This is best seen in Figure 22.
3.5.2. The age of rms at their IPO
According to the third innovation-spawning characteristic, when a GPT arrives it
gives rise to new projects that are unusually protable. When such projects arrive, rms
will be more impatient to implement them. When it is new rms that come upon such
projects (rather than incumbents), they will feel the pressure to list sooner. This argu-
ment is developed and tested in Jovanovic and Rousseau (2001). We argue there that the
Electricity- and IT-era rms entered the stock market sooner because the technologies
that they brought in were too productive to be kept out of the market for very long.
Figure 23 shows HP-ltered average waiting times from founding, rst product or
process innovation, and incorporation to exchange listing based upon individual com-
pany histories and our backward extension of the CRSP database.
The vertical dis-
tance between the solid and dotted lines shows that rms often have their rst innovation
soon after founding, but that it then takes years, even decades, to list on a stock ex-
We interpret this delay as a period during which the rm and possibly its
lenders learn about what the rms optimal investment should be. But when the tech-
nology is highly innovative, the incentive to wait is reduced and the rm lists earlier,
which is what the evidence shows.
Table 4 lists the rst product or process innovation for some of the better-known
companies, along with their dates of founding, incorporation, and stock exchange list-
ing. It also includes the share of total market capitalization that can be attributed to each
rms common stock at the end of 2003. The rms appearing in the table separate into
roughly 3 groups: those based upon electricity and internal combustion, those based
upon chemicals and pharmaceuticals, and those based upon the computer and Internet.
Let us consider a few of the entries more closely:
comes from Microsofts enormous price appreciation in 1999, when it was worth more than 5 percent of GDP
on its own, and its rapid decline in 2000, which transferred the full 5 percent share to GE. The two rms split
the 5 percent share in 2001.
Listing years after 1925 are those for which rms enter CRSP. For 18901924, they are years in which
prices rst appear in the NYSE listings of The Annalist, Bradstreets, The Commercial and Financial Chron-
icle or The New York Times. The 6,632 incorporation dates used to construct Figure 23 are from Moodys
Industrial Manual [Moodys Investors Service (1920, 1928, 1955, 1980)], Standard and Poors Stock Market
Encyclopedia [Standard and Poors Corporation (1981, 1988, 2000)] and various editions of Standard and
Poors Stock Reports [Standard and Poors Corporation (19712003)]. The 4,221 foundings are from Dun and
Bradstreets Million Dollar Directory [Dun and Bradstreet, Inc. (2003)], Moodys, Kelley (1954), and indi-
vidual company web sites. The 482 rst innovations were obtained by reading company histories in Hoovers
Online [Hoovers, Inc. (2000)] and company web sites. We linearly interpolate the series between missing
points before applying the HP-lter to get the time series in Figure 23.
Figure 23 includes several years in the 1970s and early 1980s for which it appears that the average time
from rst innovation to listing exceeds that from founding to listing. This is a result of differences in the
sample sizes used to construct each line.
Ch. 18: General Purpose Technologies 1211
Figure 23. Waiting times to exchange listing, 18902003.
Electricity/Internal Combustion Engine Two of largest companies in the United
States today are General Electric (GE) and AT&T. Founded in 1878, GE accounted
for 2.1 percent of total stock market value at the end of 2003, and had already es-
tablished a share of over 2 percent by 1910. AT&T, founded in 1885, contributed
4.6 percent to total market value by 1928, and more than 8.5 percent at the time
of its forced breakup in 1984. Both were early entrants of the Electricity era. GEs
founding was based upon the invention of the incandescent light bulb in 1879,
while AT&T established a long-distance telephone line from New York to Chicago
in 1892 to make use of Bells 1876 invention of the telephone. Both technologies
represented quantum leaps in the modernization of industry and communications,
and both rms brought these technologies to the NYSE about 15 years after found-
ing. General Motors (GM) was an early entrant to the automobile industry, listing
on the NYSE in 1917 nine years after its founding. By 1931 it accounted for
more than 4 percent of stock market value, and its share would hover between
4 and 6.5 percent until 1965, when it began to decline gradually to its share in 2003
of only 0.2 percent. These examples suggest that many of the leading entrants at
the turn of the 20th century created lasting market value. Further, the ideas that
sparked their emergence were brought to market relatively quickly.
Chemicals/Pharmaceuticals Procter and Gamble (P&G), BristolMyers Squibb
and Pzer are both leaders in their respective industries, but took much longer to
list on the NYSE than the Electrication-era rms. In fact, P&G and Pzer were
established before 1850, and thus predate all of them. Despite P&Gs early start
and the creation of the Ivory soap brand in 1879, it was not until 1932 that the
company took its place among the largest U.S. rms by exploiting advances in
radio transmission to sponsor the rst soap opera. Pzers dening moment came
when it developed a process for mass-producing the breakthrough drug penicillin
1212 B. Jovanovic and P.L. Rousseau
Table 4
Key dates in selected company histories
Company name Founding
1st major product or
process innovation
% of stock market
in 2003
General Electric 1878 1880 1892 1892 2.09
AT&T 1885 1892 1885 1901 0.11
Detroit Edison 1886 1904 1903 1909 0.04
General Motors 1908 1912 1908 1917 0.20
Coca Cola 1886 1893 1919 1919 0.83
Pacic Gas & Electic 1879 1879 1905 1919 0.08
Burroughs/Unisys 1886 1886 1886 1924 0.03
Caterpillar 1869 1904 1925 1929 0.19
KimberlyClark 1872 1914 1880 1929 0.20
Procter & Gamble 1837 1879 1890 1929 0.87
BristolMyers Squibb 1887 1903 1887 1933 0.37
Boeing 1916 1917 1916 1934 0.23
Pzer 1849 1944 1900 1944 1.81
Merck 1891 1944 1934 1946 0.69
Disney 1923 1929 1940 1957 0.32
HewlettPackard 1938 1938 1947 1961 0.47
McDonalds 1948 1955 1965 1966 0.21
Intel 1968 1971 1969 1972 1.40
Microsoft 1975 1980 1981 1986 1.99
America Online 1985 1988 1985 1992 0.52
Amazon 1994 1995 1994 1997 0.14
E-Bay 1995 1995 1996 1998 0.28
Source: Data from Hoovers Online, Kelley (1954), and company web sites.
Note. The rst major products or innovations for the rms listed in the table are: GE 1880, Edison patents
incandescent light bulb; AT&T 1892, completes phone line from New York to Chicago; DTE 1904, in-
creases Detroits electric capacity six-fold with new facilities; GM 1912, electric self-starter; Coca Cola 1893,
patents soft-drink formula; PG&E 1879, rst electric utility; Burroughs/Unisys 1886, rst adding machine;
CAT 1904, gas driven tractor; KimberlyClark 1914, celu-cotton, a cotton substitute used in WWI; P&G
1879, Ivory soap; BristolMyers Squibb 1903, Sal Hepatica, a laxative mineral salt; Boeing 1917, designs
Model C seaplane; Pzer 1944, deep tank fermentation to mass produce penicillin; Merck 1944, cortisone
(rst steroid); Disney 1929, cartoon with soundtrack; HP 1938, audio oscillator; McDonalds 1955, fast food
franchising begins; Intel 1971, 4004 microprocessor (8088 microprocessor in 1978); Microsoft 1980, devel-
ops DOS; AOL 1988, PC-Link; Amazon 1995, rst online bookstore; E-Bay 1995, rst online auction
during World War II, and the good reputation that the rm earned at that time
later helped it to become the main producer of the Salk and Sabin polio vaccines.
In Pzers case, like that of P&G, the companys management and culture had
been in place for some time when a new technology (in Pzers case antibiotics)
presented a great opportunity.
Computer/IT Firms at the core of the recent IT revolution, such as Intel, Mi-
crosoft and Amazon, came to market shortly after founding. Intel listed in 1972,
only four years after starting up, and accounted for 1.4 percent of total stock mar-
Ch. 18: General Purpose Technologies 1213
ket value at the end of 2003. Microsoft took eleven years to go public. Conceived
in an Albuquerque hotel room by Bill Gates in 1975, the company, with its new
disk operating system (MS-DOS), was perhaps ahead of its time, but later joined
the ranks of todays corporate giants with the proliferation of the PC. In 1998, Mi-
crosoft accounted for more than 2.5 percent of the stock market, but this share fell
to 1.5 percent over the next two years in the midst of antitrust action. By the end
of 2003 its share had recovered somewhat to nearly 2 percent of the stock market.
Amazon caught the Internet wave from the outset to become the worlds rst on-
line bookstore, going public in 1997 only three years after its founding. As the
complexities of integrating goods distribution with an Internet front-end came into
sharper focus over the ensuing years, however, and as competition among Internet
retailers continued to grow, Amazons market capitalization by 2003 had fallen to
0.14 percent of total stock market value.
These rms, as well as the others listed in Table 4, are ones that brought new tech-
nologies into the stock market and accounted for more than 13 percent of its value at the
close of 2003. The rms themselves also seem to have entered the stock market sooner
during the Electricity and IT eras, at opposite ends of the 20th century, than rms based
on mid-century technologies.
When rms gather less information before investing, the investments that they under-
take will be riskier. One may conjecture that if new entrants waited less before investing
during the GPT eras, then incumbents also undertook projects earlier than they would
have normally. In these cases, the resulting investments would be riskier than if more
time were allowed to plan them. Moreover, the newness of the GPT would add further
risk. On all these grounds, we would expect interest rate differentials on the average
investment to be higher in the GPT eras.
Figure 24, which shows the spread between interest rates on riskier and safe invest-
ments since 1885, shows that this has been for the most part the case.
It is important to
note that we formed the series in Figure 24 by joining three different spreads together,
and that the safe asset is a long-term U.S. government bond before 1920 and a short-
term U.S. Treasury bill thereafter, yet the uctuations in this series should still reect
risk perceptions reasonably well, at least to the extent that term premia rather than risk-
iness are the main factors that lead to yield differentials among the various government
In Figure 24, we use the spread between the interest rates on Baa-rated corporate bonds (from Moodys
Investors Service) and three-month T-bills [from the FRED database of Federal Reserve Bank of St. Louis
(2004) for 19342003 and the Board of Governors of the Federal Reserve System (1976) for 19201934]
for the period from 1920 to the present. For 19001920, we join the spread between the interest rate on
prime commercial paper with 6090 days until maturity [Homer and Sylla (1991, table 49, p. 358)] and the
redemption yields on the U.S. government consol 2s of 1930 [Homer and Sylla (1991, table 46, p. 343)] with
the Baa T-bill spread. Finally, for 18851899, we join the spread between the commercial paper rate [Homer
and Sylla (1991, table 44, p. 320)] and the redemption yields on U.S. government refunding 4s of 1907 [Homer
and Sylla (1991, table 43, p. 316)] with the previous result.
1214 B. Jovanovic and P.L. Rousseau
Figure 24. Nominal interest rate spreads between riskier and safer bonds, 18852003.
During the Electrication period, spreads rose between 1894 and 1907, which is when
uncertainty about the usefulness and possibilities for adoption of the new technology
was greatest. Spreads fell after that as the future of Electricity became clearer. In the
IT era, spreads have a generally-upward trend throughout, though they did fall for a
while in the late 1990s. This may well reect the lag in the widespread adoption of IT.
The spreads sharp rise in 1930 and very slow decline over the next 15 years probably
has to do with the macroeconomic instability induced by events prior to and during
the Great Depression, and then the heavy borrowing by the U.S. government to nance
World War II, which raised rates on T-bills.
Another measure of risk perceptions can be obtained from the distribution of rat-
ings for issues of new corporate bonds. Figure 25 uses data from Hickman (1958,
pp. 153154) and Atkinson (1967, p. 97) for the period from 19081965 to show four-
year averages, starting at the dates shown on the horizontal axis, of the percent of the
total par value of rated new corporate bond issues that received a Moodys rating of
single-A or lower and Ba or lower. In other words, the solid line excludes the highest
rated bonds (i.e., classes Aaa and Aa), but includes some investment grade bonds (i.e.,
A and Baa) along with the sub-investment grades (i.e., Ba and lower). The dashed line
includes only the sub-investment grades.
The dashed line in Figure 25 indicates that subinvestment grade bonds made up a
larger part of the value of total rated new issues during the Electrication era than af-
ter the start of the Great Depression, and though these data end in 1965, we note that
subinvestment grade issues began to rise again only on the eve of the IT revolution
in the mid-1960s. The solid line shows that issues of bonds not receiving the highest
Moodys ratings actually rose during the latter part of the Electrication era, peaking
in the 19241927 period, which was when a host of Electricity-related innovations and
appliances were being brought to market. This does not imply an increase in junk-bond
Ch. 18: General Purpose Technologies 1215
Figure 25. Percent of rated corporate bond offerings with Moodys ratings of A or lower and Ba or lower,
four-year averages, 19081965.
issuance at this time, but rather is consistent with the view that investors recognized the
risks involved with large-scale use of the new technology and were a bit more cautious
about overpaying for debt securities associated with it.
3.5.3. The stock market performance of the young vs. old after entry
Young rms are smaller. If creative destruction does indeed mean that old rms give
way to young rms, then we should see signs of it in Figure 26, which depicts the rela-
tive appreciation of the total market value of small versus large rms since 1885.
dene small rms as those in the lower quintile of CRSP, and large rms as those
in the upper quintile. The regression line in Figure 26 (with t -statistics in parenthe-
ses) shows small rms outperforming large ones in the long run and an annual growth
premium of about 7.5 percent. But the two GPT eras do not show a faster rise in rela-
tive appreciations than other times, and this is puzzling. Surprisingly, recessions do not
seem to hurt the long-term prospects of small rms: The relative index rises in 10 of the
23 NBER recessions.
The two periods that we wish to focus on are 19291931 and the early 1970s. In
both periods, the small-capitalization rms lost out relative to the large-capitalization
ones. The rst period comes at the end of the Electrication era and the relative decline
of smaller rms is what one would have expected. But the early 1970s come at the
beginning of a new GPT, and small rms should have outperformed the large rms at
Being a total value index, this differs from the relative stock price index that is plotted in Figure 8 of Hobijn
and Jovanovic (2001). For the post-1925 period, in which they overlap, the qualitative behavior of the two
series is essentially the same.
1216 B. Jovanovic and P.L. Rousseau
Figure 26. The relative capital appreciations of small vs. large rms, 18852001.
that time. Yet the opposite happened. It is only after 1974 that the small-capitalization
rms start to perform better.
Regression evidence on age and stock market performance. If the GPT is brought in
by young rms, then the capital loss imposed by the GPTs arrival should fall more
heavily on old rms. To test this using data on individual rms, let
= age since listing of rm i in 1970;
= share (in rm is sector) of IT capital in the capital stock in 2001.
This measures a rms exposure to the impact of the new technology within its sector.
We use the change in a rms stock price over intervals that start in 1971 and end
in 1975, 1980, 1985, 1990 and 1995 as measures of expected performance. These should
reect the markets assessment of how well the rm will handle the consequences of
the GPT. The regressions take the form


= c
+ c
+ c
We summarize the rm-level results in Table 5.
The interaction between the rms age (A) and its exposure to the new technology (S)
is negative and signicant only when the period during which we measure price ap-
preciation extends to 1990 and 1995. We would have expected this coefcient to be
negative always, since older rms in sectors where IT would become important would
be less able to adjust to the new technology than newer rms. The interaction term has
a positive coefcient for the 19711975, 19711980 and 19711985 periods, but it is
Ch. 18: General Purpose Technologies 1217
Table 5
Age and stock market performance
Dependent variable: ln(P
t +i
19711975 19711980 19711985 19711990 19711995
constant 0.737 0.143 0.152 0.057 0.577
(24.3) (2.96) (2.58) (0.59) (6.06)
A 0.007 0.001 0.001 0.003 0.002
(6.40) (0.46) (0.55) (0.97) (0.51)
S 3.497 2.266 1.035 0.602 2.719
(7.60) (3.37) (1.20) (0.46) (1.88)
A S 0.047 0.043 0.016 0.122 0.106
(2.22) (1.14) (0.39) (2.09) (1.76)
0.089 0.009 0.003 0.006 0.012
N 2218 1814 1367 981 843
Note. The table presents coefcient estimates for the subperiods included in the column headings with
t -statistics in parentheses. The R
and number of observations (N) for each regression appear in the nal
two rows.
statistically signicant only for the 19711975 period. It thus seems that IT rms took a
long time to realize gains in the market after the technologys arrival. There are not very
many rms with continuous price data prior to 1900, but we have enough observations
to attempt the same regression for the Electrication era. In this case, we got


= 2.111
+ 0.213
with t -statistics in parentheses and R
= 0.015, N = 56. In this very small sample, we
do not see a direct effect of age on capital depreciation as Electrication got underway,
and the interaction term is not statistically signicant.
3.6. Consumption, interest rates, and the trade decit
If it is unanticipated, the arrival of a GPT is good news for the consumer because it
brings about an increase in wealth. How quickly wealth is perceived to rise depends on
how quickly the public realizes the GPTs potential for raising output. The rise in wealth
would raise desired consumption. But to implement the GPT rms would also need to
increase their investment. Therefore aggregate demand would rise, and in a small open
economy this would lead to a trade decit. In a closed economy, on the other hand, since
income does not immediately rise, the rise in aggregate demand would cause the rate of
interest to rise so that the rise in aggregate demand would be postponed.
How much consumption rises depends on two factors. The rst is the GPTs perva-
siveness worldwide if the entire world is equally affected then consumption could not
1218 B. Jovanovic and P.L. Rousseau
rise right away and the main effects would be transmitted though the rate of interest.
The second is the openness of the U.S. economy. Even if, say, the United States were
the only country affected by the GPT, the rise in consumption would be related to how
easily capital could ow in.
In these respects, the IT episode differs from the Electrication episode in several
important respects. Capital inows into the United States simply were not in the cards
during a large part of the Electrication episode. World War I exhausted the European
nations and the United States could not borrow from the rest of the world to nance its
electrication-led expansion it was instead a creditor during this period. Moreover,
even if the war had not taken place, it is not clear whether the United States could
have borrowed much from the rest of the world because Britain, Germany, France, and
several other countries were undergoing the same process Electrication was more
synchronized across the developed world than IT has so far been.
In sum, we would expect the United States to have behaved more like a closed econ-
omy during the Electrication era and more like a small open economy during the
IT era. Specically, we would expect to see
(1) a larger rise in the trade decit during the IT era than during the Electrica-
tion era,
(2) a smaller rise in consumption during the Electrication era then during the IT era,
(3) a larger rise in the rate of interest during the Electrication era.
3.6.1. The trade decit
Figure 27, which plots the trade decit as a percentage of GNP since 1790 along with
an HP trend, shows sharply-rising trade decits at the start of the IT revolution, though
not in the early years of Electrication.
The trade decit indeed opens up fairly dra-
matically during the IT era, whereas during the Electrication era we see a surplus. As
we mentioned, this surplus was driven by the various Colonial wars that took place at
the turn of the century and, of course, by World War I.
3.6.2. The consumptionincome ratio
We expected to see a smaller rise in consumption during the Electrication era than
during the IT era, and after we adjust for the downward long-run trend, this is indeed
what has happened. Private consumption rises gradually during each GPT era, and this
is set against a long-run secular trend for private consumption that is negative. Figure 28
shows the ratio of consumption to GDP since 1790.
As our GPT hypotheses would
GDP and total imports and exports of goods and services are from the U.S. Bureau of Economic Analysis
(2004) for 19292003. For 17901920, imports and exports are from U.S. Bureau of the Census (1975, series
U-8 and U-1, p. 864, respectively), and the GDP series are from Kendrick (1961) and Berry (1988).
The series for consumption and GDP are from the U.S. Bureau of Economic Analysis (2004) for
19292003, Kendrick (1961, table A-IIb, cols. 4 and 11, pp. 296297) for 18891929, and Berry (1988,
Ch. 18: General Purpose Technologies 1219
Figure 27. The trade decit as a percent of GDP, 17902003.
Figure 28. The ratio of consumption to income, 17902003.
suggest, the arrival of Electricity in 1890 seems to mark the end of a long-term decline
in the ratio that been underway for a century. And though the level of the series falls
during the Great Depression and World War II, never to return to its pre-1930 levels,
consumption takes another sharp upward turn near the start of the IT revolution and
continues to rise.
table 9, pp. 2526) for 17901889. The BEA gures are for personal consumption, but the Kendrick and
Berry gures include the government sector as well. Since consumption in the government sector was much
smaller prior to World War I, we suspect that the downward trend in the 19th century is a result of changing
private consumption patterns rather than a reduction in the government sectors consumption.
1220 B. Jovanovic and P.L. Rousseau
3.6.3. Interest rates
We expected a larger rise in the rate of interest during the Electrication era than during
the IT era. Relative to HP trends, the evidence is not favorable. Figure 29 shows that
ex-post real interest rates were about the same during the two GPT eras, and much
lower in the middle 40 unshaded years of the 20th century.
The dashed line is the HP
detrended series. The averages are presented in Table 6. We note that the ex-post rate
is quite high in the rst era, before 1894. If the arrival of electricity and its impact was
foreseen prior to 1894, interest rates would have risen earlier, but this probably does
not explain why they were so high then. More likely, the pre-1894 era reects a lack of
nancial development: The stock market was small then, and the nancial market not
as deep. This may have given rise to an overall negative trend in interest rates over the
134-year period as a whole.
Figure 29. The ex-post real interest rate on commercial paper, 18702003.
Table 6
Era Ex-post real interest rate
18701893 7.78
18941930 2.61
19311970 0.16
19712003 2.75
Commercial paper rates are annual averages from the FRED database for 19342003 and from Homer
and Sylla (1991) for earlier years. We compute the ex-post return by subtracting ination as computed by
the growth of the implicit price deator for GNP from the U.S. Bureau of Economic Analysis (2004) for
19292003 and Berry (1988) for earlier years.
Ch. 18: General Purpose Technologies 1221
4. Conclusion
Technological invention is uneven, and comes in bursts; that much has for a long time
been clear to students of growth. Electricity and IT are, to most observers, the two most
important GPTs to date, or at least they seem so according to the three criteria that
Bresnahan and Trajtenberg proposed. In this chapter we have analyzed how the U.S.
economy reacted to the creation of these two GPTs. Having discussed in detail GPTs
with reference to the Electrication and IT eras, we believe that we have shown that the
concept is a good way to organize how we think of technological change and its effects.
The Electricity and IT eras differ in some important ways. Electrication was more
broadly adopted, whereas IT seems to be technologically more revolutionary. The pro-
ductivity slowdown is stronger in the IT era but the ongoing spread of IT and its
continuing precipitous price decline are reasons for optimism about growth in the com-
ing decades relative to what happened in the middle of the 20th century following the
spread of Electricity. But it is the similarities between the two epochs that are the most
instructive and that will guide our expectations about how the next GPT will affect
economic life when it comes along.
We thank Jason Cummins, Bart Hobijn, Josh Lerner and Gianluca Violante for provid-
ing us with some of the data used here. This research was supported in part by National
Science Foundation Grant No. 30-3207-00-0079-286.
Atkeson, A., Kehoe, P.J. (1993). Industry evolution and transition: The role of information capital. Staff
Report No. 162. Federal Reserve Bank of Minneapolis, Minneapolis, MN.
Atkinson, T.R. (1967). Trends in Corporate Bond Quality. With the assistance of E.T. Simpson. Columbia
University Press, New York, for National Bureau of Economic Research.
Bahk, B.H., Gort, M. (1993). Decomposing learning by doing in plants. Journal of Political Economy 101,
Bartel, A., Lichtenberg, F. (1987). The comparative advantage of educated workers in implementing new
technology. Review of Economics and Statistics 69, 111.
Berndt, E.R., Dulberger, E.R., Rappaport, N.J. (2000). Price and quality of desktop and mobile personal
computers: A quarter century of history. Working paper. MIT Sloan School, Cambridge, MA.
Berry, T.S. (1988). Production and population since 1789: Revised GNP series in constant dollars. Bostwick
Paper No. 6. The Bostwick Press, Richmond, VA.
Bessen, J. (2002). Technology adoption costs and productivity growth: The 70s as a technology transition.
Review of Economic Dynamics 5, 443469.
Board of Governors of the Federal Reserve System (1976). Banking and Monetary Statistics 19141941.
Board of Governors of the Federal Reserve System, Washington, DC.
Boldrin, M., Levine, D.K. (2001). Growth cycles and market crashes. Journal of Economic Theory 96,
1222 B. Jovanovic and P.L. Rousseau
Bradstreet Co. (18851925). Bradstreets. Bradstreet Co., New York.
Brady, D.S. (1966). Price deators for nal product estimates. In: Brady, D.S. (Ed.), Output, Employment,
and Productivity in the United States After 1800. Columbia University Press, New York, pp. 91116.
Bresnahan, T.F., Trajtenberg, M. (1996). General purpose technologies: engines of growth?. Journal of
Econometrics, Annals of Econometrics 65, 83108.
Caballero, R.J., Hammour, M.L. (1994). The cleansing effect of recessions. American Economic Re-
view 84, 13501368.
Comin, D. (2002). Comments on James Bessens Technology adoption costs and productivity growth: The
70s as a technology transition. Review of Economic Dynamics 5, 470476.
The Commercial and Financial Chronicle (18851925).
Cowles, A., Associates (1939). Common Stock Price Indexes. Cowles Commission for Research in Eco-
nomics, Monograph No. 3, second ed. Principia Press, Bloomington, IN.
Cummins, J.G., Violante, G.L. (2002). Investment specic technical change in the United States (1947
2000): Measurement and macroeconomic consequences. Review of Economic Dynamics 5, 243284.
David, P. (1991). Computer and dynamo: The modern productivity paradox in a not-too-distant mirror. In:
Technology and Productivity: The Challenge for Economic Policy. OECD, Paris, pp. 315347.
Devine, W.D. (1983). From shafts to wires: Historical perspectives on electrication. Journal of Economic
History 43, 347372.
Dow Jones and Company, Inc. (19972004). The Wall Street Journal. Dow Jones, New York.
DuBoff, R.B. (1964). Electric power in American manufacturing, 18891958. Ph.D. Dissertation. Univer-
sity of Pennsylvania.
Dun and Bradstreet, Inc. (2003). D&B Million Dollar Directory. Dun and Bradstreet, Inc., Bethlehem, PA.
Feder, B. (1988). Advances in drugs, courtesy of computers. New York Times (August 3), 5.
Federal Reserve Bank of St. Louis (2004). FRED Database. Federal Reserve Bank of St. Louis, St. Louis,
Gates, B. (1999). Business @ the Speed of Thought. Warner Books, New York.
Goldin, C., Katz, L.F. (1999a). The shaping of higher education: The formative years in the United States,
1890 to 1940. Journal of Economic Perspectives 13, 3762.
Goldin, C., Katz, L.F. (1999b). The returns to skill in the United States across the twentieth century. Work-
ing Paper No. 7126. National Bureau of Economic Research, Cambridge, MA.
Gordon, R.J. (1990). The Measurement of Durable Goods Prices. University of Chicago Press, Chicago, IL.
Gordon, R.J. (2000). Does the new economy measure up to the great inventions of the past?. Journal of
Economic Perspectives 14, 4974.
Gort, M. (1969). An economic disturbance theory of mergers. Quarterly Journal of Economics 94, 624642.
Greenwood, J., Jovanovic, B. (1999). The information-technology revolution and the stock market. Ameri-
can Economic Review Papers and Proceedings 89, 116122.
Greenwood, J., Jovanovic, B. (2001). Accounting for growth. In: Hulten, C.R., Dean, E.R., Harper, M.J.
(Eds.), New Developments in Productivity Analysis. University of Chicago Press, Chicago, IL, for Na-
tional Bureau of Economic Research, pp. 179222.
Greenwood, J., Seshadri, A., Yorukoglu, M. (2002). Engines of liberation. Economie dAvant Guarde 2.
Greenwood, J., Yorukoglu, M. (1997). 1974. CarnegieRochester Conference Series on Public Policy 46,
Griliches, Z. (1957). Hybrid corn: An exploration in the economics of technological change. Economet-
rica 25, 501522.
Griliches, Z. (1969). Capital-skill complementarity. Review of Economics and Statistics 51, 465468.
Helpman, E., Trajtenberg, M. (1998a). A time to sow and a time to reap: Growth based on general purpose
technologies. In: Helpman, E. (Ed.), General Purpose Technologies and Economic Growth. MIT Press,
Cambridge, MA, pp. 5583.
Helpman, E., Trajtenberg, M. (1998b). The diffusion of general purpose technologies. In: Helpman, E.
(Ed.), General Purpose Technologies and Economic Growth. MIT Press, Cambridge, MA, pp. 85119.
Hickman, W.B. (1958). Corporate Bond Quality and Investor Experience. Princeton University Press, Prince-
ton, NJ, for National Bureau of Economic Research.
Ch. 18: General Purpose Technologies 1223
Hobijn, B., Jovanovic, B. (2001). The IT revolution and the stock market: Evidence. American Economic
Review 91, 12031220.
Homer, S., Sylla, R. (1991). A History of Interest Rates, third ed. Rutgers University Press, New Brunswick,
Hoovers, Inc. (2000). Hoovers Online: The Business Network. Hoovers, Inc., Austin, TX.
Hopenhayn, H.A. (1992). Entry, exit, and rm dynamics in long run equilibrium. Econometrica 60, 1127
Hornstein, A., Krusell, P. (1996). Can technology improvements cause productivity slowdowns?. NBER
Macroeconomics Annual 1996, 209259.
Jovanovic, B., Nyarko, Y. (1996). Learning by doing and the choice of technology. Econometrica 64, 1299
Jovanovic, B., Rousseau, P.L. (2001). Why wait? A century of life before IPO. American Economic Review
Papers and Proceedings 91, 336341.
Jovanovic, B., Rousseau, P.L. (2002a). Moores law and learning-by-doing. Review of Economic Dynam-
ics 4, 346375.
Jovanovic, B., Rousseau, P.L. (2002b). The Q-theory of mergers. American Economic Review Papers and
Proceedings 92, 198204.
Jovanovic, B., Rousseau, P.L. (2002c). Mergers as reallocation. Working Paper No. 9277. National Bureau
of Economic Research, Cambridge, MA.
Kelley, E.M. (1954). The Business Founding Date Directory. Morgan and Morgan, Scarsdale, NY.
Kendrick, J. (1961). Productivity Trends in the United States. Princeton University Press, Princeton, NJ.
Kortum, S., Lerner, J. (1998). Stronger protection or technological revolution: What is behind the recent
surge in patenting?. CarnegieRochester Conference Series on Public Policy 48, 247304.
Krusell, P., Ohanian, L.E., Rios-Rull, J.V., Violante, G.L. (2000). Capital-skill complementarity and inequal-
ity: A macroeconomic analysis. Econometrica 68, 10291053.
Kuznets, S. (1961a). Capital in the American Economy: Its Formation and Financing. Princeton University
Press, Princeton, NJ.
Kuznets, S. (1961b). Annual estimates, 18691955. Manuscript. Johns Hopkins University, Baltimore, MD.
Laitner, J., Stolyarov, D. (2003). Technological change and the stock market. American Economic Re-
view 93, 12401267.
Lerner, J. (2000). 150 years of patent protection. Working Paper. National Bureau of Economic Research,
Cambridge, MA.
Lichtenberg, F., Siegel, D. (1987). Productivity and changes in ownership of manufacturing plants. Brook-
ings Papers on Economic Activity 3, 643673. Special Issue on Microeconomics.
McGuckin, R., Ngyen, S. (1995). On productivity and plant ownership change: New evidence form the
longitudinal research database. RAND Journal of Economics 26, 257276.
Moodys Investors Service (1920, 1928, 1955, 1980). Moodys Industrial Manual. Moodys Investors Service,
New York.
Nelson, R.L. (1959). Merger Movements in American Industry, 18951956. Princeton University Press,
Princeton, NJ, for National Bureau of Economic Research.
Nelson, R., Phelps, E. (1966). Investment in humans, technological diffusion, and economic growth. Amer-
ican Economic Review 56, 6979.
The New York Times Co. (18971928). The New York Times. The New York Times Co., New York.
The New York Times Co. (19131925). The Annalist: A Magazine of Finance, Commerce, and Economics.
The New York Times Co., New York.
Rousseau, P.L. (1999). Share liquidity and industrial growth in an emerging market: The case of New
England, 18541897. Historical Working Paper No. 103. National Bureau of Economic Research, Cam-
bridge, MA.
Schoar, A. (2002). Effects of corporate diversication on productivity. Journal of Finance 57, 23792403.
Schwert, G.W. (2000). Hostility in takeovers: In the eyes of the beholder?. Journal of Finance 55, 2599
1224 B. Jovanovic and P.L. Rousseau
Standard and Poors Corporation (1981, 1988, 2000). Stock Market Encyclopedia. Standard and Poors Cor-
poration, New York.
Standard and Poors Corporation (19712003). Stock Reports. Standard and Poors Corporation, New York.
United States Bureau of the Census, Department of Commerce (1975). Historical Statistics of the United
States, Colonial Times to 1970. Government Printing Ofce, Washington, DC.
United States Bureau of the Census, Department of Commerce (1980, 1992, 2003). Statistical Abstract of the
United States. Government Printing Ofce, Washington, DC.
United States Bureau of the Census, Department of Commerce (20002004). Current Population Survey.
Government Printing Ofce, Washington, DC.
United States Bureau of Economic Analysis (2002). Survey of Current Business. Government Printing Ofce,
Washington, DC.
United States Bureau of Economic Analysis (2004). Survey of Current Business. Government Printing Ofce,
Washington, DC.
University of Chicago Center for Research on Securities Prices (2004). CRSP Database. University of
Chicago Center for Research on Securities Prices, Chicago, IL.
Wiggins, S.N., Raboy, D.G. (1996). Price premia to name brands: An empirical analysis. Journal of Indus-
trial Economics 44, 377388.
Williamson, J.G., Lindert, P.H. (1980). American Inequality: A Macroeconomic History. Academic Press,
New York.
Wilson, J.W., Jones, C.P. (2002). An analysis of the S&P 500 index and Cowless extensions: Price indexes
and stock returns, 18701999. Journal of Business 75, 505533.
Abstract Attention has increasingly shifted towards the long-run perspective on
technological innovation, which suggests that progress comes in waves, each one
originating with a major breakthrough or general purpose technology (GPT). This
paper seeks to assess whether nanotechnology is likely to be (or become) a GPT, a
characteristic that other researchers have sometimes assumed though not necessarily
documented. Based on a survey of existing literature, this paper will explore the
extent to which nanotechnology addresses three primary characteristics of a GPT:
pervasiveness, innovation spawning, and scope for improvement. The paper draws
on patent and patent citation databases to highlight the types of quantitative and
qualitative information that would be necessary, and in some instances is still
lacking, to characterize fully the nature of nanotechnology.
Keywords Nanotechnology General purpose technology Patent analysis
JEL Classications 0330 0300 0340
1 Introduction
Whenever a new class of technologies emerges, conjectures are advanced on how
likely it is that they will change rms productivity, household production, con-
sumption patterns, and socio-economic relationships. If a core technology has a
J. Youtie (&)
Georgia Tech Enterprise Innovation Institute, Atlanta, GA 30332-0640, USA
e-mail: jan.youtie@innovate.gatech.edu
M. Iacopetta
School of Economics, Georgia Institute of Technology, Atlanta, GA 30332-0615, USA
e-mail: maurizio.iacopetta@econ.gatech.edu
S. Graham
College of Management, Georgia Institute of Technology, Atlanta, GA 30332-0520, USA
e-mail: stuart.graham@mgt.gatech.edu
J Technol Transfer
DOI 10.1007/s10961-007-9030-6
Assessing the nature of nanotechnology: can we uncover
an emerging general purpose technology?
Jan Youtie Maurizio Iacopetta Stuart Graham
Springer Science+Business Media, LLC 2007
substantial and pervasive effect across the whole of society, it is often termed a
General Purpose Technology (GPT). The dissemination of electricity at the turn
of the 19th century is often said to have the character of a GPT, with reference
made to the long wave of downstream innovations spawned by the electric dynamo
that reshaped the functioning of the economy. Similarly, the dissemination of
microelectronics in the last quarter of the 20th century has in it the hallmarks of a
GPT in that it led to new forms of organizations, new products, and has increased
the level of competition in service goods that were traditionally produced and
consumed locally.
The question we explore in this paper is whether the family of nanotechnologies
has the potential of inducing changes in the economy that are comparable in scope to
electricity, information and communications technology (ICT), and others that have
been previously documented as major breakthroughs.
We discuss the question by drawing on techniques and ideas from two interrelated
streams of research. One line of research has hypothesized that the long-run
behavior of the nancial market and of macro aggregates are best understood by
investigating the conditions that have favored the arrival and the process of
dissemination of major technologies. The main idea from this literature is that
technological change follows a sequence of events in which a major technological
innovation is preceded by a number of smaller inventions that expand the range of
applicability of the core technology, the so-called General Purpose Technology.
In this paper we will briey summarize the main features an innovation should have
to be part of the club of GPTs and discuss the prediction of theories that explain the
rise and fall of productivity and of rms value as the outcome of the dissemination
of a GPT.
We also draw from a stream of research that describes and characterizes tech-
nological developments by means of quantitative data taken from patent datasets. In
particular, we propose a comparison of the level of generality of nanotechnologies
relative of that of ICT (usually considered a GPT) and innovations in the drug
industry (not considered a GPT). We suggest that the kind of tests proposed in the
literature are not easily applicable to emerging technologies because these tests have
been devised for situations in which a considerable amount of historical data has
been recorded. Nevertheless, the estimations that we perform seem to suggest that
nanotechnology satises at least one major feature of a GPT, namely that of gen-
The paper begins with an introduction of the main attributes of a GPT and briey
explains the extent to which the existing literature may be used to test whether
nanotechnology has one or more of these attributes. Section 3 shows how macro
aggregates are predicted to respond to the arrival and the dissemination of a GPT.
Section 4 illustrates strategies that have been used to identify a GPT. Section 5
introduces the Generality Index and estimates it for nanotechnology and two other
classes of technologies. A conclusion follows.
2 The GPT concept and nanotechnology
Previous research has suggested that a GPT must have at least three attributes:
pervasiveness, an innovation spawning effect, and scope for improvement (Helpman
& Trajtenberg, 1994). Pervasiveness is intended to reect the performance of some
J. Youtie et al.
function that is vital to the functioning of a large segment of existing or potential
products and production systems. Bresnahan and Trajtenberg (1995, p. 4) argue that
continuous rotary motion and binary logic are the pervasive elements of steam
power and ICT, respectively, each of which is considered a GPT.
A pervasive technology would have relatively little visibility in the functioning of
the economy unless it fostered new inventions that directly or indirectly result from
the early major invention. For instance, the dynamo led to the invention of both the
light bulb and electric motor, and stimulated major innovation in plant and urban
design (David, 1990). Similarly, the microchip led to an explosion of imaging
technologies, memory devices, and digital technologies.
Helpman and Trajtenberg (1994) suggest that such widespread adoption of a core
technology is a consequence of a variety of actors coordinating their beliefs about
the promise of the technology. Complementary technologies are developed as long
as the various actors involved share beliefs that the GPT is spawning innovations in
multiple technological areas. Indeed, widespread market adoption may be a con-
sequence of the settling of beliefs among scientists, entrepreneurs, established
businesses, government, and consumers.
It remains a theoretical and empirical question whether the core technology of
these breakthroughs could be improved substantially. Evidence for the scope of
improvement for ICT was cleverly summarized by Moores Law which predicted
that the force of competition would stimulate the semiconductor industry to double
the number of transistors per chip every 1824 months. While the regularity of
Moores Law has been observed, it is not clear whether its regularity results from
technological factors or from industry coordination around a smooth and predicable
trajectory with clear transaction-cost benets.
Theory suggests that all three aspects will be present in the true breakthrough
technologies, those most widely used by rms and households. The mere fact that
an innovation can be applied in several areas of production (pervasiveness) does
not mean that it will be used. In order for society to employ the technology
pervasively, its adoption must be convenient from a cost-consideration point of
view, that is, it must reach a certain level of efciency (scope for improvement),
and it must lead to the development of new secondary or complementary
technologies (innovation spawning). Some authors add a fourth element to the
denition of a GPT, that of wide dissemination (Lipsey, Bekar, & Carlaw, 1998),
although this element is often considered a logical consequence of the other three
Are there indications that these three basic GPT attributes might be present in
nanotechnology? Can we say that nanotechnology performs (or will perform) a
generic function, whose efciency will be greatly improved over time, perhaps as
much as that of the microprocessor, and that it stimulates the appearance of new
kinds equipment comparable to the modem, or memory storage devices?
With a new technology it is hard to conjecture what aspect of it, if any, will
perform a generic function. Although most scholars who are engaged in nano-
science agree that nanotechnology is very small in scale, (in the range of
1100 nanometers (nm), with one nanometer equaling one billionth of a meter), only
a few seem to embrace the notion that if a technology is small in scale it should be
considered nanotechnology. From the perspective of this analysis such a change in
scale could be paralleled to a generic function, notwithstanding the disagreements
among nano-scientists. This would be the case, for instance, if a new scientic
Assessing the nature of nanotechnology
principle or a new methodology allows a signicant drop in scale that leads to a
radical transformation in the range of inputs used in production.
For instance, many industries have voiced concerns over the limits to Moores
Law, recognizing that any stop to the exponential growth of computing capability
would have economic consequences (Harriott, 2000). Nanotechnology, however, has
the potential to sustain circuit density increases through small scale lithography
alternatives, such as nanoimprint lithography, or eventually self-assembly (Arnold,
1995). Moreover, matter at the nanoscale has been shown to exhibit novel properties
that cannot be projected from larger or smaller scales (Kostoff et al., 2006;
Tannenbaum, 2005). These novel behaviors, and the human skills to manipulate and
engineer them, could form the basis for future pervasive applications.
The GPT characteristic of innovation spawning may be embodied in evidence
of a nanotechnology-oriented value chain of initial, intermediate, and downstream
innovations. Lux Research (2006) identies one such nano-value chain, consisting of
an initial set of nanomaterials (such as carbon nanotubes), employed by market
intermediaries to create coatings that enhance properties of nishes. The nal
products then integrate these coatings into a diverse set of products (which may
include automobiles, airplanes, electronics displays, nano-treated clothes, refriger-
ator surfaces with microorganism growth inhibitors, and self-cleaning windows that
oxidize organic matter, among others). Lux Research suggests that this nano-value
chain is supported by a set of tools including scanning probe microscopes,
nanofabrication tools, and computer modeling systems (See also Meyer, 2006 for an
alternative perspective on the nano-value chain).
Figure 1 presents a schematic, sequencing both science and commercial tech-
nologies. The sequence begins with scientic and technological discoveries in
instruments (such as scanning tunneling microscopes or STM and atomic force
microscopes or AFM) and nanomaterials (such as buckminsterfullerenes and carbon
nanotubes), forming the basis for a value chain. Intermediate and complementary-
goods producers are offered, including nanocoatings and composites manufacturers,
nano-core processing and memory. The end of the value chain shows a broad range
of end-use goods. The gure shows that the boundaries between positions in the
value chain overlap.
It is fair to assume that the development of a complete value chain will more
likely follow a coordination of beliefs. In nanotechnology, this coordination of
Fig. 1 Potential sequence of science, intermediate goods, and nal products in nanotechnology
J. Youtie et al.
beliefs appears to be taking hold. Evidence of coordination in science includes
Richard Feynmans legendary talk at the American Physical Societys annual
meeting in December 1959 (Theres Plenty of Room at the Bottom), Eric
Drexlers Engines of Creation (1986) and subsequent formation of The Foresight
Institute, the launch of Nanotechnology by the Institute of Physics as a multidisci-
plinary science and engineering journal in 1990, the creation of the Feynman Prize
rst awarded in 1993 to recognize eminent research in nanotechnology, and the
notion that advances at the nanoscale are situated in a convergence of disciplines
(Roco & Bainbridge, 2003).
Coordination in the public and private sector may be inferred from the intro-
duction of specic nanotechnology patent classes and cross-referencing categori-
zations: International Patent Class B82, the Japanese Patent and Trade Ofce
(PTO) Class ZNM, the US Patent and Trademark Ofce (USPTO) Class 977, and
the European Patent Ofce Class Y01N. In addition, there are multiple profes-
sional associations (e.g., NanoBusiness Alliance) and trade magazines (e.g., Small
Times) dedicated to nanotechnology. Signicant coordination of consumer beliefs
may be decades away from occurring, although several references to nanotech-
nology are evident in popular press and public policy documents. For instance,
popular media such as Michael Crichtons Prey (2002) has portrayed risks from
nanotechnology, while public policy has responded with legislation. In 2003, the US
Congress enacted The 21st Century Nanotechnology Research and Development
Act (Public Law 108-153) which includes specic mention of societal concerns. This
statutory mandate has been followed by the NSFs allocation of resources to create
a network of science museums and research centers to help educate the public
about nanotechnology and social change (NSF, 2005).
Scope for improvement in the family of nanotechnologies may lie in reductions
in size, lower costs, and greater complexity. Although nanotechnology is still at a
relatively early stage, advances have occurred in semiconductor manufacturing
technology (advancing from 90 nm to 45 nm during 20052007) (Kanellos, 2005,
2006), and in instrument costs (Atomic Force Microscopes can be obtained at
lower prices and/or with greater availability of features at existing prices) (Lux
Research, 2006). It is also expected that the eld of nanotechnology will like ICT
evolve in terms of complexity, with Roco (2004, 2005) suggesting that nanotech-
nology will undergo four generations of development over the next 20 years.
Armed with such information, several researchers have proposed that nano-
technology is a GPT. Huang et al. (2003) demonstrate through patent analysis
that nanotechnology covers a wide range of classes, although Porter, Shapira, and
Youtie (2006) criticize the use of an overly broad denition. Moreover, Shea
(2005) suggests that nanotechnology is a GPT because it is likely a disruptive and
radical technology, but the authors approach does not specically address the
concepts of pervasiveness, innovation spawning, and scope for improvement.
While Palmberg and Nikulainen (2006) do examine whether nanotechnology
exhibits these three characteristics of a GPT, they do not apply methods
commonly used to test for them.
Counting the number of patents by year (showing increases over time) or patents
by patent classication (showing increasing diversity) are commonly used to claim
The disciplinary convergence hypothesis is not without its detractors (See for example, Khushf,
Assessing the nature of nanotechnology
that nanotechnology is a GPT. However, such information may not in and of itself be
adequate to make any such determination. Hall and Trajtenberg (2004), for instance,
nd that GPTs (as measured by patent citations) do not necessarily have dispro-
portionately higher growth rates in terms of newly issued patents. Moreover, they
argue that some patent classications tend to be more broad-based than others,
particularly chemical-related classications. They suggest that patent classes by
themselves do not provide sufcient substantiation of breadth and pervasiveness of a
candidate GPT.
3 Why GPTs are important
In the rest of this paper, we explore the gap between studies like Palmberg and
Nikulainen (2006), which have claimed that nanotechnology is a GPT, and Hall and
Trajtenberg (2004), which nd that there are problems with using counting
methodologies. In so doing, we rst nd it necessary to justify the assessment of
whether a technology is a GPT. We see three chief reasons for engaging in this
exercise, contending that there is value to understanding (1) the returns to effective
government innovation policy; (2) the technological drivers of economic growth; and
(3) the manner in which society could most effectively prepare for these broad
technological changes.
3.1 R&D policy
An effective R&D policy would have in it some element of spurring innovation.
It is commonly believed that measures aimed at making intellectual property rights
stronger, or inventors appropriability of the social surplus generated by inventions
greater, tend to alter the supply of innovation, without regard to missed oppor-
tunities for diffusion. In the GPT theory, the improvement of the core technology
goes hand in hand with the range of applications it stimulates in various sectors of
the economy. The lower the price of capital embodying complementary technol-
ogieswhich facilitate the dissemination of the core technologythe wider the
range of adoption, and the brighter the prospects of returns on R&D investments
directed at improving and expanding the scope of the GPT (Brenhan &
Trajtenberg, 1995). However, this argument does not hold in a context in which
innovations are unrelated to each otheras it is assumed by most growth theories
(Aghion & Howitt, 1992; Romer, 1990). In such circumstances, any positive
spilloversgoing from the inventors to the usersare detrimental to the
innovation rate.
It is not difcult to nd cases suggesting that improvements in a GPT are asso-
ciated, to some extent, with the level of appropriability (i.e., how easily the prots
from innovations can be captured). In the semiconductor industry, the level of
appropriability is considered low in comparison to other industries, and yet the
industry has exhibited spectacular improvements in both invention and economic
growth (Irwin & Klenow, 1994). Clearly, the diffusion of information technologies in
many economic sectors has been facilitated by the steep decline in the production
cost of the microchip. Therefore, if a technology is a GPT it may be more efcient to
resolve the classical tension between creating monetary incentives for innovators
J. Youtie et al.
and fostering the diffusion of innovation by opting for a relatively high level of
3.2 Economic change
The second reason for investigating the nature of a technology is that any such
analysis provides insight into the source of economic expansions or slowdowns. In
the United States, the economy experienced sustained high-output growth during
the 1960s, while from the early 1970s to the early 1980s output growth was low
relative to the post-WWII average. Since the mid-1990s there has been, for the most
part, a return to strong growth. Helpman and Trajtenberg (1994) argue that these
patterns are associated with the diffusion of GPTs.
When considering economic growth, one view suggests that the appearance of a
GPT is followed by two distinct phases. In the rst phase, resources are diverted
from existing production activity to the creation of new technologies complementary
to the GPT. This redeployment of resources from one sector to the other would
cause wage and labor productivity rates to stagnate or even decline. This phase is
often called the time to sow, for the economy is developing technologies that are
unproductive in the short-run either because they are not yet efcient or because
adopters do not possess the necessary skills and knowledge to use them efciently.
This period of economic slowdown may persist.
It has been argued that the productivity slowdown that lasted for almost 25 years
(starting sometime in the rst half of the 1970s in the US and other advanced
countries) was partly due to the spread of computers. The introduction of computers
would have rendered obsolete existing skills and would have caused the abandon-
ment of existing routines. This view sees the slowdown of productivity as the cost the
economy suffered to modernize production and upgrade the types of skills needed in
the New Economy. Paul David (1990) draws a parallel between the diffusion of
electricity and computers, arguing that in both cases there was a long delay between
the introduction of the GPT and the corresponding productivity surge. It is natural
to pose a similar parallel between ICT and nanotechnology, and we explore such a
relationship below.
3.3 Societal synchronization
To the extent that a GPT can produce economic benets, and it requires a high level
of synchronization in society, the identifying of a GPT may be benecial in allowing
society to plan for a higher level of needed synchronization. Under the theory of
GPTs, coordination between inventors and users expectations about the usefulness
of the emerging technology is considered vital. Improvements in the GPT requires
R&D investments, which will likely be made only if the investors expect the
development of new complementary technologies or the renement of existing
technology in downstream sectors. In turn, complementary technologies will emerge
only if inventors are optimistic about the prospect that the GPT will be widely
Helpman and Trajtenberg (1994) developed a formal framework showing the
dynamic links between the core and complementary technologies. One important
aspect that emerges from their analysis is that at any given point in time, researchers
Assessing the nature of nanotechnology
must decide whether to devote resources to developing complementary technologies
associated with an existing technology, or to a new GPT. The choice depends not
only on what happened in the past but also on expectations about the future. The
decision is also affected by the past, because a new complementary technology is
more productive in an economy in which a wide range of other complementary
technologies have been developed.
For these reasons, an incumbent GPT has advantages over a new one. The
conjecture is that although a new GPT is more productive when it is combined with
the same number of technologies that complement an existing GPT, it is never-
theless not as productive when the range of new complementary technologies is
small. If investors believe that at a certain point a large number of complementary
technologies associated with the new GPT are forthcoming, then it is more likely
that these technologies will be developed. Otherwise, the economy may become
trapped in an equilibrium where the new GPT is never adopted and the few rms
who ventured to invest in it will fail.
4 Strategies to identify a GPT
In the foregoing, we have (1) suggested that identifying whether a technology is a
GPT is a valuable exercise, and (2) cataloged other studies nding that nanotech-
nology exhibits the characteristics of a GPT. In this section, we identify what we
believe is a more systematic approach, one that has been undertaken in recent years
by empirical scholars and economic historians in assessing whether a technology is a
GPT, and apply it to nanotechnology. The objective of this discussion is to both
examine whether nanotechnology holds up as a GPT, and also to identify tools
that may be used to test other emerging technologies.
Jovanovic and Rousseau (2005) propose new ways to use historical evidence to
test for the existence of scope for improvement, wide range of use, and the
likelihood of spawning complementary innovations. They capture a technologys
scope for improvement by analyzing the decline of equipment prices (nding that
this decline was greater for ICT than for electricity). The authors suggest that wide
range of use (pervasiveness) can be analyzed, in their case by comparing the
amount of electric power as a percentage the total horsepower in several manu-
facturing sectors with an estimate of the percentage of ICT investments across the
same sectors. Electricity exhibited dissemination across a broader range of industries
than did ICT over a comparable time period.
The authors also examine the likelihood that a technology will spawn comple-
mentary innovations by using patent analysis. They rely on a notion that patents
presage investment in new technologies, representing the rise in initial public
offerings (IPOs) and the subsequent change in the structure of capital in favor of
new technologies. The authors nd innovation spawning in the surge of
IPO activity that followed the introduction of both electricity and information
Hall and Trajtenberg (2004) and Moser and Nicholas (2004) venture beyond a
simple tally of the number of patents to contend that the data can reveal much more
on the question of whether a technological development is a GPT. Instead of
measuring the pervasiveness of a GPT by looking at how a new technology affects
the composition of capital across sectors, these authors look at the extent to which
J. Youtie et al.
patents associated with a GPT are cited outside the technology area or industry in
which the GPT originated. To address the diffusion delay, which has been associated
with a prolonged productivity slowdown followed by an acceleration of productivity,
this literature measures the citation lags (the amount of time between the issue of a
focal patent and the issue of a future patent that cites back to it), which they contend
should be longer for GPTs than for incremental technologies. These authors
measure the scope for improvement not as a reduction in production cost, but
instead with the number of citations within the technological area to which the GPT
belongs. An interesting question that the patent literature addresses is the date of
arrival of a GPT. Historians often pick a specic event, usually identied with a
major investment (e.g., for electricity it has been identied as the construction of a
hydro-electric facility at Niagara Falls, New York). Macroeconomists tend to
consider an arbitrary threshold in the data, for instance investments in the new
technologies rising above a certain percentage of overall investment. In contrast, the
patent literature utilizes an Originality measure, which allows for a determination
of the date of arrival of inuential innovations based on forward citations.
Does the patent approach lead to the same conclusions as the macro approach?
Moser and Nicholas contend that in the case of electricity it does not. These authors
nd that patented inventions associated with electricity led in the 1920s were not as
pervasive as were chemical and mechanical inventions, because these latter
inventions were cited by later patents outside their technological areas more often
than were electricity inventions. Moreover, chemical and mechanical inventions
were cited more often in general, indicating to the authors a stronger propensity to
spur innovation.
These differing conclusions may stem from chemicals and mechanical industries
being more science-based than was electricity. This observation is an important
warning for all nanotechnology investigations that rely exclusively on patents data.
There is a risk that the traditional test developed by Hall and Trajtenberg (2004)
may be biased in the sense that a science-based technology is more likely to be
designated as a GPT, even when such a determination is a spurious result, and not
borne out by a judgment made after considering the economic effects that it actually
5 Evidence from patent data
By and large, employing patent data to uncover evidence of a general purpose
technology is a problematical exercise when studying emerging technologies.
Patent data, by its nature, offers information about the current state of a technology,
and more commonly about the past development of that technology. Because new
technologies are in the process of emerging, the patent characteristics that have
traditionally been collected are either not available, or numbers are rather small and
thus prone to statistical error. So, empirical research studying whether technologies
are or were GPTs has been undertaken only after the technology has matured
sufciently so as to allow researchers to collect an adequate body of information
from the patent record (Hall & Trajtenberg, 2004 (ICT); Moser & Nicholas, 2004
Because nanotechnology is an emerging technology, we are faced with the same
limitations. To demonstrate the embryonic nature of this technology area, we
Assessing the nature of nanotechnology
present data on the patenting of nanotechnology inventions 19832005 (Fig. 2). Four
trends of granted patents are plotted in Fig. 2, based on two different denitions of
nanotechnology patents.
One denition is derived from the classication system
used at the USPTO, while another is based upon the keyword system built by the
CNS-ASU team (Porter et al., 2006). Note that the latter denition produces sub-
stantially more US nanotechnology patents during the 19832005 period (12,553
NSF-CNS patents versus 2,639 USTPO-dened patents). Figure 2 presents alter-
native counts within each denition, measuring each patent both in its year of
application and grant.
The time trends demonstrate that nanotechnology has been a slowly growing
technology space. Signicant application activity did not begin until the late 1980s
(under either denition), with relatively small numbers of patents being issued until
the 1990s. This relative paucity in the patent record is particularly problematic for
the uncovering of GPTs since the primary measurement of this characteristic of
technologies has been made with large numbers of patents, using the patents
forward citation stream.
All patent applicants are required by law to disclose the prior art (patents and
non-patents) upon which the instant invention builds. Following some give-and-take
with the patent ofce (Graham, 2006), and after the patent examiner has had an
Fig. 2 US nanotechnology patents, 19832005. Note: NSF CNS nanotechnology patents selected
according to a keyword list generated at Georgia Tech in cooperation with the CNS at Arizona State
University. USPTO nanotechnology patents selected according to US Patent and Trademark Ofce
classications. iss denotes patents graphed according to issue date. app denotes patents graphed
according to application year, and thus the seeming decline in applications is not a true decline but
instead a consequence of truncation because the data source is issued patents
These plots are subject to substantial right-side truncation due to the data being collected from the
granted patentsthese gures do not include information available in the US since 2001 on
published applications.
J. Youtie et al.
opportunity to add some references (Alcacer and Gittleman, 2006), the patent issues
with a list of patent backward citations to prior art. Researchers have used these
backward citations by looking to patents that issue in the future, and which cite
back to the focal patent, to measure the pervasiveness of the technology.
Hall and Trajtenberg (2004) suggest using these so-called forward citations in
uncovering GPTs by employing the HerndahlHirschman Index (HHI) of the
patent classes assigned to the focal patents forward citations. The resulting
measure of pervasiveness is termed a patents Generality Index and is dened
by the formula
Generality G
where S
= share of patent is forward citations in class j. The theory behind using
this measure is that it captures information on the extent to which the focal patent is
being applied in a wide range of technologiesthe so called pervasiveness of a
patent. As a patents Generality score approaches 1.0, we know that the patent is
being cited to by patents in a broader set of classications, and thus we can infer
that the patent is being applied more broadly in distant applications.
If we examine the Generality scores across all patents in a particular tech-
nologyin nanotechnology for instanceand compare these against the scores for
patents in other technologies, we may infer something about the pervasiveness of
the technologys application throughout the economy. Obviously, any measure built
in this way will be very sensitive to right truncation. In the study of new and
embryonic technologies in which the patent record is slowly developing, the absence
of a sufcient forward time window will pose great difculties in calculating a
useful Generality Index for individual patents, and thus entire patented
The trends depicted in Fig. 2 demonstrate that, in the emerging nanotechnologies,
substantial numbers of patents began to issue from the USPTO in the 1990s, thus
giving us a sufciently long forward window to develop credible Generality
scores on the earliest patents issued in this new technology space. We present these
data below with one important caveat: These generality scores are only represen-
tative to the extent that these early patents, and their characteristics, are represen-
tative of later patents issued in this emerging technology area.
Table 1 reports generality scores for nanotechnology patents as dened by the
NSF-CNS keyword method, and compares these with scores for drug and
computer patents dened by international patent classications (Graham, 2006).
Scores are reported for the index built from three different measures of patent
forward citation breadth: US patent classications, International patent classica-
tions, and NBER patent-database aggregated technology classes (Hall, Jaffe, &
Trajtenberg, 2001). Generality scores for 19901993 irrespective of classication
scheme demonstrate that the nanotechnology patents are more general than drugs
patents issued in the relevant year, and compare favorably with, and indeed are
higher at every reading than, computer patents. Computer patents are representative
of ICT, a technology that we previously mentioned has been found to be a GPT
(Hall & Trajtenberg, 2004).
Assessing the nature of nanotechnology
These scores on early nanotechnology patents provide us with limited evidence
that nanotechnology exhibits the pervasiveness characteristic of a GPT. As
mentioned previously, conventional measurement techniques for assessing the
existence of a GPT require sufcient time to elapse to allow for forward citations
to develop in the patent record. So, are researchers at an impasse as regards
nanotechnologyis the evidence of GPT pervasiveness hidden in the latent
nature of patent information?
Not necessarily. If ICT is a guide, then it is possible to determine whether this
technology, in its early development, also showed associated patents with signi-
cantly higher generality scores. And indeed, in 19751979, patents designated in
primary international class G06F, a class broadly representative of computer soft-
ware technologies (Graham & Mowery, 2003) showed relatively high generality
scores as compared to all other patents.
Thus, for this GPT, the patents that
appeared early in the technological trajectory demonstrated this measure of per-
vasiveness and thus we can take some condence in our nanotechnology patents
showing the same characteristic.
But, as outlined above, a GPT is not characterized by pervasiveness alone. Can
the patent record help us in determining whether nanotechnologyand indeed, any
candidate GPTexhibits evidence of coordination of beliefs among actors or a
sequential development of complementary technologies? We theorize here that
the patent record may contain information that could help in uncovering these two
characteristics of GPTs.
As regards coordination of belief, the key insight behind this characteristic of a
GPT is that different actors in society are conforming to a set of beliefs con-
cerning the wide applicability of the technology. The set of actors needed to coor-
dinate are necessarily broad, but include at least marketplace actors (such as
Table 1 Comparison of Generality Index scores across three technologies, 19901993
Nanotechnology Drugs Computers
Variable Count Mean Count Mean Count Mean
1990 Gen US 287 0.620 2188 0.386 1961 0.612
Gen IC 287 0.642 2187 0.385 1961 0.443
Gen TC 287 0.540 2187 0.273 1961 0.424
1991 Gen US 293 0.623 2405 0.394 2306 0.610
Gen IC 293 0.617 2405 0.389 2306 0.445
Gen TC 293 0.507 2405 0.278 2306 0.431
1992 Gen US 411 0.596 2349 0.387 1956 0.612
Gen IC 411 0.582 2349 0.388 1956 0.405
Gen TC 411 0.487 2349 0.268 1956 0.417
1993 Gen US 364 0.608 2499 0.380 2999 0.609
Gen IC 364 0.605 2498 0.376 2999 0.398
Gen TC 364 0.511 2498 0.264 2999 0.423
Variable denition: Gen US = Generality based on USPTO-classes; Gen IC = Generality based on
International Patent Classes; Gen TC = Generality based upon NBER patent database technology
classes (Hall et al., 2001).
These scores (built from USPTO classes) for computer patents 19751979 were 0.63, 0.65, 0.66,
0.68, and 0.67, respectively, and are signicantly higher than the scores for all patenting in those
years (0.49 in each year).
J. Youtie et al.
entrepreneurs) and government actors. Since 2001, the patent record has contained
information concerning the identity of the actor who included the citation in the
patent references: the inventor (and the inventors attorney agent), or the patent
ofce examiner (Alcacer & Gittleman, 2006). While the use of these data would
entail care due to possible endogeneity (the idea that information is owing between
the parties that leads to the citation placement),
nevertheless to the extent that the
generality measure is built from the citation record, and that citation record can be
tested to determine whether both marketplace actors (inventors) and government
actors (examiners) concur about the pervasiveness of a technology, it may prove
fruitful to use these records as an indicator of coordination of beliefs.
Finally, GPTs are characterized by the sequential development of complementary
technologies. This sequential development may be seen in the patent record: Cita-
tion lags (the time between the patent grant and the arrival of its forward citations)
tend to be longer as complementary technologies require more time to develop,
citation counts themselves are higher, as GPTs are characterized by a burst of
invention (Hall & Trajtenberg, 2004). In a GPT, we are also likely to see in the
patent record the technologies being adopted in a wide range of industrial sectors, as
market actors throughout the economy begin to deploy the technology generally, to
many different industrial and technology settings. Therefore, patent records may
again be used to offer a window into the development ex post of technologies that
build upon, and are complementary to, the focal technology.
6 Conclusion
This study has considered the notion that nanotechnology may be a breakthrough
innovation with long-run economic and societal affectswhether it is a General
Purpose Technology (GPT), and has explored appropriate methodologies to ad-
dress this question. GPTs are a phenomenon of much interest because these tech-
nologies have a profound impact on growth and the productivity levels in the
economy and of individual rms, despite initial periods of slowdown as resources are
devoted to the development of complementary technologies. Hence, GPTs peak the
attention of public and private sector decision makers.
Our literature review demonstrates that there is a growing body of work that
considers nanotechnology a GPT. As we show, however, few if any of these studies
claiming a nanotechnology-GPT link have developed a systematic approach for
determining if this designation is tting.
The core premise inherent in the characterization of a GPT is that it must meet
three criteria: pervasiveness, innovation spawning, and scope for improvement. Our
study has highlighted qualitative evidence suggesting that nanotechnology may
exhibit characteristics of a GPT. More importantly, we put forth the application of
quantitative methods to uncover the nature of nanotechnology. Through the use of
patent data, citations analysis, and a Generality index derived from patent clas-
sications, we demonstrate that nanotechnology exhibits similar pervasiveness
levels to that of ICT, an existing GPT. We further propose that patent data may be
effectively used to examine innovation spawning attributes such as: (1) identifying
Another problem for any such analysis would arise if the placement of citations by examiners was
driven by some internal USPTO job incentive.
Assessing the nature of nanotechnology
a critical number (burst) of technologies in intermediate sectors, and (2) the
types of convergence of believes needed for diffusion to occur. The question is
still open regarding how to use quantitative data to solicit information about the
scope for future improvements in nanotechnology.
We believe that an effort to evaluate the currently available data through the lens
of existing theory and empirical methodology is important for understanding the
possible socio-economic implications of nanotechnology during the early phases of
dissemination. We conjecture that if nanotechnology is more of an incremental
technology, there will be little change in short or long run investment decisions. But
if it is a major breakthrough innovation, as a good part of the evidence we explored
suggests, disruptions of business routines are to be expected, existing skills may
become obsolete more rapidly, and capital shifts from established rms to younger
rms of investors that are more likely to embrace the new technology may occur.
Moreover we observed the governments optimal strategy to spur innovation is
drastically different when an emerging technology has the character of a GPT than
when it is an incremental technology. The level of appropriability of technologies
complementary to the core innovation should be lower in the former case than in the
latter one. The fast dissemination of complementary technologies associated with an
emerging technology is a necessary condition to reach a critical level of acceptability
which induces even relatively moderate risk-takers to invest in the emerging
Acknowledgements Authors contributed equally to this work. This research was undertaken at
Georgia Tech through the Center for Nanotechnology in Society (CNS-ASU), supported by the
National Science Foundation under NSF NSEC CNS SES #0531194, in collaboration with research
sponsored under NSF NSEC CNS SES #0531146. We gratefully acknowledge helpful comments we
received from Philip Shapira and participants at the Technology Transfer Society Annual Confer-
ence, September 27, 2006, Atlanta, Georgia, USA.
Aghion, P., & Howitt, P. (1992). A model of growth through creative destruction. Econometrica,
60(2), 323351.
Alcacer, J., & Gittleman, M. (2006). Patent citations as a measure of knowledge ows: The inuence
of examiner citations. Review of Economics and Statistics 88(4), 774779.
Arnold, W. (1995). The SIA lithography roadmap. Microlithography World, Winter, 711.
Bresnahan, T. J., & Trajtenberg, M. (1995). General purpose technologies: Engines of Growth?
NBER Working Paper No. 4148, Cambridge, MA: National Bureau of Economic Research.
Crichton, M. (2002). Prey. New York: HarperCollins.
David, P. (1990). The dynamo and the computer: An historical perspective on the modern pro-
ductivity paradox. The American Economic Review, 80(2), 355361.
Drexler, E. (1986). Engines of creation: The coming era of nanotechnology. New York: Anchor
Graham, S. J. H. (2006). The determinants of patentees use of continuation applications in the
United States patent and trademark ofce, 198099. In Andersen (Ed.), Intellectual property
rights: Innovation, governance and the institutional environment. Edward Elgar Publishers.
Graham, S. J. H., & Mowery, D. C. (2003). Intellectual property protection in the software industry,
with David C. Mowery. In Wesley Cohen & Steven Merrill (Eds.), Patents in the knowledge-
based economy: Proceedings of the science, technology and economic policy board. National
Academies Press.
J. Youtie et al.
Hall, B. H., Jaffe, A. B., & Trajtenberg, M. (2001). The NBER patent citation data le: Lessons,
insights and methodological tools. NBER Working Paper No. 8498. Cambridge, MA: National
Bureau of Economic Research.
Hall, B., & Trajtenberg, M. (2004). Uncovering GPTS with patent data. NBER Working Paper No.
10901. Cambridge, MA: National Bureau of Economic Research.
Harriott, L. (2000). Limits of lithography. Working Paper, Bell Labs-Lucent Technologies.
Helpman, E., & Trajtenberg, M. (1994). A time to sow and a time to reap: Growth based on general
purpose technologies. NBER Working Paper No. 4854. Cambridge, MA: National Bureau of
Economic Research.
Huang, Z., Hsinchun, C., Yip, A., Ng, G., Guo, F., Chen, Z., & Roco, M. (2003). Longitudinal patent
analysis for nanoscale science and engineering: Country, institution, and technology eld.
Journal of Nanoparticle Research, 5, 333363.
Irwin, D. A., & Klenow, P. J. (1994). High tech R&D subsidies: Estimating the effects of sematech.
NBER Working Paper No. 4974. Cambridge, MA: National Bureau of Economic Research.
Jovanovic, J., & Rousseau, P. (2005). General purpose technologies. Working Paper 11093.
Cambridge, MA: National Bureau of Economic Research.
Kanellos, M. (2005). AMDs 65 nano question solved: Its a second half thing. CNET News.com,
November 2, http://news.com.com/2061-10791_3-5929097.html?tag=nl.
Kanellos, M. (2006). Intel shows test chips made on future processes. CNET News.com,
January 25, http://news.com.com/Intel+shows+test+chips+made+on+future+processes/2100-
Khushf, G. (2004). A hierarchical architecture for nano-scale science and technology: Taking stock
of the claims about science made by advocates of NBIC convergence. In Baird, D., Nordmann,
A., & Schummer, J. (Eds.), Discovering the nanoscale (pp. 2133). Amsterdam: IOS Press.
Kostoff, R., Stump, J., Johnson, D., Murday, J., Lau, C., & Tolles, W. (2006). The structure and
infrastructure of global nanotechnology literature. Journal of Nanoparticle Research, 8(34),
Lipsey, R. G., Bekar, C., & Carlaw, K. (1998). What requires explanation? In Helpman, E. (Ed.),
General purpose technologies and economic growth (pp. 1454). Cambridge: MIT Press.
Lux Research (2006). The nanotech report: 4th edition, volume 1. New York: Lux Research.
Meyer, M. (2006). What do we know about innovation in nanotechnology? Some propositions about
an emerging eld between hype and path-dependency. Paper presented at the 2006 Technology
Transfer Society Conference, September 2729, Atlanta, Georgia.
Moser, P., & Nicholas, T. (2004). Was electricity a general purpose technology? The American
Economic Review, Papers and Proceedings, 94(2), 388394.
Palmberg, C., & Nikulainen, T. (2006). Industrial renewal and growth through nanotechnology?
Helsinki: ETLA, The Research Institute of the Finnish Economy.
Porter, A., Shapira, P., & Youtie, J. (2006). Rening search terms for nanotechnology. Brieng
paper, prepared for presentation at national science foundation, rst year review for the center
for nanotechnology in society (CNS-ASU), Arizona State University, Tempe, AZ.
Roco, M. (2004). Nanoscale science and engineering: Unifying and transforming tools. AIChE
Journal, 50(5), 890897.
Roco, M. (2005). International perspective on government nanotechnology funding in 2005. Journal
of Nanoparticle Research, 7(6), 707712.
Roco, M., & Bainbridge, W. (Eds.) (2003). Converging technologies for improving human perfor-
mance: Nanotechnology, biotechnology, information technology and cognitive science. Boston:
Kluwer Academic Publishers.
Romer, P. (1990). Endogenous technical change. Journal of Political Economy, 98, 71103.
Shea, C. (2005). Future management research directions in nanotechnology: A case study. Journal of
Engineering and Technology Management, 22, 185200.
Tannenbaum, R. (2005). Advances in materials and self-assembly. In Advanced Technology and the
Future of Manufacturing, Proceedings. Atlanta, Georgia: Georgia Institute of Technology.
US National Science Foundation (2005). New grants are awarded to inform the public and
explore the implications of nanotechnology, Press Release 05-179, http://www.nsf.gov/news/
Assessing the nature of nanotechnology
c h a p t e r 3
walter w. powell
sti ne grodal
3. 1 Introduction
In February of 2001, two rival consortia published rough draft (roughly 90 per cent
complete) sequences of the human genome in Nature and Science. The public
Human Genome Project consisted of Wve key institutions and eleven collaborators,1
supported by the US National Institutes of Health, Department of Energy, and the
Wellcome Trust in the United Kingdom. The rival private consortia, led by the
biotechWrmCelera, included both commercial Wrms andacademic researchers from
the Universityof California, PennState, Case Western, Johns Hopkins, Cal Tech, Yale,
Rockefeller, as well as scientists in Spain, Israel, and Australia. These projects have
been acclaimed for their remarkable scientiWc achievement; they were also the
product of considerable organizational innovation. In contrast to the Manhattan
Project or Project Apollo, both of which were hierarchically organized, national
projects, the Human Genome Project (HGP) and the Celera team were pluralist,
multiorganizational, multinational confederations. These twogroups were intensely
rivalrous, but collaborated intensively within their own groups (Lambright 2002).
HGP involved management by two government agencies and a private British
We are grateful to David Mowery and Jan Fagerberg for their careful readings of earlier drafts.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 56
foundation that coordinated activities in government labs, universities, and
nonproWt institutes in the US and England. As the lead Wrm, Celeras organization
was more focused, but its research team included scientists and state-of-the-art
equipment at private Wrms, public and private universities, and nonproWt institutes
in four countries.
Both projects were organized as large-scale networks, and their rivalry spurred
each side to engage in a high-stakes learning race. While the cost, scale, and
distributed nature of these projects may have been unusual, the form of organiza-
tioncollaboration across multiple organizational boundaries and institutional
formsis no longer rare. Indeed, many analysts have noted that the model of
networks of innovators has become commonplace over the past twodecades (Powell
1990; Rosenbloom and Spencer 1996; Roberts and Liu 2001; Chesbrough 2003).
Collaboration among ostensible rivals was once regarded as a provisional or
transitional step taken to enter new markets, spread risks, or to share early stage
R&D costs (Mowery 1988). Such forays were often followed by mergers as the
transitory activities became incorporated inside the boundaries of the Wrm. Recent
studies suggest, however, that various forms of interorganizational partnerships are
now core components of corporate strategy. Even where these linkages endure for
relatively lengthy periods of time, they do not entail vertical integration (Gomes-
Casseres 1996; Hagedoorn 1996; Noteboom 1999; Ahuja 2000a). Contemporary
studies of industrial performance are replete with reports of a signiWcant upsurge
invarious types of interorganizational collaboration. While these collaborations can
take a number of forms (including research consortia, joint ventures, strategic
alliances, and subcontracting) and span a wide range of key functions, a National
Research Council analysis of trends in industrial research and development (R&D)
suggests that the innovation process has undergone the most signiWcant transform-
ation over the past decade (Merrill and Cooper 1999). In a survey of the period
196098, Hagedoorn (2002) Wnds a sharp growth in R&D collaborations, beginning
in the late 1970s and continuing through the mid-1990s.
A National Research Council assessment of eleven US-based industries, pur-
posefully diverse in character and technology but all resurgent in the 1990s, observes
in every sector an increased reliance on external sources of R&D, notably univer-
sities, consortia and government labs, and greater collaboration with domestic and
foreign competitors, as well as customers in the development of new products
and processes (Mowery 1999: 7). Other surveys also point to the enhanced centrality
of interorganizational collaboration, especially in R&D. For example, National
Science Foundation data show a marked increase in the number of international
alliances between US and Western European countries between 1980 and 1994;
but by the mid-1990s, the formation rates for intranational alliances linking US
Wrms with their domestic competitors outpace international linkages (National
Science Board 1998). The former collaborations were motivated largely by concerns
with market access, while the latter focus more on the development of new
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 57
networks of innovators 57
Similarly, there is now ample research illustrating the growing links between US
Wrms and universities (Powell and Owen-Smith 1998), and greater involvement by
Wrms and government labs in research joint ventures (Link 1996, 1999). In the realm
of science, Hicks and Katz (1996) Wnd that research papers are much more likely to
be co-authored and involve authors with multiple institutional aYliations that
span universities, government, and industry. Distributed networks of practice
are the organizing basis for many technical communities, suggesting both that
sources of knowledge are now more widely dispersed and that governance mechan-
isms are emerging to orchestrate distributed knowledge. The open source software
movement is but one highly visible example of this trend (OMahony 2002; Weber
2003), which illustrates how advances in information technology have greatly
facilitated virtual networks. In short, as Mowery (1999: 9) observes, the diversity
of institutional actors and relationships in the industrial innovation process has
increased considerably. Complex networks of Wrms, universities, and government
labs are critical features of many industries, especially so in Welds with rapid
technological progress, such as computers, semiconductors, pharmaceuticals, and
Our goal in this chapter is to assess the state of scholarly research on the role of
networks in the innovation process. We begin with a review of the factors that have
triggered the increased salience of networks. We discuss diVerent types of networks,
distinguishing between networks that are based more on contractual or market
considerations, and those that are based on less formal, and more primordial
relationships, such as common membership in a technological community or a
regional economy. We then turn to a discussion of the analytical leverage provided
by the tools of network analysis. This stream of research, which spans sociology,
social psychology, organizational behavior, and business strategy, highlights key
distinctions between highly clustered, dense networks, steeped in overlapping ties
and high in trust, and weak-tie networks, that provide access to novel, non-redun-
dant information. We next reviewa number of empirical studies of the contribution
of networks to the innovative output of Wrms. We take up the issue of knowledge
transfer, examining how the codiWcation of knowledge can shape what is transmit-
ted through networks. We brieXy discuss the governance of networks, and then
conclude with an assessment of what types of organizations and settings derive the
greatest impact on innovation from participation in networks.
Research on the relationship between networks and innovation is a relatively
recent area of inquiry. While there is a good deal of work underway, direct analyses
measuring the impact of interWrm networks on performance are limited. Much of
the extant research focuses on the eVects of networks on patenting, access to infor-
mation, and the generation of novel ideas. Moreover, the studies often examine
high-tech industries, where investment in R&D is pronounced. Attention to
the consequences of network ties for the Wnancial performance of Wrms is relatively
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 58
58 walter w. powell and stine grodal
3. 2 Why have Networks Grown
in Importance?
The advantages of a heterogeneous group of contacts are well established in both
social theory and network analysis. A strong tradition of theory and research,
running from Simmel (1954) to Merton (1957) to Granovetter (1973) to Burt
(1992), makes abundantly clear that there are informational, status, and resource
advantages to having broad and diverse social circles. Below we review an array of
recent empirical studies that demonstrate how interorganizational relationships
lead to various beneWts with respect to information diVusion, resource sharing,
access to specialized assets, and interorganizational learning. In science and tech-
nology-based Welds, the advantages that accrue fromdiverse sources of information
and resources are considerable. Not surprisingly, then, as the commercialization of
knowledge has assumed greater importance in economic growth, collaboration
across organizational boundaries has become more commonplace. Interorganiza-
tional networks are a means by which organizations can pool or exchange resources,
and jointly develop new ideas and skills. In Welds where scientiWc or technological
progress is developing rapidly, and the sources of knowledge are widely distributed,
no single Wrm has all the necessary skills to stay on top of all areas of progress and
bring signiWcant innovations to market (Powell and Brantley 1992; Powell, Koput,
and Smith-Doerr 1996; Hagedoorn and Duysters 2002). In such settings, networks
can become the locus of innovation, as the creation of knowledge is crucial to
improving competitive position.
Collaborative networks have long been central to the production process in craft-
based industries (Eccles 1981), in industrial districts (Brusco 1982; Piore and Sabel
1984), and in Welds such as aerospace where assembly depended upon key inputs
from diverse participants. The growth of knowledge-intensive industries has
heightened the importance of networks in R&D as well as product development
and distribution. A persistent Wnding from a diverse set of empirical studies is that
internal R&D intensity and technological sophistication are positively correlated
with both the number and intensity of strategic alliances (Freeman 1991; Hagedoorn
For organizations in rapidly developing Welds, heterogeneity in the portfolio of
collaborators allows Wrms to learn from a wide stock of knowledge. Organizations
with broader networks are exposed to more experiences, diVerent competencies,
and added opportunities (Beckman and Haunschild 2002). Such access creates an
environment in which creative abrasion, the synthesis that is developed from
multiple points of view, is more likely to occur. In this view, innovation occurs at
the boundaries between mind sets, not within the provincial territory of one
knowledge and skill base (Leonard-Barton 1995: 62). By having access to a more
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 59
networks of innovators 59
varied set of activities, experiences, and collaborators, companies broaden the
resource and knowledge base that they can draw on. By developing more multiplex
ties with individual partners, either through pursuing multiple collaborations or
expanding an existing R&Dpartnership into downstreamdevelopment, companies
increase the points of contact between them. When relationships are deepened,
greater commitment and more thorough knowledge sharing ensue. Organizations
with multiple and/or multifaceted ties to others are likely to have developed better
protocols for the exchange of information and the resolution of disputes (Powell
1998). Parties that develop a broader bandwidth for communication are, in turn,
more capable of transferring complex knowledge. In science-driven Welds such as
biotechnology, organizations that develop ties to diVerent kinds of organizations
and carry out multiple types of activities with these organizations are central players
in industry networks (Powell et al. 2004). These centrally positioned organizations
are both capable of pulling promising new entrants into the network and collabor-
ating with a wide assortment of incumbents. Moreover, research shows that in
biotechnology, organizations lacking such connections fail to keep pace and fall by
the wayside (Powell et al. 2005).
3. 3 Varieties of Networks
The literature onnetworks emphasizes that they are most pronouncedinthe domain
between the Xexibility and autonomy of markets and the force and control of
organizational authority (Powell 1990). Networks thus combine some of the incen-
tive structures of markets with the monitoring capabilities and administrative
oversight associated with hierarchies (Mowery, Oxley, and Silverman 1996). For
our purposes, we include networks based on formal contractual relations, such as
subcontracting relationships, strategic alliances or participation in anindustry-wide
research consortium, and informal ties, based on common membership in a pro-
fessional or trade association, or even a looser aYliation with a technological
One can diVerentiate networks with respect to their duration and stability, as well
as whether they are forged to accomplish a speciWc task or evolve out of pre-existing
bonds of association. Networks vary from short-term projects to long-term rela-
tionships, and the diVerent temporal dimensions have important implications for
governance. Some networks are hierarchical, monitoredby a central authority; while
others are more heterarchical, with distributed authority and strong self-organizing
features. Grabher and Powell (2004) focus on temporal stability and forms of
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 60
60 walter w. powell and stine grodal
governance to diVerentiate four key types: informal networks (based on shared
experience); project networks (short-term combinations to accomplish speciWc
tasks); regional networks (where spatial propinquity helps sustain a common
community); and business networks (purposive, strategic alliances between two
parties). These types do not represent essentialist categories; rather they may overlap
and interweave with one another. Consider these forms as useful coordinates to
locate networks with respect to diVerent combinatory elements.
Several key concepts provide potent analytical tools that apply across diVerent
types of networks and permit assessment of their eVects. First, consider the diVer-
ences between strong and weak ties (Granovetter 1973). In interpersonal terms, a
strong tie is a personwithwhomyou interact on a regular basis, while a weak tie is an
acquaintance, or a friend of a friend. Strong ties are important for social support, but
much of the novel information that a person receives comes fromweak ties. Strong
ties are based on common interests, consequently most information that is passed
reinforces existing views. Weak ties introduce novelty in the form of diVerent ideas
or tastes, and by introducing new information they are, for example, invaluable in
job searches and other circumstances where a small amount of new information is
highly useful. Weak ties have a longer reach, but a much narrower bandwidth than
strong ties. The latter are more cohesive, and often prove to be more eVective at the
exchange of complex information. Figure 3.1 illustrates the diVerence between
strong and weak ties.
Much of the research on interWrm networks extrapolates from interpersonal
relations.2 In general, this is a plausible analytical move; however, it elides the
question of whether relationships at the Wrm level are dependent on ongoing
interpersonal ties, and whether the business relationship would be harmed or
severed if the key participants were to depart. The extent to which interorganiza-
tional ties are contingent upon relations among individuals is a key question for
scholarly research, as well as a critical challenge for business strategy (Gulati 1995;
Powell 1998).
A second notable contrast is the distinction between networks as bridges and as
structural holes (Burt 1992). Bridges are points of connection between parties that
Strong tie
Weak tie
Fig. 3.1 Strong and weak ties
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 61
networks of innovators 61
lack ties, such as when Aknows B, and Cknows B but not A. B is the bridge between
A and C, thus the gateway to a linkage between A and C. Granovetter (1973) argued
that bridges are the links that make weak ties possible. Burt (1992) deepened the
argument by moving fromthe who question(i.e., which position ina network is best
situated) to the question of how certain structural arrangements generate beneWts
and opportunities. He coined the term structural holes as a potential connection
between clusters of units that are not connected. The possibility of making such a
connection provides leverage, or opportunities for arbitrage. Those positioned to
take advantage of structural holes can broker gaps in the social structure. See Figure
3.2 for illustration.
There is debate as to whether strong or weak ties, or bridges or structural holes,
oVer greater opportunities for innovation (Ahuja 2000a; Ruef 2002). Clearly, vari-
ation in network structures is associated with diVerent content in relationships.
Strong ties between two parties may restrict information gathering in terms of the
breadth of search, but the information that is exchanged is thick, or detailed and
rich. Weak ties are thinner and less durable, but provide better access to non-
redundant information. There is also disagreement as to whether networks can be
designed or pruned to produce optimal shapes, without triggering repercus-
sions. Whether location in a network is highly malleable or not, position in a
network both empowers and constrains opportunities.
A third point of contrast is between networks formed intentionally across a
market interface to accomplish a task and emergent networks that grow out of
ongoing relationships. The former may be considered an instrumental or strategic
relation, while the latter stems from more primordial relations, such as common
ethnicity, friendship, or location. These diVerent starting points matter, but in the
Xuid world of networks, the point of origin does not Wx the evolution of a relation-
ship. Consider two cases. There is a global trend toward vertical disaggregation in
manufacturing, as Wrms are relying on suppliers for design and component inputs
Redundant tie
Structural hole
Fig. 3.2 Structural holes and redundant ties
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 62
62 walter w. powell and stine grodal
in a variety of industries (see Womack, Jones, and Roos 1990, for autos; McKendrick,
Doner, and Haggard 2000, for disk drives). Often these outsourcing decisions are
driven by the need to reduce costs, save time, and enhance Xexibility, while the large
Wrm concentrates on those activities in which it has some form of competitive
advantage. But as many analysts have noted, there is no natural stopping point in
this relation (Sabel 1994; Helper, McDuYe, and Sabel 2000). The subcontractor can
end up involved in design issues, doing critical R&D, or become central to eVorts to
improve quality. What began as a choice to outsource can, in some circumstances,
become either a deep, mutually dependent collaboration or a highly ambiguous and
opportunistic partnership. Helper et al. (2000) and Dyer and Nobeoka (2000)
illustrate the marked trend for automobile subcontracting to evolve into interde-
pendent, bilateral relationships.
Or consider the contemporary life sciences, where many R&D partnerships
emerge out of ongoing intellectual relationshipsco-authorships, mentormentee
relationships, and common training (Murray 2002). These informal personal rela-
tionships may, however, come to involve signiWcant intellectual property in the form
of patents, and thus become highly formalized contractual agreements between
organizations. We oVer these examples as illustrations that networks forged out of
strategic purposes can take on strong relational elements, while more personal ties
can become contractual and highly speciWed. While it is possible to assign networks
to either a transcational (i.e., based on a consideration of business opportunities
without regard to prior social relations) or relational (i.e., embedded in ongoing
social relationships) category, it is inappropriate to assume that relations remain
Wxed. As networks evolve, there is considerable give and take. Intense competition
can render calculative strategic alliances more embedded, while the prospect of great
Wnancial reward can turn a handshake relationship between individuals into a
formal legal linkage between Wrms.
Figure 3.3 provides a typology of diVerent forms of networks, with the horizontal
axis representing degree of purposiveness, ranging from informal to contractual.
The vertical axis represents the extent of embeddedness, varying from open, epi-
sodic, or Xuid to recurrent, dense connections among a fairly closed group (Gran-
ovetter, 1985). We illustrate each of the four cells with examples of types of
innovation networks. In the lower left cell we place informal networks, such as a
scientiWc invisible college, that emerge out of sharedexperience or common interest.
Although these relations tend to be temporary and short-lived, the gray arrows are
intended to show that these informal linkages can evolve into formal business
alliances or more enduring primordial relations, where participation is more con-
stant andless Xuid. The primordial network inthe upper left cell is characterized by a
common social identity, continuous participation, and close ties. All these features
are often found in professional networks, craft-based occupations, ethnic commu-
nities, and industrial districts. The upper right cell is typiWed by involvement in a
common project. Membership in such a network is typically restricted and often
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 63
networks of innovators 63
governed by a lead Wrm. A supply-chain network or a large construction project
are apt examples. The gray arrows are meant to illustrate that supply-chain relations
can evolve into either occupational communities and industrial districts or into
formal business partnerships. The most purposive andinstrumental type of network
is the strategic alliance, represented by the lower right cell. We turn now to a
discussion of the relationship between these forms and the innovation process.



Supply Chain
Invisible College
Biotech Venture
Supply chain
Nodes similar
Common social identity
Multifunctional, craft based,
Dense, since the network exists
prior to activity, work relationships
are nested in ongoing personal
Film, construction, ethnic
community, diamond trade
Nodes dissimilar
Common work identity
Horizontal or vertical specialization
Division of labor
Incremental innovation
Spider web
The establishment of ties creates the
network. The network can become a
form of social identification over
time, as for example in the Toyota
supply chain network.
Nodes both similar and dissimilar
Common interest
Fast access to news and novel ideas
Ties reinforce the structure of the
research collaboration, information
Nodes both similar and dissimilar
Identity different
Division of labor
Purposive network, emerges through
active tie creation
Common in high technology
industries like semiconductors,
biotechnology, and chemicals
Fig. 3.3 Network typology
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 64
64 walter w. powell and stine grodal
3. 4 Empirical Studies of the Role
of Networks in Innovation
3.4.1 Formal Ties
Most empirical studies of the relationship between networks and innovation focus
on formal ties established among organizations. This streamof research documents
a strong positive relationship between alliance formation and innovation, across
such diverse industries as chemicals (Ahuja 2000a), biotechnology (Powell et al.
1996, 1999; Walker, Kogut, and Shan 1997; Baum, Calabrese, and Silverman 2000),
telecommunications (Godoe 2000), and semiconductors (Stuart 1998, 2000). The
diversity of the research contexts suggests the eVects of network structure may be
generalizable. Nevertheless, most research has focused on high technology indus-
tries, and uses patents as a proxy for innovation. More direct measures of innovative
outputs are needed. Some of the important themes that emerge from this research
highlight speciWc tie characteristics, technological uncertainty, and network evolu-
tion. In addition, researchers have emphasized the increased beneWts in the form of
resources and knowledge that alliances provide to entrepreneurial Wrms.
Tie characteristics. One line of research focuses on how diVerent types of ties
inXuence the beneWts derived from alliances. Vinding (2002) identiWed 548 Danish
manufacturing Wrms that developed one or more new products over a two-year
period. In interviews with a subset of the companies, he Wnds that the impact of a
collaboration on innovation is signiWcantly related to both the type of partner and
the pattern of previous collaborative relationships. The importance of prior inter-
action with partners points to the signiWcance of relationship building, and how
elements such as trust and cognitive understandings require time to develop.
Domestic partners were found to have a greater positive impact on innovative
performance than foreign partners, possibly due to the higher costs, both psycho-
logical andWnancial, associatedwithmore distant collaborations. Vindings research
emphasizes beneWts derived fromstrong local ties. Similarly, in a ten-year (198090)
case study of the R&D portfolio of a Norwegian telecommunications organization,
Godoe (2000) reports comparable results with respect to strong ties. His analysis
suggests that radical innovations were more likely to emerge from intimate and
prolonged interaction. But in the Norwegian case, the aYliations were not local, but
instead based on membership in international telecommunications associations.
Powell et al. (1999) emphasize that experience with collaboration and centrality in
the network derived from a diverse set of ties are important determinants of
innovation among biotechnology Wrms over the period 198899. Their analyses
suggest that centrality and experience resulted in more patenting. The most conse-
quential connections in terms of patenting were R&Dpartnerships. The diversity of
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 65
networks of innovators 65
network ties also had a positive inXuence on rates of patenting. Powell et al. (1999)
found that while network experience had a positive inXuence on patenting, the rate
of increase diminished with additional experience, suggesting possible declining
returns to network connectivity. The question of whether there are limits to con-
nectivity needs to be investigated further.3 These results suggested a cycles of
learning process in which R&D collaborations generate attention that attracts
other partners, who collaborate in developing novel ideas. This enhanced diversity
of aYliations increases a Wrms experience at managing collaborations and transfer-
ring knowledge, and increases their centrality in the industry network. Greater
centrality is associated with a higher rate of patenting, and both centrality and
higher volumes of patenting trigger subsequent R&D partnerships, restarting the
cycle for centrally placed Wrms.
Most research has looked at the presence or absence of a formal collaboration.
Ahuja (2000a), however, developed a more nuanced analysis including both direct
and indirect ties, and the level of indirectness. Drawing data from a sample of 97
Wrms in the global chemicals industry, he used the number of patents as a measure of
innovative output, while collaboration was measured through formal ties. More
distant connections through aYliates of partners were coded as weak or indirect ties.
The results show that both direct and indirect ties have a positive inXuence on
innovation, though the impact of indirect ties is smaller than the impact of direct
ties. The number of direct ties also negatively moderates the impact of indirect ties.
In contrast to Burts (1992) arguments about the arbitrage opportunities available
through non-redundant contacts, Ahuja shows that a network with many structural
holes can reduce innovative output, as measured by rates of patenting. A key
advantage of close-knit networks may be due to their superior ability to transfer
tacit knowledge (Van Wijk, Van den Bosch, and Volberda 2003). In an analysis of the
exchange of information across project teams in a large multinational computer
company, Hansen (1999) also illustrates that complex knowledge is transferred most
easily through tightly knit networks.
Entrepreneurial Wrms. One active area of research concerns the eVects of networks
on survival chances of newly founded Wrms. Larsons (1992) ethnographic study of
how a startup Wrm grew and prospered by drawing on external resources and
support for key business functions illustrates how relationships are forged and
sustained as startup Wrms grow. While not explicitly looking at innovative output,
Larson added insight into the signal importance of networks in obtaining resources
necessary to fuel a startup Wrms success. Shan, Walker, and Kogut (1994) examined
whether biotechnology startupWrms cooperative relationships withother Wrms had
a positive eVect on patenting. Their results oVer support for the argument that
collaborative relationships increased innovation, because formal cooperative rela-
tionships explained innovative output, while innovative output did not account for
the pattern of alliances. The salience of alliances for young and small Wrms is further
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 66
66 walter w. powell and stine grodal
emphasized in Stuarts (2000) study of innovation in the semiconductor industry.
His dataset includes 150 Wrms, followed by the consultancy Wrm Dataquest over the
period 198591. Drawing on sales Wgures, patterns of strategic alliances, and
patenting activity, Stuart shows that Wrms possessing technologically sophisticated
alliance partners patented at a substantially greater rate than those that lacked such
ties. Firms establishing strategic alliances with large partners also grew at a higher
rate than Wrms without access to such partners. The returns to networks with regard
to patenting were greatest for both young and small Wrms.
Baum, Calabrese, and Silverman(2000) pursue a similar question, asking howthe
composition of a startup Wrms alliance portfolio aVects its performance. Using data
on 142 biotechnology Wrms founded in Canada between 1991 and 1996, they Wnd a
positive eVect of alliance formation on startup innovation. Network eYciency,
deWned as the diversity of information and capabilities per alliance, showed a large
positive eVect on the number of biotech patents. Alliances with direct competitors
had a negative eVect on innovation, however. These results were moderated when
the rival biotechnology Wrm had a larger share of the relevant market or if the rival
biotechnology Wrm was highly innovative. Of the various performance measures
used, the number of patents and the volume of R&D expenditures were most
signiWcantly inXuenced by rates of alliance formation.
Network dynamics. Drawing on Cohen and Levinthals (1990) ideas about absorp-
tive capacity, Powell et al. (1996) argued that Wrms utilize external collaborations to
stay abreast in rapidly developing technological Welds. But organizations cannot be
passive recipients of new knowledge. What can be learned is crucially aVected by
what is already known (Powell et al. 1996: 120). To understand the news generated
externally, organizations have to make news internally. In this fashion, the rate of
acquisition of skills and resources fromthe outside is closely linked to the generation
of expertise internally. In their work on the global biotechnology industry, they Wnd
that Wrms that develop experience at managing collaborative R&D relationships
garner faster access to centrally positioned organizations. As experience at collabor-
ating grows, Wrms widen the network of organizations withwhomthey partner. As a
Wrms experience with collaboration and its diversity of partners increase, the more
central and visible the Wrmbecomes in the industry. This centrality leads, in turn, to
growthinthe size of the Wrm, andtothe ability tocoordinate more alliances, creating
a feedback cycle. This cycle of learning has been shown to be associated with positive
Wnancial performance (Powell et al. 1999), and a greater ability to collaborate with
diverse kinds of organizations, which permits Wrms to retain a leadership position in
the industry (Powell et al. 2005).
The general picture that emerges from research in organizational sociology and
business strategy is one in which networks and innovation constitute a virtuous
cycle. External linkages facilitate innovation, and at the same time innovative
outputs attract further collaborative ties. Both factors stimulate organizational
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 67
networks of innovators 67
growth, and appear to enhance further innovation. Ahuja (2000b) and Stuart
(2000), for example, demonstrate that Wrms with many prior patents are more likely
to form alliances than Wrms lacking patents, suggesting a recursive process of
innovation and growth in which collaborative ties play a central role. Further
attention needs to be given, however, to such issues as the eVects of the duration
of linkages, experience with collaboration, and the consequences of broken ties on
rates of innovation.
Technological uncertainty. Two additional aspects of the innovation process involve
the relationship between strategy and alliance formation and the level of techno-
logical uncertainty in the Weld. Eisenhardt and Schoonhoven (1996) studied the
population of semiconductor Wrms launched in the US between 1978 and 1985, and
found that the more risk-taking a companys strategy, the more alliances a company
formed. One explanation is that as Wrms gain credibility for developing pioneering
technologies, access to Wnancial and other resources for developing innovative
technology is secured through alliances. An alternative view is that alliances are
necessary to share the attendant risks inhigh-velocity environments. Sarkar, Echam-
badi, and Harrisons (2001) analysis of managers in a range of high-tech industries
revealed that an active strategy of alliance formation enhanced performance, as
measured by market share, sales growth, market development, and product innov-
ation. They also report that managers who perceived the environment as more
uncertain were more likely to pursue alliances. In addition, smaller Wrms derived
more value from network linkages than larger Wrms, presumably because smaller
companies viewed the technological landscape as more uncertain.
Rosenkopf and Tushman (1998) examined the role of technical communities in
the Xight simulation industry, where cooperative technical organizations play a
critical role in developing standards and advancing the state of the art. In a study
covering the years 195892, they found long periods marked by incremental change
punctuated by shorter eras of ferment. They show that the rate of founding of
technical networks increases during periods of discontinuity, and stabilizes into
core cliques when ferment declines and a dominant design emerges. Subsequent
technological change disrupts dominant cliques, and triggers the formation of new
networks, restarting the cycle. Thus, both technological strategy and industry
evolution are linked to patterns of network formation, with external networks
assuming greater importance during periods of technical discontinuity and for
Wrms with more risk-taking strategies. The importance of industry technical com-
mittees in standards setting has also been emphasized in the computer industry
(Farrell and Salomer 1988) and videocassette recorders (Cusumano et al. 1992).
The overall conclusion of this group of studies is that networks provide access to
more diverse sources of information and capabilities than are available to Wrms
lacking such ties, and, in turn, these linkages increase the level of innovation inside
Wrms. Younger and smaller Wrms may beneWt more fromcollaborative relationships
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 68
68 walter w. powell and stine grodal
than do larger Wrms. Most notably, Wrms with a central location within networks
generate more innovative output. Both direct and indirect ties provide a positive
contribution to innovation, but the eVect of indirect ties is moderated by the
prevalence of direct ties. The evidence for the beneWts of structural holes is not
uniform; where structural holes might be beneWcial is in the search for new infor-
mation, but the knowledge transfer process appears to be facilitated by closer-knit
networks. From the view of the dynamics of collaboration, successful external
relations appear to beget more ties, which fuel Wrmgrowth and innovation. Clearly,
there are limits to this cycle, but research has not addressed this question in depth
thus far.
The majority of the studies reviewed in this section have been carried out using
patents as the dependent variable and formal relationships as the independent
variable. Patents provide a measure of novelty that is externally validated through
the patent examination process, hence they are a useful indicator of knowledge
creation (Griliches 1990). But patents have some limitations. Some kinds of innov-
ations are not patented, and there is variation in the extent of patenting across
industries. (See Chapter 14 by Malerba on inter-industry variation in innovation
processes.) On the other hand, the focus of many of these studiessemiconductors,
chemicals, biotechnologyis in Welds where patenting is commonplace, and com-
petitors in these sectors are active patentors. The attention to these high-technology
industries raises questions, however, as to the generalizability of the results to other
less knowledge-intensive industries.
One study that speaks to diVerences across industries is Rowley, Behrens, and
Krackhardts (2000) analysis of strong and weak ties in the steel and semiconductor
industries. This study made a notable eVort to distinguish between strong aYlia-
tions, where alliances entailed signiWcant resource commitments and regular inter-
actions, and more arms-length transactions, where there was a rapid exchange,
and the relationship was characterized by less frequency and depth. For example,
equity alliances, joint ventures, and R&D partnerships were categorized as strong
ties, while licensing, patent agreements, and marketing relations were classiWed as
weak ties. Recognizing that weak ties serve as bridges to novel information, while
strong ties are useful for both social control and the exchange of tacit knowledge,
they Wnd divergent results. In the steel industry strong ties are positively associated
with performance; while in semiconductors weak ties are more eYcacious. They
suggest these Wndings reXect the importance of search and product innovation in
semiconductors, and a focus on improvements in the production process for steel.
Much of the research on buyersupplier relations and subcontracting has
focused on more traditional industries, such as automobiles or textiles. To be sure,
these industries make considerable use of technological advances, but they are less
science-driven. As a consequence, the sources of relevant knowledge are not as
widely dispersed. Strong ties thus tend to predominate over weak ties. But the
content of those ties can evolve, changing from contractual to relational. Consider
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 69
networks of innovators 69
subcontracting relations, particularly one of the more notable examplesthe
Toyota auto production network. Researchers stress how the density of overlapping
ties that connect this chain of production facilitates knowledge sharing, mutual
learning, and fast responsiveness (Dyer 1996; Dyer and Nobeoka 2000). But the trust
and reciprocity that characterize this dense network are the outcome of a long
developmental process. In the 1950s and early 1960s, when Japanese Wrms competed
on the basis of lowest cost, relationships with subcontractors were hierarchical and
asymmetric. As Wrms increasingly competed on the basis of quality and innovation,
however, complex multitiered supply relationships underwent signiWcant change.
These relationships can remain hierarchical in two key respects: the larger lead Wrm
often has a signiWcant Wnancial stake in a supplier or aYliate, and it initiates the
production process. But the asymmetry has been sharply reduced. Suppliers, in an
eVort to remain competitive, make signiWcant investments in new equipment,
constantly upgrade workers skills, and take on more critical aspects of the assembly
process (Helper et al. 2000). In turn, the larger Wrms oVer long-termcontracts, share
employees and provide technical assistance, and make Wnancial investments to fund
equipment upgrades.
Too explicit a focus on formal, contractual linkages, however, neglects the myriad
informal ties that connect organizations. All kinds of informal interactions take
place between organizations, including participation in ad hoc industry commit-
tees, or executive education programs, conferences, trade association activities, and
the like. Personnel mobility and common educational backgrounds may also foster
informal linkages across Wrms. Suchinformal connections may be the basis onwhich
more formal, contractual alliances are forged. Indeed, the success of formal aYlia-
tions may hinge on the strength of informal ties. Thus we turn to a discussion of
noncontractual relations.
3.4.2 Informal Ties
Informal patterns of aYliation have long been a central topic in sociology and
anthropology, where studies of friendship networks, advice and referral networks,
and communities are common. There is also a well-established strand of research in
organization theory that points out howinformal relations within organizations are
often not closely aligned with formal authority (Dalton 1959; Blau 1963). Asmall line
of work focuses on the impact of informal networks in large, multinational com-
panies (Ghoshal and Bartlett 1990; Hansen 1999). Relatively few studies, however,
link informal ties to the innovation process, and there is scant research on informal
interorganizational relations.
Scholars have often argued that the sharing of complex information is enhanced
by embedded ties, which suggests that informal ties have the potential to make a
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 70
70 walter w. powell and stine grodal
signiWcant contribution to innovation. There is a strong sense among researchers
that informal relations undergird formal ties. Powell et al. (1996) argue that, in the
life sciences, beneath most formal ties lie a sea of informal ties. Nevertheless, many
organizations are largely unaware of the extent to which formal activities are
buttressed by informal connections (Cross, Borgatti, and Parker 2002).
One of the key studies of informal networks among Wrms was Von Hippels (1987)
work on the sharing of proprietary information among US steel mini-mill produ-
cers. Based on interviews with plant managers and other engineers with direct
knowledge of manufacturing processes, he found that the trading of proprietary
knowledge withboth cooperating andrival Wrms was commonplace. He was initially
surprised that proprietary knowledge was so leaky, but he came to recognize that
information exchange was highly reciprocal and conditioned on expectations that
requests for help would be met. Much of the information that was shared focused on
production problems, matters of pollution control and safety, and issues dealing
with industry-wide concerns. But when relationships among engineers in rival Wrms
were particularly close, more proprietary information was exchanged. Von Hippel
also found that engineers had strong norms of membership in a professional
community that cut across Wrms, and that information trading was a means to
secure reputation and status in that community. He provides numerous examples of
howthe sharing of complex information by engineers contributed to the productiv-
ity of mini-mills.
The cluster of individuals that share a similar set of skills and expertise has been
dubbed a community of practice (Wenger 1998), or a network of practice
(Brown and Duguid 2001). Similar in some respects to a technical community, or
a sophisticated hobby club, these loose groups are engaged in related work practices,
thoughthey do not necessarily work together. SuchXuidgroups are important to the
circulation of ideas. Saxenian (1994) observed ample sharing of proprietary know-
ledge among engineers in Silicon Valley, many of whom have as strong a commit-
ment to their peers within the same occupational group as to their companies.
Saxenian argues that informal knowledge sharing, widely institutionalized as a
professional practice in Silicon Valley, is one of the crucial factors contributing to
its fertile innovative climate. Cohen and Fields (1999) stress that professional ties in
Silicon Valley are forged in complex collaborations between entrepreneurs, scien-
tists, Wrms, and associations, focused on the pursuit of innovation and its commer-
cialization. This collaborative process generates and reWnes the intangible raw
material of technical changeideas.
Kreiner and Schultz (1993) analyze the importance of informal ties through in-
depth interviews with university researchers and industry research directors in the
Danish biotech Weld. They stress that successful collaborative R&D alliances within
the Danish biotech industry are often based on informal ties. A barter economy,
where materials, laboratory tests, chemicals, etc. are exchanged, was pervasive in this
sector. They show that norms of sharing information on the frontier of research aid
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 71
networks of innovators 71
in the formation of more formal networks. As in the mini-mills, information
exchange is not under managerial control, even though such reciprocal Xows can
be channeled by managerial actions.
Many studies of informal relationships stress the signiWcance of trust. Tsai and
Ghoshal (1998) studied the association between intraWrmnetworks and innovation in
Wfteen business units of a multinational electronics company. They found, not
surprisingly, that social ties led to a higher degree of trustworthiness among business
units. Trust increasedresource-exchange andcombinationbetweenthe business units,
which contributedto product innovation. The importance of trust also looms large in
Uzzis (1997) analysis of the diVerence between arms-length ties (a deal in which
costs are everything) and embedded ties (you become friends with these people
business friends. You trust them and their work. Theyre part of the family). Uzzi
conducted interviews and ethnographic observations at twenty-three womens better-
dress Wrms in the NewYork City apparel industry. His study is notable not only for the
quality of his data, but also for his attention to the performance consequences of
diVerent kinds of exchange relations. Uzzi found that organizational performance
increases withthe use of embeddedties tonetworkpartners, as these ties were superior
at conveying complex, context-dependent knowledge. He argued, however, that a
balance between a Wrms embedded ties and a Wrms arms-length ties needed to be
struck, because a network structure comprising only arms-length ties or embedded
ties decreased organizational performance.
The signiWcance of a balanced network structure, mixing formal and informal
aYliations, is also emphasized in Ruef s (2002) analysis of entrepreneurship. He
found that individuals positioned in heterogeneous networks, comprising both
strong and weak ties, are more likely to be regarded as innovative by peers, in
comparison to entrepreneurs in more homogeneous networks. Rosenkopf has
found a similar interweaving of formal and informal relations in her research on
industry-wide expert communities in the areas of Xight simulation and mobile
phones. Rosenkopf, Metiu, and George (2001) analyze joint participation by cellular
service Wrms in technical committees, Wnding that such membership facilitates
subsequent formal interWrm alliance formation. The eVect of participation in
technical committees decreases when Wrms have already established prior alliances,
suggesting that the eVect of informal ties is more catalytic when Wrms do not already
have established alliance partners.
3.4.3 Multi-Party Relationships
Most studies of networks and innovation have examined either dyadic ties, or a focal
Wrm in the context of an overall network. Rosenkopf and Tushmans (1998) work
on expert communities emphasized the importance of studying multi-party
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 72
72 walter w. powell and stine grodal
relationships that connect technical professionals across organizations. Akera (2001)
uses archival data to map the importance of the IBM user group Share in the early
days of computing, oVering a portrait of a large multi-party collaboration. Share was
formed soon after IBMs release of the Wrst mainframe computer in 1953, to enable
IBMs costumers to swap programs, and collaborate on programming so that
duplicative eVort was avoided. Some of the main innovations that came out of
Share provided the basis for both systems programming and operating systems,
which today form the backbone of modern computer use. Share conveyed infor-
mation about hardware changes that substantially improved the design of IBM
computers and peripherals. One of the main contributions of the Share network
was the creation of technical standards, but a second-order eVect was also the
diVusion of knowledge across companies and between users.
Amodern analogue to Share is the Linux community, founded as a group of users
trying to develop an alternative to the operating system supplied by Microsoft. The
network of Linux programmers has proven eVective in developing software in a
highly distributed fashion. In the beginning, most programmers had never met each
other and only kneweach other virtuallyby the username they used when coding.
The Linux community has a very modest organizational structure, relying on a
combination of interpersonal networks and an individuals reputation as a skilled
programmer toserve as the admissionticket tothe network(OMahony 2002; Weber
2003). One diVerence between the Share network and the Linux network is that
corporate interests drove the Share network, while Linux has been primarily driven
by the end users. Nevertheless, in both cases, the decomposability of programming
tasks is an important factor in facilitating distributed networks.
Another large network that has been widely studied is the network of scientists,
often termed invisible colleges (Crane 1972). An invisible college is an informal
network of researchers who form around a common problem or paradigm. By
studying invisible colleges, Crane (1972) hoped to understand how knowledge
grows and how the structure of scientiWc communities aVects the expansion of
knowledge. There are now numerous studies of scientiWc networks, mapping the
structure of co-authorship and citations (Newman 2003 provides a good overview),
though few attend explicitly to the issue of innovation. David (2001), however,
develops a formal model to show that the liberal sharing of knowledge within the
scientiWc community is a major driver of scholarly innovation. One of the historic
characteristics of scientiWc communities is that information and research results
have been distributed openly among members of the relevant community. The shift
toward increasing research commercialization by universities has led some scholars
to question whether the innovative beneWts of invisible colleges will persist, or if
commercial interests will block informal knowledge sharing among scientists
(Powell and Owen-Smith 1998; Owen-Smith 2003). Chapter 8 by Mowery and
Sampat on universityindustry interfaces oVers a more detailed discussion of the
role of universities in the innovation process.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 73
networks of innovators 73
Another line of research that has attended to scientiWc and technological net-
works, dubbed ActorNetwork Theory (ANT), examines howparticular deWnitions
or conWgurations of science and technology triumph over alternative conceptions.
ActorNetwork Theory is unique in its treatment of artifacts and technologies, as
well as people and organizations, as members of a network (Callon 1998; Latour
1987). The primary contribution of ANT to the relationship between networks and
innovation is to showthat not only can networks facilitate innovation, but they also
constrain it by determining the kind of innovations produced, their subsequent
interpretation, and their Wnal use (Callon 2002).
A related line of work that looks at networks as systems of activity is the markets-
as-networks approach, developed by Scandinavian marketing researchers (Hakans-
son and Snehota 1995). This approach examines the multiple relationships among
organizations, and shows how these diVerent aspects of interorganizational rela-
tionships transform and evolve over time. For example, supplier networks may
change frequently, with diVerent elements of production being either outsourced
or insourced (Waluszewski 1995). The resources that are exchanged among the
partners in a production network are constantly changing. What determines
whether an entity is a resource depends on the situation, and its use in combination
with other resources. Resources are, thus, always polyvalent in both use and value.
The participants in a production network, both individually and collectively, de-
velop bonds characterized by trust and commitment. These bonds also have an
organizing eVect on networks, as they shape the identities of actors, and account for
diVerent levels of commitment among participants. This rich vein of qualitative
research has not explicitly focused on innovation, however.
The various studies of multi-party networks tend to emphasize the processual
aspects of collaboration. This attention to content is welcome, but it sometimes
comes at the expense of measuring the output of relationships, particularly how the
sharing and processing of information by members of a network can determine the
generation of novelty. A fuller understanding of the innovation process needs to
examine the topic of information sharing, a subject to which we now turn.
3. 5 Knowledge Transfer
The role of knowledge transfer is clearly central to the innovation process. Research
has highlightedtwo diVerent aspects of the knowledge-transfer process, each of which
inXuences innovation, albeit in diVerent respects. One explanation for the exchange
of information through networks emphasizes the importance of complementary
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 74
74 walter w. powell and stine grodal
assets in the division of innovative labor (Mowery, Oxley, and Silverman 1996). If
WrmAis good at producing a speciWc component and WrmB is capable of using that
component to produce an engine, they collaborate in a joint production in which
their capabilities reinforce one another. In biotechnology, for example, small Wrms
with close ties to university scientists may excel at drug development, but lack the
skills and resources to manage or fund costly clinical trials. By working closely with a
research hospital and an established Wrm that has a limited pipeline of new medi-
cines, the parties collaborate in a division of labor that is mutually rewarding, and
can result in the participants learning from one another, and accomplishing tasks
they could not do individually.
A second form of knowledge sharing occurs when existing information within a
network is recombined in novel ways. Indeed, novelty is often the unanticipated
result of reconWguring existing knowledge, problems, and solutions (Nelson and
Winter 1982; Fleming and Sorenson 2001). As a consequence of such collisions or
transpositions, Wrms can generate something they were unable to create on their
own. Both forms of knowledge transfer depend on some manner of successful
exchange of ideas, however.
An oft-used distinction is drawn between tacit and explicit knowledge (Cowan,
David, and Foray 2000). Interest in tacit knowledge stems from Polanyis (1956)
argument that we frequently know a good deal more than we can express verbally.
Explicit knowledge is highly codiWed, as in blueprints, recipes, manuals, or in the
form of training. Tacit knowledge lacks such extensive codiWcation (Nonaka and
Takeuchi 1995). Valuable, productive knowledge often demands considerable eVort
to acquire, and suchknowledge is frequently altered inthe process of acquisitionand
application. Perhaps the most vivid example is the continuing eVort of US auto-
makers to acquire, understand, and implement the Japanese system of lean produc-
tion (Womack, Jones, and Roos 1990; Dyer and Nobeoka 2000). Knowledge of
complex production technologies is rarely obtained in a fully digestible form;
understanding inevitably entails learning by using. The distinction between codiWed
and tacit is key because the latter demands considerably more trial-and-error
learning to apply the new knowledge in a diVerent setting.
Many studies point to the relatively easy transferability of explicit knowledge in
contrast to tacit knowledge. Simonin (1999) shows that knowledge transfer within
alliances is negatively aVected by both the nature of knowledge and diVerences in
organizational culture. He observes important diVerences in knowledge exchange
between long and short-lived alliances. Older alliances develop a common language
and shared mental models between partners, suggesting a learning curve within
alliances where the negative eVects of lack of experience and knowledge complexity
subside as the alliance matures. Thus, as an alliance ages and participants develop
relationship-speciWc understanding, there is the opportunity to convey more subtle
forms of information more eVectively. Complex tacit knowledge can become more
explicit as partners develop a wider bandwidth of communication.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 75
networks of innovators 75
If knowledge tacitness is a limiting factor in the transfer of knowledge, then the
cost of transferring knowledge is proportional to the type of knowledge transferred.
Easily transferred knowledge is widely dispersed at a lowcost (Boisot 1998), thus the
likelihood that explicit knowledge contains novel elements that would lead to
innovation is lower. On the other hand, when knowledge is very sticky (von
Hippel 1998) and contains a large tacit component, the degree of diYculty and the
costs of transfer are high. Consequently, the expected gains realized from this
information are uncertain, as the cost of obtaining information may exceed its
value. This suggests that when knowledge involves a moderate level of complexity,
the beneWts derived fromtransfer may be greatest. Figure 3.4 suggests a hypothetical
inverted U-shaped relationship between innovation and codiWcation. Here we
assume that there is variability in the cost of information transfer, and that the
greatest value may be derived when novel ideas are transmitted without too much
Szulanskis (1996) analysis of the transfer of internal benchmarking eVorts in
eight companies suggests key dimensions along which knowledge transfer can be
The costs of
The value of the
innovative output
exceeds the costs of
transfer and application.
There is a medium range of
knowledge codification where the
value of the innovative output
exceeds the costs of knowledge
transfer and recombination. This
domain includes information that
has important elements of novelty
but there are not significant
barriers to transfer.
When knowledge is highly
codified, the value of the
information is low; thus,
while transfer is inexpensive,
the gains are small.
When knowledge is very
tacit, the difficulty of
transfer and implementation
is high, and the payoffs are
uncertain and the process is
Fig. 3.4 Knowledge codication and innovation
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 76
76 walter w. powell and stine grodal
distinguished. He demonstrates that relationships between sender and receiver are
important, in that both parties need mutual awareness of state-of-the-art practices.
Obviously, communication is critical to information exchange. But even when
relationships function well, some knowledge is causally ambiguous, or sticky, and
thus not easily transferred. Moreover, information exchange is hindered when the
parties have diVerential levels of absorptive capacity, that is, the ability to recognize
the value of newinformation, assimilate it, and apply it to commercial ends (Cohen
and Levinthal 1990). This capacity is essential to innovative capability. For example,
internal success with R&Dand R&Dexpenditures positively aVect a Wrms ability to
exploit the opportunities presented in external relations (Cohen and Levinthal
1990; Powell et al. 1996).
The productive transfer of knowledge is also essential whentwoor more organiza-
tions are able to combine their diVerent capabilities, and create a product or service
that they would not be able to construct on their own. A good illustration is the
Italian motorcycle industry where the locus of innovation is broadly dispersed,
and through cooperation, the participants beneWt from specialization and variety
generation (Lipparini, Lorenzoni, and Zollo 2001). Because all the participants
provide valuable inputs, there is a high commitment to knowledge generation.
The lead Wrms develop relational capacities aimed at pooling the skills of specialized
participants, helping the overall Xow of information and resources in the network.
A parallel analysis of the Italian packaging machinery industry stresses the creation
of a supplier network in which specialized roles are highly complementary
(Lorenzoni and Lipparini 1999). Over time, managers of the core companies de-
veloped a specialized supplier network and each participant focused on a narrow,
but highly competitive set of core competencies. This network structure enabled
every step of the supply chain to specialize in improvements of their speciWc
component, thereby increasing the responsiveness of the participants to market
3. 6 Governance and Incentive Issues
Many studies of interWrm networks draw data from a single point in time and thus
do not examine how collaborations unfold over time. Even studies that do look at
dynamics tend to do so from the perspective of a dyad. Initially, the choice by a
young Wrmof whomto partner with is often driven by resource needs. As both Wrms
and the network mature, various dyadic choices increasingly reXect structural
properties of the network. Thus, the existing network structure shapes search
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 77
networks of innovators 77
behavior. Consequently, networks both enable information to become knowledge
and determine the nature of knowledge (Kogut 2000). But we do not, as yet, have a
parallel understanding of the management and governance of networks to accom-
pany analyses of structure and topology. Concerns with how the parties in a
relationship adapt to changing circumstances, or attend to the incentives to adjust
the relationship to make improvements remain largely unexamined.
Not surprisingly, then, many studies assume that as long as a relationship
persists, the participants are achieving their goals, and view termination as a sign
of failure. Such a view misses the point that the relationship may have completed
its goal or outlived its usefulness. Inkpen and Ross (2001) Wnd evidence that
the termination of alliances might not signify failure, but simply be a sign of the
conclusion of collaborative activity (for example, a new product is launched).
Nevertheless, some collaborations seem to persist even though they have stagnated
or outlived their usefulness. Inkpen and Ross suggest several reasons why interWrm
alliances may persist beyond their optimal duration.4 Alliances can be diYcult
and expensive to form; thus, once established, there may be reluctance to disband
them. Moreover, as more Wrms pursue alliances, a bandwagon eVect is created
and many Wrms jump on it out of fear of being left without a partner. There are
also challenges and costs associated with managing a partnership. If the relationship
is poorly coordinated, the costs can outweigh the beneWts. Alliances can also
become synonymous with a Wrms values, making them diYcult to discontinue.
Finally, there may be costs associated with closing an alliance. All of these factors
may contribute to alliances existing beyond the period when they create value for
a Wrm.
Several studies have pointed to problems of stagnation that can occur in some
long-term associations. Although this work does not deal directly with rates of
innovation, focusing instead on viability and survival, the general point is apt.
When the participants in a network become too tightly knit and information
circulates only among a small group, networks can become restrictive and ossiWed.
Information that cycles back and forth only among the same participants can lead to
lock-in or sclerosis. When networks turn inward and become restricted in terms of
access to new members, the possibility of group think increases. Grabhers (1993)
study of steelmaking in the Ruhr illustrates how a highly cohesive, homogeneous
region became so overembedded that no producers opted for alternative strategies.
This cognitive lock-in eventually led to the decline of steelmaking in Germany.
Powell (1985) shows how the failure of editors to renew and update their networks
leads to a decline in the quality of a book publishers list and reputation. Portes
and Sensenbrenner (1993) illustrate how ethnic community networks can become
restrictive and subject successful entrepreneurs to ostracism when they deviate
too much from community standards. In sum, the ties that bind economic
actors together can become the ties that blind, thwarting recognition of preferable
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 78
78 walter w. powell and stine grodal
3. 7 Summary
Interorganizational networks have grown considerably in importance over recent
decades. Networks contribute signiWcantly to the innovative capabilities of Wrms by
exposing them to novel sources of ideas, enabling fast access to resources, and
enhancing the transfer of knowledge. Formal collaborations may also allow a
division of innovative labor that makes it possible for Wrms to accomplish goals
they could not pursue alone. Research on alliances documents that investments in
mutual learning and a portfolio of diverse collaborations are associated with in-
creased patenting. While patenting is only an input to the innovation process, the
strength of these empirical results highlights the importance of access to hetero-
geneous contacts.
We have argued that the nature of knowledge, conceptualized in terms of tacitness
or explicitness, is animportant factor indetermining whether members of a network
can eVectively share information and skills. Networks that are rooted in a division of
innovative labor logic may Wnd it easier to transfer tacit knowledge in the form of
Wnished inputs, while networks involved in the co-creation of novel ideas may
succeed or fail on the basis of their ability to convey and transfer ideas that are not
easily codiWed.
Another central challenge to networks of innovators is developing the capacity to
simultaneously enhance the Xowof information among current participants and be
open to new entrants. The twin tests of increasing cohesion within the network and
recognizing promising sources of newideas are diYcult to surmount. Some research
suggests that a mixture of strong and weak ties aVords the proper blend of reliability
and novelty.
Much more research is needed, however, to ascertain how mixtures of thick,
reliable aYliations can be combined with novel linkages to with newcomers. More-
over, since some aYliations are essentially person to person ties, a greater under-
standing is needed of howrelations among individuals are aggregated to productive
relations between corporate actors. Or, in cases where this issue is neglected, as in
studies of contractual relations between Wrms where the informal relational under-
pinnings are not analyzed, the opportunity to examine the intertwining of the
careers of individuals and the strategies of Wrms is missed. Equally important, the
network literature has not focused explicitly on diVerent measures of innovative
output, whether it is newproducts or services, newmodes of organizing production,
or more rapid response to competitive demands. The standard measures are based
on either patents, an input to the innovation process, or problem solving, without
suYcient attention to either the timeliness or the optimality of the solution. The
network literature is still relatively young, however, so it may be premature to expect
such sophisticated answers. We look forward to future research that oVers a more
compelling analysis of the speciWc ways inwhichnetworks shape innovative outputs.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 79
networks of innovators 79
1. The Wve that formed the core of the human sequencing component were the Whitehead
Institute, Washington University, Baylor College of Medicine, the Joint Genome Institute
(a cluster of three national laboratories at the Department of Energy), and the Sanger
Institute in England.
2. Notable exceptions include Ahuja 2000a, Rowley et al. 2000, Powell et al. 2005.
3. Owen-Smith and Powell (2003) found that US universities with strong ties to a limited set
of commercial partners had fertile patent portfolios, with fertility measured by the
impact of patent citations. Those universities with few ties also had les fertile patents and
patented much less. The optimal strategy for research universities, with respect to patent
volume and impact, appears to be one of diverse ties to a wide array of industrial partners.
Diversity mitigates possible capture from too close relations with commercial Wrms.
4. Inkpen and Ross (2001) do not specify how to measure the appropriate duration of
relationships. Instead, they assume that some relationships become stale and dutiful
over time, and no longer generate beneWts that outweigh the costs of sustaining them.
*Ahuja, G. (2000a), Collaboration Networks, Structural Holes, and Innovation: A Longi-
tudinal Study, Administrative Science Quarterly 45: 42555.
(2000b), The Duality of Collaboration: Inducements and Opportunities in the For-
mation of InterWrm Linkages, Strategic Management Journal 21: 31743.
Akera, A. (2001), The IBM User Group, Share, Technology and Culture 42: 71036.
Baum, J. A. C., Calabrese, T., and Silverman, B. S. (2000), Dont Go It Alone: Alliance
Network Composition and Startups Performance in Canadian Biotechnology, Strategic
Management Journal 21: 26794.
Beckman, C., and Haunschild, P. (2002), Network Learning: The EVects of Partners
Heterogeneity of Experience onCorporate Acquisitions, Administrative Science Quarterly
47: 921724.
Blau, P. M. (1963), Consultation Among Colleagues, in Dynamics of Bureaucracy, Chicago:
University of Chicago Press, 15769.
Boisot, M. H. (1998), Knowledge Assets: Securing Competitive Advantage in the Information
Economy, Oxford: Oxford University Press.
Brown, J. S., and Duguid, P. (2001), Knowledge and Organization: A Social Practice
Perspective, Organization Science 12(2): 198213.
Brusco, S. (1982), The EmilianModel: Productive DecentralizationandSocial Integration,
Cambridge Journal of Economics 6: 16784.
*Burt, R. S. (1992), Structural Holes, Cambridge, Mass.: Harvard University Press.
Callon, M. (1998), Actor-Network TheoryThe Market Test, in J. Law and J. Hassard
(eds.), Actor Network Theory and After, Cambridge, Mass.: Blackwell, 18696.
* Asterisked items are suggestions for further reading.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 80
80 walter w. powell and stine grodal
(2002), From Science as an Economic Activity to Socioeconomics of ScientiWc Re-
search, in P. Mirowski and E. M. Sent (eds.), Science Bought and Sold, Chicago: University
of Chicago Press.
Chesbrough, H. (2003), The Era of Open Innovation, Boston: Harvard Business School
Cohen, S., and Fields, G. (1999), Social Capital and Capital Gains in Silicon Valley,
California Management Review 41(2): 10830.
Cohen, W., and Levinthal, D. (1990), Absorptive Capacity: A New Perspective on Learn-
ing and Innovation, Administrative Science Quarterly 35: 12852.
Cowan, R., David, P. A., and Foray, D. (2000), The Explicit Economics of Knowledge
CodiWcation and Tacitness, Industrial and Corporate Change 9: 21153.
Crane, D. (1972), Invisible Colleges: DiVusion of Knowledge and ScientiWc Communities,
Chicago: University of Chicago Press.
Cross, R., Borgatti, S., and Parker, A. (2002), Making Invisible Work Visible, California
Management Review 44(2): 2546.
Cusumano, M., Mylonadis, Y., and Rosenbloom, R. (1992), Strategic Maneuvering
and Mass-Market Dynamics: The Triumph of VHS over Beta, Business History Review
66: 5194.
Dalton, M. (1959), Power Struggles in the Line, in Men Who Manage, New York: Wiley,
David, P. A. (2001), Cooperation, Creativity, and Closure in ScientiWc Research Networks:
Modeling the Simpler Dynamics of Epistemic Communities, Conference at Centre Saint-
Gobain, Paris.
*Dyer, J. H. (1996), Specialized Supplier Networks as a Source of Competitive Advantage:
Evidence from the Auto Industry, Strategic Management Journal 17(4): 27191.
and Nobeoka, K. (2000), Creating and Managing a High-Performance Knowledge-
Sharing Network: The Toyota Case, Strategic Management Journal 21: 34567.
Eccles, R. (1981), The QuasiWrm in the Construction Industry, Journal of Economic
Behavior and Organization 2: 33557.
Eisenhardt, K. M., and Schoonhoven, C. B. (1996), Resource-based View of Strategic
Alliance Formation: Strategic and Social EVects in Entrepreneurial Firms, Organization
Science 7: 13650.
Farrell, J., and Salomer, G. (1988), Coordination through Committees and Markets,
RAND Journal of Economics 19(2): 23552.
Fleming, L., and Sorenson, O. (2001), Technology as a Complex Adaptive System:
Evidence from Patent Data, Research Policy 30: 101939.
*Freeman, C. (1991), Networks of Innovators: A Synthesis of Research Issues, Research
Policy 20: 499514.
Galunic, D. C., and Rodan, S. (1998), Resource Recombinations in the Firm: Knowledge
Structures and the Potential for Schumpeterian Innovation, Strategic Management Jour-
nal 19: 1193201.
Ghoshal, S., and Bartlett, C. (1990), The Multinational Corporation as an Interorgani-
zational Network, Academy of Management Review 15: 56185.
Godoe, H. (2000), Innovation Regimes, R&D and Radical Innovations in Telecommuni-
cations, Research Policy 29: 103346.
Gomes-Casseres, B. (1996), The Alliance Revolution: The New Shape of Business Rivalry,
Cambridge, Mass.: Harvard University Press.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 81
networks of innovators 81
Grabher, G. (1993), The Weakness of Strong Ties: The Lock-in of Regional Development in
the Ruhr Area, in G. Grabher (ed.), The Embedded Firm, London: Routledge.
and Powell, W. W. (2004), Introduction, in Critical Studies in Economic Institutions:
Networks, London: Edward Elgar.
*Granovetter, M. (1973), The Strength of Weak Ties, American Journal of Sociology 78:
(1985), Economic Action and Social Structure: The Problem of Embeddedness,
American Journal of Sociology 91: 481510.
Griliches, Z. (1990), Patent Statistics as Economic Indicators: A Survey, Journal of
Economic Literature 28: 1661707.
Gulati, R. (1995), Does Familiarity Breed Trust? The Implications of Repeated Ties for
Contractual Choice in Alliances, Academy of Management Journal 38: 85112.
Hagedoorn, J. (1995), Strategic Technology Partnering During the 1980s: Trends,
Networks, and Corporate Patterns in Non-Core Technologies, Research Policy 24:
(1996), Trends and Patterns in Strategic Technology Partnering since the Early Seven-
ties, Review of Industrial Organization 11: 60116.
(2002), Inter-Firm R&D Partnerships: An Overview of Major Trends and Patterns
since 1960, Research Policy 31: 47792.
, and Duysters, G. (2002), External Sources of Innovative Capabilities: The
Preference for Strategic Alliances or Mergers and Acquisitions, Journal of Management
Studies 39(2): 16788.

kansson, H., and Snehota, I. (1995), Developing Relationships in Business Networks,

London: Routledge.
Hansen, M. T. (1999), The SearchTransfer Problem: The Role of Weak Ties in Sharing
Knowledge across Organization Subunits, Administrative Science Quarterly 44: 82111.
Helper, S., MacDuffie, J. P., and Sabel, C. (2000), Pragmatic Collaborations: Advancing
Knowledge While Controlling Opportunism, Industrial and Corporate Change 9(3): 443
Hicks, D. M., and Katz, J. S. (1996), Where is Science Going? Science, Technology and
Human Values 21(4): 379406.
Inkpen, A. C., and Ross, J. (2001), Why do Some Strategic Alliances Persist beyond their
Useful Life? California Management Review 44: 13248.
Kogut, B. (2000), The Network as Knowledge: Generative Rules and the Emergence of
Structure, Strategic Management Journal 21: 40525.
Kreiner, K., and Schultz, M. (1993), Informal Collaboration in R&D. The Formation of
Networks across Organizations, Organization Studies 14: 189209.
Lambright, W. H. (2002), Managing Big Science: A Case Study of the Human Genome
Project, Washington, DC: Pricewaterhouse Coopers Endowment for the Business of
Larson, A. (1992), Network Dyads in Entrepreneurial Settings: A Study of the Governance
of Exchange Processes, Administrative Science Quarterly 37: 76104.
Latour, B. (1987), Science in Action, Cambridge, Mass.: Harvard University Press.
Leonard-Barton, D. (1995), Wellsprings of Knowledge: Building and Sustaining the Sources of
Innovation, Boston: HBS Press.
Link, A. (1996), Research Joint Ventures: Patterns from Federal Register Filings, Review of
Industrial Organization II (October): 61728.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 82
82 walter w. powell and stine grodal
(1999), Public/Private Partnerships inthe United States, Industry and Innovation 6(2):
Lipparini, A., Lorenzoni, G., and Zollo, M. (2001), Dual Network Strategies: Managing
Knowledge-based and EYciency-based Networks in the Italian Motorcycle Industry,
Working paper.
Lorenzoni, G., and Lipparini, A. (1999), The Leveraging of InterWrm Relationships as a
Distinctive Organizational Capability, Strategic Management Journal 20: 31738.
McKendrick, D. G., Donner, R. F., and Haggard, S. (2000), From Silicon Valley to
Singapore: Competitive Advantage in the Hard Disk Drive Industry, Stanford, Calif.: Stan-
ford University Press.
Merrill, S. A., and Cooper, R. S. (1999), Trends in Industrial Research and Development:
Evidence from National Data Sources, in Securing Americas Industrial Strength, Wash-
ington, DC: National Academy Press, 99116.
Merton, R. K. (1957), The Role-Set: Problems in Sociological Theory, British Journal of
Sociology 8: 11020.
Mowery, D. C. (ed.) (1988), International Collaborative Ventures in US Manufacturing,
Cambridge, Mass.: Ballinger.
(1999), Americas Industrial Resurgence? An Overview, in U.S. Industry in 2000:
Studies in Competitive Performance. Washington, DC: National Academy Press, 116.
Oxley, J. E., and Silverman, B. S. (1996), Strategic Alliances and InterWrm Know-
ledge Transfer, Strategic Management Journal 17: 7791.
Murray, F. (2002), Innovation as Co-evolution of ScientiWc and Technological Networks:
Exploring Tissue Engineering, Research Policy 31: 1389403.
National Science Board (1998), Science and Engineering Indicators1998. Arlington, Va.:
National Science Foundation.
Nelson, R., and Winter, S. (1982), An Evolutionary Theory of Economic Change, Cambridge,
Mass.: Harvard University Press.
Newman, M. (2003), The Structure and Function of Complex Networks, SIAMReview45:
Nonaka, I., and Takeuchi, N. (1995), The Knowledge-Creating Company, NewYork: Oxford
University Press.
Noteboom, B. (1999), Inter-Firm Alliances: Analysis and Design, London: Routledge.
OMahony, S. (2002), The Emergence of a New Commercial Actor: Community Managed
Software Projects, Ph.D. dissertation, Department of Management Science and Engineer-
ing, Stanford University.
Owen-Smith, J. (2003), From Separate Systems to a Hybrid Order: Accumulative Advan-
tage Across Public and Private Science at Research One Universities, Research Policy 32(6):
and Powell, W. W. (2003), The Expanding Role of University Patenting in the Life
Sciences, Research Policy 32(9): 1695711.
Piore, M., and Sabel, C. (1984), The Second Industrial Divide, New York: Basic Books.
Polanyi, M. (1956), Personal KnowledgeTowards a Post-Critical Philosophy, Chicago:
University of Chicago Press.
Portes, A., and Sensenbrenner, J. (1993), Embeddedness and Immigration: Notes on the
Social Determinants of Economic Action, American Journal of Sociology 98: 132050.
Powell, W. W. (1985), Getting Into Print: The Decision-Making Process in Scholarly Publish-
ing, Chicago: University of Chicago Press.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 83
networks of innovators 83
Powell, W. W. (1990), Neither Market Nor Hierarchy: Network Forms of Organization, in
B. M. Straw and L. L. Cummings (eds.), Research in Organizational Behavior, Greenwich,
Conn.: JAI Press, 12: 295336.
(1996), Inter-Organizational Collaboration in the Biotechnology Industry, Journal of
Institutional and Theoretical Economics 120(1): 197215.
(1998), Learning fromCollaboration: Knowledge and Networks in the Biotechnology
and Pharmaceutical Industries, California Management Review 40(3): 22840.
and Brantley, P. (1992), Competitive Cooperation in Biotechnology: Learning
Through Networks? in R. Eccles and N. Nohria (eds.), Networks and Organizations,
Boston: Harvard University Press.
*Koput, K. W., and Smith-Doerr, L. (1996), Interorganizational Collaboration
and the Locus of Innovation in Biotechnology, Administrative Science Quarterly 41(1):
and Owen-Smith, J. (1999), Network Position and Firm Performance, in S.
Andrews and D. Knoke (eds.), Research in the Sociology of Organizations Greenwich,
Conn.: JAI Press, 16: 12959.
and Owen-Smith, J. (1998), Commercialism in Universities: Life Sciences
Research and Its Linkage with Industry, Journal of Policy Analysis and Management 17
(2): 25377.
, White, D. R., Koput, K., and Owen-Smith, J. (2005), Network Dynamics and Field
Evolution: The Growth of Interorganizational Collaboration in the Life Sciences, Ameri-
can Journal of Sociology.
Roberts, E. B., and Liu, W. K. (2001), Ally or Acquire? How Technology Leaders Decide,
Sloan Management Review 43: 2634.
Rosenbloom, R. S., and Spencer, W. J. (1996), The Transformation of Industrial Research,
Issues in Science and Technology 12(3): 6874.
*Rosenkopf, L., and Tushman, M. (1998), The Coevolution of Community Networks and
Technology: Lessons from the Flight Simulation Industry, Industrial and Corporate
Change 7: 31146.
Metiu, A., and George, V. P. (2001), From the Bottom Up? Technical Committee
Activity and Alliance Formation, Administrative Science Quarterly 46: 74872.
Rowley, T., Behrens, D., and Krackhardt, D. (2000), Redundant Governance Struc-
tures: An Analysis of Structural and Relational Embeddedness in the Steel and Semicon-
ductor Industries, Strategic Management Journal 21: 36986.
Ruef, M. (2002), Strong Ties, Weak Ties and Islands: Structural and Cultural Predictors of
Organizational Innovation, Industrial and Corporate Change 11: 42749.
Sabel, C. F. (1994), Learning by Monitoring: The Institutions of Economic Development,
in N. Smelser and R. Swedberg (eds.), Handbook of Economic Sociology, Princeton:
Princeton University Press, 13765.
Sarkar, K. B., Echambadi, R., and Harrison, J. S. (2001), Alliance Entrepreneurship and
Firm Market Performance, Strategic Management Journal 21: 36986.
*Saxenian, A. (1994), Regional Advantage: Culture and Competition in Silicon Valley and
Route 128. Cambridge, Mass.: Harvard University Press.
Shan, W., Walker, G., and Kogut, B. (1994), InterWrm Cooperation and Startup Innov-
ation in the Biotechnology Industry, Strategic Management Journal 15: 38794.
Simmel, G. (1954), ConXict and the Web of Group AYliations, Glencoe, Ill.: The Free Press.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 84
84 walter w. powell and stine grodal
Simonin, B. L. (1999), Ambiguity and the Process of Knowledge Transfer in Strategic
Alliances, Strategic Management Journal 20: 595623.
Stuart, T. E. (1998), Network Positions and Propensities to Collaborate, Administrative
Science Quarterly 43: 66898.
(2000), Interorganizational Alliances and the Performance of Firms: A Study of
Growth and Innovation Rates in a High-technology Industry, Strategic Management
Journal 21: 791811.
Szulanski, G. (1996), Exploring Internal Stickiness: Impediments to the Transfer of Best
Practice with the Firm, Strategic Management Journal 17: 2743.
Tsai, W., and Ghoshal, S. (1998), Social Capital and Value Creation: The Role of IntraWrm
Networks, Academy of Management Journal 41: 46476.
*Uzzi, B. (1997), Social Structure and Competition in InterWrm Networks: The Paradox of
Embeddedness, Administrative Science Quarterly 42: 46476.
Van Wijk, R., van den Bosch, F. A. J., and Volberda, H. W. (2003), Knowledge and
Networks, in M. Easterby-Smith and M. A. Lyles (eds.), Handbook of Organizational
Learning and Knowledge Management, Oxford: Blackwell.
Vinding, A. L. (2002), Interorganizational DiVusion and Transformation of Knowledge in
the Process of Product Innovation, Ph.D. Thesis, Aalborg University.
Von Hippel, E. (1987), Cooperation between Rivals: Informal Know-how Trading,
Research Policy 16: 291302.
(1998), Economics of Product Development by Users: The Impact of Sticky Local
Information, Management Science 44: 62944.
*Walker, G. B., Kogut, B., and Shan, W. (1997), Social Capital, Structural Holes and the
Formation of an Industry Network, Organization Science 8: 10925.
Waluszewski, A. (1995), Glulam, in H. Hakansson and I. Snethota (eds.), Developing
Relationships in Business Networks, London: Routledge.
Weber, S. (2003), The Success of Open Source, Cambridge, Mass.: Harvard University Press.
Wenger, E. (1998), Communities of Practice, New York: Cambridge University Press.
Womack, J. P., Jones, D. T., and Roos, D. (1990), The Machine That Changed the World: The
Story of Lean Production, New York: Harper & Row.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 6:48pm page 85
networks of innovators 85
Technovation 24 (2004) 563571
The small worlds of strategic technology alliances
Bart Verspagen

, Geert Duysters
Eindhoven University of Technology, Faculty of Technology Management, PO Box 513, Eindhoven 5600 MB, Netherlands
This paper analyzes the phenomenon of strategic technology alliances. It is proposed that the concept of small worlds, which
has been adopted from mathematical graph theory, is a useful model to combine two theoretical streams that have previously
analyzed this phenomenon. These are the theory of social capital and the theory of structural holes. We outline a small worlds
model, and apply it to data on strategic technology alliances. We nd that networks of strategic technology alliances can indeed
be characterized as small worlds, and that this has favorable implications for knowledge transfer. There are, however, also important
differences between two different technology elds that we consider: chemicals and food, and electricals.
2002 Elsevier Ltd. All rights reserved.
Keywords: Strategic technology alliances; Models of network dynamics; Social capital; Structural holes
1. Introduction
In the past decades, strategic alliances have become
an important mechanism for (quasi) external acquisition
of technological know-how. The strong upheaval in the
use of strategic technology alliances has led to the
emergence of complex inter-organizational research net-
works. The growing strategic importance of these net-
works has been acknowledged in many contributions
(see, e.g., Burt, 1992; Gulati, 1998; Duysters and Hage-
doorn, 2002). The quest for a better understanding of
these networks subsequently led to the use sophisticated
network analysis techniques and conceptions arising
from graph theory for the analysis of these networks,
both from the point of view of policy and strategic man-
agement within rms (Vonortas, 1997).
Recently, the theory of small worlds has emerged in
the eld of graph theory (Watts, 2000; Cowan and Jon-
ard, 2000). This paper will attempt to apply this theory
to the case of sectoral technology alliance networks. It
will be argued that the theory of small worlds can be
seen as a model that unites two distinct perspectives
found in the strategic management literature. These are
the notions of social capital and structural holes (Walker

Corresponding author. Tel.: +31-40-2475613; fax: +31-40-

E-mail address: b.verspagen@tm.tue.nl (B. Verspagen).
0166-4972/$ - see front matter 2002 Elsevier Ltd. All rights reserved.
et al., 1997). Using the small-worlds idea in this way, it
is argued here that one may use this concept as a way to
judge the overall efciency of a network of technology
alliances. This argument is applied to empirical data on
two major RTD networks.
The next section of this paper will briey review the
social capital and structural holes theory of network for-
mation as it is found in the strategic management litera-
ture. In Sect. 3 we formulate two specic research ques-
tions and we outline the theory of small worlds, and
argue how it can be applied to the theory discussed in
Sect. 2. Section 4 will present the empirical results. The
argument will be summarized and concluded in Sec-
tion 5.
2. Strategic technology alliances and networks
One of the key trends over the past decades has been
the remarkable growth in the number of strategic
alliances (see Fig. 1). According to many authors, this
trend is likely to continue in the near future. Major con-
sultancy organizations such as BoozAllen Hamilton
(2000) predict that within ve years from now, the value
of alliances will be in the range of $30$50 trillion.
The rapid proliferation of alliances is in large part due
to an increase in the number of strategic technology
alliances (Hagedoorn, 1996; Duysters and Hagedoorn,
2002). Whereas, for a long time rms have relied heavily
564 B. Verspagen, G. Duysters / Technovation 24 (2004) 563571
Fig. 1. Number of newly established strategic alliances per year (19852000), three-year moving averages. Source: Thomson Financial.
on the internal development of know-how and new tech-
nology, rms have come to realize that internal develop-
ment is no longer sufcient to deal with their continu-
ously changing technological environment. A growing
number of rms seems to recognize that the use of stra-
tegic technology alliances allows them to increase their
exibility and to tap into other companys technologi-
cal resources.
The growth in the number of strategic technology
alliances has led to the emergence of complex tech-
nology networks. The growing importance of these net-
works induced a shift of focus in the alliance literature,
away from a dyadic relationship level towards a network
perspective. Firms which are embedded in these net-
works are increasingly affected by the relationships of
their partners and the overall structure of the network
(Granovetter, 1992). Embeddedness, on the one hand,
restrains rms in their behavior and on the other hand
creates opportunities for companies to tap into rich
The central notion in the network literature is what
Nooteboom (1999) has coined indirect access. Because
an alliance partner is also likely to be connected to
others, a rm does not only gain access to resources of
its immediate partner, but also, to a certain extent, to
those with whom the partner collaborates. The issue of
partner-seeking then becomes one of picking partners
not only on the basis of their own capabilities and
resources, but also those of the indirect actors it gives
(partial) access to in terms of its network.
This gives rise to the idea of a set of strategic alliances
as a network of actors that are directly or indirectly con-
nected to each other. A direct connection results from
participation in a specic alliance. Indirect connections
result when information or knowledge exchanged in one
partnership is also (implicitly) entered in other partner-
ships. Such networks have been the topic of the math-
ematical branch of graph theory, and have been studied
since the 1950s (Watts, 2000 provides a brief overview
of the history of the eld). A crucial element in this
theory is the structure of the network and the position
that actors occupy in them.
The focus on the importance of indirect ties gave rise
to two different perspectives on the dynamics governing
the formation of networks and their efciency as infor-
mation or knowledge transmitters.
The rst perspective emphasizes the importance of
social capital (Bourdieu, 1980; Coleman, 1990). Social
capital can be dened as the sum of the resources,
actual or virtual, that accrue to an individual or a group
by virtue of possessing a durable network of more or
less institutionalized relationships of mutual acquaint-
ance and recognition (Bourdieu and Wacquant, 1992,
as cited by Walker et al., 1997, p. 109). Putting the con-
cept of social capital in a network context, Walker et al.
have argued that the density of network relations is a
good proxy: Some rms occupy positions that are
embedded in regions lled with relationships, indicating
a high level of available social capital, but other pos-
itions are located in regions with few relationships, sug-
gesting a low social capital (Walker et al., 1997, p.
111). In general, one can argue that rms with more
social capital will have access to a larger pool of infor-
mation sources and will be able to attract better partners
(Gulati, 1998).
The second perspective on networks stems from the
work by Burt (1992), who argued that a rm interested
in using networks as a source of information should
attempt to ll the structural holes in between the dense
areas of a network rather than replicate existing partner-
ships that are already in place. In other words, in strategi-
cally selecting ones partners for cooperation, one should
look for partners that have strong links to other actors
with whom oneself does not yet have strategic links, in
order to try to bridge holes in ones own network. From
this perspective, the primary strategic aim of forming
partnerships becomes the desire to serve as bridges
between two relatively unconnected parts of a network.
565 B. Verspagen, G. Duysters / Technovation 24 (2004) 563571
Burt (1992) argues that this strategy will provide access
to knowledge or information that has a high yield. The
roots of this perspective can be found in social network
literature that emphasizes the importance of weak ties
and the importance of overarching structural holes.
McEvily and Zaheer (1999) have argued that bridging
ties are characterized by:
Non-redundancy, operationalized as the degree to
which the contacts in a rms ego network are not
linked to one another.
Infrequency of interaction with the partners.
Geographic dispersion the more distant, the more
Overall, Burt argues that redundant contacts often
carry the same information. Therefore, organizations
should focus on establishing non-redundant contacts that
enable rms to bridge dense areas in the network.
Confronting these two perspectives, one can clearly
see two distinct tendencies emerging. Strategies built
solely on the idea of enhancing social capital would lead
to networks with highly dense local environments.
When, on the other hand, in an extreme case, rms
mainly try to pursue a strategy of lling structural holes
in their own network, a tendency towards less densely
connected local environments would result.
3. Research question and theory
3.1. Research question
Intuitively, the social capital strategy and the struc-
tural holes strategy will lead to quite different networks
of strategic technology alliances. Firms dominated by a
social capital view of the world will seek out a limited
number of partners with whom they build strong and
repeated ties. This will result in an overall network struc-
ture in which local environments of a rm are densely
populated with alliances, but where path length to other
local environments is long. Alternatively, in a world
dominated by strategies aimed at bridging structural
holes, local environments will tend to be populated less
densely, while paths through the network will tend to
be shorter.
This intuitive argument would suggest that there is
some degree of trade-off involved between social capital
(density of network) and path length, where higher social
capital leads to longer paths. We will use a simple model
presented by Watts (2000) in order to investigate the
nature of this trade-off. The model will show that there
is a particular class of networks, which are known in the
literature as small worlds, in which relatively high social
capital can go hand in hand with relatively short path
length. The mechanism involved in shaping the network
as a small world will be the existence of so-called short-
cuts in the network. We will relate this concept to the
notion of bridge-building in Burts theory.
Our second research question is how actual networks
of strategic technology alliances compare to our hypo-
thetical network structures. In other words, we will
investigate whether these actual networks resemble one
of our extreme network structures (social capital world,
or world dominated by bridges to cover structural holes),
or whether they can be classied as a small world. In
this way, we will be able to put a value on the relative
efciency of the whole network as a generator of social
capital, or as a transmitter of information and knowl-
edge. Our special interest will be in nding differences
in this respect between different technology elds, or
over time.
3.2. Concepts and denitions
In order to describe our two extreme cases, two central
concepts will sufce. These are the concepts of charac-
teristic path length of a network, and that of clustering.
An assumption in our treatment of the concepts and the
theory hereafter will be that we deal with a network in
which all nodes are connected through each other
(directly or indirectly), i.e., that there are no parts of
the network that are unconnected to other parts. In our
empirical applications below, we will construct the net-
work by dening a direct link to be present if two rms
participate in at least one alliance together.
For any pair of actors in the networks, dene the path
length between them as the minimum number of inter-
mediaries necessary to connect them, plus one. It follows
from this denition that actors that are directly connected
have a path of length of one between them. Two actors
that are not directly connected, but share a third rm as
a direct connection, have path length equal to two, etc.
The particular indicator that has been used in much
theoretical work in order to arrive at a single measure
of path length for the network as a whole is called
characteristic path length. This is dened as the median
of average path length of all actors in the network. Aver-
age path length of an actor is dened as the average of
path length to all other actors in the network.
Characteristic path length of the network is interesting
because it gives an indication of the relative (potential)
efciency of the network. The shorter path length in a
network, the easier and quicker knowledge or infor-
mation may diffuse through the network. Form the point
of view of a single actor in the network, shorter path
length implies easier access to the knowledge of other
actors in the network.
Characteristic path length will depend on a number
The exposition here draws heavily on Watts (2000).
566 B. Verspagen, G. Duysters / Technovation 24 (2004) 563571
of network characteristics. The number of actors in the
network is one prime determinant, as is the average num-
ber of direct connections that each actor has (the latter
is called average degree of the network). In addition to
these quantities, which can be easily measured for any
network, it is the network topology that determines
characteristic path length. A prime factor of interest with
regard to network topology is the way in which the edges
are distributed over the network as a whole.
The concept of network clustering can be used to
quantify some interesting factors in network topology.
In order to dene network clustering, one must rst
dene the neighborhood of actor i. The term will be used
to describe the subset of actors that have a direct connec-
tion to actor i (this is also called ego network of actor
i). Obviously, the number of actors in the neighborhood
of actor i is identical to the degree of actor i (i is dened
not to be a member of its own neighborhood). Now
dene clustering of the neighborhood of i as the number
of edges in the neighborhood i as a fraction of the
maximum possible edges in that neighborhood. The lat-
ter is simply the number of combinations of two distinct
actors one can draw from a subset of k actors, where k
is the degree. Clustering at the level of the network as
a whole is then dened as the average of clustering of
all neighborhoods i. Obviously, clustering is a direct
indicator for social capital, where high clustering indi-
cates high social capital.
A nal concept that will be necessary to set out the
theory of small worlds is that of a shortcut. Shortcuts
are edges that complete triads, or, alternatively, shortcuts
are edges that connect two vertices that would otherwise
be (widely) separated. We will use the parameter to
denote the fraction of edges in a graph that are shortcuts.
Clearly then, shortcuts are exactly the bridges that ll
in structural holes in a network members environment.
The fraction of such shortcuts or bridges in the total
of all connections can thus be seen as a degree of pres-
ence of the structural holes strategy in overall network
3.3. The theory of small worlds
We have already introduced the intuition that high
clustering goes hand in hand with long characteristic
path length, and vice versa. Mathematical work in net-
work theory (e.g., Watts, 2000) tends to support this
intuition. In fact, two specic network typologies have
attracted much attention, each representing one extreme
of the average clusteringcharacteristic path length
The rst typology is the so-called connected caveman
world. Such a world consists of a number of distinct
caves that are connected to each other by a single edge.
A cave is a subset of actors that is fully connected. Each
cave is connected to exactly two other caves by two dis-
tinct members, each of whom are connected to one other
actor in a different cave. In such a network, which is
depicted in Fig. 2, all paths to actors in the same cave
have the minimum length of 1, but paths to actors out-
side the cave are much longer because they depend on
only a few actors that can act as intermediaries. Since
the cave is almost fully connected internally, the degree
of clustering of this neighborhood is close to one. This
high clustering leads to a relatively high value for
characteristic path length (we will present values for
these two variables below).
The connected cavemen world is one that can be inter-
preted as emerging from pure social capital strategies.
The individual caveman obviously has a high amount of
social capital, because (s)he is a member of a densely
populated environment. However, the world is larger
than a single local environment (cave), and the high
resulting characteristic path length implies that the high
social capital comes hand in hand with a large average
distance to other parts of the world. This implies that it
would take a long time for information emerging in one
cave to spread through the world as a whole.
The other extreme network typology of interest is that
of the Moore Graph. This is a network in which every
vertex is adjacent to exactly k other vertices, and none
of these other vertices are connected to each other. The
Moore Graph is a purely theoretical construction, and
cannot even be obtained for many combinations of n and
k. This is why Fig. 2 only displays a local view around
a single vertex for the Moore Graph. It can easily be
veried that in the Moore Graph, clustering is zero, and
one may conclude that social capital strategies played no
role in network formation. On the other hand, there is
always a relatively efcient path to other members of
the network, because every node is in fact operating as
a bridge.
The Moore Graph is of interest here because it puts
a lower limit on characteristic path length. An approxi-
mate expression for characteristic path length of the
Moore Graph with given n and k can be obtained, and
it can be shown (Watts, 2000, Sect. 4.1.2 and the refer-
ences there) that it is impossible to obtain shorter charac-
Fig. 2. The Connected Caveman World (left) and a local view of the
Moore Graph (right).
567 B. Verspagen, G. Duysters / Technovation 24 (2004) 563571
teristic length for any graph with identical n and k pro-
vided that the degree of the individual vertices does not
differ too widely. Hence, the Moore Graph has much
lower characteristic path length than the Connected
Caveman World.
In practice, one can only approximate the perfect
Moore Graph by means of a graph in which the links
are distributed randomly. This will yield values for the
clustering coefcient that are in fact positive, but very
close to zero. By drawing random graphs in which the
probability of shortcuts is varied, it is possible to con-
struct graphs that provide a more or less smooth tran-
sition between the highly clustered and long connected
caveman world and the short and not very highly clus-
tered Moore Graph. This will be shown below.
Summarizing, two main typologies of graphs have
been introduced: one for which the characteristic path
length is high, and hence information takes a long time
to travel from one (average) network member to another;
and one for which the characteristic path length is much
lower, and hence information travels much more rapidly.
The long network is characterized by high clustering,
the short one by low clustering. The extreme case of
high clustering and long path length can be seen as a
representation of the pure social capital perspective on
network formation. The other extreme (short path length
and low clustering) can be seen as an extreme network
structure associated with the structural holes perspective.
Is the negative relationship between clustering and
length that results from comparing these two stereotypes
a general phenomenon?
It can be shown that high clustering and long charac-
teristic path length need not always go together. A spe-
cial class of networks can be identied in which clus-
tering is relatively high, but characteristic path length is
relatively short. This type of networks has been called
small worlds (Watts, 2000). This term has been derived
from the hypothesis that although most people in the
world mainly know other people that belong to a fairly
clustered set of friends, there are ve to six intermedi-
aries necessary to connect the largest part of the popu-
lation of the globe. In other words, even though a person
knows mainly people in her own environment, she will
know indirectly most other people in the world through
only a small number of indirect steps.
Specically, a small world has been dened as a net-
work with n actors and average degree k that displays
characteristic path length approximately equal to a
Moore Graph with the same n and k, but has much larger
clustering than such a graph. The relevance of the idea
of small worlds becomes immediately obvious in the
context of the comparison of the structural holes per-
spective and the social capital perspective. Suppose one
would start from a situation of the connected cavemen
world, i.e., a world in which social capital dominates the
formation of networks completely. Then imagine that in
a controlled experiment, network members are allowed
to ll in structural holes in their personal network, by
engaging in partnerships with rms in other parts of the
network. How would this affect the characteristic path
length of the network, and hence the efciency of knowl-
edge transfer? How many structural holes would have
to be lled before a certain degree of network efciency
would be reached? It turns out the theory of small worlds
provides a very clear-cut answer to these questions.
In order to see how the small world is located in
between the Connected Caveman World and the Moore
Graph, Watts (2000) has proposed a number of formal
models describing the construction of (random) graphs.
The exact construction of these graphs need not concern
the reader here. What is interesting, however, is that
these networks can be tuned by the single parameter
to either side of the Connected Caveman World or the
Moore Graph. Moreover, this parameter has a clear
interpretation in terms of the structural holes perspective:
it measures the tendency for network connections to ll
in structural holes. Fig. 3 illustrates the model.
The horizontal axis of Fig. 3 displays the parameter
, which tunes the graph, and which has an obvious
interpretation as the degree to which strategies aimed at
bridging structural holes play a role in partner seeking.
Fig. 3. Length and clustering as a function of the graph tuning para-
meter .
Instead of using , one may also use a slightly different parameter
denoted . This parameter measures the fraction of all pairs of edges
in the network that are not connected and have one and only one com-
mon neighbor. Such pairs of edges are called contractions, and can be
seen as an alternative conceptualization of the idea of bridges lling
in structural holes. Watts (2000, p. 73) argues that is an analogous
parameter to , although it is more general, as most shortcuts result
in contractions, but not the reverse. In the networks depicted in Fig.
2, there is an exact quadratic relationship between and , which is
568 B. Verspagen, G. Duysters / Technovation 24 (2004) 563571
The gure shows that for a low fraction of shortcuts in
the set of the networks edges, one observes a high value
for characteristic length of the network, and also a high
value for the clustering coefcient of the network. This,
in other words, corresponds to the connected caveman
world network, or, alternatively, to a world in which
social capital completely dominates network formation.
At the other extreme, where the large majority of all
edges are shortcuts and the bridging of structural holes
is very important in strategy, one observes low values
for clustering and characteristic length. This corresponds
to the Moore Graph. Thus, we clearly nd support for
the intuition that a social-capital-dominated strategy
increases path length, while strategies aimed at bridging
structural holes decrease overall network length.
What is interesting, however, is that the path from
high to low values of , i.e., the path from a pure social
capital perspective to a pure structural holes perspective,
is quite different for the two variables in the graph. The
clustering coefcient does not descend very much in the
beginning, but shows rapid decline towards the end.
Characteristic path length shows more or less the
opposite path: it declines rapidly at rst and reaches very
low levels already at relatively low levels of . Small
worlds are dened as networks in the limited range (say
between 0.01 and 0.1) for which characteristic path
length is already at levels comparable with the Moore
Graph (right extreme), but the clustering coefcient is
still relatively high. This world thus combines high
social capital with an efcient ow of information and
Note that the curves in Fig. 3 are drawn for some
specic values of other networks parameters. These are
the size of the network (number of vertices) and the
average number of direct connections that each of them
has (average degree). Also, the underlying model
assumes that the vertices do not differ too widely in their
actual degree (see Watts, 2000, Chaps. 3 and 4 for more
details). When applying the theory to a specic real-
world case, as will be done in the next section, violation
of this assumption may lead to observed values of ,
length and clustering that are off the curves.
Also, the model applied to draw Fig. 3 does not con-
tain any behavioral foundation. It is just a mathematical
construct useful to provide some benchmark against
which we can compare values of , length and clustering
observed in actual practice. The benchmark enables us
to assess how important the two network strategies
(structural holes and social capital) are in a real-world
governed by only one additional parameter. This is the so-called bundle
size, which is roughly equivalent to the number of parties involved in
a single partnership or alliance. However, because the quadratic term
in this relationship adds little for the values of n and k that characterize
the networks of interest in this paper, using or is more or less
equivalent from a theoretical point of view.
case, and whether we nd values of path length and clus-
tering that are in broad accordance with these strategies.
Coming back to our research questions, the model
shows that very dense networks in which social capital
dominates network formation indeed lead to inefcient
networks in terms of the overall speed of transmission
of knowledge ows through the network. Compared to
this extreme case, the application by network members
of a strategy of trying to ll structural holes will gener-
ally decrease characteristic path length, and hence facili-
tate knowledge ows in the network as whole. The
theory also shows how frequently the structural holes
strategy needs to be applied in order to approach the
theoretical lower boundary on characteristic path length.
Specically, the model suggests that the small worlds
range of the parameter is relatively efcient. Forming
shortcuts beyond the upper limit of the small worlds
range does not lead to a noticeable further decrease of
characteristic path length, but it does decrease social
Note that this line of reasoning refers to overall net-
work path length and clustering. In any real-world net-
work there will be some rms with shorter (or longer)
path length. In other words, even in the small worlds
range, there will be an incentive for some rms to invest
further in alliances that ll in structural holes in their
own local environment.
The analysis will now proceed by investigating how
the real-world networks of strategic technology alliances
compare against the benchmark theory.
4. Empirical results
We use the CATI database as our source of infor-
mation on strategic technology alliances. The data covers
alliances from the period 19801996. In constructing our
networks, we dened two rms to be connected when
they are present in at least one joint alliance. A moving
window in time was used to date the networks, so that
the network at time t covers alliances in the period tl
to t, where l+1 is the window length. We experimented
with l=4, l=2 and l=17 (the latter means we include the
whole period for which we have alliances data). The
results are largely invariant to the value for l. The
alliances were also assigned to a technology eld, and
we made calculations for all technology elds together,
as well as two separate elds. In order to have sufcient
observations, we dened the two technology elds for
which separate calculations were done relatively
broadly. The two elds were chemicals and food, and
electricals (including electronics and ICT). When we
consider a network, we concentrate on the portion of the
network that is connected. In all cases, this involves by
far the largest proportion of rms that is present in the
database. The parts of the network that are not connected
569 B. Verspagen, G. Duysters / Technovation 24 (2004) 563571
usually consist of a fairly high number of pairs of rms
that have only one alliance. Details on this are available
from the authors on request.
Table 1 gives an overview of the main results. All
ve cases show a similar pattern: characteristic network
length (L) has the order of magnitude corresponding to
a Moore Graph, while the clustering parameter () has
an order of magnitude that is quite a lot higher than that
of a Moore Graph, although also somewhat lower than
that corresponding to the connected caveman world
(CCW). On the basis of these results, we can classify
the network of strategic alliances that we investigated as
small worlds.
The rst three lines consider the total network, i.e.,
no split-up has been made with regard to technology
eld. Here we see that the general tendency of a low
value for L and a fairly high value for is surprisingly
invariant to the value for l, although obviously the size
of the network depends on this. At the level of the two
separate technology elds, larger differences arise, as is
obvious from the two last lines in the table. Here we see
that the electricals eld is characterized by a relatively
low value for L, as compared to chemicals and food, as
well as the total network, although the value for is
quite similar for the two elds.
Fig. 4ac show how, for the networks in Table 1 with
l=4, the network structure compares to the benchmark
of a random network with varying , as in Fig. 3 above.
The gure gives both the theoretical network structure
(indicated by the lines) for n and K given for the specic
case in Table 1, and the observed values of L and
(dots). Hence the distance between the dots and the lines
is an indication of how much the observed network
structure differs from the theoretical one. Note that we
have no reason to expect that the observed values will
match the theoretical case, since the model used to con-
struct the latter is a random model without behavioral
basis. Thus, the theoretical lines cannot be seen as pre-
dictions, but rather as benchmarks corresponding to ran-
dom partner seeking. The deviations from these bench-
marks are then indications of strategic networking
behavior leading to a specic tendency for higher or
lower L and/or .
For the network for all technology elds, we observe
Table 1
Basic network statistics, networks of strategic alliances from CATI
Actual CCW Moore
n K L L L
Total, l=4, t=1996 0.406 1886 4.08 5.02 0.33 185.2 0.95 5.69 0.002
Total, l=2, t=1996 0.422 1300 3.82 5.34 0.30 134.5 0.94 5.75 0.002
Total, l=17, t=1996 0.330 5504 5.29 4.20 0.34 437.1 0.97 5.45 0.0008
Electricals, l=4, t=1996 0.295 837 4.75 3.75 0.28 72.37 0.97 4.45 0.004
Chemicals and food, l=4, t=1996 0.760 639 2.79 5.45 0.31 83.94 0.88 7.73 0.003
that the value for L is almost on the rightmost at part of
the benchmark line. Hence we conclude that somehow,
network length has been minimized, and the fraction of
shortcuts in the network () is a good predictor for this.
Clustering is somewhat higher than the benchmark
value, indicating a tendency for the building-up of
social capital.
The gure for electricals shows a somewhat similar
situation, although here the observed value for L is a bit
more below the benchmark. The value for clustering (),
on the other hand is somewhat closer to the benchmark,
although still above. This again indicates a tendency for
strong social capital. For chemicals and food, the devi-
ation from the benchmark is most extreme. This is the
eld where the value for (shortcuts) is very high. The
benchmark would predict low clustering (low social
capital) on the basis of this, but the contrary is the case.
In this technology eld, clustering is almost as high as
the maximum value in the benchmark (for low ).
How can this be interpreted? The relatively high
values for clustering () can be taken as an indication of
the importance of social capital. On the other hand, the
relatively high values of and (partly as a result) low
value of L is an indication of strategic partner seeking
in line with a structural holes perspective on alliance
formation. In other words, the networks under consider-
ation seem to combine features of both theoretical per-
spectives that we have outlined above. The result of this
combination of perspectives is a network structure simi-
lar to a small world, in which information ow is rela-
tively efcient (low L), but social capital () still high.
How exactly these two strategies are combined in
terms of alliance formation is an issue that is open to
future research, which could possibly lead to a more
adequate (mathematical) model of network development.
Our results indicate that in such a theory, there would
certainly have to be a role for factors that are specic
to technology elds or industrial sectors. This is clear
from the large differences in networks in the two tech-
nology elds in Table 1, as well as from the observed
values for in these two elds. The electricals eld dis-
plays a value for that is broadly comparable to the
total network. The chemicals and food eld, on the other
570 B. Verspagen, G. Duysters / Technovation 24 (2004) 563571
Fig. 4. Network benchmark for the total network (l=4, t=1996), b:
Network benchmark for the network in electricals (l=4, t=1996). c:
Network benchmark for the network in chemicals and food (l=4,
hand, displays a much higher value for in combination
with a low degree of centrality.
Thus, the chemicals and food technology eld is one
in which a rather peculiar networking strategy is found.
Strategic partnering () is rather strong, and although,
according to our benchmark, this induces a smaller role
for social capital (low ), we nd exactly the opposite:
a high value for social capital (). As a preliminary
explanation of this nding, one may point to the nature
of the knowledge base underlying this technology eld.
Especially in pharmaceuticals and biotechnology, the
close relations to science have induced an industry struc-
ture of many small rms, which are highly specialized
in research. Each of these rms covers a rather small
eld in a very in-depth way. Large companies tend to
focus their research on specic major diseases and have
adopted a strategy of tapping into this knowledge base
by means of many partnerships with these small special-
ized rms. Further research specically on the compari-
son between partnering strategies in different industries
may obviously enlighten this issue more.
5. Conclusions
This paper has argued that the recent theory of small
worlds provides a natural way of uniting two perspec-
tives found in the strategic management literature on net-
works and strategic alliances. One perspective, that of
the theory of social capital, leads to a view in which
networks consist of densely connected local environ-
ments, corresponding to high values of social capital.
Firms are considered to seek such dense local environ-
ments in building their strategic alliances. The theory of
small worlds shows that in the extreme case of very
dense local environments, the average path length for
knowledge ows between two actors in the network
becomes very large. This can be considered as an
impediment to the efcient spread of knowledge and
information within the network, and thus an undesirable
property both from the point of view of the networking
rm, and the policy point of view.
The other perspective found in the strategic manage-
ment literature is the one of structural holes. This theory
argues that rms will seek partnerships that form bridges
between their own part of the network (local environ-
ment which may be densely connected) and other inter-
esting parts of the network. Rather then extending their
network based on an argument of enhancing social capi-
tal (i.e., dense local relations), rms are expected to pick
and choose partnerships based on the strategic position
of potential partners in the network. From a structural
holes perspective one might argue that rms should be
engaged in a strategy of bridging structural holes. This
will lead to short average path length in the network,
and hence to efcient information ows.
571 B. Verspagen, G. Duysters / Technovation 24 (2004) 563571
The theory of small worlds provides a useful way of
representing the tendency for this bridging behavior in
terms of a single parameter. This parameter measures
the number of shortcuts (links that complete a triad)
as a fraction of all links in the network. The theory then
shows how the characteristics of the network in terms
of clustering (dened as density of local environments)
and average path length of the network develop as a
function of this tendency. For very dense local environ-
ments (i.e., a world dominated by network ties resulting
from a social capital view of the world), average path
length is very long, while for extreme levels of bridg-
ing (a structural holes world) path length approaches
a lower limit. In between these two extreme cases one
nds a small range of the bridging parameter for which
clustering is still relatively high, but average path length
is already close to the lower limit implied by the extreme
structural holes world. This is the range of small
worlds as it has been dened by Watts (2000). It is
argued here that these small worlds can be characterized
as relatively efcient networks in terms of knowledge
transfer. In contrast to a pure structural holes or social
capital perspective, the small-worlds perspective is able
to assess the optimal conguration of ties within cohes-
ive sub-groups and bridging ties.
The theory of small worlds thus enables us to compare
the efciency of empirically observed networks to a
theoretical optimal conguration of these networks. In
such an empirical application, the analysis has shown
that networks of strategic technology alliances between
large multinational rms indeed show small-worlds
properties. It can thus be concluded that these tech-
nology networks are relatively efcient means of knowl-
edge transfer. When splitting our total network between
two distinct technology elds (chemicals and food, and
electricals, the latter including electronics and ICT), it
was found that especially the chemicals and food eld
shows extreme properties of networking structure.
Here the fraction of shortcuts (i.e., bridge building) is
very high, while this still does not lead to a signicant
decline in social capital building.
As a suggestion for further work, we propose more
research into the relative role of social capital and bridge
building in network-partner seeking. The results summa-
rized above suggest that especially the chemicals and
food sector would be an important eld for this research.
Bourdieu, P., 1980. Le Capital Sociale: Notes Provisaires. Actes de la
Recherche en Sciences Sociale 2-3.
Bourdieu, P., Wacquant, L., 1992. An Invitation to Reexive Soci-
ology. University of Chicago Press, Cicago.
Burt, R.S., 1992. Structural Holes. The Social Structure of Compe-
tition. Harvard University Press, Cambridge MA.
Coleman, J., 1990. Social Capital in the Creation of Human Capital.
American Journal of Sociology 94, S95S120.
Cowan, R. and Jonard, N. 2000 The Dynamics of Collective Invention,
MERIT Resarch Memorandum # 00-018.
Duysters, G., Hagedoorn, J., 2002. External Appropriation of Innov-
ative Capabilities: The Choice Between Strategic Partnering and
Mergers and Acquisitions. Journal of Management Studies 39 (2),
Granovetter, M., 1992. Economic action and social structure: the prob-
lem of embeddedness. In: Granovetter, M., Swedberg, R. (Eds.),
The sociology of economic life. Westview Press, Boulder, CO, pp.
Gulati, R., 1998. Alliances and Networks. Strategic Management Jour-
nal 19, 293317.
Hagedoorn, J., 1996. Trends and patterns in strategic technology part-
nering since the early seventies. Review of Industrial Organization
11, 601616.
McEvily, Zaheer, 1999. A Source of Firm Heterogeneity in Competi-
tive Capabilities. Strategic Management Journal.
Nooteboom, B., 1999. Inter-rm Alliances. Routledge, London.
Vonortas, 1997. Cooperation in Research and Development. Kluwer
Academic Publishers, Boston.
Walker, G., Kogut, B., Shan, W., 1997. Social Capital, Structural
Holes and the Formation of an Industry Network. Organization
Science 8, 109125.
Watts, D.J., 2000. Small Worlds. The Dynamics of Networks Between
Order and Randomness. Princeton University Press, Princeton.
Geert Duysters (1966) received his PhD from the Maastricht Economic
Research Institute on Innovation and Technology (MERIT) at Maastricht
University. He is currently full professor of Organization Science at the
Department of Technology Management of the Eindhoven University of
Technology. He is also acting as the Scientic Director of the Eindhoven
Centre for Innovation studies (ECIS). His main areas of interest are in
innovation strategies, mergers and acquisitions, network analytical
methods and strategic alliances.
Bart Verspagen (1966) received his PhD from the Maastricht Economic
Research Institute on Innovation and Technology (MERIT) at Maastricht
University. He is currently full professor of the Economics of Innovation
and New Technology at the Department of Technology Management of
the Eindhoven University of Technology, as well as Visiting Professor at
the Centre for Technology, Innovation and Culture (TIK) at the University
of Oslo, Norway. His area of interest includes evolutionary economic
theory, economic growth, dynamics of science and technology.
p a r t ii
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 179
Introduction to Part II
A central Wnding in innovation research is that Wrms seldom
innovate in isolation. Interaction with customers, suppliers, com-
petitors and various other private and public organizations is very
important, and a system perspective is useful in understanding
and analyzing such interaction. In Chapter 7, Edquist traces the
development of such an innovation system perspective, particu-
larly with respect to nations (so-called national systems of innov-
ation), and discusses the achievements, shortcomings and
potential of this approach. He also considers the role of some
central activities of such systems, e.g., R&D and education.
Following this, Mowery and Sampat (Chapter 8) examine one of
the central organizations in national systems of innovations: uni-
versities. Chapter 9 by OSullivan focuses on another aspect of
innovation systems, Wnance, discussing the varied approaches
adopted on the relationship between Wnance and innovation. An
essential feature in innovation systems is the conditions for appro-
priation of the economic returns to innovation. Granstrand
(Chapter 10) provides a historical overviewof intellectual property
rights and surveys the extensive literature on this topic. The next
two chapters deal with the boundaries of systems of innovation. As
Edquist notes in Chapter 7 these boundaries do not have to be
national, but may be regional, global or sectoral (sectoral systems
are discussed in more detail in Part III of this volume). Regional
systems of innovation are analyzed by Asheim and Gertler (Chap-
ter 11) and Narula and Zanfei in Chapter 12 examine the globaliza-
tion of innovation and the role of multinational enterprises in this
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 180
c h a p t e r 7
charles edqui st
7. 1 Introduction1
This chapter presents an overview and assessment of the systems of innovation
approach. I focus mainly on national systems of innovation, but in addition address
sectoral and regional systems of innovation to a limited extent.2 The chapter
addresses the emergence and development of the systems of innovation (SI)
approach, its strengths and weaknesses, the criticism that it is undertheorized,
the constituents of SIs, the main function and activities in SIs, the boundaries of SIs,
and proposals for further research. I also discuss howthe rigour andspeciWcity of the
SI approach could be increased.3 The most central terms used in this chapter are
speciWed in Box 7.1.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 181
7. 2 The Emergence and Development
of the SI Approach
The chapter by Fagerberg in this volume highlights the systemic nature of innov-
ation processes, noting that Wrms do not normally innovate in isolation, but in
collaboration and interdependence with other organizations. These organizations
may be other Wrms (suppliers, customers, competitors, etc.) or non-Wrm entities
such as universities, schools, and government ministries. The behavior of organiza-
tions is also shaped by institutionssuch as laws, rules, norms, and routinesthat
constitute incentives and obstacles for innovation. These organizations and insti-
tutions are components of systems for the creation and commercialization of
knowledge. Innovations emerge in such systems of innovation.
The innovation concept used in this chapter is wide and include product innov-
ations as well as process innovations. Product innovations are newor better
material goods as well as new intangible services. Process innovations are new ways
of producing goods and services. They may be technological or organizational
(Edquist, Hommen, and McKelvey 2001).
Box 7.1 Systems of innovation: main terms used
Innovations product innovations as well as process innovations. Product innovations
are newor bettermaterial goods as well as new intangible services. Process innov-
ations are new ways of producing goods and services. They may be technological or
SI system of innovation the determinants of innovation processes all important
economic, social, political, organizational, institutional, and other factors that inu-
ence the development, diffusion, and use of innovations.
Constituents of SIs components relations among the components.
Main components in SIs organizations and institutions.
Organizations formal structures that are consciously created and have an explicit
purpose. They are players or actors.
Institutions sets of common habits, norms, routines, established practices, rules, or
laws that regulate the relations and interactions between individuals, groups, and
organizations. They are the rules of the game.
An SI has a function, i.e. it is performing or achieving something. The main function in
SIs is to pursue innovation processes, i.e. to develop, diffuse and use innovations.
Activities in SIs are those factors that inuence the development, diffusion, and use of
innovations. The activities in SIs are the same as the determinants of the main
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 182
182 charles edquist
The expression national system of innovation (NSI) was, in published form,
Wrst used in Freeman (1987). He deWned it as the network of institutions in the
public and private sectors whose activities and interactions initiate, import, and
diVuse new technologies (Freeman 1987: 1). Two major books on national systems
of innovation (NSI) are Lundvall (1992) and Nelson (1993), which employ diVerent
approaches to the study of NSIs. Nelson (1993) emphasizes empirical case studies
more heavily than theory development4 and some of the studies focus narrowly on
nations R&D systems. By contrast, Lundvall (1992) is more theoretically oriented
and seeks to develop an alternative to the neo-classical economics tradition by
placing interactive learning, userproducer interaction and innovation at the center
of the analysis (Lundvall 1992: 1).
Lundvall argues that the structure of production and the institutional set-up
are the two most important dimensions that jointly deWne a systemof innovation
(Lundvall 1992: 10). In a similar way, Nelson and Rosenberg single out organizations
supporting R&D, i.e. they emphasize those organizations that promote the creation
and dissemination of knowledge as the main sources of innovation (Nelson and
Rosenberg 1993: 5, 913).5 Lundvalls broader approach recognizes that these
narrow organizations are embedded in a much wider socio-economic system
in which political and cultural inXuences as well as economic policies help to
determine the scale, direction and relative success of all innovative activities
(Freeman 2002: 195).
Both Nelson and Lundvall deWne national systems of innovation in terms of
determinants of, or factors inXuencing, innovation processes.6 However, they single
out diVerent determinants in their actual deWnitions of the concept, presumably
reXecting what they believe to be the most important determinants of innovation.
Hence, they propose diVerent deWnitions of the concept, but use the same term. This
reXects the lack of a generally accepted deWnition of a national systemof innovation.
A more general deWnition of (national) systems of innovation includes all
important economic, social, political, organizational, institutional and other factors
that inXuence the development, diVusion and use of innovations (Edquist 1997b:
14). If all factors that inXuence innovation processes are not included in a deWnition,
one has to argue which potential factors should be excludedand why. This is quite
diYcult, since, at the present state of the art, we do not know the determinants of
innovation systematically and in detail. It seems dangerous to exclude some poten-
tial determinants, since these might prove to be very important, once the state of the
art has advanced. For example, twenty-Wve years ago it would have been natural to
exclude the interactions between organizations as a determinant of innovation
processes. Included in this general deWnition are the relationships among the factors
listed and the actions of both Wrms and governments.
There are other speciWcations of systems of innovation than national ones.
Carlsson and colleagues focus on technological systems, arguing that these are
unique to technology Welds (Carlsson 1995). The sectoral approach of Breschi and
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 183
systems of innovation 183
Malerba (1997) similarly focuses on a group of Wrms that develop and manufacture
the products of a speciWc sector and that generate and utilize the technologies of
that sector. The concept of regional innovation system has been developed and
used by Cooke et al. (1997) and Braczyk et al. (1998), Cooke (2001), and Asheim and
Isaksen (2002).
The three perspectivesnational, sectoral and regionalmay be clustered as
variants of a single generic systems of innovation approach (Edquist 1997b: 3,
1112). Much of the discussion in this chapter is relevant for the generic approach,
and is based on the premise that the diVerent variants of systems of innovation
coexist and complement each other. Whether the most appropriate conception of
the system of innovation, in a certain context, should be national, sectoral or
regional, depends to a large extent on the questions one wants to ask.7
7. 3 Strengths and Weaknesses
of the SI Approach
7.3.1 The DiVusion of the SI Approach
The diVusion of the SI approach has been surprisingly rapid, and is nowwidely used
in academic circles. The approach also Wnds broad applications in policy contexts
by regional authorities and national governments, as well as by international organ-
izations such as the OECD, the European Union, UNCTADandUNIDO. InSweden,
a public agency has even been named after the approach, i.e. the Swedish Agency for
Innovation Systems (VINNOVA). The practice of VINNOVA is strongly inXuenced
by the SI approach, an approach that appears to be especially attractive to policy
makers who seek to understand diVerences among economies innovative perform-
ance, and develop ways to support technological and other kinds of innovation. The
next section brieXy addresses some of the strengths of the generic SI approach.
7.3.2 The Strengths of the SI Approach
The SI approach places innovation and learning processes at the center of focus. This
emphasis on learning acknowledges that innovation is a matter of producing new
knowledge or combining existing (and sometimes new) elements of knowledge in
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 184
184 charles edquist
new ways. This focus distinguishes the SI approach from other approaches that
regard technological change and other innovations as exogenous.
The SI approach adopts a holistic and interdisciplinary perspective. It is holistic in
the sense that it tries to encompass a wide arrayor allof the important determin-
ants of innovation, and allows for the inclusion of organizational, social, and
political factors, as well as economic ones. It is interdisciplinary in the sense that
it absorbs perspectives from diVerent (social science) disciplines, including eco-
nomic history, economics, sociology, regional studies, and other Welds.
The SI approach employs historical and evolutionary perspectives, which makes the
notion of optimality irrelevant. Processes of innovation develop over time and
involve the inXuence of many factors and feedback processes, and they can be
characterized as evolutionary. Therefore, an optimal or ideal system of innovation
cannot be speciWed. Comparisons can be made between diVerent real systems (over
time and space), and between real systems and target systems, but not between real
systems and optimal ones. Although this is a complex view of the innovation
process, it is far richer and more realistic than its alternatives.
The SI approach emphasizes interdependence and non-linearity. This is based on the
understanding that Wrms normally do not innovate in isolation but interact with
other organizations through complex relations that are often characterized by
reciprocity and feedback mechanisms in several loops. Innovation processes are
not only inXuenced by the components of the systems, but also by the relations
between them. This captures the non-linear features of innovation processes and is
one of the most important characteristics of the SI approach.
The SI approach can encompass both product and process innovations, as well as
subcategories of these types of innovation. Traditionally, innovation studies have, to
a large extent, focused upon technological process innovations and to some extent
upon goods product innovations, but less on non-technological and intangible
ones, i.e. service product innovations and organizational process innovations (as
speciWed in Section 7.2). As argued in this Handbook, there are good reasons to use a
comprehensive innovation concept,8 and the systems of innovation approach is well
suited to this comprehensive perspective, since all the categories of innovations
speciWed in this chapter can be analyzed within it. That non-technological forms of
innovation deserve more attention is also argued in OECD (2002a: 24.d).
The SI approach emphasizes the role of institutions. Practically all speciWcations of the
SI concept highlight the role of institutions, rather than assuming them away from
the list of determinants of innovation. This is important since institutions strongly
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 185
systems of innovation 185
inXuence innovation processes. There is, however, no agreement about what the
term institutions means (see Section 7.3.3).
These six characteristics are often considered to be strengths of the SI approach by
academic analysts, policy makers, andincreasinglyby Wrmstrategists, andpartly
explain its rapid diVusion. However, the SI approach also has weaknesses, which
represent challenges for future research on systems of innovation.
7.3.3 The Weaknesses of the SI Approach
The SI approach is still associated with conceptual diVuseness. One example is the
term institution, which is used in diVerent senses by diVerent authors: it is
sometimes used to refer to organizational actors as well as to institutional rules.
Sometimes the word means diVerent kinds of organizations or players (according
to the deWnitions in Section 7.4.2). At other times, the term means laws, rules,
routines, and other rules of the game. For Nelson and Rosenberg (1993), insti-
tutions are basically diVerent kinds of organizations, while for Lundvall (1992) the
terminstitution means primarily the rules of the game. Hence institution is used
in several diVerent senses in the literature (Lundvall 1992: 10; Nelson and Rosenberg
1993: 5, 913; Edquist 1997b: 268).
Another example of conceptual diVuseness is that the originators of the SI
approach did not indicate what exactly should be included in a (national) system
of innovation; they did not specify the boundaries of the systems (Edquist 1997b:
1315). Nelson and Rosenberg provided no sharp guide to just what should be
included in the innovation system, and what can be left out (Nelson and Rosenberg
1993: 56). Lundvall insisted that a deWnition of the system of innovation must be
kept open and Xexible (Lundvall 1992: 13).
With regard to the status of the SI approach, it is certainly not a formal theory, in
the sense of providing speciWc propositions regarding causal relations among
variables. It can be used to formulate conjectures for empirical testing, but this has
beendone only toa limiteddegree (see Section7.5). Because of the relative absence of
well-established empirical regularities, systems of innovations should be labeled
an approach or a conceptual framework rather than a theory (Edquist 1997b: 289).
Scholars disagree onthe seriousness of these weaknesses of the SI approach andon
howthey should be addressed. According to some, the approachshould not be made
too rigorous; the concept should not be overtheorized and it should remain an
inductive one.9 Another position argues that the SI approach is undertheorized,
that conceptual clarity should be increased and that the approach should be made
more theory-like.10
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 186
186 charles edquist
Hence, the international community within innovation studies is currently
divided on this issue. In what follows, I try to increase the rigor and speciWcity of
the SI approach. This eVort is intended as a step towards developing the approach
further. If it reveals weaknesses associated with the approach, this is a good thing.
Acknowledging such weaknesses may lead to additional research and new insights
into the operation of innovation systems.
7. 4 The Constituents, Function,
Activities and Boundaries of SIs
7.4.1 What is a System?
In an eVort to develop the SI approach, it might be useful to relate it explicitly to
general systems theory, which has been used much more in the natural sciences
than in the social sciences. In everyday language, as well as in large parts of the
scientiWc literature, the termsystem is used generously and with limited demands
for a precise deWnition. To the question What is a system? there is, however,
a common answer in everyday language as well in scientiWc contexts (Ingelstam
2002: 19):
. A system consists of two kinds of constituents: There are, Wrst, some kinds of
components and, second, relations among them. The components and relations
should forma coherent whole (which has properties diVerent fromthe properties
of the constituents).
. The system has a function, i.e. it is performing or achieving something.
It must be possible to discriminate between the system and the rest of the world;
i.e. it must be possible to identify the boundaries of the system. If we, for example,
want to make empirical studies of speciWc systems, we must, of course, knowtheir
Making the systems of innovations approach more theory-like does not require
that all components and all relations among them must be speciWed. Such an
ambitionwouldcertainly be unrealistic. At present, it is not a matter of transforming
the SI approach into a general theory of innovation, but rather we need to make it
clearer and more consistent so it can better serve as a basis for generating hypotheses
about relations between speciWc variables within SIs (which might be rejected or
supported through empirical work). Even the much more modest objective of
specifying the main function of SIs, the activities and components in them and
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 187
systems of innovation 187
some important relations among these, would represent a considerable advance in
the Weld of innovation studies. Used in this way, the SI approach can be useful for the
creation of theories about relations between speciWc variables within the approach.
7.4.2 The Main Components of SIs
Organizations and institutions are often considered to be the main components of
SIs, although it is not always clear what is meant by these terms (as argued in Section
7.3.3). Let me therefore specify what organizations and institutions mean here.
Organizations are formal structures that are consciously created and have an explicit
purpose (Edquist and Johnson 1997: 467). They are players or actors.12 Some
important organizations in SIs are Wrms, universities, venture capital organizations
and public agencies responsible for innovation policy, competition policy or drug
Institutions are sets of common habits, norms, routines, established practices, rules, or
laws that regulate the relations and interactions between individuals, groups, and
organizations (Edquist and Johnson 1997: 46). They are the rules of the game.
Examples of important institutions in SIs are patent laws, as well as rules and
norms inXuencing the relations between universities and Wrms. Obviously, these
deWnitions are of a Northian character (North 1990: 5), discriminating between
the rules of the game and the players in the game.
SIs may diVer from one another in many respects. For example, the set-ups of
organizations and institutions, constituting components of empirically existing SIs,
vary among them. Research institutes and company-based research departments
may be important R&D performers in one country (e.g., Japan) while research
universities may play a similar role in another (e.g., the United States). In some
countries, such as Sweden, most research is carried out in universities, while the
independent public research institutes are weak. In Germany, the latter are much
more important. That the organizational set-up varies considerably among NSIs is
shown in proWles of the national systems in Austria, Belgium, Finland, Germany,
Spain, Sweden, Switzerland, and the United Kingdom, presented in OECD (1999a:
Annex 3).
Institutions such as laws, rules, and norms also diVer considerably among na-
tional SIs. For example, the patent laws are diVerent between countries. In the USA,
an inventor can publish before patenting, whereas this is not possible according to
European laws. With regard to the patent rights of university teachers, individuals in
this category own their patents outright in Sweden, thanks to the so-called univer-
sity teachers privilege. However, this is not the case inthe USA, where diVerent laws
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 188
188 charles edquist
apply. In Denmark and Germany, new laws have recently transferred ownership
from the teachers to the universities, while in Italy, a transfer has occurred in the
opposite direction. Many OECD governments are currently experimenting with
changes inthe ownershipof knowledge createdinuniversities, inthe belief (basedon
little evidencesee the chapter by Mowery and Sampat) that such changes will
inXuence the propensity to patent and accelerate the commercialization of econom-
ically useful knowledge.
In summary, there seems to be general agreement that the main components in SIs
are organizationsamong which Wrms are often considered to be the most import-
ant onesand institutions. However, the speciWc set-ups of organizations and
institutions vary among systems.
7.4.3 Functions and Activities in SIs
Although a systemis normally considered to have a function, this was not addressed
in a systematic manner in the early work on SIs. Somewhat later, some hints in this
direction were made by Galli and Teubal (1997: 3467). As we will see below, some
recent contributions to the literature have started to address this theoretical gap. Functions and Activities in SIs and Determinants
of Innovation Processes
Xielin LiuandStevenWhite (2001) address what they call a fundamental weakness of
national innovation system research, namely the lack of system-level explanatory
factors (Liu and White 2001: 1092). To remedy this, they focus upon the activities
in the systems, activities being related to the creation, diVusion and exploitation
of technological innovation within a system (Liu and White 2001: 1093). On this
basis, they compile a list of Wve fundamental activities in innovation systems.13
Johnson and Jacobsson (2003) emphasize that, for an innovation system to
support the growth of an industry, a number of functions have to be served within
it, e.g. the supply of resources (JohnsonandJacobsson2003: 2). They suggest that a
technology or product speciWc innovation system may be described and analysed in
terms of its functional pattern, i.e. in terms of how these functions are served
(Johnson and Jacobsson 2003: 3). These authors present a list of Wve functions.14
Rickne (2000: 175) provides a list of eleven functions that are important for new
technology-based Wrms (i.e. not for innovations in an immediate sense).15 Clearly,
thereis noconsensus as towhichfunctions or activities shouldbeincludedinasystem
of innovation and this provides abundant opportunities for further research.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 189
systems of innovation 189
One way of addressing what happens in SIs is the following. At a general level,
the main functionor the overall functionin SIs is to pursue innovation
processes, i.e. to develop, diVuse and use innovations. What I, from now on, call
activities in SIs are those factors that inXuence the development, diVusion, and use
of innovation.16 Examples of activities are R&D as a means of the development of
economically relevant knowledge that can provide a basis for innovations, or the
Wnancing of the commercialization of such knowledge, i.e. its transformation into
A satisfactory explanation of innovation processes almost certainly will be multi-
causal, andtherefore shouldspecify the relative importance of various determinants.
These determinants cannot be expected to be independent of each other, but instead
must be seen to support and reinforceor oVsetone another. Hence, it is import-
ant to also study the relations among various determinants of innovation processes.
One way to try to do that would be to establish a hierarchy of causes a` la E. H. Carr.
Carr argues that the study of history is a study of causes and that the historian
continuously asks the question, Why? Further, the historian commonly assigns
several causes to the same event (Carr 1986: 81, 83). He continues: The true
historian, confronted with this list of causes of his own compiling, would feel a
professional compulsion to reduce it to order, to establish some hierarchy of causes
which would Wx their relation to one another, perhaps to decide which cause, or
which category of causes, should be regarded in the last resort or in the Wnal
analysis. . . as the ultimate cause, the cause of all causes (Carr 1986: 84). I do not
believe that we will ever reach such an objective in a detailed and systematic manner
or that we will be able to identify all determinants of innovationbecause of the
complexity of the task. However, there are good reasons to try to strive in this
direction by developing theories about relations between speciWc variables within
the approach in a pragmatic way (as proposed in Section 7.4.1.).
I believe that it is important to study the activities (causes, determinants) in SIs in
a systematic manner. The hypothetical list of activities presented below is based
upon the literature, e.g. the lists mentioned earlier, and on my own knowledge about
innovation processes and their determinants. The activities listed are not ranked in
order of importance, but start with knowledge inputs to the innovation process,
continues with the demand side factors, the provision of constituent of SIs, and ends
with support services for innovating Wrms.
The following activities can be expected to be important in most SIs:
(1) Provision of Research and Development (R&D), creating new knowledge,
primarily in engineering, medicine, and the natural sciences.
(2) Competence building (provision of education and training, creation of
human capital, production and reproduction of skills, individual learning)
in the labor force to be used in innovation and R&D activities.
(3) Formation of new product markets.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 190
190 charles edquist
(4) Articulation of quality requirements emanating from the demand side with
regard to new products.
(5) Creating and changing organizations needed for the development of new
Welds of innovation, e.g. enhancing entrepreneurshipto create newWrms and
intrapreneurship to diversify existing Wrms, creating new research organiza-
tions, policy agencies, etc.
(6) Networking through markets and other mechanisms, including interactive
learning between diVerent organizations (potentially) involved in the in-
novation processes. This implies integrating new knowledge elements de-
veloped indiVerent spheres of the SI andcoming fromoutside withelements
already available in the innovating Wrms.
(7) Creating and changing institutionse.g. IPR laws, tax laws, environment
and safety regulations, R&Dinvestment routines, etcthat inXuence innov-
ating organizations and innovation processes by providing incentives or
obstacles to innovation.
(8) Incubating activities, e.g. providing access to facilities, administrative sup-
port, etc. for new innovative eVorts.
(9) Financing of innovation processes and other activities that can facilitate
commercialization of knowledge and its adoption.
(10) Provision of consultancy services of relevance for innovation processes, e.g.
technology transfer, commercial information, and legal advice.
This list is provisional and will be subject to revision as our knowledge about
determinants of innovation processes increases. In addition to a set of activities that
is likely to be important in most SIs, there are activities that are very important in
some kinds of SIs andless important inothers. For example, the creationof technical
standards is critically important in some (sectoral) systems, such as mobile telecom-
The systematic approach to SIs suggested here does not imply that they are or can
be consciously designed or planned. On the contrary, just as innovation processes
are evolutionary, SIs evolve over time in a largely unplanned manner. Even if we
knew all the determinants of innovation processes in detail (which we certainly do
not now, and perhaps never will), we would not be able to control them and design
or build SIs on the basis of this knowledge. Centralized control over SIs is
impossible and innovation policy can only inXuence the spontaneous development
of SIs to a limited extent. Three Kinds of Learning in the SI Approach
Regarding competence building as an important activity in SIsand given that
R&D has earlier been a central activity in SI studiesmeans that the SI approach
focuses on three kinds of learning:
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 191
systems of innovation 191
. Innovation (in new products as well as processes) takes place mainly in Wrms and
leads to the creation of structural capital, which is a knowledge-related asset
controlled by Wrms (as opposed to human capital); it is a matter of organiza-
tional learning.
. Research and Development (R&D) is carried out in universities and public research
organizations as well as inWrms andleads topublicly available knowledge as well as
knowledge owned by Wrms and other organizations, as well as by individuals.
Competence Building (e.g. training and education) which occurs in schools and
universities (schooling, education) as well as in Wrms, and leads to the creation of
human capital. Since human capital is controlled by individuals, it is a matter of
individual learning.
An important area for further research in the SI tradition concerns the relationship
among these three kinds of learning, whichappear to be closely relatedtoone another.
One objective of such studies would be to address what types and levels of education
and training are most important for speciWc kinds of innovationse.g. process
innovations and product innovations, or incremental and radical innovations.
As exempliWcation, I will now discuss two central learning activitiesR&D and
competence buildingin some greater detail. This discussion constitutes the be-
ginning of an examination of the relations among activities and constituents in SIs
that is continued in section 7.4.4. It also serves as a guide to and synthesis of some of
the recent work on these issues within the systems of innovation literature.
A. Research and Development. Considerable work on NSIs has been carried out
within the OECD. However, although most of the OECD contributions mentioned
here have systems of innovation in the title, many of them actually use this
approach more as a label than as an analytical tool. The Wrst phase of this work
included the development of quantitative indicators, country case studies, and work
within six focus groups on innovative Wrms, innovative Wrm networks, clusters,
mobility of human resources, organizational mapping, and catching-up economies.
Some of the empirical results were presented in (OECD 1998a) and a synthesis is
found in (OECD 1999a).
The second phase provided a deepening of the analysis in three areas: innovative
clusters (OECD 2001a and 1999b), cooperation in national innovation systems
(OECD2001b), and mobility of skilled personnel inNSIs (OECD2001c). Yet another
study (OECD2002a) summarizes the Wndings of the second phase of the project and
derives policy implications. In the studies mentioned, R&D as well as competence
building is addressed to some extent.
In most countries, universities are the most important public organizations
performing R&D(OECD1998a: ch. 3). Governments fund university R&Dactivities
in a number of ways. Traditionally, they have provided general support via block
grants fromthe Ministries of Education, part of whichwas used by university staVto
carry out R&D. Such funding is still very important in small, highly R&D-intensive
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 192
192 charles edquist
countries such as the Netherlands, Sweden, and Switzerland. Governments may also
provide grants to encourage research for the advancement of knowledge or grants
to obtain the knowledge needed for government missions such as defense or health
care. In most countries, block grants have declined and direct support has grown in
importance (OECD 1998a: ch. 3).
In certain countries, universities fall under the responsibility of the national
government. In others, such as Germany, they are the responsibility of the regional
governments. Whatever the form of organization, a growing regional inXuence can
be observed in most countries. In Germany, the universities are Wnancially very
autonomous. In the UK, Wnancial support is provided by research councils to
individual projects selected on a competitive basis.
In many countries, the science system also includes public research institutes (or
national laboratories) which carry out the same type of R&D activities as univer-
sities, as well as more applied research and technical development work. Although
the relative importance of universities in terms of performing R&Dhas increased in
most countries (see Mowery and Sampat in this volume), public research organiza-
tions remain important. These organizations may be linked to the universities and
included in the higher education sector, or they may be independent of them. The
largest single case in the OECD area is the Centre National de la Recherche Scien-
tiWque (CNRS) in France, which receives the lions share of direct funding of R&Din
the higher education sector. The CNRS provides support for projects that are
normally carried out in collaboration with university researchers. In this regard,
the CNRS can be clearly distinguished from its counterparts in Germany (Max
Planck Gesellschaft), Italy (CNR) and Spain (CSIC) (Laredo and Mustar 2001c: 502).
In the United States, the higher education sector contains a large number of public
research laboratories. (OECD 1998a: 834). Other countries with a large institute
sector include Norway, Taiwan and Germany (e.g. Max Planck Gesellshaft and
Fraunhofer). A number of national governments have tried to change these organ-
izations and promote their links with the rest of the economy and society. This has,
for example, been done in quite diVerent ways in France and the UK (Laredo and
Mustar 2001c: 503).
As this short discussion suggests, diVerent kinds of public organizations (such as
universities and public research institutes) can perform the same activity (R&D) in
an NSI. NSIs diVer signiWcantly with regard to which organisations that perform
public R&D and with regard to the institutional rules that govern or inXuence these
organisations (Laredo and Mustar 2001b: 67).
In most NSIs, especially in low- and medium-income nations, only modest sums
are invested in R&Dand most of the R&Dis performed by public organizations. The
few countries that invest heavily in R&D are all rich, and much of their R&D is
carried out by private organizations. This group includes some large countries, such
as the USA and Japan, but also some small and medium-sized countries, such as
Sweden, Switzerland, and South Korea. There are also some rich countries that do
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 193
systems of innovation 193
rather little R&D, e.g. Denmark and Norway. As mentioned, a considerable part of
the R&D in many rich countries is carried out and Wnanced by the private sector,
primarily Wrms (although there are also public Wnancial support schemes to stimu-
late Wrms to perform R&D). The proportion of all R&D performed in high-income
OECDmember states that is Wnancedby Wrms ranged between21 per cent (Portugal)
and 72 per cent (Japan) in 1999 (OECD2002b). Acknowledging such diVerences may
help to distinguish between diVerent types of NSIs.
Most of the R&D carried out by private organizations may be characterized as
development work rather thanresearch. Innovationcertainly does not depend solely
on R&D results, but requires also other actions, such as technical experimentation,
technology adoption, market investigations, and entrepreneurial initiative. R&D
and innovation activities are normally driven by diVerent rationales and motiv-
ationsi.e. the advancement of knowledge and the quest for proWts, respectively.
One implication of the complex interface between research and innovation is
that links between universities/public research organizations and innovating Wrms
are especially important to the performance of NSIs.18 Innovating Wrms often need
to collaborate with public research organizations or universities. Here, publicly
created institutions are important. Governments may, for example, support collab-
orative centers and programs, remove barriers to cooperation, and facilitate the
mobility of skilled personnel among diVerent kinds of organizations. This might
involve the creation of institutional rules, such as those in Sweden stating that
university professors should perform a third task in addition to teaching and
doing researchi.e. interact with the society surrounding the university, including
Wrms. However, such linkage activity is carried out in diVerent ways and to
diVerent extents in diVerent NSIs.
B. Competence building. The early work within the SI approach largely neglected
learning in the form of education and training.19 However, competence building is
increasingly considered to be an important activity in systems of innovation,
reXecting the importance of skilled personnel for most innovative activities
(Smith 2001: 8).20 But no rigorous analyses of competence building have, to my
knowledge, been conducted as part of the analysis of innovation systems.
Nevertheless, there is a large literature on various aspects of competence building
outside the SI context. Competence building (e.g. training and education) is the
same as enhancement of human capital and is carried out largely, though not
exclusively, in schools and universities. Competence building also occurs in Wrms
(in the form of training, learning-by-doing, learning-by-using, and individual
learning) often throughout working life.
Arecent OECDstudy analyzed vocational andtechnical education andtraining in
some detail in Australia, Austria, Denmark, England (including Wales and Northern
Ireland), France, Germany, Italy, the Netherlands, Quebec, and Switzerland (OECD
1998b). This study pointed out many diVerences across countries with regard
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 194
194 charles edquist
to vocational and technical training. One diVerence concerns the stage prior to
vocational and technical training, i.e. the structure of middle and lower secondary
education. This structure is uniWed in most countries, but is divided into distinct
programs in Germany, Austria, the Netherlands, and Switzerland. One of these
programs is the beginning of the academic pathway, while the others lead essentially
to vocational and technical training. Another diVerence concerns the relative nu-
meric importance of vocational and technical training as opposed to academic
pathways in upper secondary education. In the countries of the British Common-
wealthAustralia, Canada, and the UKthe academic pathway is very much in the
majority, while in the countries of continental Europe, vocational and technical
training dominates (OECD 1998b).
The ways in which people access skilled jobs (and then climb the career ladders
of enterprises) diVer greatly among NSIs:
This may occur at a certain time after recruitment, when the young person has proved
himself; after a Wxed and codiWed period of service, according to a speciWc labour contract; or
on recruitment, depending on the qualiWcations previously acquired. For vocational
training, these three modes of access lead to three broad traditions: on-the-job training,
formal apprenticeship, school training. (OECD 1998b: 12)
These practices coexist in various countries, but their relative importance varies
considerably; frequently, one of them dominates and determines training policy.
The models for transition from education to employment also diVer across coun-
tries. Apprenticeshipis important insome countriese.g. inGermany, where it caters
to about two-thirds of the age group (OECD 1996: 48). In other countries, school-
based learning and productive work are combined in alternative wayse.g. in
Sweden, Australia, France, the United Kingdom, and Korea (OECD 1996: 146).
The organizational and institutional contexts of competence building thus vary
considerably among NSIs. There are particularly signiWcant diVerences between the
systems in the English-speaking countries and continental Europe. However,
scholars and policy makers lack good comparative measures on the scope and
structure of such diVerences. There is little systematic knowledge about the ways
in which organization of education and training inXuences the development, diVu-
sion and use of innovations. Since labor, including skilled labor, is the least mobile
production factor, domestic systems for competence building remain among the
most enduringly national elements of NSIs.
7.4.4 The Relations Between Activities and Components
and among Components
This chapter has placed greater emphasis on activities than much of the early
work on SIs. Nonetheless, this emphasis does not mean that we can disregard or
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 195
systems of innovation 195
neglect the components of SIs and the relations among them. Organizations or
individuals perform the activities and institutions provide incentives and obstacles
inXuencing these activities. In order to understand and explain innovation pro-
cesses, we need to address the relations between activities and components, as well as
among diVerent kinds of components.
What then are the relations between the components and the activities in SIs? As
we saw in Section, the activity of research (the creation of new knowledge)
can be carried out by research institutes, universities, or research-oriented Wrms.
Most of the other activities mentioned earlier can also be performed by diVerent
organizations. Further, many categories of organizations can perform more than
one activity. For example, universities provide new knowledge and educate people
(human capital). Hence, there is not a one-to-one relation between activities and
organizations.21 However, there are limits to this Xexibilityfor instance, primary
schools cannot carry out basic research. The relations between activities and insti-
tutions are less direct, since institutions inXuence whether or not, and how, certain
organizations perform certain activities. It seems that the set-up of activities can be
expected to vary less across NSIs than the set-up of organizations performing them
and the set-up of institutions inXuencing those organizations. However, the quan-
tity of each activity and the eYciency with which each activity is performed might
vary considerably among NSIs.22
As we sawinsection7.3.2, the SI approachemphasizes the relations or interactions
among the components in SIs. Interactions among diVerent organizations may be of
a market or non-market kind. That market, as well as non-market, relations should
be addressed in SI research is stressed in a recent OECDreport. There the concept of
interaction is speciWed as including:
Competition, which is an interactive process wherein the actors are rivals, and
which creates or aVects the incentives for innovation.
. Transaction, which is a process by which goods and services, including technol-
ogy-embodied and tacit knowledge, are traded between economic actors.
Networking, which is a process by which knowledge is transferred through collab-
oration, cooperation and long term network arrangements. (OECD 2002a: 15)
With regard to interactions among organizations in their pursuit of innovations,
empirical work inspired by and designed on the basis of the SI approach has been
carried out in many countries. One example is the Community Innovation Surveys
(CIS) that have beencoordinated by Eurostat of the EuropeanUnionandcarried out
in all EU countries and in several additional countries (see Smith, Chapter 6 in this
volume, for a detailed discussion of CIS). The CIS results include data on collabor-
ation among innovating organizations, and indicate that such collaboration is very
common and important. This result is supported by other surveys which have
shown that between 62 per cent and 97 per cent of all product innovations
were achieved in collaboration between the innovating Wrm and some other
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 196
196 charles edquist
organizations in Austria, Norway, Spain, Denmark, and the region of East Gothia in
Sweden (Christensen et al. 1999; O

rstavik and Nas 1998; Edquist, Ericsson, and

Sjogren 2000: 47).
These Wndings constitute empirical support for one of the main tenets of the SI
approach, i.e. that interactive learning among organizations is crucial for innovation
processes. This also illustrates the dynamics of this Weld of research over time. The
emergence of the SI approach in its Danish version (Lundvall 1992) took inspiration
from case studies indicating that userproducer interaction was very important for
innovations, e.g. in the Danish dairy industry; the SI approach was formulated
partly on this basis. One of its central elementsthe importance of relations of
interactive learning among organizationshas since been veriWed by systematic
empirical research in Denmark and elsewhere. This is a good example of fruitful
interaction between theoretical and empirical work.
Another example of empirical work partly based on the NSI approach is Furman,
Porter, and Stern (2002), who introduce the concept of national innovative capacity,
which is the ability of a country to produce and commercialize a Xowof new-to-the-
world technologies over the long term. This concept is explicitly based upon ideas-
drivenendogenous growth theory a` la Romer (see Verspagen, Ch. 18 in this volume),
the cluster-based approach a` la Porter (1990), and the NSI approach. On this basis,
they estimate the relationship between international patenting (patenting by foreign
countries in the USA) and observable measures of national innovative capacity.
Their results suggest that a small number of observable factors describe a countrys
national innovative capacityi.e. they identify determinants of the production of
new-to-the-world technologies. They Wnd that a great deal of variation in patenting
across countries is due to diVerences in the level of inputs devoted to innovation
(R&D manpower and spending). They also Wnd that an extremely important role is
played by factors associated with diVerences in R&Dproductivity, e.g. policy choices
such as the extent of protection of intellectual property and openness to inter-
national trade, the share of research performed by the academic sector and funded
by the private sector, the degree of technological specialization, and each individual
countrys knowledge stock (Furman et al. 2002).
The relations between organizations and institutions are important for innov-
ations and for the operation of SIs. Organizations are strongly inXuenced and
shaped by institutions, so that organizations can be said to be embedded in an
institutional environment or set of rules, which include the legal system, norms,
routines, standards, etc. But institutions are also embedded in and develop within
organizations. Examples are Wrm-speciWc rules with regard to bookkeeping or
concerning norms with regard to the relations between managers and employees.
Hence, there is a complicated relationship of mutual embeddedness between insti-
tutions and organizations (Edquist and Johnson 1997: 5960).
Some organizations create institutions that inXuence other organizations.
Examples are organizations that set standards and public organizations that
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 197
systems of innovation 197
formulate andimplement those rules that we might call innovationpolicy. Examples
are the NMT 450and the GSMmobile telecomstandards. The NMT 450was created
by the Nordic public telephone operators, which were state-owned monopolies at
the time. The development and implementation of NMT 450was an example of the
importance of userproducer relations in innovation processes, which is stressed so
strongly in the SI approach. The public organizations provided a technical frame-
work for private equipment producers and thereby decreased uncertainty. The
Nordic equipment producers, Ericsson and Nokia, greatly beneWted from this, and
it was animportant factor contributing to their leading role in mobile telecommuni-
cations equipment production today. In essence, the NMT 450 provided the cradle
for the development of mobile telecommunications in Europe (Edquist 2003: 213).
Institutions may also be the basis for the creation of organizations, as when a
government makes a law that leads to the establishment of an organization.
Examples of such organizations include patent oYces or public innovation policy
There may also be important relations betweendiVerent institutions, for example,
between patent laws and informal rules concerning exchange of information
between Wrms. Institutions of diVerent kinds may support and reinforce each
other, but may also contradict and be in conXict with each other, as discussed in
some detail by Edquist and Johnson (1997). This work has been carried forward by
Coriat and Weinstein (2002), who discuss diVerent levels of institutions and focus
on the principle of a hierarchy among rules themselves (Coriat and Weinstein
2002: 280).23
Our knowledge about the complex relationscharacterized by reciprocity and
feedbackbetween organizations and institutions is limited. Since the relations
between two phenomena cannot be satisfactorily analyzed if they are not conceptu-
ally distinguished from each other, it is important to make a clear distinction
between organizations and institutions when specifying the concepts.24
7.4.5 Boundaries of SIs: Spatial, Sectoral and
in Terms of Activities
The distinction betweenwhat is inside and outside a systemis cruciali.e., the issue
of the boundaries of SIs cannot be neglected (see Section 7.4.1). It is therefore
necessary to specify the boundaries if empirical studies of speciWc SIsof a national,
regional, or sectoral kindare to be carried out. As will be discussed later, one way
to identify the boundaries of SIs is to identify the causes or determinants of
Although national systems of innovation is only one of several possible spe-
ciWcations of the generic SI concept, it certainly remains one of the most relevant.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 198
198 charles edquist
One reason is the fact that the various case studies in Nelson (1993) show that there
are sharp diVerences among various national systems in such attributes as insti-
tutional set-up, organizational set-up, investments in R&D, and performance. For
example, the diVerences in these respects between Denmark and Sweden are
remarkablein spite of the fact that these two small countries in northern Europe
are very similar in many other respects (Edquist and Lundvall 1993: 56).
Another reason to focus on national SIs is that most public policies inXuencing
innovation processes or the economy as a whole are still designed and implemented
at the national level. For very large countries, the national SI approach is less relevant
than for smaller countries, but institutions such as laws and policies are still mainly
national, even in a country such as the USA. In other words, the importance of
national SIs has partly to do with the fact that they capture the importance of the
policy aspects of innovation. It is not only a matter of geographical delimitation, as
the state, and the power attached to it, is also important.
SIs may be supranational, national, or subnational (regional, local)and at the
same time they may be sectoral within any of these geographical demarcations.25 All
these approaches may be fruitful, but for diVerent purposes or objects of study.
Generally, the variants of the generic SI approach complement rather than exclude
each other and it is useful to consider sectoral and regional SIs in relation toand
often as parts ofnational ones.
There are three ways in which we can identify boundaries of SIs:
(1) spatially / geographically;
(2) sectorally; and
(3) in terms of activities.
1. To deWne the spatial boundaries is the easiest task, although it also has its
problems. Such boundaries have to be deWned for regional and national SIs, and
sometimes also for sectoral ones. The problem of geographical boundaries is some-
what more complicated for a regional than for a national SI. One question is which
criteria should be used to identify a region.
For a regional SI, the speciWcationof the boundaries should not only be a question
of choosing or using administrative boundaries between regions in a mechanical
manner (although this might be useful fromthe point of viewof availability of data).
It should also be a matter of choosing geographical areas for which the degree of
coherence or inward orientation is high with regard to innovation processes.
One possible operationalization of this criterion could be a suYcient level of
localized learning spill-overs (among organizations), which is often associated
with the importance of transfer of tacit knowledge among (individuals and) organ-
izations. A second could be localized mobility of skilled workers as carriers of
knowledge, i.e. an operationalization which shows that the local labor market is
important. A third possibility could be that a minimum proportion of the innov-
ation-related collaborations among organizations should be with partners within
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 199
systems of innovation 199
the region. This is a matter of localized networks, i.e. the extent to which learning
processes among organizations are contained within regions.
For a national SI, the countrys borders normally provide the boundaries. How-
ever, it could be argued that the criteria for regional SIs are as valid for national ones.
In other words, if the degree of coherence or inward orientation is very low, the
country might not reasonably be considered to have a national SI. It was also
mentioned above that the national SI approach is less relevant for large than for
smaller countries. In Germany, for example, the appropriate unit of analysis may be
Lander. The choice of approach may not only be a question of size of the country,
but also whether it is federally organized or not.
2. Leaving the geographical dimension, we can also talk about sectorally
delimited SIs, i.e., systems that include only a part of a regional, national or
international system. These are delimited to speciWc technological Welds (generic
technologies) or product areas. The technological systems approach belongs to
this category (although it did not initially use language associated with systems of
innovation) (Carlsson and Stankiewicz 1995: 49).
According to Breschi and Malerba, a Sectoral Innovation System (SIS) can be
deWned as that system (group) of Wrms active in developing and making a sectors
products and in generating and utilising a sectors technologies (Breschi and
Malerba 1997: 131; see also Ch. 14 by Malerba in this volume). SpeciWc technologies
or product areas are used to deWne the boundaries of sectoral systems, but they must
also normally be geographically delimited (if they are not global). However, it is not
self-evident what a sector is, i.e., the sectoral boundaries are partly a theoreticalor
socialconstruction, which may reXect the speciWc purpose of the study. It should
also be noted here that the speciWcation of sectoral boundaries is particularly
diYcult with regard to new sectors or sectors going through radical technological
3. Within a geographical area (and perhaps also limited to a technology Weld or
product area), the whole socio-economic systemcannot, of course, be considered to
be included in the SI. The question is, then, which parts should be included? This is a
matter of deWning the boundaries of SIs in terms of activities. These have to be
deWned for all kinds of SIs: national, regional, and sectoral. This is more complicated
than in the cases of spatial and sectoral boundaries.
Early work in the SI approach did not address the activities in SIs in a systematic
way, and therefore failed to provide clear guidance as to what should be included in a
system of innovation. Nor have the boundaries of the systems in terms of activities
been deWned in an operational way since then.
In Section 7.2, a system of innovation was deWned as including all important
economic, social, political, organizational, institutional, and other factors that
inXuence the development, diVusion, and use of innovations. If the concept of
innovations has been speciWed (e.g., as in the beginning of Section 7.2), and if we
know the determinants of their development, diVusion, and use, we will be able to
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 200
200 charles edquist
deWne the boundaries of the SIs in terms of activities. This is one reason why it is so
important to identify the activities in SIs. Admittedly, this is not as easy in practice as
in principle, since we simply do not know in detail and systematically all the
activities in SIs or determinants of innovation processes. As pointed out in Section, any list of activities in an SI must be treated as provisional and subject to
change as our knowledge increases.
7. 5 Research Gaps and Opportunities
In innovation studies, there has traditionally been a tendency to focus much more
on technological process innovations and goods product innovations than on
organizational process innovations and service product innovations. There are
strong reasons to use a comprehensive innovation concept and give more attention
to non-technological and intangible kinds of innovation (as proposed in Section
7.3.2). Such an orientation is implicit in the fact that we talk about systems of
innovation and not systems of technological change.
More research should be done on the activities in SIs, i.e., on the determinants of
the development, diVusion, and use of innovations. One particular task may be to
revise and restructure the preliminary list of important activities in SIs presented in
Section Such a list can provide an important point of entry for empirical
innovation studies.
A stronger focus on activities would increase our knowledge of, and capacity for,
explaining innovation processes. Given our limited systematic knowledge about
determinants of innovations, case studies of the determinants of speciWc innovations
or speciWc (and narrow) categories of innovations would be very useful. In particu-
lar, I believe that comparative case studies have great potential, comparing innov-
ations systems of various kinds as well as the determinants of innovation processes
within them. Relevant questions to ask would include: Which activities of which
organizations are important for the development, diVusion, or use of speciWc
innovations? Is it possible to distinguish between important activities and less
important ones? Which institutional rules inXuence the organizations in carrying
out these activities? Suchwork could further develop the SI approach and contribute
to the creation of partial theories about relations between variables within SIs. Such
theories would also improve our ability to specify the boundaries of innovation
In this chapter, I have accounted for many of the existing empirical studies that
claim to have been carried out within a SI framework. The result has, on the whole,
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 201
systems of innovation 201
been rather disappointing in the sense that many of the studies cited have not been
related to the SI approach in a profound way, although there are exceptions. The SI
approach has often been used more as a label than as an analytical tool. It has not
inXuenced the empirical studies in depth; for example, it has not been used to
formulate hypotheses to be confronted to empirical observations. This has made a
virtuous fertilization between conceptual and empirical work, that is so important
to scholarly progress in this andother Welds of research, diYcult to achieve. The state
of the art of the SI approach is partly responsible for this: it is often presented in too
vague and unclear a way.
Clearly deWned concepts are necessary in order to identify empirical correspond-
ents to theoretical constructs and to identify the data that should be collected.
Therefore, conceptual speciWcations are crucial for empirical studies and it is
important to increase the rigor and speciWcity of the SI approach. This can be
done by clarifying the meaning of key concepts such as innovation, function,
activities, components, organizations, and institutions, as well as the relations
among them. Moving in this direction does not mean transforming social science
into something similar to natural science. For example, one cannot abstract from
time and space, since there are no universal laws in the social sciences. It is also
important to continue the work of specifying the boundaries of SIs of various kinds.
There are strong reasons to integrate conceptual and theoretical work with
empirical studies in an eVort to identify determinants of the development,
diVusion, and use of innovations. Such integration can be expected to lead to
cross-fertilizationjust as in the case of work on interactive learning referred to
in Section 7.4.4. The SI approach should be used as a conceptual framework in
speciWc empirical analyses of concrete conditions. Testable statements or hypotheses
should be formulated on the basis of the approach and these should be investigated
empirically, by using qualitative as well as quantitative observations. Theoretically
based empirical work is the best way to straighten up the SI approach conceptually
and theoretically; the empirical work will, in this way, serve as a disciplining device
in an eVort to develop the conceptual and theoretical framework. Such work would
increase our empirical knowledge about relations between the main function,
activities, organizations, and institutions in SIs. This knowledge could then be a
basis for further empirical generalizations to develop the frameworkincluding
theory elements. In other words, empirically based theoretical work is also very
fruitful. Independently of where one starts, the important thing is that there should
be a close relationship between theoretical and empirical work.
The array of determinants of innovations and the relations among them can be
expected to vary over time and space, i.e. between innovation systems, as well as
among diVerent categories of innovation. For example, the determinants will
probably vary between process and product innovations as well as between incre-
mental and radical innovations (and between subcategories of these). It is therefore
important to pursue the explanatory work at a meso- or micro-level of aggregation.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 202
202 charles edquist
Taxonomies of diVerent categories of innovations can therefore be expected to be an
important basis for this work.
Innovation studies have traditionally included research on R&D and its signiW-
cance for innovation processes. A well-educated labor force is necessary for both
R&D and innovation, and competence building therefore should receive greater
emphasis in innovation studies and in the SI approach. We should not only address
those learning processes that lead directly to process and product innovations, but
also address the knowledge infrastructure and learning in a more generic way.
This widening might eventually transcend the SI approach and move into
thinking along the lines of Systems of Learning rather than Systems of Innov-
ation. Systems of Learning would include individual learning (leading to the
creation of human capital) as well as organizational learning (leading to the creation
of structural capital, e.g. innovations). It would include work on three kinds of
learning: R&D, innovation, and competence building, and, above all, the relations
between them. This also points out one direction in which the SI approach is
currently developing.
1. For comments on previous versions of this chapter, I want to thank the editors of this
Handbook, my discussant at the Lisbonworkshop (John Cantwell), and my discussants at
the Roermond workshop (Jan Fagerberg, Bill Lazonick, and Rikard Stankiewicz). I have
also greatly proWted from comments by other participants in the workshops and from
Pierre Bitard, Susana Borras, David Doloreux, Leif Hommen, Bjorn Johnson, Rachel
Parker, Lars Mjoset, and Annika Rickne. I also want to thank The Swedish Agency for
Innovation Systems (VINNOVA) for supporting my work with this chapter. However,
I remain responsible for the contents.
2. The regional and sectoral versions are dealt with in more detail in Ch. 11 by Asheim and
Gertler and in Ch. 14 by Malerba in this volume.
3. In this sense, this chapter is a continuation along the same trajectory as earlier attempts,
e.g. Edquist (1997b), Edquist and Johnson (1997) and Edquist (2001).
4. [T]he orientation of this project has been to carefully describe and compare, and try to
understand, rather thanto theorise Wrst and thenattempt to prove or calibrate the theory
(Nelson and Rosenberg (1993: 4).
5. They mention organizations such as Wrms, industrial research laboratories, research
universities, and government laboratories.
6. Their deWnitions of NSIs do not include, for example, consequences of innovation
which does not, of course, exclude the fact that innovations, emerging in innovation
systems, have tremendously important consequences for socio-economic variables such
as productivity growth and employment. To distinguish between determinants and
consequences does not, of course, exclude feedback mechanisms.
7. It should also be mentioned that the publications mentioned in Section 7.2 by no
means exhaust the stock of literature addressing or using the SI approach. Edquist and
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 203
systems of innovation 203
McKelvey (2000) is a reference collection containing forty-two articles on SIs, some of
whichare reviewedinthis chapter. Other contributions will be addressedinlater sections
of this chapter.
8. There are chapters inthe Handbookonservice product innovations (Ch. 16 by IanMiles)
and on organizational process innovations (Ch. 5 by Alice Lam).
9. See Lundvall et al. (2002: 221) and Lundvall (2003: 9), where it is argued that the
pragmatic and Xexible character of the concept may be seen to be a great advantage.
However, Lundvall et al. (2002: 221) also argue that eVorts should be made to give the
concept a stronger theoretical foundation.
10. Such a view has, for example, been expressed by the OECD: There are still concerns in
the policy making community that the NIS approach has too little operational value and
is diYcult to implement (OECD2002a: 11). Asimilar position is taken by Fischer (2001:
11. Only in exceptional cases is the system closed in the sense that it has nothing to do with
the rest of the world (or because it encompasses the whole world). Like the SI approach,
general systems theory might rather be considered to be an approach than a theory.
12. Although there are other kinds of actors than organizationse.g. individualsthe
terms organizations and actors are used interchangeably in this chapter.
13. The Wve activities are R&D, implementation, end-use, education, and linkage.
14. These are: to create new knowledge, to guide the direction of the search process, to
supply resources, to create positive external economies, and to facilitate the formation of
markets (Johnson and Jacobsson 2003: 34). Anna Johnsonnow Anna Bergek
previously discussed these issues in Johnson (1998). There she identiWed functions
mentioned or implicitly addressed invarious previous contributions to the development
of the SI approach. She also listed and stressed various beneWts of using the concept of
function in SI studies.
15. These functions are to create human capital, to create and diVuse technological oppor-
tunities, to create and diVuse products, to incubate (provide facilities, equipment, and
administrative support), to facilitate regulation that may enlarge the market and facili-
tate market access, to legitimize technology and Wrms, to create markets and diVuse
market knowledge, to enhance networking, to direct technology, market and partner
search, tofacilitate Wnancing, andtocreate a labor market that the newtechnology-based
Wrms can utilize.
16. The activities in SIs are the same as the determinants of the main function. Analternative
termto activities could have been subfunctions. I chose activities in order to avoid
the connotation with functionalism or functional analysis as practiced in sociology,
whichfocuses onthe consequences of a phenomenon rather than onits causes, which are
in focus here.
17. The activities in this sectoral system are discussed in Edquist (2003: 11).
18. SpeciWc ways in which knowledge transfer takes place between universities and Wrms are
analyzed in detail for the case of Austria in Schibany and Schartinger (2001).
19. When designing the anthology edited by Lundvall (1992), the Aalborg group planned to
have a chapter on the education system. However, in the end it was not included
(Lundvall and Christensen (1999: 3).
20. Competence building has also been addressed in some OECD publications, including a
study on knowledge management in the learning society, managed by the Centre for
Educational Research and Innovation (CERI) (OECD 2000). Another CERI study
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 204
204 charles edquist
includes a conceptual framework which tries to integrate individual learning (e.g.
education) and organizational learning (e.g. innovation) into a generic conceptual
framework on learning. It also contains empirical studies of the respective roles of
education and innovation for economic growth at a regional level (OECD 2001d).
Another contribution is the DISKOproject in Denmark as reported by Lundvall (2002).
21. In Rickne 2000: ch. 7, there is a more detailed discussion of the relations between
activities and organizations.
22. As we saw in Section, there are also important relations between activities, i.e.
relations between determinants of innovations processes.
23. Coriat and Weinstein address the relations between organizations and institutions as
well, although they consider Wrms to be both institutions and organizations (Coriat and
Weinstein 2002: 279).
24. The so-called varieties of capitalism literature has a wider perspective and focuses on a
broader range of institutions and organizations. Examples are Hollingsworth and Boyer
(1997), Whitley (1999), Hall and Soskice (2001), and Whitley (2002). Space limitations
prevent me fromgoing intothis literature here. The same applies tothe social systems of
innovation perspective (e.g. Amable 2000) and the Triple-Helix perspective (e.g.
Etzkowitz and LeydesdorV (2000).
25. An industrial complex or cluster as used by Porter (1990, 1998) can, if it is regionally
delimited, be seen as a combination of a sectoral and a regional SI.
Amable, B. (2000), Institutional Complementarity and Diversity of Social Systems of
Innovation and Production, Review of International Political Economy 7(4): 64587.
Asheim, B., and Isaksen, A. (2002), Regional Innovation Systems: The Integration of Local
Sticky and Global Ubiquitous Knowledge, Journal of Technology Transfer 27: 7786.
*Braczyk, H.-J., Cooke, P., and Heidenreich, M. (eds.) (1998), Regional Innovation
Systems: The Role of Governance in a Globalised World, London and Pennsylvania: UCL.
*Breschi, S., and Malerba, F. (1997), Sectoral Innovation Systems: Technological
Regimes, Schumpeterian Dynamics, and Spatial Boundaries, in Edquist, 1997a: 13056.
*Carlsson, B. (ed.) (1995), Technological Systems and Economic Performance: The Case of
Factory Automation, Dordrecht: Kluwer.
and Stankiewicz, R. (1995), On the Nature, Function and Composition of Techno-
logical Systems, in Corlson 1995: 2156.
Carr, E. H. (1986), What is History? Harmondsworth: Penguin.
Christensen, J. L., Rogaczewska, A. L., and Vinding,A. L., (1999) Summary Report of the
Focus Group on Innovative Firm Networks, OECD home page.
Cooke, P. (2001), Regional Innovation Systems, Clusters, and the Knowledge Economy,
Industrial and Corporate Change 10(4): 94574.
*Gomez Uranga, M., and Etxebarria, G. (1997). Regional Systems of Innovation:
Institutional and Organisational Dimensions, Research Policy 26: 47591.
* Asterisked items are suggestions for further reading.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 205
systems of innovation 205
Coriat, B., and Weinstein, O. (2002), Organisations, Firms and Institutions in the
Generation of Innovation, Research Policy 31(2): 27390.
*Edquist, C. (ed.) (1997a), Systems of Innovation: Technologies, Institutions and Organiza-
tions, London: Pinter.
(1997b), Systems of InnovationApproachestheir Emergence andCharacteristics, in
Edquist 1997a: 135. (The book is out of print, but this chapter has been republished in
Edquist and McKelvey 2000.)
(2001), The Systems of Innovation Approach and Innovation Policy: An Account of
the State of the Art, Lead paper presented at the DRUID Conference, Aalborg, June 1215
2001. Unpublished.
(2003), The Fixed Internet and Mobile Telecommunications Sectoral System of In-
novation: Equipment, Access and Content, in C. Edquist (ed.), The Internet and Mobile
Telecommunications Systemof Innovation: Developments in Equipment, Access and Content,
Cheltenham: Edward Elgar, 139.
Ericsson, M.-L., and Sjogren, H. (2000), Collaboration in Product Innovationinthe
East GothiaRegional Systemof Innovation, Enterprise&InnovationManagement Studies, 1.
Hommen, L., and McKelvey, M. (2001), Innovation and Employment: Process versus
Product Innovation, Cheltenham: Edward Elgar.
and Johnson, B. (1997), Institutions and Organisations in Systems of Innovation, in
Edquist 1997a: 4163. (The book is out of print, but this chapter has been republished in
Edquist and McKelvey 2000.)
and Lundvall, B.-A

. (1993), Comparing the Danish and Swedish systems of innov-

ation, in Nelson 1993: 26598.
and McKelvey, M. (eds.) (2000), Systems of Innovation: Growth Competitiveness and
Employment, Cheltenham: Edward Elgar.
Etzkowitz, H., and Leydesdorff, L. (2000), The Dynamics of Innovation: FromNational
Systems and Mode 2 to Triple Helix of UniversityIndustryGovernment Relations,
Research Policy 29: 10923.
Fischer, M. F. (2001), Innovation, Knowledge Creation and Systems of Innovation,
Regional Science 35: 199216.
Freeman, C. (1987), Technology Policy and Economic Performance: Lessons from Japan,
London: Pinter.
*(2002), Continental, National and Sub-national Innovation SystemsComplemen-
tarity and Economic Growth, Research Policy 31(2): 191211.
Furman, J. L., Porter, M. E., and Stern, S. (2002), The Determinants of National
Innovative Capacity, Research Policy 31: 899933.
Galli, R., and Teubal, M. (1997), Paradigmatic Shifts in National Innovation Systems, in
Edquist 1997a: 34270.
Hall, P. A., and Soskice, D. (eds.) (2001), Varieties of Capitalism. The Institutional Founda-
tions of Comparative Advantage, Oxford: Oxford University Press.
Hollingsworth, J. R., and Boyer, R., (eds.) (1997), Contemporary Capitalism: The Embed-
dedness of Institutions, Cambridge: Cambridge University Press.
Ingelstam, L. (2002), Systematt tanka over samhalle och teknik (Systems: To ReXect over
Society and Technologyin Swedish), Energimyndighetens forlag.
Johnson, A. (1998), Functions in Innovation SystemApproaches, Mimeo, Department of
Industrial Dynamics, Chalmers University of Technology.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 206
206 charles edquist
*and Jacobsson, S. (2003), The Emergence of a Growth Industry: A Comparative
Analysis of the German, Dutch and Swedish Wind Turbine Industries, in S. Metcalfe and
U. Cantner (eds.), Transformation and Development: Schumpeterian Perspectives, Heidel-
berg: Physica/Springer.
Laredo, P., and Mustar, P. (eds.) (2001a), Research and Innovation Policies in the NewGlobal
Economy: an International Comparative Analysis, Cheltenham: Edward Elgar.
(2001b), General Introduction: AFocus on Research and Innovation Policies, in
Laredo and Mustar 2001a: 113.
(2001c), General Conclusion: Three major Trends in Research and Innovation
Policies, in Laredo and Mustar 2001a: 497509.
*Liu, X. and White, S. (2001), Comparing Innovation Systems: A Framework and Appli-
cation to Chinas Transitional Context, Research Policy 30: 1091114.
*Lundvall, B.-A

. (ed.) (1992), National Systems of Innovation: Towards a Theory of Innov-

ation and Interactive Learning, London: Pinter.
(2002), Innovation, Growth and Cohesion: The Danish Model, Cheltenham: Edward
(2003), National Innovation Systems: History and Theory, in Elgar Companion to
Neo-Schumpeterian Economics, Cheltenham: Edward Elgar.
Johnson, B., Andersen, E. S., and Dalum, B. (2002), National Systems of Produc-
tion, Innovation and Competence Building, Research Policy 31(2): 21331.
and Lindgaard Christensen, J. (1999), Extending and Deepening the Analysis of
Innovation Systemswith Empirical Illustrations fromthe DISKO-project, Paper for the
DRUID Conference on National Innovation Systems, Industrial Dynamics and Innov-
ation Policy, Rebild, 912 June 1999.
*Nelson, R. R. (ed.) (1993), National Systems of Innovation: A Comparative Study, Oxford:
Oxford University Press.
and Rosenberg, N. (1993), Technical Innovation and National Systems, in Nelson
1993: 321.
North, D. C. (1990), Institutions, Institutional Change and Economic Performance, Cam-
bridge: Cambridge University Press.
OECD (1996), Lifelong Learning for All, Paris: OECD.
(1998a), Technology, Productivity and Job Creation: Best Policy Practices, Paris: OECD.
(1998b), Pathways and Participation in Vocational and Technical Education and Training,
Paris: OECD.
(1999a), Managing National Innovation Systems, Paris: OECD.
(1999b), Boosting Innovation: The Cluster Approach, Paris: OECD.
(2000), Knowledge Management in the Learning Society, Paris: OECD (CERI).
(2001a), Innovative Clusters: Drivers of National Innovation Systems, Paris: OECD.
(2001b), Innovative Networks: Co-operation in National Innovation Systems, Paris:
(2001c), Innovative People: Mobility of Skilled Personnel in National Innovation Systems,
Paris: OECD.
(2001d), Cities and Regions in the NewLearning Economy, written by C. Edquist, G. Rees,
K. Larsen, M. Lorenzen, and S. Vincent-Lancrin, Paris: OECD (CERI).
(2002a), Dynamising National Innovation Systems, Paris: OECD.
(2002b), Main Science and Technology Indicators, vol. 2002/2, Paris: OECD.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 207
systems of innovation 207

rstavik, F., and NA

s, S.-O. (1998), The Norwegian InnovationCollaboration Survey,

Oslo, the STEP Group, STEP Working Paper A-10.
Porter, M. E. (1990), The Competitive Advantage of Nations, New York: Free Press.
(1998), Clusters and the New Economics of Competition, Harvard Business Review
Rickne, A. (2000), New Technology-Based Firms and Industrial Dynamics: Evidence from the
Technological Systems of Biomaterials in Sweden, Ohio and Massachusetts, Department of
Industrial Dynamics, Chalmers University of Technology.
Schibany, A., and Schartinger, D. (2001), Interactions between Universities and Enter-
prises in Austria: An Empirical Analysis at the Micro and Sector Levels, in OECD 2001b:
Smith, K. (2001), Human Resources, Mobility and the Systems Approach to Innovation, in
OECD 2001c: ch. 1.
Whitley, R. (1999), Divergent Capitalisms: The Social Structuring and Change of Business
Systems, Oxford: Oxford University Press.
(ed.) (2002), Competing Capitalisms: Institutions and Economics. An Elgar Reference
Collection, 2 vols., Cheltenham: Edward Elgar.
g g p p g Fagerberg / The Oxford Handbook of Innovation Revised Proof 31.7.2004 5:10pm page 208
208 charles edquist
American Economic Association
B&B SpiIIovevs and lIe OeogvapI oJ Innovalion and Fvoduclion
AulIov|s) Bavid B. AudvelscI and Mavann F. FeIdnan
Bevieved vovI|s)
Souvce TIe Anevican Econonic Beviev, VoI. 86, No. 3 |Jun., 1996), pp. 630-640
FuIIisIed I American Economic Association
SlaIIe UBL http://www.jstor.org/stable/2118216 .
Accessed 11/04/2012 1710
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to The
American Economic Review.
Progress in Human Geography
The online version of this article can be found at:
DOI: 10.1191/0309132504ph469oa
2004 28: 31 Prog Hum Geogr
Harald Bathelt, Anders Malmberg and Peter Maskell
knowledge creation
Clusters and knowledge: local buzz, global pipelines and the process of
Published by:
can be found at: Progress in Human Geography Additional services and information for

http://phg.sagepub.com/cgi/alerts Email Alerts:

http://phg.sagepub.com/subscriptions Subscriptions:
http://www.sagepub.com/journalsReprints.nav Reprints:

http://www.sagepub.com/journalsPermissions.nav Permissions:

http://phg.sagepub.com/content/28/1/31.refs.html Citations:

by guest on May 9, 2011 phg.sagepub.com Downloaded from
Clusters and knowledge: local buzz,
global pipelines and the process of
knowledge creation
Harald Bathelt,
Anders Malmberg
Peter Maskell
Faculty of Geography, Philipps-University of Marburg, Deutschhausstrae 10,
D-35032 Marburg, Germany
Department of Social and Economic Geography, and Centre for Research on
Innovation and Industrial Dynamics (CIND), Uppsala University, PO Box 513, S-751
20 Uppsala, Sweden
Danish Research Unit for Industrial Dynamics (DRUID), Department of Industrial
Economics and Strategy (IVS), Copenhagen Business School (CBS), Howitzvej 60,
DK-2000 Frederiksberg, Denmark
globe almost frictionlessly. The paper highlights the conditions under which both tacit and
onehand,thelearningprocessestakingplaceamongactorsembedded inacommunitybyjust
channels ofcommunicationcalledpipelines toselected providers locatedoutside thelocal
firmslocatedinoutward-lookingandlively clusterswithastringofparticularadvantagesnot
available to outsiders. Finally, some policy implications, stemming from this argument, are
Progress in Human Geography 28,1 (2004) pp. 3156
Arnold 2004 10.1191/0309132504ph469oa
by guest on May 9, 2011 phg.sagepub.com Downloaded from
32 Clusters and knowledge
I Introduction
The riddle we are dealing with in this paper is concerned with spatial clustering of
economicactivityanditsrelationtothespatialityofknowledgecreation.A condensed
have contributed in various papers in recent years,
goes as follows. Innovation,
knowledge creation and learning are all best understood if seen as the result of
organizational, commercial orintellectual problems. Such exchange andinteraction
less space-sensitive should these processes tend to be. If on the other hand the
knowledge involved is diffuse and tacit, the argument is that such interaction and
moresubtle formsofinformationbeexchanged. Thishasbeen proposedasthemain
mechanism that makes it beneficial for a firm to be located in a spatial cluster,
also, Gertler, 2003), asit doesnotexplain whyinteractions andtransactions between
of the simple tacit = local versus codified = global model, by highlighting the
conditionsunder which both tacit and codified knowledge can be exchanged locally
1 Codified and tacit knowledge, local and global
traderegimes,emergingmarketsforintellectual propertyrightsandimprovementsin
have sometimes led to the assumption that knowledge, once codified, is almost
instantly available to all firms at zero cost regardless of their location.
However, in
ilating and applying codified knowledge already in existence and use. Attaining
almost omnipresent it may become valuable only if fused with less transitory
knowledge whether proprietary or embedded in a local environment in tacit forms
(Maskellet al.,1998;Asheim,1999).
One of the main distinguishing features of spatial clusters of similar and related
external knowledge new value can be created. It is the quest for superior rents that
by guest on May 9, 2011 phg.sagepub.com Downloaded from
Harald Bathelt et al. 33
compelfirmsinclustersnottorelyoninternal orlocalassetsonly,buttopursuesys-
tematically and sometimes vigorously potentially useful knowledge pools residing
Thislineofreasoningmakesussuspectthattheparticularly successfulclustersare
theindustrialdistrictsliterature(e.g.,Goodmanet al.,1989;Pykeet al.,1990)tostudies
of innovative milieus (Camagni, 1991; Ratti et al., 1997) and transaction cost-based
analyses (e.g., Scott, 1988; 1998) that external sources of knowledge are often
that deals with the question of how the structure of interactive and knowledge-
empirical observation and conceptualization has stimulated debate in recent years.
emphasizes theimportance oflocal-level governance.. . .Linkswiththewider world
are [in such literature] frequently acknowledged, but they are weakly theorized
(emphasis added).
Using theseand otherstudies asa point ofdeparture, we would
2 Aim and structure of the paper
We maintain that the skills and efforts required when attending to the local
environmentareratherdifferentfromtheonesnecessarytomaximize theinflowand
configurations.Werefertothespecificcasesofclustersbasedonknowledge creation
Ourargument proceedsinthefollowingway.Inthenextsectionweidentify some
important properties ofknowledge creation within firms while emphasizing reasons
geographical agglomerations or spatial clusters of similar and related economic
activities. InsectionIVwemoveontowhatisthepivotalargumentofthispaper:the
residingelsewhere. Thissectiondiscusseshowthismaybedoneandthebalance that
by guest on May 9, 2011 phg.sagepub.com Downloaded from
34 Clusters and knowledge
II Knowledge creation within and across firms
A main argument in the contemporary literature on learning and innovation is that
thesearetheresultofinteractive processesinwhichdifferentactorscometogetherto
collaborate in solving particular problems. While much of this literature focuses on
network relations between firms, we believe that it is important to start out by
1 Knowledge creation within firms
(Fuchs, 2001; Clark and Tracey, 2004). A firm constitutes a common interpretative
experience which help to ensure that what each employee learns is in some way
Lorenz, 1999). However, as knowledge is in itself an important source for further
knowledgecreation,smallinitial individual differencesincreaseovertimeevenwhen
sharing common experiences. As the firm matures, its knowledge stock will, conse-
thefirmbecomesthefewerexperiences are shared acrossall sectionsandemployees
for this is that each body of knowledge, which a firm possesses, is often linked to
distinct technologies and associated with the performance of certain tasks that may
require a particular set of criteria for decision-making and a specific style of
management. Furthermore, each field of competence usually requires some sort of
2 Cross-departmental knowledge creation
ofcourse,tosomedegree bekeptatbay by taking refuge totheformation ofadivi-
managerial controllossproblems(Teece,1980), thebenefitsareoftenobtainedatthe
expenseofthecross-divisionalsynergiesthatconstitutedtheraisondtre fordiversifi-
cation.Theensuing difficulties incrossingdepartmental layers within thedivisional-
ized companymight atleast partly help toexplain the . . . somewhatironicfact that
by guest on May 9, 2011 phg.sagepub.com Downloaded from
Harald Bathelt et al. 35
manymanagersconsiderinternal transactionstobemoredifficultthanexternalones,
to combine expertise from a broad range of fields. When experts from different
the presentation ofdifferent types of knowledge relevant forthis new development.
Thisrequiresthatvarioussourcesoftacitknowledgearearticulated andexplicated to
stage involves the re-combination and connection of the various explicit knowledge
of structured discussion and brainstorming. Once this stage has resulted in a new
systems and employees routines to build and test a prototype. Finally, this tacit
knowledge is constantly being transformed and perfected through processes of
learning andsocializing. Throughthis,productioncanbeorganized withlowercosts
different internal organizational structures to be executed efficiently (Nonaka et al.,
2000). The process of articulating various types oftacit knowledge and recombining
3 Knowledge creation across firms
The insistence that radical knowledge creation is usually an interactive process
quite some time (Rosenberg, 1982; Freeman, 1982; 1991; Kline and Rosenberg, 1986;
Lundvall, 1988; Hkansson, 1989; Hagedoorn and Schakenraad, 1992; OECD, 1992;
Gertler, 1993; 1995). The National Innovation Systems approach uses this insight
as its most basic building block (Lundvall and Maskell, 2000), but the idea that the
division of labour is a device for developing knowledge is, of course, much older
how scientific as well as nonscientific knowledge becomes more specialized as it
develops, leading to the apprehension of detailed anomalies that would easily
be overlookedand thus contributing to an acceleration ofthe growth ofknowledge.
Even when specializing in performing some particularly trivial tasks, individuals
find solutions and notice peculiarities otherwise overlooked. By creating an
appropriate differentiation, a group of firms can therefore develop knowledge far
new economic activities become possible; the economy progresses; and the resulting
process of knowledge creation resulting from the division of labour contributes
to the variation needed for the future reassembly of knowledge. The division of
by guest on May 9, 2011 phg.sagepub.com Downloaded from
36 Clusters and knowledge
labour among firms thus impacts directly on the (possible) level of learning in the
thecognitivedistance thatfirmshavetoovercomewhenutilizing differentbodiesof
knowledge in their ongoing activities, particularly when they engage in interfirm
product development projects (Storper and Venables, 2002). The knowledge hetero-
make interaction worthwhile,hence allowing learning processestotake place.Atthe
same time, if the cognitive distance becomes too great or the knowledge bases too
dissimilar, then interfirm learning will cease (Nooteboom,2000). Firms build external
thatclustering ofrelatedeconomicactivities maybeaparticularly usefulstrategy for
III Knowledge creation within clusters: the nature of local buzz
acluster canrange fromasingle cityorstatetoacountryorevenagroupofneigh-
bouring countries.
This broad characterization says, however, very little about the
advantages firms havewhenlocating within a cluster.
Toanswer this questionone
may distinguish between the horizontal and vertical dimensions of a cluster and
identify theadvantages oflocal orregional asopposedtoextra-local orinterregional
1 Cluster dimensions
and compete with one another. This dimension can play a decisive role in the early
andproductdifferentiation.Thesefirmsdonotnecessarily haveclosecontactstoone
and costoftheproductionfactorsthatthey use.Advantages ofproximityarise from
continuous monitoring and comparing. Due to their co-presence, the production
conditions are basically the same for all regional firms. This enables the firms to
effectively compare their performance with that of their competitors. Overall, this
by guest on May 9, 2011 phg.sagepub.com Downloaded from
Harald Bathelt et al. 37
Marshall (1920: 225) described the process of how variety at the horizontal level
stimulates growth in the vertical dimension: [. . .] if one man starts a new idea, it is
bourhood, supplying it with implements and materials, organizing its traffic, and in
aspecialized industryclusterhasbeenestablished, thefirmsofthisclusterdevelopa
markets, thesuppliers can gain economiesofscale anddistribute large partsoftheir
productionatlowcosts(i.e., transactionandtransportationcosts).Asaconsequence,
one would expect the development of dense networks of transaction and material
seldom characterized by strong internal input-output linkages.
Karaskas (1969)
classical study of input-output linkages in the Philadelphia manufacturing sector
revealedthatonlyarelativelysmallpercentageofmateriallinkages tookplacewithin
the region. This modest importance of regional input-output linkages has been
confirmed in many studies of older industrial regions (e.g., Gilmour, 1974; Erikson,
2 Localized capabilities
Recent contributions by Storper (1995; 1997), Maskell and Malmberg (1999a; 1999b),
other kinds of traded interdependencies is simply not enough to understand the
processes behind spatial clustering. In emphasizing localized capabilities and
untraded interdependencies, it has been shown that socioinstitutional settings,
interfirmcommunicationandinteractive processesoflocalized learning playdecisive
rolesinprocessesofinnovationandgrowth(Maskellet al.,1998;GordonandMcCann,
Overall, thesharedknowledge basis enables cluster firmstocontinuouslycombine
and re-combine similar and nonsimilar resources to produce new knowledge and
innovations.Thisstimulates economicspecialization within theclusterand resultsin
A location within an industrial cluster brings further advantages that are not
available to firms situated elsewhere. Again, Marshall (1927) expressed this in his
tothepeoplewithinaparticular regionorplace. Throughobservationsinthecutlery
industry of Sheffield and Solingen, Marshall (1927: 284) concluded that such places
facturersofcutlery great advantages, thatare noteasily tobehadelsewhere: and an
by guest on May 9, 2011 phg.sagepub.com Downloaded from
38 Clusters and knowledge
3 Atmosphere, broadcasting, noise, buzz: the exclusive quality of the cluster
Marshallsindustrial atmosphereconcernedtheindustry-specificclusteradvantages
upandextendedtobecomeamoregeneral statementabouttheadvantagesthatmay
ization economies; cf. Hoover,1970). Storper and Venables (2002) have, forinstance,
recently identified what they see as a particularly important subset of urbanization
economies, which theylabel buzz. Ina similar way, Owen-Smith and Powell (2002)
use the notionoflocal broadcasting and Grabher (2002a) the term noise todenote
somethingsimilar: theideathatacertainmilieu canbevibrantinthesensethatthere
are lots of piquant and useful things going on simultaneously and therefore lots of
and Maillat, 1991; Maillat, 1998). Buzz refers to the information and communication
ecology created by face-to-face contacts, co-presence and co-location of people and
firms within the same industry and place or region. This buzz consists of specific
informationand continuousupdates ofthisinformation,intended andunanticipated
learning processes inorganized and accidental meetings, the application ofthe same
as well as shared cultural traditions and habits within a particular technology field,
Participating in the buzz does not require particular investments. This sort of
informationandcommunicationismoreorlessautomatically received bythosewho
are located within the region and who participate in the clusters various social and
economic spheres. In this context, actors are not deliberately scanning their
concoction of rumours, impressions, recommendations, trade folklore and strategic
information . . .(Grabher,2002a:209).Itisalmostunavoidabletoreceiveinformation,
rumours and news about other cluster firms and their actions. This does not imply
relevant tothefirmsofacluster.Infact,thediffusionofbuzzwithinaclustercango
smoothly, but it can also be somewhat blocked depending on the structure of social
relations between the local actors and firms and the history of interactions between
them(BatheltandGlckler,2002). Especially incaseswheredistrustandmalfeasance
tions oflocalized productionsystems (e.g., Markusen, 1996), we can also expect that
different types of buzz develop. Of course, buzz is never just saying hello to some
which develop within a cluster. This occurs in negotiations with local suppliers, in
phone calls during office hours, while talking to neighbours in the garden or when
fluid. Co-presence within the same economic and social context generates manifold
opportunities for personal meetings and communication. These meetings can be
planned oroccurspontaneously.Theycanbenondesigned,nontargetedandmoreor
by guest on May 9, 2011 phg.sagepub.com Downloaded from
Harald Bathelt et al. 39
less accidental. Hence, as pointed outby Uzzi (1997: 52), network ties link actors in
multiple ways(asbusinesspartners, friends,agents, mentors),providingameans by
which resources from one relationship can be engaged for another. In investment
situations, these factors increase an actors capacity to access resources, adjust to
unforeseenevents,andevaluate risks.Thisisparticularly thecaseinthecontextofa
cluster which has a rich history of social interaction and offers opportunities for
of social relations stimulate fine-grained information transfer, joint problem-solving
arrangements and the development oftrust andreciprocity (Granovetter,1985; Uzzi,
Therefore, different modes of communication operate in a clusters social and
economic context (e.g., chatting, gossiping, brainstorming, in-depth discussions,
problem analysis).Co-locationandvisibility generate potentials forefficientinterper-
sonal translation of important news and information between the cluster actors and
firms, as has been emphasized by actor-network theory (Latour, 1986; Allen, 1997).
Coherence within the cluster is established by particular learning processes, path
ronments (Murdoch, 1995). This is supported by the actual movement of employees
development of a particular institutional structure shared by those who participate.
Firms develop similar language, technology attitudes and interpretative schemes
(LawsonandLorenz,1999).Also,ashasbeensuggestedelsewhere(Maskellet al.,1998),
by default.
This processofinstitution building is triggered by the establishment of
communities of practice (Brown and Duguid, 1991; Wenger, 1998). These consistof
agents which are bound together through day-to-dayinteraction, based onthe same
expertise, a common set of technological knowledge and similar experience with a
particular setofproblem-solvingtechniques.Suchcommunitiescandevelopwithina
firm but also span a single organization and include other firms of a value chain
(Gertler,2001b). Sharedexperience inthesame technologiesandongoingattempts to
ical paradigms, support the development of mutual engagement, joint enterprise,
generation of distinct routines, conventions and other institutional arrangements.
tencies, shared language and attitudes and the like (Wenger, 1998). This enables
particular decisionstofollowearlier particular decisionsandspecialized communica-
tion to follow earlier specialized communication. Clusters can become important
catalysts fortheformation ofsuchcommunities. Inthis case, they develop into local
by guest on May 9, 2011 phg.sagepub.com Downloaded from
40 Clusters and knowledge
IV Knowledge creation across clusters: the nature of pipelines
emphasized the importance of local networking (e.g., Scott, 1988; Saxenian, 1994;
Maillat et al., 1997, Ratti et al., 1997) relatively few empirical studies have actually
provided convincing empirical evidence of the superiority of local over nonlocal
interaction, aside from some well-known case studies on industrial districts and
has provided evidence that even in regions such as the San Francisco Bay area and
internal transactions are by nomeans dominant over external relations (Oakey et al.,
1988; GrotzandBraun,1993). Notsurprisingly,anincreasingnumberofstudieshave
begun to question the seemingly dominant character of local learning processes
give support to the claims about localized learning. As learning processes are not
empirically documented, the mere agglomeration of successful firms in an industry
seemstobeassumedtosignifylocalized learning.Further,itseemsevident thatthe
creation of new knowledge (learning) might be best viewed as a result of a
combinationofcloseanddistantinteractions(p. 365).
1 The need for pipelines
community that, even though knowledge spillovers may be more effective within a
regional network than across its borders, physical distance is notthe only influence.
Decisive, nonincremental knowledge flows are often generated through network
this concept, Owen-Smith and Powell (2002) have shown in the case of the Boston
biotechnology industry that access tonew knowledge doesnotjust result fromlocal
gional and international reach. Bostons biotechnology firms are thus embedded not
regionally orbyanyotherdiscretespatiallevel.Grabhers(2002b)studyofSohosad
and entertainment cluster, also emphasize the importance of extra-local linkages to
While interaction andinformationexchange withintheclusterthelocal buzz is
the firms. Unlike in the case of local relations between cluster firms (Maskell and
by guest on May 9, 2011 phg.sagepub.com Downloaded from
Harald Bathelt et al. 41
canbenefit.Instead,theestablishment ofglobal pipelines withnewpartners requires
uptrusttakestimeandinvolvescosts(Harrison,1992). Lorenz(1999)haspointedout
transactions and interactions wherein small risks are followed by larger ones and
Now, one could indeed argue that the extra-local knowledge flows (through the
high-technology industry in Silicon Valley it could be expected that the two are
mutually reinforcing. Themorefirmsofacluster engage inthebuildup oftranslocal
into internal networks and the more dynamic the buzz from which local actors
Because of their potential to intensify local interaction, global pipelines
TheGREMIapproachtoinnovativemilieus(Maillat,1998;Camagni,1991;Rattiet al.,
external information sources to maintain the flow of important information about
market trends and new technologies into the milieu (Perrin, 1991; Quvit, 1991).
Otherwise, it is claimed, a milieu runs the risk of stagnating (Maillat, 1998). From a
et al.(2001)have,furthermore,foundthatthereasonsbehindtheestablishmentofanew
found that external effects, agglomeration factors and networking synergies did not
ability to tap into new areas outside the established technologies and markets have
been decisive forthe genesis ofthoseclusters studied.Thesuccessofthese entrepre-
neurial endeavours largely depended on the firms ability to access major markets
outside the cluster in their early stages. In addition, such clusters have been able to
attract specialized skills from other localities and regions. According to the work of
Bresnahan et al. (2001), the openness of cluster relations and active search for large
In a similar vein, Scott (1998) has emphasized that the performance of localized
productionsystems depends onthe right mix oflocal and nonlocal transactions and
that strong growth can only result if external markets are linked to the production
as both are closely intertwined and stimulate one another there still seems tobe a
missinglinkintheconceptualization oftheconnectionbetweeninternal andexternal
by guest on May 9, 2011 phg.sagepub.com Downloaded from
42 Clusters and knowledge
empirical study of the New York textile industry, Uzzi (1996; 1997) has found a
stronger a large group of suppliers is embedded with the same set ofcustomersthe
refers to these network relations as plumbing through which information and
resourcesare being transmitted. Kern(1996) emphasizes thatattempts toconsciously
certain amountofdistrustwithrespecttotraditional solutions,is important toavoid
lock-in, while Malecki (2000) draws attention to the significant difference between
introvert and extrovert firms. Even if a firm over time achieves a sufficiently
nevertheless only able to handle a limited number of external linkages, as Grabher
(2001; 2002b), for instance, observed when studying Londons advertising industry.
This is because the establishment and maintenance of external linkages requires
Overall, communicationprocessesinglobal pipelines arecontingentbynatureand
are plagued by great uncertainty. Common institutions and procedural rules are
time thisenables fine-grained informationtransfer andcooperationin morecomplex
projects. Because global pipelines encompass firms fromdifferent parts of the world
which are embedded in different socio-institutional and cultural environments they
changesandnewandbetterproductsandprocessesevolve.Onceapipeline hassuc-
theroutinesofthelocalcluster.A firmmaychoosebetweendifferenttechnologicaland
organizational optionsandselectthosewhicharerobusttowardschangesintheinsti-
variety of spontaneous and unanticipated situations where firms interact and form
interpretativecommunities(Nonakaet al.,2000).Theadvantagesofglobalpipelinesare
instead associated with the integration of multiple selection environments that open
differentpotentialities and feedlocal interpretation and usage ofknowledge hitherto
residing elsewhere. Malecki (2000: 341) concludesthisline ofreasoning whenstating
that[s]omeplacesareabletocreate,attract, andkeepeconomicactivity. . .[particu-
larly]becausepeopleinthoseplacesmakeconnectionswithotherplaces . . ..
by guest on May 9, 2011 phg.sagepub.com Downloaded from
Harald Bathelt et al. 43
2 The limitations to pipeline formation
firms requires conscious efforts. In contrast to the type of communication and
interaction that often occurswithin the cluster, costconsiderations tend tomake the
knowledge flowsandinteraction inglobal pipelines targetedtowardsacertain, often
the local buzz, information flowing through global pipelines has an intrinsic bias
towards filtering information offailures even when knowledge ofwhat went wrong
mustbepredesigned andplanned inadvance, andtheyrequire specific investments.
ofpotential partners isusually truncatedandtheknowledge ofthesefirmsandtheir
Firms have to be able to understand different institutional regimes in order to
communicate and interact with actors in other parts of the world through global
notthateasy toachieve asfirmstapintoparticular cultures whichcanvarybetween
places, regions and states. The case studies reviewed by Schoenberger (1999) and
systematic influences of institutions, especially between different national environ-
ments, prevent the diffusion of universal operational standards or a single best
practice. The nation-state (whether home or host) is still a primary source of
influenceoverindustrial practices(Gertler 2001b: 16). Infact,itisquitepossible that
through a lack of understanding of different institutional contexts communication
might be almost impossible. In this case, firms would continue speaking different
languages. It would be impossible totranslate messages between these institutional
Tosuccessfully establish a global pipeline therefore requires the development of a
shared institutional context which enables joint problem-solving, learning and
knowledge creation. This involves intense efforts to develop joint action frames and
projects. Through experience trust may gradually develop over time enabling more
can develop and handle. To avoid the risk of costly failure in establishing global
to properly select potential partners for the establishment of global pipelines. We
Glckler andArmbrster, 2003). Alternatively, firms can also scan their environment
through a mobilization of weak ties (Granovetter, 1973) or use regular conventions
by guest on May 9, 2011 phg.sagepub.com Downloaded from
44 Clusters and knowledge
and trade fairs to establish contact with potential partners which they have known
in distant clusters in order to develop local linkages within new locations and their
respective markets. Thenewlinkages mayhave the character ofextra-local pipelines
but they may also benefit from knowledge transfers between the buzz of different
Since theestablishment ofapipeline isaconsciousattempt toovercomeidentified
shortcomings in the local knowledge base and fulfil certain goals and expectations,
actors are also prepared to make special efforts to bridge cognitive distance. While
interaction requires thatactorsatbothendsofthepipeline havesomemixofsimilar
and nonsimilar knowledge (Nooteboom, 2000; Nonaka et al., 2000), we may assume
buzz (Lawson and Lorenz, 1999). It seems particularly important in the selection of
a minimum amount of nonsimilar knowledge. Only in this case will the necessary
3 Absorptive capacity
Identifying the value and location of external knowledge and building pipelines to
a firms innovative capability. An equally immense task is to establish the ability to
assimilate the information arriving through pipelines and to apply it successfully
towardscommercial ends.A firmsabsorptive capacity(CohenandLevinthal, 1990)
acrossand within departments and subunits which may be removed fromthe point
spanners becomescrucial fortranslating externally producedknowledge into aform
betweenthefirmsenvironmentanditsorganizational adaptation.
Leipzig media industry cluster, actors and firms are clearly lacking this capability to
systematically access external markets and establish pipelines with extra-local
customers (Bathelt, 2001). This lack of absorptive capacity, which is related to the
theaptnesstocopewithknowledge thatcannoteasily bebrokendownintoseparate
by guest on May 9, 2011 phg.sagepub.com Downloaded from
Harald Bathelt et al. 45
expertise also affects how knowledge, which arrives through the pipelines and is
who receive it in the various departments andsubunitsofthefirm.The background
knowledge whichisrequired bythegroupofemployees asawholewhenaiming at
maximizing thevalueofinformationconsistsofsharedlanguageandsymbols;but,if
allindividualsinthefirmsharepreciselythesamespecialized languageandsymbolic
representations, they will notbe able totapinto diverse external knowledge sources
V A buzz-and-pipeline model of cluster competitiveness: concluding discussion
We can now sum our argument in terms of a number of points which form an
other firms and organizations which are involved in similar and related types of
Figure 1 exemplifies a cluster of interrelated actors and firms which are bound
Thisbuzz encourages thedevelopment ofshared values, attitudes and interpretative
interactive learning and problem-solving, and give meaning to complex information
about changes in markets and technologies. Everyday communication patterns and
routines also serve to create coherence within the cluster through the powers of
association, in the language of actor-network theory (Latour, 1986). Figure 1 also
Spaces of shared meaning and identity are established through ongoing interactions
al arrangements may easily include actors which are located outside the region. In
1 Main propositions of the model
by other actors with relevant similar and nonsimilar skills and competencies. These
for the buzz to be valuable, it has to be comprehensible and include enough useful
tary and heterogeneous knowledge, skills and information reside, provides a great
pipelinestructure.A well-developedsystemofpipelinesconnectingthelocalclusterto
by guest on May 9, 2011 phg.sagepub.com Downloaded from
46 Clusters and knowledge
state-of-the-art knowledge creation. New and valuable knowledge will always be
createdinotherpartsoftheworldandfirmswhocanbuildpipelines tosuchsitesof
global excellence gain competitive advantage. Second, itseemsreasonable toassume
that the information that one cluster firm can acquire through its pipelines will spill
themoredevelopedthepipelines betweentheclusteranddistantsitesofknowledge,
The third argument highlights the intrinsic trade-off between a toomuch inward-
looking and a too much outward-looking organizational structure. In the first case,
knowledge is easily transmitted throughout the firm, but new external knowledge
ly useful knowledge. However, if the buzz is sufficiently intense some such derelict
ofcommunicationusingunconventionalinterpretative schemes.Somemayevenspill
receiver everbenefiting fromtheparticular transfer.Theinvestmentsneededinother
Figure1 Thestructureanddynamicsoflocalbuzzandglobalpipelines
by guest on May 9, 2011 phg.sagepub.com Downloaded from
Harald Bathelt et al. 47
pipelines thatanindividual firmcanmanagesimultaneously.Onecouldhypothesize
that a large number of related independent firms in a cluster can manage a larger
2 Countervailing forces and limitations
notionofanupperlimittothebenefitsofspatial clustering.Thelargerthenumberof
similar and related firms in aspatial cluster,themore vibrant and valuable thelocal
betterdevelopedthesepipelines, themorerefinedthebuzz. Couldwethinkofcoun-
tervailing forceswhichlimit thebenefits ofspatial clustering? Inthatcase, theabove
hypotheseswouldhavetobemodifiedinsuchawaythatabalance betweenvibrant
buzz andinformationoverloadorbetweenglobalopennessandinternal coherenceis
so overcrowded that there is too much buzz? Then it would be difficult for an
individual actortomakesenseofthebuzzandidentifyimportantinformation.Cana
cal, physical and social problems which showup in a lack of direction, absence ofa
commoninterpretativeframework,paralysisofanalytical capacity,ongoingsearchfor
moreinformation,increased anxiety andpoordecision-making (BuchananandKock,
unable totake action because they do nothave that information necessary toenable
them to act efficiently. The information-rich actors are also paralysed simply due to
include the use of various types of filters to separate important and relevant
information from irrelevant, unclear and inaccurate fragments of information (cf.
processes (Witt, 1999), based on the common cognitive frame developed over time
among the firms in the cluster. Knowledge that has passed this cognitive filter is
by guest on May 9, 2011 phg.sagepub.com Downloaded from
48 Clusters and knowledge
evaluated every time it is used by critical local actors. Unreliable or otherwise low-
quality knowledge tends to be weeded out before it enters into general usage. Even
face to face has already been tested for relevance and customized to the receiver.
Another possible countervailing force revolves around the relation between well-
developed pipelines and the quality of the local buzz. While external linkages can
supportregional growthprocesses,whentheyare toostrongtheycouldthreaten the
long-termexistenceofacluster.Strongexternallinkages couldthenprovokesegmen-
term future (Bathelt and Taylor, 2002). The argument here is that when actors focus
primarily on external linkages global pipelines begin to dominate the local milieu.
Therefore, lessattention is being paid tolocal communication andinformationflows
Still, thisdanger isreal.A cluster which is moreorlessempty because itsimportant
actorsareconstantlytravelling theworldinordertobuildandmaintain anextensive
3 Policy implications
spatial clustersofsimilar andrelatedeconomicactivityhavebeenabundantinrecent
encouraging and developing mechanisms that promote the potential for interactive
learning and knowledge creation across firms and other organizations within a
needs to be addressed in local development policies. The local buzz is certainly
of buzz will automatically result (even in prisons, where the inmates are kept apart
Incontrast,itisespecially thedevelopmentofglobalpipelineswhichrequiresinsti-
tutional and infrastructure support. Of course, we do not suggest that cluster firms
shouldbeexclusively outward-orientated asthis wouldreduceinternal cohesiveness
by guest on May 9, 2011 phg.sagepub.com Downloaded from
Harald Bathelt et al. 49
so predisposed toward local networking that the importance of external, translocal
the possibilities of stimulating pipeline development rather than to make extensive
efforts in generating and promoting local buzz through various forms of social
Earlier versions of this paper were presented in September 2002 at the Second
at theAnnual Meeting oftheAssociationofAmerican Geographers in New Orleans.
The opinions and ideas presented in this paper are solely those of the authors, who
contributedequally tothepaper.Theauthorsofthispaper acknowledge thesupport
Foundation and Vinnova, and Centre for Economic and Business Research (CEBR),
1. Cf.Malmberg and Maskell (1997;2002),Maskell et al. (1998),Maskell and Malmberg (1999a,
1999b),Maskell (2001a;2001b),Bathelt (2001;2002),Bathelt andGlckler (2002),Bathelt andTaylor
2. We are aware that tacit and codified knowledge are certainly not independent categories of
knowledge (e.g., Nonaka and Takeuchi, 1995;Maskell and Malmberg, 1999a).In contrast, they are
interdependent, complementaryanddialecticallyinterwoven.Forourargument,however,itsuffices
to start with a simple tacit-versus-codified knowledge dichotomy and its problematic local-global
implications in order to move on to a more developed argument tackling the complex interdepen-
3. This is,forinstance, the casein recent theories ofincreasing returns andendogenous growth
andis spilling over intothemodelsdeveloped within the branchofeconomicswhich issometimes
referredtoasneweconomicgeographyorgeographicaleconomics(Fujitaet al.,1999).
4. Ifneitherdecisionsnorinvestments wereneeded,codifiedknowledgewouldindeedbecomea
ubiquity and, as realized since Weber (1909)published his seminal work, would be without much
5. Theremaybeseveral,mutuallyreinforcing,reasonsfortheeasewithwhichknowledgeiscom-
develop a common language, joint interpretative contexts and a shared knowledge basis. Grabher
1998).Suchfeatureshave,furthermore, been proposed asageneral explanationfortheexistence of
6. Thetendencyinsomecontemporarywritingsonclustersandknowledgecreationtoemphasize
localizedlearning andagglomeration factorsratherthanextra-localinfluences, thatarenotdefined
by guest on May 9, 2011 phg.sagepub.com Downloaded from
50 Clusters and knowledge
7. Despite this conclusion,Humphrey and Schmitz(2002)primarily deal with this shortcoming
fromaglobalvaluechainperspectiveproviding ananalysisofthelocalizedeffectsandprospectsof
8. Thelineofargumentinthisandthetwofollowingsectionsstems,ofcourse,fromAdamSmiths
(1776)initialconjectureregarding thelearningadvantagesassociatedwiththedivisionoflabourand
9. SeealsoKieandHynes(1996)foranempiricallyfoundedcritiqueoftheverticaldisintegration
10. Thisgeneraltendencydoesnot,ofcourse,implythetotalabsenceofreal-lifesituationswhere
11. Nonaka et al. (2000)use the Japanese concept of ba to refer to the organizational contexts
within which individuals interact at a specific time and place over a certain time period when
12. Foracritique,seeMartinandSunley(2003).
13. Forachallengingdifferentinterpretation,seeKlepper(2002).
14. Marshallsinitialreflectionshavebeenfollowedbycontributionsfromavastrangeofscholars
15. Theclusterdimensionsarenotindependentfromoneanother.Theymayreinforceoneanother
buttherecanbealsotrade-offsbetweenopenness andembeddedness, cooperationandcompetition,
16. Ofcourse,casestudiesdoexistwhichprovideevidenceofimportantintraregionaltransactions
betweenaclustersactors,suchasthoseinsomeItalianindustrialdistricts(e.g.,Goodmanet al.,1989;
Pykeet al.,1990;AminandThrift,1992).
17. Even in the San Francisco Bay area, which encompasses Silicon Valley, local transactions
between high-technology firms are much lower than would be expected. In a survey of high-
technology small firms, Oakeyet al.(1988)foundthatmore than45%and70%oftheir respondents
the cost advantageofintraregional input-outputlinkages, whatarethen the incentives for firms to
18. FollowingLorenz(1999:305),weviewtrustasthejudgementonemakesonthebasisofones
past interactions with others thatthey will seek to actin waysthat favourones ideas,rather than
harmthem . . ..Trustmeansthatanactorwithinaclusterknowsthathe/shecaninallprobabilityrely
on others in terms of production quality and capabilities.Firms would of course also be aware of
19. In a case study of the restructuring in the Lyon machine-building industry, Lorenz (1999)
shows howtrustcandevelop through suchproceduralrules betweenproducers andsuppliers asa
20. Ofcourse,theparticularmixoflocalbuzzandglobalpipelinescanvaryacrossvaluechains,
technologies and markets segments. Some industries (e.g., fashion products) might, for instance,
require more buzz than others (e.g., steel production), while other industries would need more
continued growth and innovation. A particularly interesting combination of localbuzzand global
pipelines which is mutually reinforcing can be found in New York Citys womens-wear industry
which forms a distinct cluster in the Garment District. The creativity of this cluster is based on
interaction between the buyers, designers, specialized producers and services, enriched by
impressions from fashion magazines, designer schools, trade shows and the like (Uzzi,1996).This
location provides an advantage for firms as it generates openness and swift access to external
influences just by being there. Due to its creativity and leadership in design, the womens-wear
industryattractsdesigners andothercreativesfromallovertheworld.Unlikeinmanyotherplaces,
by guest on May 9, 2011 phg.sagepub.com Downloaded from
Harald Bathelt et al. 51
21. Onecouldarguethatthebuzz-and-pipelineargumentcarriedoutinthispaperisatoddswith
Granovetters (1973) classic work on the strength of weak ties. Therein, Granovetter (1973) has
innovations which wouldotherwise beslowed down by strong ties in the network core.Ofcourse
the actorsof acluster. For the same reasons, weakties are alsoextremely important when making
decisions aboutwhich outside technologies andmarketstotapintoandwhich externalpartners to
22. Incontrastwiththeinformationflowsinglobalpipelines,thelocalbuzzspreadsinformation
andengageinglobalpipelines is,tosomeextent,morerisky becauseinformationflowsaboutother
firms are biased towards successful endeavours at this level and tend to overlook the not-so-
23. SeevandenBoschet al.(1999).Aage(2001)appearstobethefirstwhointroducedtheliterature
Aage, T. 2001: External relations and industrial
districts. Paper presented at the Danish Re-
search Unit on Industrial Dynamics (DRUID)
Nelson and Winter Summer Conference in
dk/conferences/nw/paper1/aage.pdf (last
Lee, R. and Wills, J., editors, Geographies of
Almeida,P. and Kogut, B.1999: Localizationof
knowledge and the mobility of engineers in
regional networks. Management Science 45(7):
Amin, A. and Thrift, N. 1992: Neo-Marshallian
Asheim, B. 1999: Interactive learning and
localised knowledge in globalising learning
media industry. Institut fr Wirtschafts- und
Sozialgeographie. Universitt Frankfurt Main
www. uni - f r ankf ur t . de /f b11 /wi geo /
Bathelt, H. and Glckler, J. 2002:
Wirtschaftsgeographie: konomische Beziehungen
in rumlicher Perspektive (Economic geography:
economic relations in spatial perspective).
Bathelt,H.and Taylor,M.2002:Clusters,power
Belussi, F. and Pilotti, L. 2002: Knowledge
creation, learning and innovation in Italian
industrial districts. Geografiska Annaler 84B,
Bresnahan, T., Gambardella, A. and Saxenian,
A. 2001: Old economy inputs for new
economy outcomes: cluster formation in the
new Silicon Valleys. Industrial and Corporate
by guest on May 9, 2011 phg.sagepub.com Downloaded from
52 Clusters and knowledge
Brown,J.S.and Duguid,P.1991:Organizational
unified view of working, learning, and
Buchanan,J.and Kock,N.2001:Multiplecriteria
decision making in the new millennium.
Lecture notes in Economics and Mathematical
of competition. Cambridge, MA, and London:
Camagni, R., editor 1991: Innovation networks:
spatial perspectives. London and New York:
Chandler, A.D. 1962: Strategies and structure.
Chapters in the history of the international
Chapman, K. and Walker, D. 1987: Industrial
Clark,G.L.and Tracey,P.2004:Globalcompetitive-
ness and innovation: an agent-centred perspec-
tive. Houndsmill and New York: Palgrave
Cohen, W.M. and Levinthal, D.A. 1990:
Absorptive capacity: a new perspective on
Crevoisier, O. and Maillat, D. 1991: Milieu,
industrial organization and territorial
production system: towards a new theory of
spatial development. In Camagni, R., editor,
Dicken, P. 2001: Placing firms firming places.
Paper presented at the Meeting of German
Geographers in Leipzig, 29 September5
Durham, W.H. 1991: Co-evolution: genes, culture
and human diversity. Stanford: Stanford
1992:Applicationsof evolutionarycultural
theory. Annual Review of Anthropology 19,
Eccles, R.G. 1983: Transfer pricing, fairness and
control. Working Paper 167. Cambridge,
MA: Harvard Business School, Harvard
Edmunds,A.and Morris,A.2000:Theproblem
of information overload in businessorganisa-
tions: a review of the literature. International
firms: the case of Boeing 19631968. Land
Freeman, C. 1982: The economics of industrial
Fuchs, M. 2001: Von der lernenden Region zur
lernenden Organisation (From the learning
region to the learning organization). Institut fr
Entwicklung und Frieden, Universitt
Fujita,M.,Krugman,P.and Venables,A.J.1999:
The spatial economy. cities, regions and inter-
Gertler, M.S. 1993: Implementing advanced
manufacturing technologies in mature
industrial regions: towards a social model of
technology production. Regional Studies 27,
andcultureinthe developmentandadoption
of advanced manufacturing technologies.
2001: Best practice? Geography, learning
and the institutional limits to strong
convergence. Journal of Economic Geography 1,
the economic geography of context, or the
interindustriallinkageand decisionmakingin
manufacturing. In Hamilton, F.E.I., editor,
Glckler, J. and Armbrster, T. 2003: Bridging
uncertainty in management consulting: the
mechanisms of trust and networked
Goodman,E.,Bamford,J.and Saynor,P.,editors
Gordon, I.R. and McCann, P. 2000: Industrial
clusters: complexes, agglomeration and/or
Goulding, A. 2001: Information poverty or
overload? Journal of Librarianship and
Grabher, G. 1993: The weakness of strong ties:
area.InGrabher,G.,editor,Theembedded firm.
On the socioeconomics of industrial networks,
by guest on May 9, 2011 phg.sagepub.com Downloaded from
Harald Bathelt et al. 53
group,andtheheterarchicorganisationof the
British advertising industry. Environment and
PlanningA 33,35174.
2002a: Cool projects, boring institutions:
temporary collaboration in social context.
2002b: The project ecology of advertising:
tasks, talents and teams. Regional Studies 36,
1985: Economic action and economic
structure: the problem of embeddedness.
Grotz,R.and Braun,B.1993:Networks,milieux
and individual firm strategies: empirical
evidence of an innovative SME environment.
Hagedoorn,J.and Schakenraad,J.1992:Leading
companiesandnetworksof strategicalliances
Hkansson, H. 1989: Corporate technological
behaviour: co-operation and networks. London:
Hoover, E.M. 1970: An introduction to regional
Humphrey, J. and Schmitz, H.2002: How does
insertion in global value chains affect
upgrading in industrial clusters? Regional
Karaska, G.J. 1969: Manufacturing linkages in
the Philadelphia economy: some evidence of
external agglomeration forces. Geographical
Kern, H. 1996: Vertrauensverlust und blindes
chen Handeln (Loss of trust and blind
confidence in economic action). SOFI-
Kie,C.and Hynes,M.1996:Whereisthedisinte-
gration? Growth and Change, Winter 1996,
Klepper, S. 2002: The evolution of the U.S.
automobile industry and Detroit as its capital.
Paper presented at the DRUID Winter
Conference in Aalborg, 1719 January 2002.
URL: http://www.druid.dk/conferences/
winter2002/gallery/klepper.pdf (last accessed
Kline,S.J.and Rosenberg,N.1986:Anoverview
of innovation. In Landau, R. and Rosenberg,
development network of an urban region.
Krugman, P. 1991: Geography and trade. Leuven:
2000: Where in the world is the new
M.P. and Gertler, M.S., editors, The Oxford
Latour, B. 1986: The powers of association. In
Law, J., editor, Power, action and belief: a new
sociology ofknowledge?London:Routledgeand
Lawson, C.1999: Towards a competence theory
Lawson, C. and Lorenz, E. 1999: Collective
learning, tacit knowledge and regional
Loasby, B.J. 2000: Organisations as interpretative
systems. Paper presented at the DRUID
Summer Conference in Rebild, 1517 June
2000. URL: http://www.druid.dk (last
economic. Graz Schumpeter Lectures. London:
Lorenz, E. 1999: Trust, contract and economic
Lundvall, B.-. 1988: Innovation as an
interactive process: from user-producer
interaction to the national system of
innovation. In Dosi, G., Freeman, C., Nelson,
R.R., Silverberg,G. and Soete, L.L.G., editors,
Technical change and economic theory, London:
Lundvall, B.-. and Maskell, P. 2000: Nation
states and economic development from
national systems of production to national
Clark, G.L., Feldman, M.P. and Gertler, M.S.,
editors, The Oxford handbook of economic
geography, Oxford: Oxford University Press,
Maillat, D. 1998: Vom Industrial District zum
innovativen Milieu: Ein Beitrag zur Analyse
der lokalen Produktionssysteme (From
by guest on May 9, 2011 phg.sagepub.com Downloaded from
54 Clusters and knowledge
industrial networks to innovative milieus:
towards an analysis of territorial production
Maillat,D.,Lchot,G.,Lecoq,B.and Pfister,M.
1997: Comparative analysis of the structural
the Swiss and French Jura Arc. In Ratti, R.,
Bramanti, A. and Gordon, R., editors, The
dynamics of innovative regions: the GREMI
approach. Aldershot and Brookfield: Ashgate,
Malecki,E.J.and Oinas,P.,editors1999:Making
connections. Technological learning and regional
Malmberg,A.and Maskell,P.1997:Towardsan
explanation of industry agglomeration and
regional specialization. European Planning
2002: The elusive concept of localization
of spatialclustering.EnvironmentandPlanning
A 34,42949.
Malmgren,H.B.1961:Information, expectations
Markusen, A. 1996: Sticky places in slippery
space: a typology of industrial districts.
Marshall,A.1920:Principles ofeconomics (eighth
technique and business organization; and their
nations (thirdedition).London:Macmillan.
Martin, R. and Sunley, P. 2003: Deconstructing
clusters: chaotic concept or policy panacea?
Maskell, P. 2001a: Towards a knowledge-based
theory of the geographical cluster. Industrial
geographic clusters. International Journal of
Maskell,P.and Lorenzen,M.2003:Thecluster
and other current forms of market organization.
Paper presented at the Regional Studies
Association International Conference on
Maskell, P. and Malmberg, A.1999a: The com-
1999b: Localised learning and industrial
Maskell, P., Eskelinen, H., Hannibalsson, I.,
Malmberg, A. and Vatne, E. 1998:
Competitiveness, localised learning and regional
open economies. London and New York:
Men, J. 2001: Is mobility of technical personnel a
source of R&D spillovers? Washington, DC:
Murdoch, J. 1995: Actor-networks and the
evolution of economic forms: combining
description and explanation in theories of
regulation, flexible specialization, and net-
works.EnvironmentandPlanningA 27,73157.
Nonaka,I.and Takeuchi,H.1995:Theknowledge-
Nonaka, I.,Toyama, R. and Nagata, A. 2000:A
firm as a knowledge-creating entity: a new
Nooteboom, B. 2000: Learning and innovation in
organizations and economies. Oxford: Oxford
Oakey,R.,Rothwell,R.and Cooper,S.1988:The
management of innovation in high-technology
small firms: innovation and regional development
OECD 1992: Industrial policy in the OECD
countries. Annual review. Paris: The
Organisation for Economic Co-Operation and
Oinas, P. 1999: Activity-specificity in organiza-
tional learning:implications for analysingthe
Owen-Smith, J. and Powell, W.W. 2002:
Knowledge networks in the Boston biotechnology
community.Paperpresentedat theConference
on Science as an Institution and the
Pavitt, K. 1999: Division of labour in the
innovating firm. In Dow, S.C. and Earl, P.E.,
the firm. Essays in honour of Brian J. Loasby,
Penrose, E. 1959: The theory of the growth of the
Perrin, J.-C. 1991:Technological innovation and
territorialdevelopment:anapproach interms
of networks and milieux. In Camagni, R.,
by guest on May 9, 2011 phg.sagepub.com Downloaded from
Harald Bathelt et al. 55
editor, Innovation networks: spatial perspectives.
Porter, M.E. 1990: The competitive advantage of
1998: Clusters and the new economics of
competition. Harvard Business Review 76
2000: Locations, clusters, and company
strategy. In Clark, G.L., Feldman, M.P. and
Gertler, M.S., editors, The Oxford handbook of
Pred, A. 1976: The interurban transmission of
growth in advanced economies: empirical
findings versus regional-planning
Pyke, F., Becattini, G. and Sengenberger, W.,
operation in Italy. Geneva: International
Quvit, M. 1991: Innovative environments and
local/international linkages in enterprise
strategy: a framework for analysis. In
perspectives, London and New York: Belhaven
source of variety: openness and adaptability
inNewYork CitysGarmentDistrict.Regional
Ratti, R., Bramanti, A. and Gordon, R., editors
1997: The dynamics of innovative regions: the
GREMI approach. Aldershot and Brookfield:
and economics. Cambridge and New York:
Rosenkopf,L.and Almeida,P.2001:Overcoming
local search through alliances and mobility.
Philadelphia: The Warton School (Mimeo),
Saxenian, A.L. 1994: Regional advantage. Culture
and competition in Silicon Valley and Route 128.
Cambridge, MA and London: Harvard
Schickhoff, I. 1983: Materialverflechtungen von
Industrieunternehmen. Eine empirische Unter-
firms. An empirical investigation in the Lower
Rhine Valley). Duisburg: Habilitationsschrift,
the region in the firm. In Barnes, T.J. and
Gertler, M.S., editors, The new industrial
geography: regions, regulation and institutions,
Scott, A.J. 1988: New industrial spaces: flexible
production organization andregional development
in North America and western Europe. London:
1998: Regions and the world economy: the
political order. Oxford and New York: Oxford
2002: A new map of Hollywood: the
production and distribution of American
Simon, H.A. 1985: What we know about the
in creative and innovative management,
Smith, A. 1776: An inquiry into the nature and
causes ofthewealthofnations.London: Strahan
Staber, U. 1997: Specialization in a declining
industrial district. Growth and Change 28,
Storper, M. 1995: The resurgence of regional
1997: The regional world. Territorial
Storper, M. and Venables, A.J. 2002: Buzz: the
DRUID Summer Conference on Industrial
is Embracing Whom? in Copenhagen,
Teece, D.J. 1980: Economies of scope and the
scope of the enterprise. Journal of Economic
Uzzi, B. 1996:The sources and consequences of
embeddedness for the economic performance
of organizations: the network effect. American
1997: Social structure and competition
in interfirmnetworks:the paradox of embed-
dedness. Administrative Science Quarterly 42,
van den Bosch, F.A.J., Volberda, H.W. and de
Boer,M.1999:Coevolutionof firmabsorptive
by guest on May 9, 2011 phg.sagepub.com Downloaded from
56 Clusters and knowledge
zational forms and combinative capabilities.
importance of ties to information and the institu-
tional and territorial structure of technological
of the IGU Commission on the Dynamics of
Local Development on Local Development:
Issues of Competition, Col-laboration and
von Hayek, F.A. 1937: Economics and
knowledge. Economica IV (New Series) (13),
1945: The use of knowledge in society.
Weber, A.1909:ber denStandortder Industrien.
Erster Teil: Reine Theorie des Standorts (On the
meaning, and identity. Cambridge: Cambridge
Witt, U. 1999: Do entrepreneurs need firms? A
economics. Review of Austrian Economics 11,
by guest on May 9, 2011 phg.sagepub.com Downloaded from
Research Policy 31 (2002) 14171429
Global production networks, knowledge diffusion,
and local capability formation
Dieter Ernst
, Linsu Kim
East West Center, Honolulu, HI 96848, USA
College of Business Administration, Korea University, Seoul 136-701, South Korea
Received 15 October 2001; received in revised form 7 January 2002; accepted 11 February 2002
This paper develops a conceptual framework that explores the linkage between the evolution of global production networks
(GPN), the role of network agships in transferring knowledge, and the formation of capabilities by local suppliers. GPN
are a major innovation in the organization of international business. These networks combine concentrated dispersion of the
value chain across the boundaries of the rm and national borders, with a parallel process of integrating hierarchical layers of
network participants. The network agships transfer both explicit and tacit knowledge to local suppliers through formal and
informal mechanisms. This is necessary to upgrade the local suppliers technical and managerial skills, so that they can meet
the agships specications. We also examine how GPN can act as mediators in the capability formation of local suppliers.
2002 Elsevier Science B.V. All rights reserved.
Keywords: International business strategy; Global production networks; Multinational Corporations; Organizational knowledge; Knowledge
diffusion; Capability formation; Local suppliers
1. Introduction
Multinational corporations (MNCs) have been
around for a long time (e.g. Wilkins, 1970). Until
recently, their international production has focused
on the penetration of protected markets through
tariff-hopping investments, and on the use of assets
developed at home to exploit international factor
cost differentials, primarily for labor (e.g. Dunning,
1981). This has given rise to a peculiar pattern of
international production: offshore production sites in
low-cost locations are linked through triangular trade

Corresponding author.
E-mail addresses: ernstd@eastwestcenter.org (D. Ernst),
linsukim@united.co.kr (L. Kim).
with the major markets in North America and Europe
(e.g. Dicken, 1992).
A progressive liberalization and deregulation of
international trade and investment, and the rapid de-
velopment and diffusion of information and commu-
nication technology (IT) have fundamentally changed
the global competitive dynamics, in which MNCs op-
erate. While both market access and cost reductions
remain important, it became clear that they have to
be reconciled with a number of equally important re-
quirements that encompass: the exploitation of uncer-
tainty through improved operational exibility (e.g.
Kogut, 1985; Kogut and Kulatilaka, 1994); a com-
pression of speed-to-market through reduced product
development and product life cycles (e.g. Flaherty,
1986); learning and the acquisition of specialized
external capabilities (e.g. Antonelli, 1992; Zander
0048-7333/02/$ see front matter 2002 Elsevier Science B.V. All rights reserved.
PII: S0048- 7333( 02) 00072- 0
1418 D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429
and Kogut, 1995); and a shift of market penetra-
tion strategies from established to new and unknown
markets (e.g. Christensen, 1997).
In response to the increasingly demanding re-
quirements of global competition, three interrelated
transformations have occurred in the organization
of international economic transactions. First, global
production networks (GPN) have proliferated as a
major organizational innovation in global operations
(e.g. Borrus et al., 2000). Second, these networks
have acted as a catalyst for international knowledge
diffusion, providing new opportunities for local ca-
pability formation in lower-cost locations outside
the industrial heartlands of North America, Western
Europe and Japan. Third, a long-term process of dig-
ital convergence (e.g. Chandler and Cortada, 2000),
enabling the same infrastructure to accommodate
manipulation and transmission of voice, video, and
data, has created new opportunities for organizational
learning and knowledge exchange across organiza-
tional and national boundaries, hence magnifying the
rst two transformations (Ernst, 2002c).
The combination of these three transformations has
changed dramatically the international geography of
production and innovation. We focus on the rst two
of these transformations. The rst transformation sig-
nals a new divide in industrial organization: a transi-
tion is under way from multinational corporations,
with their focus on stand-alone overseas investment
projects, to global network agships that integrate
their dispersed supply, knowledge and customer bases
into global (and regional) production networks (Ernst,
1997, 2002a). There is a growing acceptance in the
literature that, to capture the impact of globalization
on industrial organization and knowledge diffusion,
the focus of research needs to move from the indus-
try and the individual rm to the international dimen-
sion of business networks (e.g. Ghoshal and Bartlett,
1990; Rugman and DCruz, 2000). Our understanding
of these networks is limited. Most studies have focused
too narrowly on the perspective of the network ag-
ship (agship bias). We need research that explores
as well implications for network suppliers, especially
lower-tier suppliers from developing countries.
Equally important is the second transformation:
GPN in their operations reportedly disseminate im-
portant knowledge to local suppliers in low-cost loca-
tions, which could catalyze local capability formation.
Knowledge transfer, however, is not automatic. It re-
quires a signicant level of absorptive capacity on the
part of local suppliers and a complex process to in-
ternalize disseminated knowledge. Our understanding
of knowledge transfer and local capability formation
is limited. International knowledge transfer has been
extensively studied, but research has primarily fo-
cused on such formal mechanisms as foreign direct
investment (FDI) and foreign licensing (FL) (Reddy
and Zhao, 1990). These formal mechanisms, how-
ever, are only the tip of the iceberg. A larger amount
of knowledge is transferred through various informal
mechanisms (Westphal et al., 1985; Kim, 1991; Ernst,
2000a). Research on informal knowledge transfer is
scarce. The importance of local capabilities in assim-
ilating, adapting, and improving imported technology
has long been recognized, but few studies exist on
the complex process of local capability formation in
developing countries (e.g. Kim, 1997).
GPN transform the production and use of knowl-
edge, with far-reaching implications for an evolu-
tionary theory of economic change. There is a fun-
damental trend towards an increasing mobility of
knowledge, yet little do we know about drivers and
implications. A major constraint is a lack of com-
munication between research on GPN, research on
international knowledge diffusion, and research on
local capability formation. While all three are highly
relevant strands of research, their lack of interaction
obstructs our understanding of how global networks
affect knowledge diffusion and the formation of local
capabilities. There is a need to bridge this gap through
appreciative theories, as dened in Nelsons (1995)
thought-provoking review of economic growth theory.
This paper develops a conceptual framework that
links together the above three areas of research, as
a rst step towards an appreciative theory. We argue
that globalization has culminated in an important or-
ganizational innovation: the spread of GPN. These
networks combine concentrated dispersion of the
value chain across rm and national boundaries, with
a parallel process of integration of hierarchical layers
of network participants. This has created new op-
portunities for international knowledge diffusion that
lower-tier network suppliers should strive to exploit.
To substantiate this argument, we proceed as follows.
Section 2 analyzes the three dynamic forces that drive
the rapid development of GPN and highlights the
D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429 1419
characteristics of the agship model of GPN. Section 3
explores the categories of knowledge, and the mecha-
nisms of knowledge transfer from agship companies
to local network suppliers. And in Section 4, we dis-
cuss how GPN can act as mediators of local capability
2. Driving forces and characteristics of GPN
2.1. Driving forces
What has driven the shift in industrial organiza-
tion from multinational corporations to global
network agships that integrate their dispersed sup-
ply, knowledge and customer bases into global (and
regional) production networks? To answer this ques-
tion, we introduce a stylized model of globalization
drivers, focusing on three inter-related explanatory
variables: institutional change through liberalization,
information technology, and competition.
The rst driving force is liberalization, which in-
cludes four main elements: trade liberalization; liberal-
ization of capital ows; liberalization of FDI policies;
and privatization. While each of these has generated
separate debates in the literature, they hang together.
Earlier success in trade liberalization has sparked an
expansion of trade and FDI, increasing the demand
for cross-border capital ows. This has increased the
pressure for a liberalization of capital markets, forc-
ing more and more countries to open their capital ac-
counts. In turn this has led to a liberalization of FDI
policies, and to privatization tournaments.
The overall effect of liberalization has been a
considerable reduction in the cost and risks of in-
ternational transactions and a massive increase in
international liquidity. Global corporations (the net-
work agships) have been the primary beneciaries:
liberalization provides them with a greater range of
choices for market entry between trade, licensing,
subcontracting, franchising, etc. (locational special-
ization) than otherwise; it provides better access to ex-
ternal resources and capabilities that a agship needs
to complement its core competencies (outsourcing);
and it has reduced the constraints for a geographic
dispersion of the value chain (spatial mobility).
The second important driver of GPN is the
rapid development and diffusion of information and
communication technology (IT). These technolo-
gies have had a dual impact: they increase the need
and create new opportunities for globalization. This
argument is based on two propositions. The rst
proposition is that the cost and risk of developing IT
has been a primary cause for market globalization:
international markets are required to amortize fully
the enormous R&D expenses associated with rapidly
evolving process and product information technolo-
gies (Kobrin, 1997, p. 149). Of equal importance
are the huge expenses for IT-based organizational
innovations. (Brynjolfson and Hitt, 2000; Ernst and
OConnor, 1992). As the extent of a companys R&D
effort is determined by the nature of its technology
and competition rather than its size, this rapid growth
of R&D spending requires a corresponding expan-
sion of sales, if protability is to be maintained. No
national market, not even the US market, is large
enough to amortize such huge expenses.
The second proposition explains why international
production rather than exports have become the main
vehicle for international market share expansion. Of
critical importance has been the enabling role played
by IT: it has substantially increased the mobility, i.e.
dispersion of rm-specic resources and capabilities
across national boundaries; it also provides greater
scope for cross-border linkages, i.e. the integration of
dispersed specialized clusters. This has substantially
reduced the friction of time and space, both with re-
gard to markets and production: a rm can now serve
distant markets equally well as local producers; it can
also now disperse its value chain across national bor-
ders in order to select the most cost-effective location.
In addition, IT and related organizational innova-
tions provide effective mechanisms for constructing
exible infrastructures that can link together and co-
ordinate economic transactions at distant locations
(Antonelli, 1992; Hagstrm, 2000). This has im-
portant implications for organizational choices and
locational strategies of rms. In essence, IT fosters
the development of leaner, meaner and more agile
production systems that cut across rm boundaries
and national borders. The underlying vision is that of
a network of rms that enable a global network ag-
ship to respond quickly to changing circumstances,
even if much of its value chain has been dispersed.
The third driving force is competition. Together
with liberalization, IT has drastically changed the
1420 D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429
dynamics of competition. Again, we reduce the com-
plexity of these changes and concentrate on two im-
pacts: a broader geographic scope of competition; and
a growing complexity of competitive requirements.
Competition now cuts across national borders, a rms
position in one country is no longer independent from
its position in other countries (e.g. Porter, 1990). This
has two implications. The rm must be present in
all major growth markets (dispersion). It must also
integrate its activities on a worldwide scale, in order
to exploit and coordinate linkages between these dif-
ferent locations (integration). Competition also cuts
across sector boundaries and market segments: mutual
raiding of established market segment efdoms has
become the norm, making it more difcult for rms
to identify market niches and to grow with them.
This growing complexity of competition has
changed the determinants of rm organization and
growth, as well as the determinants of location. No
rm, not even a dominant market leader, can gen-
erate all the different capabilities internally that are
necessary to cope with the requirements of global
competition. Competitive success thus critically de-
pends on a capacity to selectively source specialized
capabilities outside the rm that can range from sim-
ple contract assembly to quite sophisticated design
capabilities. This requires a shift from individual to
increasingly collective forms of organization, from the
multidivisional (M-form) functional hierarchy (e.g.
Chandler, 1977) of multinational corporations to
the networked global agship model (Ernst, 2002c).
Until recently, these fundamental changes in the orga-
nization of international production have been largely
neglected in the literature, both in research on knowl-
edge spill-over through FDI, and in research on the
internationalization of corporate R&D.
2.2. Characteristics of GPN
The concept of a GPN covers both intra-rm and
inter-rm transactions and forms of coordination
(Fig. 1): it links together the agships own subsidia-
ries, afliates and joint ventures with its subcontrac-
tors, suppliers, service providers, as well as partners
in strategic alliances (e.g. Ernst, 1997; 2002b). These
arrangements may, or may not involve ownership of
equity stakes. A network agship like IBM or Intel
breaks down the value chain into a variety of discrete
functions and locates them wherever they can be
carried out most effectively, where they improve the
rms access to resources and capabilities and where
they are needed to facilitate the penetration of impor-
tant growth markets.
The main purpose of these networks is to provide the
agship with quick and low-cost access to resources,
capabilities and knowledge that are complementary to
its core competencies. In other words, transaction cost
savings matter. Yet, the real benets result from the
dissemination, exchange and outsourcing of knowl-
edge and complementary capabilities.
GPN typically combine a breath-taking speed of ge-
ographic dispersion with spatial concentration: much
of the recent cross-border extension of manufacturing
and services has been concentrated on a growing, but
still limited number of specialized lower-cost clusters.
Apart from the usual suspects in Asia (Korea, Tai-
wan, Singapore, China, Malaysia, Thailand, and now
also India), this includes once peripheral locations in
Europe (e.g. Ireland, Central and Eastern Europe and
Russia), Brazil, Mexico, and Argentina in Latin Amer-
ica, some Carribbean locations (like Costa Rica), and
a few spots elsewhere in the so-called RoW (rest of
the world).
The degree of dispersion differs across the value
chain: it increases, the closer one gets to the nal prod-
uct, while dispersion remains concentrated especially
for critical precision components. Concentration of
dispersion increases, the more we move toward more
complex, capital-intensive precision components:
memory devices and displays are sourced primarily
from Japan, Korea, Taiwan and Singapore; and hard
disk drives from a Singapore-centered triangle of lo-
cations in Southeast Asia. Finally, dispersion becomes
most concentrated for high-precision, design-intensive
components that pose the most demanding require-
ments on the mix of capabilities that a rm and its
cluster needs to master: microprocessors for instance
are sourced from a few globally dispersed afliates
of Intel, two secondary American suppliers, and one
recent entrant from Taiwan.
2.2.1. Flagships
GPN typically consist of various hierarchical layers
that range from network agships that dominate such
networks, down to a variety of usually smaller, local
specialized network suppliers. This taxonomy helps to
D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429 1421
Fig. 1. The nodes of a global production network.
assess the different capacities of these rms to benet
from knowledge diffusion and to upgrade local capa-
bility formation.
We distinguish two types of global agships: (i)
brand leaders (BL), like Cisco, GE, IBM, Compaq
or Dell; and (ii) contract manufacturers (CM), like
Solectron or Flextronics, that establish their own GPN
to provide integrated global supply chain services
to the global brand leaders. Cisco is an interest-
ing example of a brand leader: its GPN connects
the agship to 32 manufacturing plants worldwide.
These suppliers are formally independent, but they
go through a lengthy process of certication to en-
sure that they meet Ciscos demanding requirements.
Outsourcing volume manufacturing and related sup-
port services enable brand leaders to combine cost
reduction, product differentiation and time-to-market.
Equally important are nancial considerations:
getting rid of low-margin manufacturing helps the BL
to increase shareholder returns.
Contract manufacturers have rapidly increased in
importance since the mid-1990s. This represents an
acceleration of a long-standing trend towards vertical
specialization that is particularly pronounced in the
electronics industry (Ernst, 2002a). During the 1990s,
global brand leaders have put up for sale a growing
number of their overseas facilities, and in some cases
whole chunks of their GPN. BL from North Amer-
ica like HP, Dell, Compaq, Motorola, Intel, IBM, Lu-
cent, Nortel were rst in pursuing such divestment
strategies. But European BL (e.g., Philips, Ericcson,
Siemens and Nokia) and, more recently, Japanese ones
(e.g., NEC, Fujitsu, Sony) have followed suit. Out-
sourcing based on contract manufacturing became the
panacea of the 1990s (Lakenan et al., 2001, p. 3).
CM have aggressively seized this opportunity: through
1422 D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429
acquisition and capacity expansion they have devel-
oped, within a few years, their own GPN that now
complement the networks established by the global
brand leaders. This gave rise to an extremely rapid
growth of the CM industry. From 1996 to 2000, capi-
tal expenditures grew 11-fold (50% CAGR), and rev-
enues increased by almost 400% (81% CAGR).
The agship is at the heart of a network: it pro-
vides strategic and organizational leadership beyond
the resources that, from an accounting perspective,
lie directly under its management control (Rugman,
1997, p. 182). The strategy of the agship company
thus directly affects the growth, the strategic direc-
tion and network position of lower-end participants,
like specialized suppliers and subcontractors. The lat-
ter, in turn, have no reciprocal inuence over the
agship strategy (Rugman and DCruz, 2000, p. 84).
The agship derives its strength from its control over
critical resources and capabilities that facilitate inno-
vation, and from its capacity to coordinate transac-
tions and knowledge exchange between the different
network nodes.
2.2.2. Local suppliers
Greatly simplifying, we distinguish two types
of local suppliers: higher-tier lead suppliers and
lower-tier suppliers. Higher-tier suppliers, like for
instance Taiwans Acer group (Ernst, 2000b), play an
intermediary role between global agships and local
suppliers. They deal directly with global agships
(both brand leaders and contract manufacturers);
they possess valuable proprietary assets (including
technology); and they have developed their own
mini-GPN (Chen, in press). With the exception of
hard-core R&D and strategic marketing that remain
under the control of the network agship, the lead
supplier must be able to shoulder all steps in the value
chain. It must even take on the coordination functions
necessary for global supply chain management.
Lower-tier suppliers are in a much more precar-
ious position. Their main competitive advantages are
low cost and speed, and exibility of delivery. They
are typically used as price breakers and capacity
buffers, and can be dropped at short notice. This
second group of local suppliers rarely deals directly
with the global agships; they interact primarily with
local higher-tier suppliers. Lower-tier suppliers nor-
mally lack proprietary assets; their nancial position is
weak; and they are highly vulnerable to abrupt changes
in markets and technology, and to nancial crises.
3. GPN and knowledge diffusion
The agships can exert considerable pressure on
local suppliers, especially in small developing coun-
tries: they can discipline suppliers by threatening to
drop them from the networks whenever they fail to
provide the required services at low price and world
class quality.
At the same time, GPN also act as powerful carriers
of knowledge. First, agships need to transfer techni-
cal and managerial knowledge to the local suppliers.
This is necessary to upgrade the suppliers technical
and managerial skills, so that they can meet the tech-
nical specications of the agships. Second, once a
network supplier successfully upgrades its capabili-
ties, this creates an incentive for agships to transfer
more sophisticated knowledge, including engineer-
ing, product and process development. This reects
the increasingly demanding competitive requirements
that we referred to earlier. In the electronics industry
for instance, product-life-cycles have been cut to six
months, and sometimes less (Ernst, 2002a). Over-
seas production thus frequently occurs soon after the
launching of new products. This is only possible if
agships share key design information more freely
with overseas afliates and suppliers. Speed-to-market
requires that engineers across the different nodes of
a GPN are plugged into the agships design debates
(both on-line and face-to-face) on a regular basis.
Of course, knowledge transfer is not a sufcient
condition for effective knowledge diffusion. Diffusion
is completed only when transferred knowledge is in-
ternalized and translated into the capability of the lo-
cal suppliers (e.g. Kim, 1997, and Ernst et al., 1998).
Much depends on the types of knowledge involved
and the mechanisms that agships use to disseminate
different types of knowledge.
3.1. The categories of knowledge
Knowledge may be classied into various categories
depending on the purpose of its use. Polanyis (1962)
classied knowledge into explicit and tacit knowledge.
Explicit knowledge refers to knowledge that is codied
D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429 1423
in formal, systematic language (encoded knowledge).
It is knowledge that can be combined, stored, retrieved,
and transmitted with relative ease and through various
But explicit knowledge is useful only when tacit
knowledge enables individuals and organizations to
make sense of and utilize it. Tacit knowledge refers to
knowledge that is so deeply rooted in the human body
and mind that it is hard to codify and communicate. It
is knowledge that can only be expressed through ac-
tion, commitment, and involvement in a specic con-
text and locality. Tacit knowledge is based on experi-
ence: people acquire it through observation, imitation,
and practice. Its diffusion requires apprentice-type
training and face-to-face interaction. It can also be
transferred, however, through the movement of hu-
man carriers of such knowledge, a fact that much of
the literature on industrial districts used to neglect.
Many have attempted to unpackage the blackbox
of tacit knowledge (e.g. Sparrow, 1998; Spender,
1996). For our purpose, the following classication,
rst coined by Collins (1993) and later expanded
by Blackler (1995), appear to be most useful. Tacit
knowledge may become part of the human body as
skills (embodied knowledge); part of human being as
cognitive capacity (embrained knowledge); routinized
in organizational practice (embedded knowledge); and
inculcated in the organization as basic assumptions,
beliefs and norms (encultured knowledge). Different
types of tacit knowledge are associated with different
aspects of organizational activities and with different
degree of difculties in transferring it.
3.2. Knowledge transfer mechanisms
Flagships transfer knowledge across borders
through various mechanisms. First, the transfer may
be mediated through the market, involving a formal
contract for terms and conditions between the knowl-
edge supplier and the knowledge buyer with payment
involved. Knowledge may also be transferred infor-
mally without any payment involved. Second, the
agship may play an active role, exercising signif-
icant control over the way in which knowledge is
disseminated to and used by the local supplier. Alter-
natively, the agship may play a passive role, having
almost nothing to do with the way the local supplier
takes advantage of available knowledge that is either
embodied in or disembodied from the physical items.
These two dimensionsmarket-mediation and the
role of agshipsoffer a useful two-by-two matrix,
as shown in Fig. 2, to identify different mechanisms
of knowledge transfer through GPN (Kim, 1991).
First, network agships use largely formal mecha-
nisms such as FDI, FL, technical consultancies, etc.
in quadrant 1 to transfer knowledge to local suppliers,
if the latter are subsidiaries or joint venture partners.
For instance, when such agships as Intel, Motorola,
Texas Instruments, and Fairchild decided to outsource
assembly operations of their semiconductor devices,
they took the mechanisms of FDI, FL, and techni-
cal consultancies to establish their subsidiaries in the
Philippines (Antonio, 2001) and other countries in
Southeast Asia. They insisted on majority ownership
in the subsidiaries, licensed and transferred a complete
production system.
Second, independent local suppliers rely heavily
on standard machinery in quadrant 2 to improve their
productivity in production operations. Machinery is
a major source of process innovation for their users
(Abernathy and Townsend, 1975). Flagships are not
necessarily the suppliers of the machinery, but they
can play an important indirect role, by forcing inde-
pendent local suppliers to purchase more sophis-
ticated equipment to improve their production
capabilities. For instance, Mando, one of the major
auto components suppliers from Korea, purchased
a series of robots to automate their production pro-
cesses. Each of the robots embodied state-of-the-art
production knowledge. The equipment suppliers,
however, had little inuence over the way Mando em-
ployees assimilate knowledge by using these robots.
Third, a more direct way for agships to transfer
knowledge to independent local suppliers are infor-
mal mechanisms in quadrant 3, largely through the
original equipment manufacturing (OEM) arrange-
ments. As in the quadrant 1, agships actively transfer
knowledge in the form of blue prints, technical spec-
ications, and technical assistance, mostly free of
charge, to independent local suppliers to ensure that
products and services produced by the latter meet the
formers technical specications. For instance, Boeing
outsources some parts of fuselage from independent
local suppliers in Japan, Taiwan, and Korea. In doing
so, Boeing actively provides the local suppliers with
1424 D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429
Fig. 2. Knowledge transfer mechanisms.
technical literature, product specications, and tech-
nical assistance to help them meet its specications.
Fourth, independent local suppliers can also rely on
knowledge transfer mechanisms in quadrant 4. Like in
quadrant 2, agships exert little direct inuence over
the way independent local suppliers use such mech-
anisms as reverse engineering, observations, and hu-
man mobility to expedite upgrading their capabilities.
In Korea, for instance, the Small Industry Promotion
Corporation and industry-related SME associations
frequently organizes observation tour of foreign rms
as a way to acquire new knowledge. Human mobil-
ity in quadrant 4 includes not only the repatriation
of top-rated engineers trained abroad (Saxenian, in
press) but also the active use of experienced foreign
engineers who are hired for short periods as so-called
To what degree do the agships use the knowl-
edge transfer mechanisms? The shift from MNCs to
global network agships has expanded both the mech-
anisms and the volume of knowledge transfer. MNCs
relied heavily on the mechanisms in quadrant 1 of
Fig. 2 in setting up their plants either for the penetra-
tion of protected markets or for exploiting differential
factor costs. In contrast, agships transfer knowledge
not only through mechanisms in quadrant 1 but also
through mechanisms in quadrant 3. Flagships also tend
to transfer more knowledge to local suppliers than ver-
tically integrated MNCs. These transfers are neces-
sary to enable local suppliers to provide the agship
with competitive products and services, in line with
the changing requirements of markets and technology.
4. Local capability formation
Local suppliers can only effectively absorb knowl-
edge disseminated by global network agships, if
they have developed their own capabilities. Knowl-
edge internalization and capability building require
individual and organizational learning. Individuals are
the primary actors in learning and knowledge creation
(Hedberg, 1981). They constitute local capabilities
that may be combined at the organizational level. Or-
ganizational learning, however, is not the simple sum
of individual learning. Only effective organizations
can translate individual learning and capabilities into
organizational learning and capabilities.
4.1. Concepts
Firms create knowledge primarily through the dy-
namic process of conversion between explicit and
tacit knowledge (Nonaka, 1991). Tacit-to-tacit con-
version (called socialization) takes place when tacit
knowledge of one individual is shared with others
through training, whereas explicit-to-explicit conver-
sion (combination) takes place when an individual or
a group combines discrete pieces of explicit knowl-
edge into a new whole. Tacit-to-explicit conversion
D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429 1425
(externalization) occurs when an individual or a group
is able to articulate the foundations of individual tacit
knowledge. Finally, explicit-to-tacit conversion (inter-
nalization) takes place when new explicit knowledge
is shared throughout the rm and other members
begin to use it to broaden, extend, and reframe their
own tacit knowledge. Such conversion tends to be-
come faster in speed and larger in scale in a spiral
process, as more actors in and around the rms be-
come involved in knowledge conversion (Nonaka and
Takeuchi, 1995). For effective knowledge conversion
to lead to productive learning, two important elements
are required: an existing knowledge base or compe-
tence (most of it tacit knowledge), and the intensity
of effort or commitment. Cohen and Lavinthal (1990)
call this absorptive capacity. How fast and success-
fully the local suppliers internalize and translate trans-
ferred knowledge into their own capability through
learning will be largely determined by their absorptive
capacity and their ability to upgrade it continuously.
Tacit knowledge enables the individual as well as
the organization to use both explicit and tacit knowl-
edge available elsewhere and to create new knowl-
edge through various knowledge conversion activities
in production and R&D. The intensity of effort, on
the other hand, determines the speed of knowledge
conversion. It represents the amount of emotional,
intellectual, and physical energy that members of
an organization invest in acquiring and converting
knowledge. Exposure of individuals and rms to rel-
evant external knowledge is insufcient, unless they
make a conscious effort to internalize and use it.
Hence, considerable time and effort must be directed
to learning (Kim, 1998). Of the two, the intensity
of effort or commitment is more important than the
knowledge base, as the former creates the latter, but
not vice versa (Ullrich, 1998).
4.2. GPN as mediators of local capability
Let us rst look at explicit knowledge. Flagships
typically provide the local suppliers with encoded
knowledge, such as machinery that embodies new
knowledge, blueprints, production and quality con-
trol manuals, product and service specications, and
training handouts. This is done to assist the suppliers
in building capabilities that are necessary to produce
products and services with the expected quality and
price. Personnel at the local suppliers read and try
to assimilate the transferred explicit knowledge into
their tacit knowledge (internalization in Fig. 3). In
most cases, the acquisition of explicit knowledge
alone is not sufcient for the local suppliers to as-
similate and use it in production, as the translation
of explicit knowledge into actual operations requires
a signicant amount of tacit knowledge. Thus, to
augment the explicit knowledge, network agships
typically invite engineers and managers of the local
suppliers to its best-practice plants, so that they can
observe how actual production systems work and to
receive a systematic training.
This can help to translate knowledge gained from
the literature into actual operations (internalization).
It also enables local engineers to internalize how the
agships organization and production systems are
managed (internalization of embedded knowledge),
and to absorb tacit knowledge directly transferred
from foreign engineers through training (socializa-
tion). Once they return home, however, these engi-
neers confront various unforeseen problems in their
attempts to translate what they have learned at the
agships into the operational systems that exist at
home. For this reason, the agships also send their
own engineers (embodied and embrained knowledge)
to help local engineers debug problems in engineering
and manufacturing systems (socialization).
Take the case of subsidiaries or joint ventures in
quadrant 1 of Fig. 2. When Sony established Hwashin
Electronics Company in Korea as a joint venture
to outsource its consumer electronics products, it
supplied not only machinery and equipment for the
mass-production system of its joint venture partner.
Sony also provided blue prints of products, product
specications, and production and quality control
manuals (encoded knowledge). In addition, the ag-
ship invited a number of Korean engineers, techni-
cians, and managers to undergo training at Sonys
plant in Japan on production, organization, and human
resource management (HRM), transferring embed-
ded and encultured knowledge. Sony also dispatched
a number of engineers and technicians to Korea to
help Korean engineers debug problems encountered
in operating and maintaining the production system
and controlling the quality of products to ensure that
Hwashin meet the technical specications of Sonys
1426 D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429
Fig. 3. The process of local capability formation.
products (embodied and embrained knowledge). Sony
had done these knowledge transfer activities formally
as part of its FDI and FL to Hwashin.
In the case of independent local suppliers in quad-
rant 3 of Fig. 2, when General Electric decided to
outsource its microwave ovens from Samsung under
the OEM arrangements, it sent its engineers to Sam-
sung to explain its technical specications (encoded
knowledge) and taught Samsung engineers master the
engineering details of the product (embrained knowl-
edge), (Magaziner and Patinkin, 1989). GE had done
all these knowledge transfer activities free of charge
to ensure that Samsungs products meet GEs technical
Second, local suppliers may attempt to translate
such explicit knowledge as production and quality
D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429 1427
control manuals, HRM handbooks, and other litera-
ture transferred from agships into their own produc-
tion and quality control manuals and HRMhandbooks,
which may be more compatible with local institutions
and business behavior. Then a combination takes place
froma set of explicit knowledge to a newset of explicit
knowledge at the local suppliers. In this process, ex-
ternalization of knowledge also takes place from tacit
knowledge of local engineers and managers to explicit
knowledge in the form a new set of manuals and hand-
books. For instance, when Volvo took over the owner-
ship of Samsungs heavy machinery division after the
Asian crisis to turn it into its Asian supplier, Volvo in-
troduced its own management systems, which reects
both Volvos requirements and those shaped by local
institutions. In developing a new set of manuals and
handbooks, the ground was laid for internalization,
combination and externalization.
Third, the link with GPN also induces knowledge
conversions within local suppliers. The key is the
diffusion of locialized and internalized knowledge ac-
cumulated by a limited number of engineers and man-
agers of the local suppliers through training provided
by the network agship. This knowledge needs to
be diffused within local suppliers through spiral pro-
cesses of socialization, as more actors in and around
the rms get involved in knowledge conversion ac-
tivities. Externalization and internalization take place
internally, as actors convert from/to explicit to/from
tacit knowledge within the local supplying rms,
gradually developing embedded knowledge. For in-
stance, Samsung Electronics recently sent a group of
HRM specialists to GE to learn the latter HRM sys-
tem. Upon return, these specialists have conducted a
series of seminars for HRM specialists in the rm to
share the knowledge, leading to the development of
new HRM policy and procedures and to the gradual
formation of new embedded knowledge.
Fourth, the effectiveness and speed of knowledge
conversion process will be determined not so much by
quantity and quality of the knowledge transferred by
the agships as by the absorptive capacity of the lo-
cal suppliers. This holds true regardless of the knowl-
edge transfer mechanisms. Tacit knowledge plays a
central role in this process. This is true even for the
conversion from explicit knowledge to explicit knowl-
edge. Once again, this highlights how important it is
for local suppliers to develop a rich tacit knowledge
base. The strength of the domestic knowledge base
determines the level of sophistication of the converted
knowledge, while the intensity of effort accelerates the
speed of the conversion processes. In turn, spiral pro-
cesses of knowledge conversion determine the level of
the companys internal knowledge base. Leading local
suppliers thus invest heavily in recruiting the cream of
the crop from universities; they also develop intensive
training programs to upgrade the existing knowledge
5. Conclusions
Liberalization, digital convergence, and intensify-
ing global competition have produced a major organi-
zational innovation: a transition from multinational
corporations that exploit labor cost differentials in
different countries to global network agships that
integrate their dispersed supply, knowledge, and cus-
tomer bases into global (or regional) production net-
works. The paper demonstrates that these networks
have boosted international knowledge diffusion, pro-
viding new opportunities for capability formation by
local suppliers in developing countries. Under pres-
sure from agships, local suppliers have a strong in-
centive to internalize transferred knowledge through
various forms of knowledge conversion. The base-
line, however, is the absorptive capacity of the local
suppliers: it determines the effectiveness of capability
There is nothing automatic about these processes.
Local suppliers need to take an active approach to
maximize their benets from network participation.
Flagships place business orders and transfer valuable
knowledge to local suppliers with only one objective
in mind: to strengthen the competitiveness of their
GPN. In response to intensifying global competition,
the agships outsourcing requirements have become
more demanding. Typically, suppliers are selected by
three criteria: a solid nancial standing; high ratings
on a quarterly scoreboard measuring performance in
delivery, quality etc. and speed of response. The latter
is of critical importance: suppliers are expected to re-
spond within hours with a price, a delivery time, and
a record on their recent performance on reliability and
product quality. This implies that local suppliers can
only upgrade or perish.
1428 D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429
To stay on the GPN, local suppliers must constantly
upgrade their absorptive capacity. Local suppliers,
therefore, must tap, develop, and retain highly skilled
human resources. Adequate incentives are required
to sustain a high intensity of effort. Ruthless global
competition implies that typically only a limited num-
ber of higher-tier local suppliers is strong enough to
cope with these challenges. But it is unclear what will
happen to the great majority of local rms that are
outside this charmed circle of higher-tier suppliers.
In short, participation in GPN is no substitute for
domestic upgrading efforts. Without the latter, net-
work integration of some higher-tier suppliers may
well increase the divide between rms and districts
that have and those that do not have access to the in-
formation and knowledge that is necessary to reap the
benets of network participation. Many people are un-
derstandably concerned that this may lead to a decline
in economic growth and welfare. This paper demon-
strates that there is cause for cautious optimism: net-
work participation may provide new opportunities for
effective knowledge diffusion to local rms and in-
dustrial districts in developing countries, provided ap-
propriate policies and support institutions are in place
that enable local suppliers to exploit the opportunities
and pressures that result from network participation.
The nature of these policies and institutions will be
the subject of a separate article.
Abernathy, W.J., Townsend, P.L., 1975. Technology, productivity,
and process change. Technological Forecasting and Social
Change 7, 377396.
Antonelli, C., 1992. The Economics of Information Networks.
Elsevier, Amsterdam.
Antonio, E.T., 2001. Country Sector Study: Philippines, a paper
presented at the Progress Review Meeting of the International
Competitiveness of Asian Economies: A Cross-Country Study,
Asian Development Bank, Manila.
Blackler, F., 1995. Knowledge, knowledge work, and organizations:
an overview and interpretation. Organization Studies 16, 121
Borrus, M., Ernst, D., Haggard, S., 2000. (Eds.) International
Production Networks in Asia. Rivalry or Riches? Routledge,
Brynjolfson, E., Hitt, L.M., 2000. Beyond Computation:
Information Technology, Organizational Transformations and
Business Performance, A Manuscript, Sloan School of
Management, MIT, Cambridge, MA.
Chandler, A.D., 1977. The Visible Hand: The managerial
Revolution in American Business, Harvard University Press,
Cambridge, MA.
Chandler, A.D., Cortada, J.W., 2000. The information age:
continuities and differences. In: Chandler, A.D., Cortada,
J.W. (Eds.), A Nation Transformed by Information. Oxford
University Press, New York. pp. 281300.
Chen, S.H. Global production networks and information
technology: the case of Taiwan. Industry and Innovation 9, in
Christensen, C.M., 1997. The Innovators Dilemma: When New
Technologies Cause Great Firms to Fail. Harvard Business
School Press, Boston.
Cohen, W.M., Lavinthal, D.A., 1990. Absorptive capacity: a new
perspective on learning and innovation. Administrative Science
Quarterly 35, 128152.
Collins, H., 1993. Structure of knowledge. Social Research 60,
Dicken, P., 1992. Global Shift: The Internationalization of
Economic Activity. Paul Chapman, London.
Dunning, J., 1981. International Production and the Multinational
Enterprise. George Allen & Unwn, London.
Ernst, D., 1997. From Partial to Systemic Globalization.
International Production Networks in the Electronics Industry,
The Data Storage Industry Globalization Project Report
19972002, Graduate School of International Relations and
Pacic Studies, University of California at San Diego.
Ernst, D., 2000a. Inter-organizational knowledge outsourcing: what
permits small taiwanese rms to compete in the computer
industry. Asia Pacic Journal of Management 17, 223255.
Ernst, D., 2000b. Placing the Networks on the Internet: Challenges
and Opportunities for Managing in Developing Asia. In:
Lundvall, B.A., Smith, K. (Eds.), Proceedings of the Paper
Presented at the Second Asia Academy of Management
Conference, 1518 December 2000, Knowledge Creation in the
Learning Economy, Edward Elgar, Cheltenham.
Ernst, D., 2002a. The Economics of Electronics Industry:
Competitive Dynamics and Industrial Organization. In: Warner,
M., Lazonick, W. (Eds.), The International Encyclopedia of
Business and Management (IEBM), International Thomson
Business Press, London, in press.
Ernst, D., 2002b. Global production networks and the changing
geography of innovation systems. implications for developing
countries. Journal of the Economics of Innovation and New
Technologies 12, 127.
Ernst, D., 2002c. The new mobility of knowledge: digital
information systems and global agship networks. In: Latham,
R., Sassen, S. (Eds.), Cooperation and Conict in a Connected
World. Routledge, London, in press.
Ernst, D., OConnor, D., 1992. Technology and Global
Competition: The Challenge for Newly Industrializing
Economies. OECD Development Centre Studies, Paris.
Ernst, D., Ganiatsos, T., Mytelka, L. 1998. Technological
Capabilities and Export Success: Lessons from East Asia.
Routledge, London.
Flaherty, T., 1986. Coordinating International Manufacturing and
Technology. In: Porter, M. (Ed.), Competition in Global
Industries. Harvard Business School Press, Boston, pp. 83110.
D. Ernst, L. Kim/ Research Policy 31 (2002) 14171429 1429
Ghoshal, S., Bartlett, C.A., 1990. The multinational corporation
as an interorganizational network. Academy of Management
Review l15, 603625.
Hagstrm, P., 2000. New wine in old bottles: information
technology evolution in rm strategy and structure. In:
Birkinshaw, J., Hagstrm, P. (Eds.), The Flexible Firm:
Capability Management in Network Organizations. Oxford
University Press, Oxford, pp. 194206.
Hedberg, B., 1981. How organizations learn and unlearn. In:
Nystrom, P.C., Starbuck, W.H. (Eds.), Handbook of Organi-
zational Design: Adapting Organizations to Their Environments,
Vol. 1. Oxford University Press, New York, pp. 326.
Kim, L., 1991. Pros and cons of international technology transfer:
a developing country view. In: Agmon, T., von Glinow, M.A.
(Eds.), Technology Transfer in International Business. Oxford
University Press. New York, pp. 223239.
Kim, L., 1997. Imitation to Innovation: The Dynamics of
Koreas Technological Learning. Harvard Business School
Press, Boston.
Kim, L., 1998. Crisis construction and organizational learning:
dynamics of capability building in catching-up at hyundai motor.
Organization Science 9, 506521.
Kogut, B. 1985. Designing Global Strategies: Proting from Opera-
tional Flexibility. Sloan Management Review, Fall, pp. 2739.
Kogut, B., Kulatilaka, N., 1994. Operating exibility, global
manufacturing, and the option value of a multinational network.
Management Science 40, 123139.
Kobrin, S.J., 1997. The architecture of globalization: state
sovereignty in a networked global economy. In: Dunning, J.H.
(Ed.), Governments, Globalization and International Business,
Oxford University Press, London, pp. 146172.
Lakenan, B., Boyd, D., Frey, E., 2001. Outsourcing and its perils.
Strategy + Business 24, 112.
Magaziner, I., Patinkin, M., 1989. Fast Heat: How Korea Won the
Microwave War. Harvard Business Review, January, pp. 8393.
Nelson, R.R., 1995. The Agenda for Growth Theory: A Different
Point of View. IIASA Working Paper.
Nonaka, I. 1991. The Knowledge-Creating Company. Harvard
Business Review, NovemberDecember, pp. 96104.
Nonaka, I., Takeuchi, H., 1995. The Knowledge Creating
Company: How Japanese Companies Creates the Dynamics of
Innovation. Oxford University Press, New York.
Polanyi, M., 1962. Personal Knowledge: Towards a Post-Critical
Philosophy. University of Chicago Press, Chicago.
Porter, M., 1990. The Competitive Advantage of Nations.
Macmillan, London.
Reddy, N.M., Zhao, L., 1990. International technology transfer: a
review. Research Policy 19, 285307.
Rugman, A.M., 1997. Canada, In: Dunning, J.H. (Ed.).
Governments, Globalization And International Business. Oxford
University Press, London, pp. 175202.
Rugman, A.M., DCruz, J.R., 2000. Multinationals as Flagship
Firms, Regional Business Networks. Oxford University Press,
Saxenian, A. Transnational communities and the evolution of
global production networks: the case of Taiwan, China and
India. Industry and Innovation 9, in press.
Sparrow, J. 1998. Knowledge in Organizations: Access to Thinking
at Work. Sage Press, London.
Spender, J.C., 1996. Making knowledge the basis of a dynamic
theory of the rm. Strategic Management Journal 17, 45
Ullrich, D. 1998. Intellectual capital = competence commitment.
Sloan Management Review, Winter, pp. 1526.
Westphal, L., Kim, L., Dahlman, C., 1985. Reections on
the Republic of Koreas Acquisition of Technological
Capability. In: Rosenberg, N., Frischtak, C. (Eds.), International
Technology Transfer: Concepts, Measures, and Comparisons.
Praeger, New York, pp. 167221.
Wilkins, M., 1970. The Emergence of Multinational Enterprise.
Harvard University Press, Cambridge, MA.
Zander, U., Kogut, B., 1995. Knowledge and the speed of
the transfer and imitation of organizational capabilities: an
empirical test. Organizational Science 6, 7692.
Research Policy 31 (2002) 569588
Market versus technology drive in R&D internationalization:
four different patterns of managing research and development
Maximilian von Zedtwitz
, Oliver Gassmann
International Institute for Management Development (IMD), P.O. Box 915, CH-1001 Lausanne, Switzerland
R&D Technology Management, Schindler Corporation, CH-6031 Ebikon, Switzerland
Received 13 January 2000; received in revised form 2 October 2000; accepted 24 April 2001
Research and development are subject to different location drivers. The analysis of 1021 R&D units, each distinguished by
its main orientation towards either research or development work, reveals that research is concentrated in only ve regions
worldwide, while development is more globally dispersed.
Our research is based on 290 research interviews and database research in 81 technology-intensive multinational companies.
We identify two principal location rationalesaccess to markets and access to scienceas the principal determinants for four
trends that lead to four archetypes of R&D internationalization: national treasure, market-driven, technology-driven,
and global. Their organizational evolution is characterized by four trends. The model is illustrated with short cases of
international R&D organization at Kubota, Schindler, Xerox, and Glaxo-Wellcome.
Differences in R&Dinternationalization drivers lead to a separation of individual R&Dunits by geography and organization.
Current belief is to integrate R&D processes; separation seems to contradict this trend. We argue that this need not be the
case, for there are good reasons to maintain some independence between research and development. 2002 Elsevier Science
B.V. All rights reserved.
Keywords: Globalization; Research and development; Archetypes; Trends; Locations
1. Introduction
1.1. Difcult internationalization of R&D
The internationalization of industrial R&D in
recent years has been interpreted as the attempt of
technology-intensive companies to exploit location-
specic innovation advantages in order to compete in
an increasingly globalized environment. Our empirical
investigation conrms that in most multinational com-
panies, globalization of R&D is typically accepted

Corresponding author. Tel.: +41-21-618-07-99;

fax: +41-21-618-07-07.
E-mail addresses: zedtwitz@imd.ch (M. von Zedtwitz),
oliver gassmann@ch.schindler.com (O. Gassmann).
more with resignation than with pleasure (De Meyer
and Mizushima, 1989, p. 139). Some practitioners
have expressed reservations to substantially invest in
international R&D because of the high execution costs
and low project efciency. At the same time, how-
ever, they recognize that the potential of international
R&D is underestimated and insufciently exploited.
Their main concern is that useful means for effective
implementation of international R&D processes are
not in place. Thus, some companies aim to create a
culture in which employees realize that cooperation
across regional and departmental boundaries, and
to combine internal expertise and know-how with
that of top-performing research facilities worldwide
(Daimler-Benz, 1997, pp. 1415).
0048-7333/02/$ see front matter 2002 Elsevier Science B.V. All rights reserved.
PII: S0048- 7333( 01) 00125- 1
570 M. von Zedtwitz, O. Gassmann / Research Policy 31 (2002) 569588
The establishment of international R&D networks
and the management of transnational R&D projects
are non-trivial and very risky endeavors. The princi-
pal challenges are imposed by the physical distance
among R&D units, as well as between R&D units
and corporate headquarters. Distance impacts com-
munication in terms of frequency and quality, raises
transaction costs, and introduces principal-agent re-
lated difculties. Thus, problems of coordination and
control prevent the exploitation of potential synergy
effects. Local specialization leads to the occurrence
of the not-invented-here syndrome and the compart-
mentalization along separate R&D functions. The
denition of organizations along geographic bound-
aries often neglects the different responsibilities of
R&D, hence creating sub-optimal resource and capital
Managing transnational R&D projects is inherently
more difcult than managing local projects. Data
and information exchange cannot be achieved at the
same quality and speed. Frequent travel puts high
personal and emotional stress on the project manager.
Despite modern communication technologies, the ex-
change of tacit knowledge, the creation of trust, and
a common working culture require direct face-to-face
Fig. 1. Increasing internationalization of R&D in Europe, Japan, and US. Sources: 1. Dunning (1994), 2. Gates (1995), 3. Roberts (1995),
4. Patel (1995), 5. Buderi et al. (1991), 6. National Science Board (1991), 7. Kumar (1995), 8. OECD (1996).
communication. Social and family ties make senior
R&D individuals geographically immobile and reluc-
tant to accept even temporary overseas assignments.
Leading R&D projects is thus mainly based on dis-
tant coordination, jeopardizing the necessary trust and
commitment among remote teams.
1.2. Trends in R&D internationalization
Nevertheless, the potential benets from interna-
tional R&D are too signicant to ignore. Research
in R&D management have resulted in a better under-
standing of the determinants in international R&D,
and many R&D organizations are being transformed
to meet the new challenges. At the national level,
estimates are in the range of 1570% for individual
European countries (average: 30%), 18% for Japan,
and 812% for the US (see Fig. 1). These numbers
vary with the investigating researcher and the unit of
analysis used (e.g. patents, expenditures, personnel).
A few exceptional companies from these countries
exceed 90% of R&D internationalization.
There is a substantial amount of research in inter-
national R&D with an economic and quantitative
background (see, e.g. Ronstadt, 1977; Behrmann and
M. von Zedtwitz, O. Gassmann / Research Policy 31 (2002) 569588 571
Fischer, 1980; Hirschey and Caves, 1981; Pavitt, 1984;
Pearce, 1989; Cantwell, 1995; Dalton and Serapio,
1995). For instance, based on the investigation of the
share of US patents by large foreign rms, Cantwell
and Janne (2000) observe a signicant increase of in-
ternational R&D activity measured by patent origin,
particularly for European and to a lesser degree for
US rms.
This work has led to a number of R&D classi-
cation schemes (see Medcof, 1997 for a concise re-
view). More and more, managerial aspects entered
the discussion (e.g. Rubenstein, 1989; De Meyer and
Mizushima, 1989; Granstrand et al., 1992; Boutellier
et al., 2000). Location factors and overseas determi-
nants have been studied as drivers for R&D decen-
tralization (e.g. Howells, 1990; Serapio and Dalton,
1993; Beckmann and Fischer, 1994; Brockhoff, 1997;
Odagiri and Yasuda, 1996); this work is extended by
contributions from business and policy strategy, R&D
management, and international management.
Often, the contracted term R&D beguiles us
into disregarding the inherent differences between re-
search and development. The necessities of science,
compared with the needs of engineering and devel-
opment, entail different managerial problems (see,
e.g. Leifer and Triscari, 1987). Most contributions
that distinguish between research and development
management neglect the international dimension (e.g.
Eldred and McGrath, 1997a,b; Iansiti, 1998). Knowl-
edge transfer and diffusion have been identied as
major management challenges in international R&D
contexts. Kuemmerle (1997) and Chiesa (1996) sug-
gested models of R&D organization that center around
the knowledge creating and transferring capabilities
of R&D laboratories.
In an earlier contribution (Gassmann and von
Zedtwitz, 1998, 1999a,b), we have outlined trends
and evolutionary patterns for international R&D
organization based on internal distribution and al-
location of R&D resources. Differences between
research and development in terms of location ra-
tionales and work culture effectuate different geo-
graphical distribution and concentration in different
regional centers. In this paper, we propose a model
of R&D internationalization that focuses on external
sources of knowledge as well as the exploitation of
home-based-generated but locally implemented forms
of knowledge.
What are the factors that inuence the differences
in internationalization between research and develop-
ment, and what managerial difculties emerge from
this phenomenon? Based on our analysis of 81 com-
panies representing 1021 R&D sites, we seek to make
the following contributions.
We identify four archetypes of international re-
search and development dispersion: national trea-
sure, market-driven, technology-driven, and global.
We highlight the importance of two principal in-
ternationalization forces in R&D: access to local
science and technology, and access to local markets
and customers.
We observe and present four trends in interna-
tionalizing R&D, each aiming to capture specic
advantages for conducting research as well as de-
velopment in particular locations.
Although seamless integration calls for collocation
and close collaboration, we argue that there are
signicant benets achieved by geographical and
functional separation.
2. Research methodology and data sample
2.1. Two phases of interview-based research
Given our research focus on the patterns of in-
ternationalization of industrial R&D, we chose the
R&D organization of individual companies to be our
unit of analysis. For each of these organizations, our
research followed three principal logical steps. First,
we sought to determine the extent of international-
ization for both research as well as development.
Although we followed the company internal denotation as
closely as possible, we had to establish a terminology that would
allow us to classify our ndings. Hence, R&D comprises cre-
ative work undertaken on a systematic basis in order to increase
the stock of knowledge, including knowledge of man, culture and
society, and the use of this stock of knowledge to devise new
applications (OECD, 1996, p. 7). Commonly accepted denitions
apply: research is dened as the process of discovering new sci-
entic knowledge which has the potential to act as a platform for
the subsequent development of commercially viable products and
manufacturing processes. There is no expectation that the outputs
of research will have immediate commercial value. Development
is the process of creating new products and processes that do have
commercial value, through the application of currently available
platforms or scientic knowledge (Medcof, 1997, p. 303).
572 M. von Zedtwitz, O. Gassmann / Research Policy 31 (2002) 569588
Secondly, we examined changes in the international
R&D structure to deduce emerging patterns of R&D
internationalization over time. Thirdly, we conrmed
these patterns in follow-up research, analyzing man-
agerial challenges of conducting multi-site R&D and
strategic decisions made by senior R&D personnel.
Different types of data were needed in each step,
and different data-collection methods were used to
ensure the quality and accuracy of subsequent data
analysis. As a rst step, R&D locations were compiled
by means of company publications and third-party
R&D databases. Unclear R&D orientations of indi-
vidual sites were resolved in follow-up interviews.
This provided us with a comprehensive overview of
the physical and geographical structure of a rms
R&D organization.
In order to determine how R&D units were inte-
grated within the wider R&D organization, we devel-
oped a semi-structured interview guideline focusing
on functional and hierarchical linkages, participation
in corporate-wide R&D programs, establishment and
ramp-up of R&D sites, as well as the management
of selected transnational R&D projects. Our intervie-
wees were R&D directors and senior R&D managers.
In some companies, we were able to participate in
workshops and R&D project meetings. Combined
with internal documentation on R&D organization,
presentations by R&D personnel as well as memos
from R&D managers, these semi-structured data
Table 1
List of 81 industrial companies considered in this study, sorted by location and industry groups
Company based in Pharmaceuticals, chemicals, food Electrical, information and
software technology
Machinery, petro, automotive
Europe Akzo-Nobel, AstraZeneca, BASF, Bayer,
Boerhinger Ingelheim, Ciba Spec.,
Glaxo-Wellcome, Hoechst, ICI, Nestle,
Novartis, Novo Nordisk,
Hoffmann-LaRoche, Schering, Unilever
Bosch, Cerberus,
Electroclux, Esec, Leica,
Nokia, Siemens, Philips, SAP
ABB, BMW, Continental,
DaimerChrysler, Elf
Aquitaine, Hilti, Mahle,
Mettler Toledo, Royal
Durch/Shell, Schindler,
USA 3M, American Home Products,
BP-Amnoco, Bristol-Myers-Squibb,
Chevron, Dow, Eli Lilly, W.R. Grace
AT&T, Compaq, General
Electric, Hewlett-Packard,
IBM, Intel, Lucent,
Microsoft, Motorola,
National Semiconductor,
Texas Instruments, Unisys,
United Technologies, Xerox
Exxon, Ford, General
Motors, Honeywell
Japan, East Asia Eisai, Kao, LG, Yamanouchi Canon, Daewoo, Hitachi,
Kobe Electric, Samsung,
NEC, Sharp, Sony, Toshiba
Fujitsu, Haier, Honda,
Hyundai, Matsushita,
Mitsubishi, Toyota
helped us (in a second step) to determine the reasons
and patterns of how companies internationalized their
R&D function. Finally, we reported our ndings to
the interviewed companies and sought their feedback
to correct erroneous interpretation and classication.
We conducted a total of 290 interviews using this
semi-structured interview guideline. Total 136 inter-
views in 24 companies were carried out in a rst
phase between March 1994 and September 1996, 154
interviews in 50 companies in a second phase be-
tween October 1996 and December 1998. We targeted
multinational companies in science and technology-
intensive industries (see Table 1), since these indus-
tries rank among the highest in terms of average R&D
to sales ratio, ranging between 4.2% for motor vehi-
cles and 12.6% for telecommunications. Furthermore,
they are characterized by a high degree of international
division of labor. Fig. 2 shows a comparison of R&D
and sales for those 46 companies in the investigation
sample. Internationalization of sales was computed
analogous to the internationalization of R&D: the
share of international sales in relation to total sales.
In the entire sample, only one company, Eisai, was
signicantly more internationalized in R&D than in
sales. This exception requires an explanation. Like
other Japanese drug makers, Eisai relied heavily on
domestic sales. The 1990 export ratio for the seven
largest Japanese pharmaceutical companies was 7.6%
as compared with 70.1% for the seven largest Western
M. von Zedtwitz, O. Gassmann / Research Policy 31 (2002) 569588 573
Fig. 2. R&D internationalization as compared to sales internationalization of 46 technology-intensive companies.
companies. Woodall and Yoshikawa (1997) argued
that the low export rations of Japanese pharmaceuti-
cal companies may be explained by their weakness in
basic research, ineffective R&D, and government in-
tervention in drug pricing. Eisais motivation for large
international research investment can thus be seen
to circumvent unfavorable domestic conditions. With
two overseas research centers, Eisais international
engagement remained relatively small.
Almost all companies rank higher in international-
ization of sales than of R&D: R&D is more centralized
than the distribution of revenues. Not surprisingly,
merger companies exhibit strong internationalization
of sales as well as R&D (ABB, Royal Dutch/Shell,
Mettler-Toledo). Moreover, companies with high
sales internationalization are headquartered in small
countries with small markets (Denmark, Finland, The
Netherlands, Sweden, Switzerland).
2.2. Concentration of research
and dispersion of development
We identied 1021 R&D locations in a total of 81
technology-intensive companies (Table 1). The home
bases of the investigated companies were Europe
(35 companies), USA (26), Japan (15), South Korea
(4), and the Peoples Republic of China (1). In 1998,
13 of the 81 investigated companies ranked among
the top 20 companies in the world, as measured by
their market capitalization in Business Week (1998),
31 were among the top 100. These companies alone
spent several dozen billion US dollars annually on
international R&D activities.
The location data of the 1021 R&D units (Fig. 3
and Table 2) exhibit a strong concentration of R&D
in the Triad regions of Europe, the US, Japan, as well
as major regional centers in South Korea, Singapore,
and other emerging economies along the Pacic Rim.
Research is more concentrated than development.
Total 73.2% of all research sites are located in the
ve regions of the northeastern USA (New Jersey,
New York, Massachusetts), California, the United
Kingdom, Western Continental Europe (in particular
Germany), and the Far East (Japan, South Korea). The
trend of research concentration is even more appar-
ent if only foreign research locations are considered:
87.4% operate in the Triad.
Although, the main regional centers for develop-
ment largely coincide with the regional centers for re-
search, development is more evenly distributed among
574 M. von Zedtwitz, O. Gassmann / Research Policy 31 (2002) 569588
Fig. 3. Concentration of R&D in Europe, Japan, and US.
European countries and the northeastern US, and ex-
tends into southeast Asia, Australia, Africa, and South
America. Only slightly more than half (53.4%) of
all development sites are located in the eight most
development-intensive countries. Development sites
from 19 countries must be considered in order to ac-
count for a similar share in worldwide development
(74.2%) as the top eight countries in research (73.2%).
Table 2
R&D orientation and location of 1021 R&D sites
N = 1021 Research Development International Domestic International
US 107 197 153 151 54 99 53 98
Europe 98 254 258 94 60 198 38 56
Japan 45 93 48 90 17 31 28 62
4 Tigers
30 86 34 82 1 33 29 53
4 New tigers
12 12 12
4 8 12 4 8
15 72 84 3 15 69 3
Total 299 722 601 420 151 450 148 272
Hong Kong, Singapore, South Korea, Taiwan.
Indonesia, Malaysia, Thailand, Vietnam.
Other advanced countries
Rest of the world.
Moreover, research and development sites of the
same company are not necessarily collocated. For
instance, AstraZeneca operates research as well as de-
velopment units in the US. Aresearch unit in Waltham,
Massachusetts focuses on infectious diseases. Since
there is no complementary development in the US,
their research ndings are transferred to a development
laboratory in Sweden. Of IBMs 10 international R&D
M. von Zedtwitz, O. Gassmann / Research Policy 31 (2002) 569588 575
locations, only the Tokyo site maintains both research
and development in the same building. Similar sepa-
ration of research and development can be observed
at DaimlerChrysler, Hewlett-Packard, Canon, and oth-
ers. Why do we observe this separation of research
and development, when everyone claims that the inte-
gration of R&D processes would be so important?
3. Four archetypes of R&D internationalization
The decision about where to establish new R&D
units is made by higher management, usually in
consultation with representatives from the R&D
and strategy departments. This decision considers
R&D-specic factors such as the quality of input at
the new site (through tapping local talent, engaging
in local scientic cooperation, etc.), the quality of
expected output (cooperation with local customers
and local development, market proximity, etc.), and
the general operating efciency (critical mass, project
hand-over, cost issues, etc.) of this R&D unit. The de-
cision is also affected by R&D-external factors, such
as tax optimization, reliability and stability of the local
political and social system, and image enhancement.
Mergers and acquisitions distort the picture of R&D
expansion based on internal growth. Ronstadt (1978)
noted that about a quarter of all R&D investments that
he had studied were incidental through M&A activity
Fig. 4. Organizational structures of internationalized R&D.
of the parent company, and none of these acquisi-
tions had pursued with the intent to gain access to
the organizations R&D resources. Ronstadts obser-
vation has been conrmed, e.g. by Pausenberger and
Volkmann (1981) and Hkanson and Nobel (1993).
During our interviews with R&D directors, we
aimed at disentangling the different factors and drivers
that had led a specic R&D conguration. This re-
quired their input and insight as people who were
intimately familiar with the history of their R&D or-
ganizations. The relative importance of science versus
engineering in a given company also appeared to play
a role. Furthermore, we analyzed principal R&D inter-
nationalization strategies and their actual implemen-
tations. Based on these considerations, we identied
two principal drivers that were responsible for natural
R&D internationalization: the quest for external sci-
ence and technology, and the quest for new markets
and new products. These gave rise to four archetypical
forms of international R&D organization (Fig. 4):
1. domestic research and domestic development: na-
tional treasure R&D;
2. dispersed research and domestic development:
technology-driven R&D;
3. domestic research and dispersed development:
market-driven R&D;
4. dispersed research and dispersed development:
global R&D.
576 M. von Zedtwitz, O. Gassmann / Research Policy 31 (2002) 569588
Most of the 81 R&D organizations that we stud-
ied fall clearly into one of these four categories:
national treasure R&D accounts for 10 companies,
technology-driven for 7, market-driven for 42,
and global for 19. As noted above, special circum-
stances in the process of R&D internationalization
give rise to inconsistencies and hence a <100% t
with the proposed categories. In total, three R&D
organizations were hard to assign to one of the four
archetypical forms. In these ambiguous or unclear
cases, international R&D was either not determined
by either science or market access, or the drivers for
internationalization were too mixed to warrant a dis-
cernible classication. Wherever we observed strong
substantiation for either market or science access, we
categorized accordingly.
The following four case studies illustrate the effects
of these drivers on the management of R&D interna-
3.1. National treasure R&D: domestic
research and domestic development
The R&D organization with both research and de-
velopment at the home base is called the national
treasure model. R&D is kept at home, because core
technologies are easier to control or critical minimum
Fig. 5. Kubotas national treasure R&D organization.
mass is important (see, e.g. Patel and Pavitt, 1992).
There is little R&D at the international level, although
important technological advances may be monitored
from home via local representative ofces and inter-
national patent scanning. Companies with a national
treasure R&D organization are either in a strong dom-
inant design position in its main technologies or their
principal market is domestic. Neither requires much
adaptation for foreign markets. R&D management is
ethno- or geocentric, with foreign experts usually lim-
ited to advisory or consulting roles. The home-based
management style is viable as long as technological
dominance can be maintained.
The Japanese machinery company Kubota owned
a strongly centralized R&D organization (see Fig. 5;
data in these examples are 1999 gures, unless other-
wise noted). Kubota established a technology develop-
ment headquarters to ensure company-wide coherence
of all R&D carried out in four central R&D labora-
tories and 49 R&D units in ve divisions. All central
R&D laboratories and most of the R&D units were lo-
cated in the Osaka-Kyoto area (central Japan). Kubota
invested about 4.5% of its sales in R&D, and 15%
of this amount was reserved for technology develop-
ment. About 430 researchers were employed in the
central R&D laboratories, and some 1600 R&D engi-
neers worked in plant- and factory-based R&D units.
M. von Zedtwitz, O. Gassmann / Research Policy 31 (2002) 569588 577
Kubota had no international R&D locations, since
the domestic market was still dominant (about 83% of
total sales). However, strong linkages existed between
global production sites and the home-based R&D or-
ganization, since the production sites were a principal
provider of new market needs and customer feedback.
A computerized information network connected the
entire corporation, rationalizing corporate functions
and accelerating the innovation process by giving em-
ployees worldwide access to Kubota researchers and
their work. Kubota thus engaged in a variety of in-
ternational research and development projects, while
retaining the efciency advantage of centralization.
3.2. Technology-driven R&D: dispersed research
and domestic development
In technology-driven companies, research is more
internationalized than development. Access to lo-
cal centers-of-scientic-excellence and the relative
scarcity of scientic personnel at home drives a sub-
stantial share of the technology identication and
creation process abroad. Development remains cen-
tralized because of a number of factors, including scale
effects in the development process (e.g. establishment
of technology platforms, access to specialized testing
equipment), proximity to central control and decision
making, protection of commercial results, synergy
Fig. 6. Xeroxs technology-oriented R&D organization.
effects (e.g. improved communication during the in-
novation process, technical cross-fertilization), or the
high information and coordination costs associated
with international R&D projects. These centraliza-
tion factors are less important in research as long as
the scientic results are easy to communicate to the
R&D center and the mission of each research lab is
sufciently focused.
Xerox is a good example of a company with a strong
focus on technology (see Fig. 6). Over the past two
decades, the document processing industry has been
substantially affected by electronic and information
technologies. As a result, Xerox Researchwhich
was originally modeled along the lines of earlier
centralized research organizationswas gradually
decentralized (see Myers and Smith, 1999). There
were ve research sites worldwide: Webster (New
York), Palo Alto (California), Mississauga (Canada),
and two locations in Europe: Grenoble (France),
Cambridge (UK). Each site had a specic research
focus and was strategically located to leverage area
industry and academic competencies relevant to that
region and research focus. Research and technology
development at Xerox was centrally managed under
the corporate research and technology group with di-
rect line-of-site to the CEO. Corporate research and
technology had an average annual headcount of 1320;
about 80 of which are in Europe.
578 M. von Zedtwitz, O. Gassmann / Research Policy 31 (2002) 569588
Technology centers developed and delivered new
technology to business groups. Although, manage-
ment of the technology centers was centralized, re-
sources were located as to align with the development
groups and manufacturing sites they delivered to.
The three primary technology centers had ofces in
El Segundo (California), Webster (New York), Palo
Alto (California). Engineering was found mainly in
the same locations. Product development teams were
geographically collocated with the primary manufac-
turing sites serving their respective product markets.
The largest manufacturing site was in Webster, New
York, responsible for approximately 65% of Xerox
manufactured end products. The second largest man-
ufacturing site in the US was El Segundo, California.
Although, there were no manufacturing facilities in
Palo Alto, California, this site was the third largest de-
velopment center in the US. Development in Palo Alto
focused primarily on software product development.
3.3. Market-driven R&D: domestic research
and dispersed development
Companies with highly dispersed development
and little internationalized research have typically
followed the call of the market; therefore they consti-
tute the market-driven model. Business development
Fig. 7. Schindlers market-driven R&D organization.
is dominated by customer demands and not by sci-
entic exploration. Research is of low signicance in
the overall R&D effort and is kept at home to retain
critical mass. Technology monitoring is carried out
from home or in association with local development
groups. The benet of conducting research internally
is often questioned, and research is under pressure to
provide added value to product development and new
business creation.
R&D at Schindler, a worldwide leader in eleva-
tors and escalators with a turnover of over 8 billion
Swiss Francs, was characterized by the market-driven
paradigm (see Fig. 7). Schindlers core mission was
to be a global service company with strong customer
orientation. In this industry, it was considered crucial
to respond to regional customer requirements such as
codes, standards, and passenger behavior. Many of the
companys 45 000 employees worked in technical ser-
vice or in direct customer contact. On the other hand,
Schindler had to realize global synergies in product
In order to combine global efciency with local ef-
fectiveness, Schindler organized R&D around three
1. Corporate R&D was responsible for system en-
gineering, development of key components, and
M. von Zedtwitz, O. Gassmann / Research Policy 31 (2002) 569588 579
breakthrough innovations. This R&D took place
mainly in Ebikon (Switzerland), Morristown (New
Jersey), Shanghai (China), and Sao Paulo (Brazil).
A special department of corporate R&D, Technol-
ogy Management was responsible for advanced
development, technology monitoring, competitor
analysis, and strategic technology management,
with the principal location in Ebikon and minor
outposts in Morristown and Shanghai.
2. Manufacturing R&D was collocated with the
component factories. Being responsible for in-
cremental product improvements, these activi-
ties were located in Spain (Zaragoza), France
(Mulhouse, Melune), Switzerland (Locarno), USA
(Gettisburgh, Sydney), Malaysia (Ipoh), and China
3. Field engineering was responsible for customized
solutions, including local adaptation of the elevator
to specic customer requirements. These engineer-
ing activities were conducted at every large eld
organization distributed over the 100 countries in
which Schindler is represented.
These three levels enabled the organization to
make best possible use of its competencies. Synergies
and platform concepts are brought in through corpo-
rate R&D, while local solutions were found locally
near downstream activities. Recently, Schindler also
Fig. 8. Glaxo-Wellcomes global R&D organization.
showed signs of more and more technology-driven
internationalization. The systematic international-
ization of corporate R&D and the establishment of
innovation outposts for technology management in-
dicated that Schindler was heading towards a global
R&D organization.
3.4. Global R&D: dispersed research
and dispersed development
Finally, global companies have distributed re-
search as well as development worldwide. These
companies aim for global coordination of their R&D
activities; most of them feature integrated R&D net-
works. Centrifugal forces have become stronger than
centralizing forces. Research is located where there is
high-quality scientic input expected from centers-of-
excellence. Development labs conform to local
demands and standards. The additional costs of main-
taining transnational R&D are offset by the creation
of business and market advantages. In global R&D
networks, local science can be quickly absorbed
and adapted for utilization elsewhere, and single
development centers can take the lead to prepare
products for global market launch. Managing R&D
in this environment is signicantly more complex
and more costly than in the traditional R&D organi-
580 M. von Zedtwitz, O. Gassmann / Research Policy 31 (2002) 569588
Global R&D organizations have been established,
for instance, in the highly competitive pharmaceutical
industry. Headquartered in the UK, Glaxo-Wellcome
had a 4.72% share of the US$ 275 billion global
pharmaceutical market in 1998. It was the second
largest pharmaceutical company both in Europe and
the US, with market shares of approximately 5 and
6%, respectively. In 1998, seven of its products were
in the world top 50. Being pharmaceutical company,
Glaxo-Wellcome identied disease areas of unmet
medical need and determined mechanisms by which
they might be addressed, mainly through developing
and registering new medicines.
Formed in 1995 by the merger of two major compa-
nies, Glaxo-Wellcomes R&D was distributed around
the world (see Fig. 8). Principal sites for R&D were
Stevenage, Greenford, and Ware in the UK; Research
Triangle Park and Palo Alto in the USA; Tsukuba
Science City, Japan; Verona, Italy; Les Ulis, France;
Madrid, Spain; Mississauga, Canada. Research loca-
tions focused on special areas. The two major research
centers in Stevenage, Hertfordshire, UK and in the
Research Triangle Park in North Carolina covered the
widest range of research areas. They were supported
by smaller locations focusing on, e.g. anti-microbial
and anti-fungal research (Madrid), bioinformatics
(Geneva), or disease models (Lausanne) and the Affy-
max Research Institute in Palo Alto, California, which
was engaged in technology development in screening
and combinatorial chemistry.
4. Two principal forces and four trends
in internationalizing corporate R&D
Although, complete integration of globally dis-
persed R&D activities is associated with high coor-
dination and social costs, few companies can afford
to ignore the benets of local product adaptation or
tapping into foreign science clusters. Operating with
constrained resource availabilities in R&D (budget,
manpower, etc.), management will prioritize its ef-
forts as to why and when to internationalize R&D.
Summarizing location factors found in the literature
as well as the analysis of our four archetypes, we can
make the following conclusions.
1. Access to local markets and customers: If forces of
local markets and customer compliance prevail, the
company will develop a decentralized development
2. Access to local science and technology: If critical
scientic knowledge is globally dispersed, interna-
tional research outposts will be established to feed
back technological information to development.
R&D internationalization, however, does not nec-
essarily stop there. Once the R&D organization has
reacheddriven by either of the two principal inter-
nationalization factors of access to science or access
to marketa satisfactory mode of operation, the re-
spective other factor may prevail in driving further
R&D internationalization. This leads to four principal
trends of internationalization in research and develop-
ment (see Fig. 9).
1. Trend 1: internationalization of research: If the
company has no or difcult access to science from
its home base, it may have to establish local re-
search units which directly tap into scientic com-
munities, centers-of-innovation, and local talent
pools. This is often the case when the companys
R&D center is located in a country with strong
engineering but relative scarce scientic research
capabilities (e.g. Daimler between 1988 and 1996,
Canon between 1988 and 1992, Eisai).
2. Trend 2: internationalization of development: If the
companys business requires local product adapta-
tion and intensive customer cooperation, it is likely
that local development units are established that
deploy and implement technology created at the
R&D center. Foreign markets have become more
important than the domestic market. However, the
main technological inputs originate at the home
base (e.g. Leica in the 1980s and 1990s, SAP and
Unisys in the 1990s).
3. Trend 3: development follows research: In science
and technology-driven industries, development
units may bring prospective new business oppor-
tunities from research to the market. These devel-
opment units are often located close to the local
research site and characterized by high personnel
and infrastructure integration. However, their or-
ganizational structure and their reporting lines are
different (e.g. Canon between 1993 and 1998, and
Xerox in France recently).
4. Trend 4: research follows development: The evolu-
tion of local development sites and the necessity to
M. von Zedtwitz, O. Gassmann / Research Policy 31 (2002) 569588 581
Fig. 9. Principal determinants and trends in internationalization of research and development.
provide more substantial research support locally
may result in the on-site establishment of research
units. In the strive to establish competence-based
R&D networks, local development units become
exclusively responsible for a particular technology,
and hence must acquire the necessary research ca-
pability to support subsequent development and
maintenance of this technology (e.g. Schindler be-
tween 1997 and 1999).
Only corporations with large resources to spend on
R&D will be able to opt for concurrent international-
ization of both research and development. They often
do so in separate research and development organi-
zations. For instance, IBM internationalized research
around the same time as development, establishing
a fairly independent research organization (IBM Re-
search) and retaining development in the IBMbusiness
units. IBM Research has followed trend 1, while de-
velopment at IBM followed trend 2. In the 1980s and
1990s, IBM developed an integrated R&D network
by aligning worldwide R&D activities and guiding
both R&D organizations towards the global model.
A similar development took place at Hewlett-Packard
and Philips. In general, however, resource constraints
and industry requirements will tip the balance towards
one of the two major internationalization trends.
Reorganizations of R&D structure (as they often
follow mergers and acquisitions of the parent com-
panies) may seemingly lead to a reversal of above
trends. Cost saving and the exploitation of synergy
effects being a principal motivation for M&A activ-
ity, redundant R&D units are closed down or col-
located. For example, after the merger of Astra and
Zeneca, Astras former Rochester, New York lab was
moved to a newly established R&D site in Waltham,
Massachusetts, which also integrated R&D activi-
ties from Zenecas development unit in Wilmington,
Delaware, and two other Astra sites in Worchester and
Cambridge, Massachusetts. However, AstraZenecas
overall investment in R&D in the US was main-
tained or even expanded, since other R&D units in
Sweden and the UK were being cut in staff simul-
Similarly, a strategic initiative to strengthen corpo-
rate research and development at Schindler has led to
a reduction of research sites. These sites however were
not eliminated but, except in one case, moved to the
supervision of business units as their work has become
increasingly product-oriented. Hence the signicance
of international development has been improved, giv-
ing corporate research the exibility to concentrate
on strategic investments in technology development in
Such events can be seen as corrections to jungle
growth internationalization in R&D. They are aimed
at reducing the costs of conducting R&D both do-
mestically and abroad, improving local effectiveness
of development (e.g. project management constraints
and synergy building) and global utility of research
582 M. von Zedtwitz, O. Gassmann / Research Policy 31 (2002) 569588
(e.g. economies of scale and platform/competence
5. Discussion
5.1. International R&D literature revisited
A number of management researchers have come
close to describing similar structures as the four
archetypes proposed in this paper. Most recently,
Kuemmerles (1997) study on the role of individ-
ual R&D sites distinguished home-base-augmenting
laboratories with the objective to create knowledge
and then transfer it back to a central R&D site, from
home-base-exploiting laboratories that commercial-
ize knowledge by transferring it from the companys
home base to the laboratory site abroad. Chiesa (1996)
recognized that in the same rm, different R&D struc-
tures are developed for experimentation and exploita-
tion activities. Both external sources of knowledge as
well as internal dispersion of R&D resources affect
the resulting R&D structure. Knowledge acquisition
and absorption is a means to reduce uncertainty. Von
Boehmer et al. (1992) applied Pearsons (1990) un-
certainty map to the location of international R&D
activities, nding differences between application
engineering and technical development.
The four archetypes are a concept based on external
factors driving R&D internationalization. If we com-
pared the archetype concept with a model based on in-
ternal coordination and competition (Gassmann, 1997;
Gassmann and von Zedtwitz, 1999a,b), the national
treasure companies appear to fall within the geo- and
ethnocentric paradigm, the technology-driven com-
panies relate best to the hub-model, market-driven
companies follow typically polycentric R&D inter-
nationalization, and the companies denoted here as
global are found to be either striving for integrated
R&D networks or retain a polycentric decentralized
R&D conguration. This latter model is based on
Bartletts (1986) work on the structure of multina-
tional companies as well as Perlmutters (1969) basic
behavioral patterns of multinational corporations.
The four archetypes also describe the evolution
of an entire R&D organization based on the roles of
its individual R&D units. While many authors deep-
ened our understanding of the multitude of different
location factors for individual units (e.g. Beckmann
and Fischer, 1994; De Meyer and Mizushima,
1989; Gassmann and von Zedtwitz, 1996; Hkanson
and Zander, 1988; Carnegie Bosch Institute, 1994;
Krubasik and Schrader, 1990; von Boehmer et al.,
1992), we have summarized these factors and con-
centrated on two principal external drivers that af-
fect the development of the R&D organization as
a whole.
But what are other possible explanations for the ob-
served R&D archetypes? The behavior of companies
is to a large degree determined by corporate culture
and the ethnic/cultural background of their decision
makers. We shall therefore rst consider differences
by geographicalcultural origin. Then business and
industry environments may necessitate a competitive
behavior inuencing the organization of international
R&D. We thus look into differences of industry by
industry. However, we believe our strongest argument
is given by the differences between scientic and en-
gineering work, which not only account for different
patterns between individual companies but also the
observed global distribution of R&D sites.
5.2. Differences by geographicalcultural origin
European, American, and Japanese companies have
shown different patterns in the way they approached
the establishment of foreign R&D operations. Perrino
and Tipping (1989, pp. 1415) argued that European
companies have been most aggressive in establishing
foreign R&D outposts, often through the acquisition
of entire rms. Japanese companies have relied on
home development, licensing, and listening posts to
acquire technology rather than expanding R&D over-
seas. American companies have tended to support new
business opportunities by setting up overseas R&D
units staffed with American researchers. Brockhoff
(1998, p. 42) noted differences between companies
from Japan and other countries in their acquisition of
foreign R&D units.
We may be able to rene the patterns of R&D
internationalization for multinational companies by
discerning differences between research and develop-
ment and their specic location drivers. Due to their
large home markets and their strength in domestic
research, many American and Japanese companies
have been found to adhere to the national treasure
M. von Zedtwitz, O. Gassmann / Research Policy 31 (2002) 569588 583
paradigm. European companies are characterized by
the market-driven model mostly because of their
small domestic markets and relative ease of estab-
lishing development units within the European con-
tinent. Based on our study of 1021 R&D locations,
we observed a substantial amount of intra-European
Of a total of 352 R&D loca-
tions in Europe, no less than 133 R&D sites were
owned by companies from other European countries.
If they were not considered as international R&D
sites, Europe would be left with 30 research sites
from extra-European companies, and 95 develop-
ment sites, or just about half of their previous rate of
Recently, many American companies have assumed
a market-driven conguration because of the ris-
ing relative importance of foreign markets. Japanese
companies are following suit. Due to their relative
weakness in domestic research, they have located
research units in Europe and North America. How-
ever, instead of having R&D follow production, many
Japanese companies have chosen to internationalize
research rst in order to prepare the ground for lo-
cal development and manufacturing. This constitutes
the technology-driven paradigm. With the relatively
strong emphasis on technology implementation in
Japanese companies, cultural and social factors like
tight personal networks and narrowly-meshed com-
pany interdependencies have made it difcult for them
to establish productive development centers outside
Similar regional attributions for global companies
are difcult to make. Only few Japanese companies
so far have established signicant overseas pres-
ence in both research and development. A number
of European and US companies have spread re-
search and development evenly in dominant regions
around the globe. Global R&D is thus less likely
determined by geographic provenience but rather by
maturity of an internationally-dispersed R&D organi-
The OECD estimates that about one-half of EC investment goes
into other EC countries (OECD, 1996, p. 32, citing Thomsen and
Woolcock, 1993). Only the UK has stronger ties to the US as
measured by inward and outward investment.
The remaining 94 locations (38 research, 58 development) are
R&D sites established by companies in their home counties.
5.3. Differences by industry
A breakdown by industry neglects national, cul-
tural, and company-specic factors. The available data
and examples suggest that pharmaceutical and some
IT/electrical companies prefer the global R&D or-
ganization, probably due to particular location advan-
tages in centers of excellence for research with global
development networks to adapt or test products locally.
The market-driven paradigm is the most preva-
lent and obvious way of R&D internationalization. It
is pursued in almost all industries, and particularly
strongly in electrical, machinery, and chemical com-
panies. In these cases, technology platforms developed
at home are the bases for local product development.
There appears to be little need for elaborate interna-
tional research networks.
Internationalization of development occurs more
naturally as a consequence of internationalization of
sales and local market support. Therefore, there are
relatively few examples where research is more de-
centralized than development. It appears that some
pharmaceutical and electrical companies internation-
alize research before development, following trend
1. The pharmaceutical industry appears to be the
most internationalized in research: it operates about
one foreign research site for every foreign develop-
ment unit. Companies of more traditional business
such as automobiles, heavy industry, and oil explo-
ration concentrate their R&D resources in the home
Based on our data, foreign R&D investment is more
than twice as likely to be development oriented than
domestic R&D investment (oil, machinery, automo-
tive, chemicals, telecommunication, food, diversied
products). The information technology and electrical
industries have a moderate emphasis towards more do-
mestic research. Only the pharmaceutical industry re-
peats its domestic R&D investment ratio abroad.
In some cases, we observed that research sites were
established as precursors to development centers.
For instance, Philips corporate research center in
Briarcliff Manor, New York, has recently spun off a
development team to their digital video and television
business groups. Xerox research center in Grenoble
has dedicated a team to more product-oriented re-
search. In general, however, the present data set is
still too limited to provide strong indications for clear
584 M. von Zedtwitz, O. Gassmann / Research Policy 31 (2002) 569588
industrial preferences in the patterns of research and
development internationalization.
5.4. Different drivers for sciences and engineering
In our understanding, neither geographical/cultural
nor industrial analysis sufciently explains the ob-
served archetypes and trends, although it is possi-
ble to argue for certain correlations of geographical
origin or industry with R&D archetypes. We have
found too many counter-examples to warrant possi-
ble hypotheses like pharmaceutical companies tend
to establish global R&D organizations. Disregard-
ing R&D-external factors such as mergers and acquisi-
tions, one could also expect the causal determinants of
differences in R&D dispersions to be rooted in differ-
ences in the very constituents of R&D work, namely
science and engineering.
Scientists aspire to nd new phenomena and create
new knowledge. Their time horizon is usually long:
some scientists devote their entire career to the ad-
vancement of a very specialized subject. Scientists
identify themselves stronger with their eld than with
the company they work for. Their main contacts are
other scientists, although there is a strong tendency to
work and obtain credit as an individual. Commercial
objectives are often absent and their success uncer-
tain, and it is a difcult task for the director of a re-
search department to match thematical freedom with
budgetary constraints.
Table 3
Location drivers for research and development
Reasons to locate research in a particular location
Reasons to establish development in a particular location
Proximity to local universities and research parks Local market requirements
Tapping informal networks Global customers request local support
Proximity to centers-of-innovation Customer proximity and lead users
Limited domestic science base Cooperation with local partners
Access to local specialists/recruiting Market access
Dissipating risks among several research units Local citizen image
Support of local development projects Simultaneous product launching
Adhering local regulations Use of different time zones
Local patenting issues Country-specic cost advantages
Subsidies Facilitating scale-up in manufacturing
Low acceptance of research in home country (e.g. genetics) Process innovation and adaptation to local production
National protection
Science and technology drivers.
Engineering and market drivers
Engineers develop new technology and products
mostly by utilizing existing knowledge in a novel way.
There is often a strong time pressure to meet techni-
cal, performance, and market targets. Almost exclu-
sively, these goals require the close cooperation with
other engineers as well as frequent contacts with sup-
pliers and lead customers. Technical specications and
commercial objectives are dened or negotiated with
production, marketing, or customers. Executing a de-
velopment project is thus more a matter of process
management than of people administration (Boutellier
et al., 1997).
These R&D-internal demands help to explain the
two principal motivations to establish R&D sites
1. Proximity to other corporate activities (e.g. manu-
facturing) and proximity to local customers, which
favor the operation and productivity of engineering
and local product development.
2. The quest for technical know-how and exper-
tise available in only a few centers-of-excellence
around the world, which favor the productivity of
research and technology monitoring.
Since science has productivity requirements dif-
ferent from engineering, the internationalization of
research follows a different rationale than the interna-
tionalization of development. The internationalization
of research is driven by access to local science and
absorption of know-how of global value (see Table 3).
M. von Zedtwitz, O. Gassmann / Research Policy 31 (2002) 569588 585
In international development, understanding and re-
acting to the local market and the efcient cooperation
with local customers (manufacturing, development
partners) are important drivers.
New R&D units are rarely set up as fully-edged
R&D centers. Most of them are established as rather
small units that have to prove their viability rst. Be-
cause of different science and engineering orientation,
their internal evolution towards more mature R&D
sites may be quite different. In many cases for which
we obtained historical data of individual R&D units,
we observed that development units evolve from local
market support units, while research units evolve from
scanning and listening posts. For instance, Leicas
product development unit in Singapore is based on
their previous technical service expertise and local
manufacturing cooperation, and Daimlers research
center in Palo Alto was originally conceived as a
listening post. Although, still largely uninvestigated,
we believe that our principal external factors (cus-
tomer and market demands as well as integration of
local science and technology) are important deter-
minants for the evolution of individual R&D labo-
6. Managerial implications
R&D is torn between the demands for scientic and
commercial results. Research is geared towards dis-
covery rather than invention, while development aims
at invention rather than discovery. Since research dif-
fers from development, the management of research
differs from the management of development. Coping
with managerial and organizational distances between
research and development are at least as challenging as
coping with geographical distances. These distances
explain why there are internal struggles about mission
and leadership, time pressures, funding sources and -
nancing, results and performance evaluation, structure
and thematical freedom.
At the same time, it is often argued that a re-inte-
gration of the two disciplines would improve overall
innovation efciency (e.g. Iansiti, 1998; Dimanescu
and Dwenger, 1996). Detz (1996, p. 31) states that
the isolation of this (research) function in separate
units, often physically separated from a companys
other technology activities, and remote geographically
and strategically from the needs of the companys
businesses made technology transfer exceed-
ingly difcult. Asking for seamless integration
of R&D processes, authors with a focus on ef-
cient project management expect shorter development
times, improved technology transfer and project hand-
over, higher customer orientation, and reduced overall
We are therefore confronted with a fundamental
tension between centrifugal forces that try to estab-
lish distance between research and development for
greater R&D effectiveness, and centripetal forces
that try to integrate research and development for
stronger customer orientation and higher economic
The current trend towards shorter time-to-market
certainly favors the quality and intensity of com-
munication and collaboration which are achieved by
frequent face-to-face contact, physical collocation
(see especially Allen, 1977), and modern information
and communication technology. However, we believe
that physical and organizational separation between
research and development has also the following
research and advanced development teams can bet-
ter evaluate radical new technologies off-line;
mid- and long-term platform teams are not just ex-
tended capacity resources for time critical projects;
pilots with visionary lead users can be developed
separately; this enables solutions that are more in-
novative than the average existing customers re-
quirement, which is normally the base for business
Thus, at the possibly high costs from inefcient
communication and long-distance collaboration pro-
cesses, research needs to be conducted in places where
scientic input can be maximized and where the en-
vironment is more conducive to carry out research.
A calm environment is needed to cultivate new
buds, including separate performance evaluation cri-
teria and a more focused and specialized research
infrastructure. Removed from distracting short-term
concerns of development managers and corporate ad-
ministration, these teams are more likely to pursue
radical ideas, possibly leading to breakthrough inno-
vations. The installation of research teams in scientic
hot-spots like Silicon Valley, science clusters such
586 M. von Zedtwitz, O. Gassmann / Research Policy 31 (2002) 569588
as Route 128 near Boston, Massachusetts, and re-
search parks allows the emergence of such a creative
Seamless integration must not lead to tearing
down all walls between research and development.
Research must be organized according to the task at
hand, not the function! This may well mean that the
entire innovation process must bridge geographical
distances, synchronize different technologies, and
overcome organizational reservations. Two funda-
mentally different attitudes in R&D must be aligned
or even merged. Thus, the dilemma between well-
separated research and development and an efcient
knowledge transfer needs specic managerial atten-
tion. Management of dispersed research must address
technology and knowledge transfer between remote
locations, the ivory-tower syndrome, and the guidance
of colorful and individualistic personalities.
7. Conclusion
International R&D does not necessarily mean a
globally integrated approach to R&D. International
R&D dispersion is often a result of non-R&D-related
merger and acquisitions of parent companies. The
analysis of 1021 R&D units, each distinguished by its
main orientation towards either research or develop-
ment work, has revealed that research is concentrated
in ve regions worldwide, while development is more
dispersed globally than research.
The establishment of new R&D units is inuenced
by two principal factors: access and support of local
markets, and access to local science and technology.
They give rise to four archetypical forms of R&D
organization in which research and development are
not necessarily collocated. Four generic trends de-
scribe the evolution of R&D organizations from one
archetype into another. More than geographic origin
or industry, the relative importance of science and en-
gineering appears to inuence the direction of R&D
internationalization. Spatial and functional distances
make the research-to-development interface more
difcult to handle. Particular in fullling its role as
a facilitator of inter-unit knowledge and technology
transfer, R&D management becomes more complex
due to limitations in knowledge mode conversion,
side effects of cultural, linguistic and behavioral
diversity, and project-internal communication impe-
We expect that advances in information and com-
munication technologies will further facilitate the
dispersion of R&D. At the same time, multinational
companies will establish talent pools of managers
who are experienced in conducting multi-site R&D
projects. With increased global presence, companies
will be able to draw more easily on local science and
technology. We do not expect, however, that interna-
tional R&D alone will improve a companys ability
to innovate. It will remain necessary to manage the
communication ows and innovation processes that
are at the core of the integrated R&D network.
Allen, T.J., 1977. Managing the Flow of TechnologyTechnology
Transfer and the Dissemination of Technological Information
within the R&D Organization. MIT Press, Cambridge.
Bartlett, C.A., 1986. Building and managing the transnational:
the new organizational challenge. In: Porter, M.E. (Ed.),
Competition in Global Industries. HBS Press, Boston,
pp. 367401.
Beckmann, C., Fischer, J., 1994. Einussfaktoren auf die Inter-
nationalisierung von Forschung und Entwicklung in der
Chemischen und Pharmazeutischen Industrie. Zeitschrift fr
Betriebswirtschaftliche Forschung 46 (7/8), 630657.
Behrmann, J.N., Fischer, W.A., 1980. Transnational corporations:
market orientations and R&D abroad. Columbia Journal of
World Business 15 (3), 5559.
Boutellier, R., Gassmann, O., von Zedtwitz, M., 1997. Innovation
and phases in transnational R&D. In: Hirasawa, R. (Ed.),
Proceedings of the 7th International Forum on Technology
Management on Challenges for the 21th CenturyNetworking
East and West, Kyoto, Japan, 37 November, pp. 481
Boutellier, R., Gassmann, O., von Zedtwitz, M., 2000. Managing
Global Innovation. 2nd Revised Edition, Springer, Berlin.
Brockhoff, K., 1997. Industrial Research for Future Competi-
tiveness. Springer, Berlin.
Brockhoff, K., 1998. Internationalization of Research and Deve-
lopment. Springer, Berlin.
Buderi, R., Weber, J., Hoots, C., Neff, R., 1991. A Tighter Focus
for R&D. Business Week, Vol. 2, December, pp. 8084.
Cantwell, J., 1995. The globalisation of technology: what remains
of the product cycle model? Cambridge Journal of Economics
19, 155174.
Cantwell, J., Janne, O., 2000. The role of multinational cor-
porations and national states in the globalization of innovatory
capacity: the European perspective. Technology Analysis and
Strategic Management 12 (2), 243262.
M. von Zedtwitz, O. Gassmann / Research Policy 31 (2002) 569588 587
Carnegie Bosch Institute, 1994. Managing International Research
and Development. Working Paper 9498, International
Executive Forum of the Carnegie Bosch Institute, Pittsburgh,
Chiesa, V., 1996. Managing the internationalization of R&D acti-
vities. IEEE Transactions on Engineering Management 43 (1),
Daimler-Benz, 1997. Producing the goods. Daimler-Benz High
Tech Report97. Daimler-Benz, Stuttgart, pp. 1215.
Dalton, D.H., Serapio, M.G., 1995. Globalizing Industrial Research
and Development. US Department of Commerce, Washington,
De Meyer, A., Mizushima, A., 1989. Global R&D management.
R&D Management 19 (2), 135146.
Detz, C.M., 1996. Corporate-supported research in a reengineered
technology organization. Research Technology Management
39 (4), 3032.
Dimanescu, D., Dwenger, K., 1996. World-Class New Product
Development. Amacom, New York.
Eldred, E., McGrath, M., 1997a. Commercializing new technology
- I. Research Technology Management 40 (1), 4147.
Eldred, E., McGrath, M., 1997b. Commercializing new technology
- II. Research Technology Management 40 (2), 2933.
Gassmann, O., 1997. Internationales F&E-Management. Potentiale
und Gestaltungskonzepte transnationaler F&E-Projekte. Olden-
bourg, Mnchen.
Gassmann, O., von Zedtwitz, M., 1996. Ein Referenzrahmen
fr das Internationale Innovationsmanagement. In: Gassmann,
O., von Zedtwitz, M. (Eds.), Internationales Innovationsmana-
gement. Vahlen, Mnchen, pp. 315.
Gassmann, O., von Zedtwitz, M., 1998. Organization of industrial
R&D on a global scale. R&D Management 28 (3), 147161.
Gassmann, O., von Zedtwitz, M., 1999a. New concepts and trends
in international R&D organization. Research Policy 28, 231
Gassmann, O., von Zedtwitz, M., 1999b. Organizing virtual R&D
teams: towards a contingency approach. In: Kocaoglu, D.F.
(Ed.), Proceedings of the Portland International Conference on
Management of Engineering and Technology (PICMET) on
Technology and Innovation ManagementSetting the Pace for
the Third Millennium, Portland, July, pp. 2529.
Gates, S., 1995. The Changing Global Role of the Research
and Development Function. Report No. 1123-95-RR, The
Conference Board, New York.
Granstrand, O., Hkanson, L., Sjlander, S. (Eds.), 1992.
Technology Management and International Business: Interna-
tionalization of R&D and Technology. Wiley, Chichester.
Hkanson, L., Nobel, R., 1993. Foreign research and development
in Swedish multinationals. Research Policy 22, 373396.
Hkanson, L., Zander, U., 1988. International management of
R&D: the Swedish experience. R&D Management 18 (3), 217
Hirschey, R.C., Caves, R.E., 1981. Internationalisation of research
and development by multinational enterprises. Oxford Bulletin
of Economics and Statistics 42 (2), 115130.
Howells, J., 1990. The location and organisation of research and
development: new horizons. Research Policy 19, 133146.
Iansiti, M., 1998. Technology Integration: Making Critical Choices
in a Dynamic World. Harvard Business Press, Boston.
Krubasik, E.G., Schrader, J., 1990. Globale Forschungs- und
Entwicklungsaktivitten. In: Welge, M.K. (Ed.), Globales
Management. Schaetfer-Poeschel, Stuttgart, pp. 1727.
Kuemmerle, W., 1997. Building effective R&D capabilities abroad.
Harvard Business Review 3/4, 6170.
Kumar, N., 1995. Intellectual Property Protection, Market
Orientation and Location of Overseas R&D by Multinational
Enterprises. UNU/INTECH Discussion Papers, March.
Leifer, R., Triscari, T., 1987. Research versus development:
differences and similarities. IEEE Transactions on Engineering
Management EM-34 (2), 7178.
Medcof, J.W., 1997. A taxonomy of internationally dispersed
technology units and its application to management issues.
R&D Management 27 (4), 301318.
Myers, M., Smith, K., 1999. Xerox: the global market and
technology innovator. In: Boutellier, R., Gassmann, O., von
Zedtwitz, M. (Eds.), Managing Global Innovation. Springer,
Berlin, pp. 299315.
National Science Board, 1991. Why US technology leadership is
eroding. Research Technology Management 34 (2), 3642.
Odagiri, H., Yasuda, H., 1996. The determinants of overseas R&D
by Japanese rms: an empirical study at the industry and
company levels. Research Policy 25, 10591079.
OECD, 1996. Science, Technology and Industry Outlook 1996.
OECD, Paris.
Patel, P., 1995. Localised production of technology in global
markets. Cambridge Journal of Economy 19, 141153.
Patel, P., Pavitt, K., 1992. Large rms in the production of the
worlds technology: an important case of non-globalisation. In:
Granstrand, O., Hkanson, L., Sjlander, S. (Eds.), Technology
Management and International Business: Internationalization of
R&D and Technology. Wiley, Chichester, pp. 5373.
Pausenberger, E., Volkmann, B., 1981. Forschung und Entwick-
lung in internationalen Unternehmungen. In: Moll, H.H.,
Warnecke, H.J. (Eds.), RKW-Handbuch Forschung, Entwick-
lung, Konstruktion (F&E). Bd. 3, Stand: Okt. 1990 (19. Lfg.),
Kennzahl 8400, Berlin.
Pavitt, K., 1984. Sectional patterns of technical change: towards
a taxonomy and a theory. Research Policy 13, 343374.
Pearce, R.D., 1989. The Internationalisation of Research and Deve-
lopment by Multinational Enterprises. Macmillan, Basingstroke,
Pearson, A., 1990. Innovation strategy. Technovation 10 (3), 185
Perlmutter, H.V., 1969. The tortuous evolution of the multinational
corporation. Columbia Journal of World Business 4, 918.
Perrino, A.C., Tipping, J.W., 1989. Global management of
technology. Research Technology Management 32 (3), 1219.
Roberts, E.B., 1995. Benchmarking the strategic management of
technology I. Research Technology Management 38 (1), 4456.
Ronstadt, R.C., 1977. Research and Development Abroad by US
Multinationals. Praeger, New York.
Ronstadt, R.C., 1978. International R&D: the establishment and
evolution of research and development abroad by seven US
multinationals. Journal for International Business Studies 9, 7
588 M. von Zedtwitz, O. Gassmann / Research Policy 31 (2002) 569588
Rubenstein, A.H., 1989. Managing Technology in the Decen-
tralized Firm. Wiley, New York.
Serapio, M., Dalton, D., 1993. Foreign R&D facilities in the United
States. Research Technology Management 36 (6), 3339.
Thomsen, S., Woolcock, S., 1993. Direct Investment and European
Integration: Competition among Firms and Governments.
Catham, London.
von Boehmer, A., Brockhoff, K., Pearson, A.W., 1992. The
management of international research and development. In:
Buckley, P.J., Brooke, M.Z. (Eds.), International Business
Studies. Oxford University Press, Oxford, pp. 495509.
Woodall, B., Yoshikawa, A., 1997. Japans Failure in
Pharmaceuticals: Why is the World Saying No to Japanese
Drugs? CIBER Working Paper.
Research Policy 34 (2005) 13501365
On the road again: Researcher mobility
inside the R&D network
Paola Criscuolo

Tanaka Business School, Imperial College London, South Kensington Campus, London SW7 2AZ, UK
Received 1 December 2004; accepted 1 May 2005
Available online 30 August 2005
This paper examines the roles of international assignments and other forms of researcher mobility inside the integrated R&D
networks of six of the largest European pharmaceutical companies. From in-depth interviews with R&D managers and scientists
it is found that, while the use of international assignments is both limited and often not aimed specically at the transfer of
knowledge, other forms of short-term mobility are widely employed. But what is transferred through such short-term visits is
narrow in focus and often related to specic projects whereas international assignments enable the transfer of broader and more
complex knowledge. The potential impact of these changes in the inter-unit mobility patterns of researchers may be to limit both
the exploitation of potential synergies across disciplines and the creation of enduring and strong personal ties among researchers
working in distant locations.
2005 Elsevier B.V. All rights reserved.
Keywords: Researcher mobility; R&D internationalisation; Multinational companies
1. Introduction
In recent years, considerable evidence has been
gathered that points to the increasing internationali-
sation of R&D activities by multinational enterprises
(MNEs) (see, for example, Dalton et al., 1999; Le Bas
and Sierra, 2002). As shown by a number of studies
(Chiesa, 1996b; Gassmann and von Zedtwitz, 1999)
this growth in R&D activities performed abroad is
being accompanied by a transformation in the way

Tel.: +44 20 75945926; fax: +44 20 75945915.

E-mail address: p.criscuolo@imperial.ac.uk.
global R&D is organised, a transformation involving
a transition from highly centralised structures to a less
hierarchical interdependent network of mutually sup-
portive facilities (Pearce, 1999). However, very few
studies have investigated the link between the interna-
tionalisation of R&D, on the one hand, and the degree
of inter-rm researcher mobility, on the other (B eret et
al., 2003, is a notable exception).
The successful adoption of an integrated network
of R&D facilities rests on the ability of the MNE
to leverage knowledge from pockets of excellence
around the world and to exploit specialised assets
within the organisation. The management of R&Dstaff
0048-7333/$ see front matter 2005 Elsevier B.V. All rights reserved.
P. Criscuolo / Research Policy 34 (2005) 13501365 1351
thus becomes important, both as a means of achieving
technology transfer among physically distant research
facilities, and as a coordination mechanismfor exploit-
ing diversity among different R&D units (Edstr om
and Galbraith, 1977). While there are a number of
recent studies (for example, Almeida and Kogut, 1999;
Rosenkopf and Almeida, 2003; Stolpe, 2002) showing
that inter-rm researcher mobility facilitates the trans-
fer of knowledge, especially tacit knowledge, through
the creation and reinforcement of personal relation-
ships, the role of researcher mobility as a means of
achieving technology transfer within an organisation
has been less analysed. As pointed out by B eret et al.
(2003), the human resource literature is mainly ori-
ented to top management or to questions of executive
stafng practice and little attention has been paid to the
management of R&D workers. Indeed rms have gen-
erally assumed that the scientic culture dominated the
local culture and that interaction between foreign R&D
units and the rest of the organisation was poor or non-
existent (Chiesa, 1996a). But, while in the traditional
centralised R&D structure there was no need to stimu-
late network building among researchers because each
centre was working in isolation with very little interac-
tion with other units, this is essential in the integrated
multi-hub R&D structure. The adoption of this organi-
sational model in R&D requires a better understanding
of the role of researchers mobility as a technology
transfer and coordination mechanism. Thus, the aim of
this paper is to ll the gap in the literature and to assess
the degree, modality and role of researcher mobility in
six of the largest European pharmaceutical MNEs.
The paper is organised as follows. Section 2 dis-
cusses the theoretical framework. Section 3 describes
the different phases of the drug discovery process. Sec-
tion 4 outlines the methodology of the study. Section 5
describes the main distribution of the innovation activ-
ities and R&D organisation across geographically dis-
persed units. Section 6 reports the empirical ndings on
the role of researchers mobility. Section 7 concludes
the paper.
2. R&D organisation and researchers mobility
Traditionally the most strategic and core innova-
tive activities were concentrated in the central R&D
unit in the home country of the MNE, while foreign
sites were mainly responsible for supporting local man-
ufacturing activities or for marketing activities. Mobil-
ity in this organisational model, dubbed the ethnocen-
tric (Perlmutter, 1965) or centralised hub (Bartlett,
1986), was very limited because of the role and the
size of the foreign R&D unit. In addition, foreign
R&D centres did not provide interesting career oppor-
tunities for researchers coming from the home coun-
try unit because qualication requirements were often
lower than in the central R&D laboratory (B eret et al.,
In this model, there was only one centre and MNEs
relied largely on one location, i.e. the home country,
as the principal and almost the only source of their
competitive advantage. However, particularly in R&D-
intensive andtechnologicallycomplexindustries, inno-
vation sources have become much more dispersed than
in the past and in order to remain internationally com-
petitive rms require access to foreign technological
developments. MNEs cannot rely only on exploit-
ing internationally those technological assets built on
home-country competences, but have to source knowl-
edge fromeach leading market and national technology
system. As a result a different R&D organisational
structure, where each R&Dunit assumes a leading role
in the creation of unique competences, has emerged.
In this integrated R&D network structure, in the tax-
onomy of Gassmann and von Zedtwitz (1999), the
creation of new technologies is not the prerogative of
the centre, but rather takes place in foreign subsidiaries
building on and exploiting host countries competitive
However, if the main advantage of implementing
a geographically dispersed R&D network structure is
the ability to tap selectively into centres of excellence,
the main drawback is the high costs of coordinating
and achieving knowledge diffusion. Communication
among different R&D units is crucial but at the same
time it is also more complex than in the centralised
hub structure. In the integrated network model techno-
logical upgrading emerges from complex external and
internal knowledge ows both between subsidiaries
and from the centre to the periphery. Thus, the rm has
to move away from the management of a set of dyadic
relationships between the centre and the foreign R&D
unit and adopt a more systemic coordination mech-
anism in order to promote intensive communication
1352 P. Criscuolo / Research Policy 34 (2005) 13501365
If coordinating these exchanges of knowledge is
complex, it is even more difcult to realise them. It
is often argued that transfer of knowledge within units
belonging to the same organisation is easier to achieve
than is the transfer of knowledge between organisa-
tions (Grant, 1996; Kogut and Zander, 1992) and that
the main competitive advantage of multinationals lies
in the possibility of transferring and integrating knowl-
edge generated by subsidiaries located in different
countries. However, several authors have pointed out
that knowledge transfer even within organisations is
far from being an automatic process. There are barri-
ers connected to the characteristics of the knowledge
to be transferred, to the inter-unit technological, geo-
graphical and organisational distance, and also to the
motivational disposition of both the sender and the
receiver units (see Kogut and Zander, 1993; Szulan-
ski, 1996; Gupta and Govindarajan, 2000).
Although early MNE theories perceived knowledge
as a public good (Buckley and Casson, 1976), it is now
well accepted that there are codied and tacit forms
of knowledge and that, while codied knowledge is
revealed by its communication, and therefore is eas-
ily transferred, tacit knowledge is revealed through its
application, implying that its diffusion is slow, costly
and uncertain.
Another barrier to internal knowledge transfer is
inter-unit geographical distance. Early studies by Allen
(1970, 1977) showedthat physical proximityaffects the
likelihood of communication among R&D staff within
technical functions and between technical functions
and other functions in a rm. Although these problems
have been mitigated by developments in information
and communication technologies (ICT), which have
facilitated the management and coordination of inter-
national research networks, geographical distance is
still a barrier to the transfer of knowledge especially if
it is tacit in nature (Howells, 1995).
Organisational and cultural distance is another fac-
tor inuencing the ease of intra-rm knowledge trans-
fer (Bartlett and Ghoshal, 1989). The adoption of a
geographically distributed R&D organisation implies
the existence of a strong cultural heterogeneity, which
creates barriers to knowledge sharing. The lack of
a common culture and greater autonomy may intro-
duce motivational barriers in the subsidiary to transfer
technology within the organisation. The motivational
element can be particularly important in determining
knowledge transfer from units that have been recently
acquired, although with time this effect should fade
A further obstacle is inter-unit technological dis-
tance. This problem is especially relevant for those
companies that have adopted a differentiated R&D
network, whereby subsidiaries have a set of unique
capabilities reecting the host countrys technological
specialisation. Knowledge sharing requires that sender
and the receiver have a common set of prior knowledge.
The increased level of technological specialisation and
diversication in the R&Dorganisational network may
reduce the amount of shared knowledge hindering the
process of knowledge transfer.
In this organisational structure characterised by
inter-unit geographical, organisational and technolog-
ical distance inter-unit researcher mobility could facil-
itate knowledge diffusion and coordination of R&D
activities. Traditionally the human resource literature
has identied three reasons for transferring person-
nel, mainly managers, to other units (see, for example,
Edstr om and Galbraith, 1977; Harzing, 2000). First,
transfers are made to ll positions when there is a per-
ceived lack of availability of qualied individuals in
a foreign country, i.e. to transfer knowledge. Second,
international assignments are used for personal devel-
opment, i.e. to give a manager the opportunity to gain
international experience in the managing of the organ-
isation. Third, transfers are used for organisational
development. In this case international assignments are
employed as a control and coordination mechanism.
This strategy consists of two elements: socialisation
of both expatriate and local managers into the corpo-
rate culture on the one hand, and, on the other, the
creation of informal communication networks and per-
sonal relationships that can provide links between the
subsidiary and the headquarters. More recently, it has
been shown that coordination by socialisation is also
conducive to increasing knowledge ows within the
MNE (Gupta and Govindarajan, 2000). From a knowl-
edge sharingperspective the underlyingrationale is that
the more that different units share similar goals, the
more likely they are to transfer resources and exchange
complementary knowledge.
Hedlund (1986) himself recognised the importance of human
resource management in the heterarchical model: In order for
P. Criscuolo / Research Policy 34 (2005) 13501365 1353
In the area of R&D, transfers of personnel are
particularly crucial because they allow face-to-face
interaction through which less codied forms of knowl-
edge can be shared. Although ICT applications such
as email, videoconferencing, instant messaging and
groupware have enabled the implementation of virtual
teams, face-to-face communication is still considered
one of the richest forms of communication (Daft
and Lengel, 1986). Face-to-face interaction appears to
solve problems and complete tasks faster than elec-
tronic communication (see, for example, Desanctis
and Monge, 1999) because it allows the transmis-
sion of multiple clues (i.e. eye contact, head nods and
tone of voice) and immediate feedback. As shown
by Orlikowskis (2002) study of a global product
development team, face-to face interactions play a
crucial role in transferring complex knowledge as
well as building trust, commitment, and establish-
ing and sustaining social relationships. This point
is also emphasised by recent studies on the social
dimension of R&D spillovers (Breschi and Lissoni,
2003; Singh, 2004; Stolpe, 2002). These contributions
have shown that inter-rm researcher mobility is very
important in explaining the occurrence of knowledge
spillovers through the formation of social ties between
researchers who have worked together in the past.
These social relationships allowthe sharing of informa-
tion among researchers even when they are no longer
part of the same organisation.
Notwithstanding the fundamental strategic role of
inter-unit researcher mobility in efciently exploiting
knowledge within the geographically dispersed R&D
network, little is known about the use of this technol-
ogy transfer and coordination mechanism. The human
resource literature has mainly focused on the mobility
of managers, while the international business literature
has not explicitly analysed this mechanism although it
has acknowledged its strategic role in both the coordi-
nation of R&D activities and in the internal knowl-
edge transfer process (Bartlett and Ghoshal, 1990;
Casson and Singh, 1993; Chiesa and Manzini, 1996;
De Meyer, 1993; De Meyer and Mizushima, 1989;
internalisation of norms to take place, a lot of rotation of personnel
and international travel and postings are necessary . . .. Advances
in information technology may help the formation of the nervous
system of the rm, but this will not be enough for building internal
cultures (p. 29, emphasis added).
Persaud et al., 2001; Teigland et al., 2000; Westney,
This paper, therefore, tries to ll the gap in the liter-
ature and report new evidence on three aspects. First it
will assess what the degree of mobility of researchers
inside the R&D network is and what types of assign-
ments are used. A number of recent studies within the
human resource literature have shown that there is an
increasing trend in the use of short-term assignments
and other alternatives to long-term assignments (such
as international commuting and frequent ying) to
contain the cost of relocating staff abroad and partly to
overcome immobility of dual-career couples (Harris,
2002; PricewaterhouseCoopers, 1999; Roberts et al.,
1998; Welch et al., 2003).
Second, it will examine
the reasons for transferring researchers to other units.
Third, it will analyse what the impact of researcher
mobility is on the internal knowledge diffusion process.
3. Research and development in the
pharmaceutical industry
This section summarises the major phases of the
drug discovery process. It explains what types of activ-
ities are undertaken in different units of the R&D
network and the industry trend towards a common
multi-hub network structure in research.
The drug discovery and development process can be
dividedintosixstages (see Fig. 1). The rst three stages
comprise the drug discovery phase, which aims to iden-
tify newcompounds; the remainder of the process com-
prises the development phase, when compounds are
tested to assess their efcacy and tolerability. The divi-
sion between research and development is not clear-cut
and certain companies classify as research part of the
clinical development up to the proof of concept, i.e.
when the compound can be demonstrated to be operat-
ing by the desired mechanism.
Harris (2002) gives the following taxonomy of international
assignments: long-term assignments involve the relocation of the
family for a period over 1 year; short-term assignments last for less
than 1 year and do not always require the relocation of the fam-
ily; international commuting entails commuting from the home to
the host country on a weekly or bi-weekly basis, while the family
remains at home; frequent-yer assignment refers to those forms of
assignments where the employee undertakes frequent international
trips but does not relocate.
1354 P. Criscuolo / Research Policy 34 (2005) 13501365
Fig. 1. The drug discovery process.
Developments in the technologies employed in the
drug discovery phase and scientic advances in cell
biology, pharmacology and enzymology have funda-
mentally changed the nature of the research process in
the pharmaceutical industry (Arora and Gambardella,
1994; Henderson and Cockburn, 1994; Nightingale,
2000; Reiss and Hinze, 2000). Traditionally, the dis-
covery of a newcompound was the result of a trial-and-
error process during which thousands of compounds
were screened in order to nd one with a specic
biological prole, because in general the mechanism
of action of a compound was not clearly under-
stood. This process required huge laboratories to con-
duct large-scale screening, and extensive nancial,
human and technological resources. Since the 1990s,
the introduction of what are known as enabling tech-
nologies, such as high-throughput screening, com-
binatorial chemistry, bio-informatics and scientic
advances in biomedical sciences have completely
transformed the experimental design and the drug dis-
covery process. The discovery of a drug is now the
result of a science-deductive method and researchers
knowwhich biochemical and molecular pathways they
want to block or stimulate, and are aware of the
basic features of the molecules that might serve this
Advances in biomedical knowledge have helped
decouple the competencies that underlie drug discov-
ery from drug development. Since the knowledge and
The initial optimistic expectations about the impact of the adop-
tion of new technologies on the drug discovery process have been
challenged by some more recent evidence on the decline in produc-
tivity in drug discovery (see Nightingale and Martin, 2004 and the
literature cited therein). Part of the reason for this decline in produc-
tivity is that the introduction of newmethods such as high-throughput
screening, combinatorial chemistry, and genomics have generated a
range of new problems connected with information overload and
statistical quality control.
P. Criscuolo / Research Policy 34 (2005) 13501365 1355
expertise that are critical to drug discovery can be easily
codied and partitioned, it is not longer the case that
discovery and development activities are more effec-
tively performed together. As shown by Chiesa (1996c)
there has been a tendency towards physical separa-
tion between development and research functions and a
reduction in size of research units. However, the devel-
opment of a drug still requires large amounts of human
and nancial resources and its efciency relies on the
achievement of a critical mass to carry out highly stan-
dardised large-scale activities.
4. Research method
The exploratory nature of the questions addressed in
this paper makes a case study approach based on data
collection the most appropriate research strategy. Data
were collected through 24 face-to-face semi-structured
interviews with R&D managers and scientists in six of
the largest European pharmaceutical companies (some
descriptive statistics of the interviewed companies are
reported in Table 1). Heads of global R&D func-
tion were contacted with a request to interview both
R&Dmanagers and scientists seconded for a minimum
periodof 6months toone of the USR&Dfacilities. This
paper is based on their experience, the impact that this
international assignment has had on their career and on
their innovative activity, and thus is not a representative
It was decided to focus on secondments to the US
because, as will be shown in the next section, these
MNEs carry out a large proportion of their R&D activ-
ities in this country in order to acquire specialised
capabilities particularly in biotechnology (Allansdottir
et al., 2002; Reger, 2000; Senker, 1998; Shan and Song,
1997; Sharp, 1999).
Between two and six interviews of 1.5 h each were
carried out in each company between June 2002 and
April 2003. The interviews were based on two sepa-
rate but overlapping sets of questions for the managers
and the scientists. This enabled results to be corrobo-
rated that might otherwise be biased depending on the
position of the individuals within the organisation. But
it should be noted that, since the interviews were only
carried out in the headquarters of the company, they
may reect only the perceptions of the central organi-
The questions addressed to the researchers focused
on the knowledge sources involved in their research
activity and the degree of interaction with other sci-
entists; on the reasons for being seconded overseas;
on the knowledge exchange and accumulation during
the assignment; and on the applicability of the knowl-
edge learned abroad. The interviews with the R&D
managers covered three themes: the R&D organisation
structure; the use of socialisation mechanisms in the
internal knowledge transfer process; and the rationale
behind the use of international assignments.
The general method used to analyse the data gath-
ered in the interviews is based on a procedure described
by Glaser and Strauss (1967) as constant comparison
method. Since the analysis aims to provide a descrip-
tion of the role of researcher mobility as a coordination
and technology transfer mechanism, this method was
used not as a way to derive theory but mainly as a way
to examine the data. The rst stage included coding
the interview material. This involved devising codes
or subheadings to link the raw data fragments such as
paragraphs, phrases or sentences from the interviews,
Table 1
Description of the interviewed companies
Company name Corporate
2002 Pharmaceutical
revenues ($)
Rank No. of employees in
2002 R&D
expenditure ($)
Rank No. of interviews
AstraZeneca London 17,841 4 11,000 3,069 4 5
Aventis Strasburg 16,639 6 5,600 3,235 5 4
GlaxoSmithKline London 27,060 2 15,000 4,108 2 5
Novartis Basel 13,547 8 3,000 2,799 6 2
Roche Basel 9,355 13 5,030 2,746 7 3
Schering Berlin 3,074 19 1,200 869 19 5
Source: Revenues, R&D expenditures and ranking data are from the Contract Pharma Ranking of top 20 pharmaceutical companies
Not all R&D employees and expenditure are in the pharmaceutical business of these companies.
1356 P. Criscuolo / Research Policy 34 (2005) 13501365
to pieces of text relevant to a particular theme. All
interviews were taped and transcribed and then pas-
sages of text were grouped into patterns according to
the codes they had been assigned. The identication of
codes was based on the research questions concerning
the role of international assignments inside the R&D
Constant comparison was used to evaluate the data
to determine the similarities and differences among
them and to identify the concepts and categories that
they represented. Through constant comparison of the
ndings from within-rm and across-rm analyses the
core categories, including themes, concepts and possi-
bly relationships between variables, began to emerge.
This process was repeated until the conceptual cat-
egories were saturated, i.e. when no additional data
could develop the properties of a category any further.
Finally, the ndings were reported back to the inter-
viewed companies and their feedback were sought to
correct erroneous interpretations.
5. The integrated R&D network and its
managerial challenges
Before turning to the ndings on the role of
researcher mobility in the internal transfer of knowl-
edge, a description of the specic R&D organisational
structures encountered is needed. Different interna-
tional R&D structures entail very different knowl-
edge transfer efforts and different levels of integration
among geographically dispersed units.
As can be seen from Table 2, overseas R&D centres
are actively engaged in drug discovery activities in one
or more therapeutic areas. Although all the companies
in the sample have an R&D presence in several conti-
nents, the actual distribution of research phases varies
across companies. In particular they have chosen dif-
ferent points of the drug discovery process on which
to centralise resources to achieve economies of scales
and on which to decentralise to achieve specialisation.
Since the merger of Glaxo Welcome and Smith
Kline Beecham in 2000, the early phase of
GlaxoSmith Klines discovery process has become a
global function located in the UK, US, Italy and Spain,
although 90%of the activity is concentrated in research
sites in the US and the UK. Once molecules have been
identied and optimised they are passed over to the
Centres of Excellence of Drug Discovery (CEDDs),
which are aligned by therapeutic areas; chemists and
biologists, expert in the disease in question, can work
closely to bring the compound to the proof of concept
phase. According to an R&D manager the CEDDs are
almost set up like independent operations if they were
small biotechnology companies. They compete for the
resources that are distributed according to how close
their molecules are to the market and how well they
have performed. Once the compound has reached the
proof of concept stage, it is transferred to other global
functions that might be located elsewhere, to be further
In Aventis, there are three drug-discovery sites that
are set up as entrepreneurial units and compete on
a global basis for resources. Each site has responsi-
bilities from the early phase of a project up to the
proof of concept phase, but the project team gets sup-
port from the so-called global functions of: (1) lead
generation, which provides support for genomics and
high-throughput screening technologies and chemical
libraries; (2) lead optimisation, which provides support
for drug metabolism, pharmacokinetics, drug safety,
clinical discovery, human pharmacology and labora-
tory animal and welfare. There is a Global Drug Devel-
opment centre in Bridgewater (US), from where all
clinical development activities subsequent to the proof
of concept stage are coordinated. Clinical trials are
carried out all over the world, monitored by regional
development centres in Paris for Europe, in Tokyo for
Japan, in Bridgewater for the US. This R&D organi-
sation has been in place since the creation of Aventis,
which resulted fromthe merger of Hoechst and Rh one-
Poulenc in 2000.
In Schering since 2001 there have been ve
Research Business Areas (in vivo diagnostic, neurol-
ogy/immunology/cardiovascular, dermatology, gender
health care and oncology), which are attached to three
research sites. The target identication phase is car-
ried out in all three locations, but the studies of lead
identication and lead optimisation are only under-
taken in Berlin and Richmond. The activity of each
research site is supported by three regional research
centres in Europe, Japan and the US. In Novartis and
Roche, each research site performs all the phases of
the drug discovery process up to the pre-clinical stage.
The compound is further developed in Basel and in
the US. In AstraZeneca each site is a fully edged
P. Criscuolo / Research Policy 34 (2005) 13501365 1357











1358 P. Criscuolo / Research Policy 34 (2005) 13501365
R&D facility engaged in activities which range from
earlydiscoverytolife-cycle management ina particular
therapeutic area. More than one R&D unit is normally
involved in different phases of the drug discovery pro-
cess supported by centres of excellence in enabling
Usingthe taxonomyproposedbyGassmannandvon
Zedtwitz (1999), it can be seen that drug discovery is
organised as an integrated-network structure. In each
location a critical mass of scientists and technicians
specialised in a set of targets for a disease area are
brought together, facilitating personal interaction and
the exchange of tacit knowledge. The activity of each
centre is supervised by a central research unit, which
could either be in the home country or in the US (as, for
example, in Aventis and Novartis), to avoid duplication
and favour knowledge integration. As pointed out by
Ramirez (2003), the organisation of research activities
according to therapeutic areas has been made possible
by the new heuristic in drug discovery that has allowed
scientists to focus on specic groups of targets in par-
ticular centres of expertise. This international division
of labour enables the rm to access multiple external
knowledge sources from centres of excellence around
the world and/or internal knowledge sources from
research units with a strong technological background
within the organisation. Centres of excellence for a par-
ticular enabling technology (such as genomics, high-
throughput screening, combinatorial chemistry and
bio-informatics) support the activity of these research
The drug development process appears, however, to
be organised as a polycentric decentralised structure,
in Gassmann and von Zedtwitzs (1999) taxonomy.
This structure is characterised by a geographically dis-
persed network of autonomous units with their own
assets and capabilities, which allows themto respond to
local demands and opportunities. Development activi-
ties are indeed concentrated in a few locations in order
to achieve critical mass and economies of scale and
tend to be located in both the home country and the
US to be near the largest markets and the regulatory
authorities. These units usually conne themselves to
developing the results from research units located in
the same region and are coordinated by a global devel-
opment centre (for example, Bridgewater for Aventis).
The degree of interaction among the development cen-
tres is not as intense as among the research network
although they frequently carry out studies for other
development centres whenever these have problems of
insufcient capacity.
6. The role of international assignements and
other forms of mobility
The existing organisation of research activities
in pharmaceuticals is, therefore, quite complex and
involves a high degree of communication among geo-
graphically dispersed research centres that try to oper-
ate in an integrated way during the drug discovery
process. A particular project will be handed-over from
one laboratory to another as it progresses through the
different phases of the drug discovery and develop-
ment process. Thus, results from the drug discovery
process must be transferred to the units carrying out
development activities. Moreover, knowledge acquired
in a particular therapeutic area also needs to be trans-
mitted not only to other units working in the same area
but also to other centres working in different therapeu-
tic areas where it could be employed.
The management and functioning of this complex
organisation relies heavily on ICT applications such as
intranets, portals, project websites, internal databases
of compounds, videoconferencingandemail. However,
face-to-face interaction is important not only for trans-
ferring less codied forms of knowledge but also for
buildingthe necessarytrust, andthus generatingknowl-
edge sharing through electronic communication. As
one scientist noted:
We coordinate our work through phone calls, emails,
but personal contact is crucial; you need to sit together
and you can work much better together if you have built
a personal relationship which is necessary for building
trust . . . the ow of information happens when there is
Given the apparent importance of face-to-face inter-
action and the creation of personal relationships it
seems natural to investigate what use is made of inter-
national assignments of researchers. Table 3 shows the
characteristics of the interviewed researchers and the
main purposes behind their international assignments.
Some important ndings should be noted. First, the
majority of researchers were seconded for a period of
P. Criscuolo / Research Policy 34 (2005) 13501365 1359
Table 3
Interviewee prole and international assignment characteristics
Job description Major activity R&D function Assignment Period Main purpose of
Proposed by
Research chemist Search for new compounds Discovery 6 months Personal development Management
Medicinal chemist Synthesis of compounds Discovery 1 year Organisational
Medicinal chemist Combinatorial chemistry Discovery 6 months Knowledge acquisition Scientist
Senior scientist Development of
high-throughput screening
Discovery 1 year Knowledge transfer Scientist
Senior scientist Synthesis of compounds Discovery 6 months Personal development Scientist
Toxicologist Toxicology studies Development 1 year Organisational
Product line extension
Creation of new formulations
and products from existing
Development 6 months Personal development Scientist
Bio-analyst Measurement of drug
metabolism in the plasma
Development 3 years Knowledge transfer Management
Genomics researcher Analysis of gene-expression
Discovery 6 months Knowledge acquisition Management
Modelling and simulation of
biological processes
Discovery 10 months Knowledge transfer Management
Analysis of how the drug
works in the human body and
how the body reacts to the
Development 5 years Knowledge transfer Management
Head of therapeutic
Development of new
Discovery 2 years Knowledge acquisition Management
6 months or 1 year, which seems in line with the overall
trend in the international mobility of the work force in
MNEs mentioned in Section 2. Second, although the
most common reason is to transfer or acquire knowl-
edge, international assignments of researchers are also
used for personal and organisational development.
Third, even though international assignments appear
to be more frequently employed in the drug discovery
process, which reects the higher level of integration
and interaction among the research units, they are also
used in the drug development process. Finally, these
secondments can either be imposed by the manage-
ment or they can be proposed by the scientist because
he/she considers the experience to be a useful part of
his/her personal development.
Even though experience abroad is not articulated as a requirement
for advancement within the company and no promise of promotion is
made before the assignment, researchers know that gaining experi-
ence on how R&D activities are organised in another facility is very
benecial to their careers.
But beyond these general trends it is also use-
ful to look one-by-one at the individual assignments
described in Table 3. One scientist spent 6 months in the
US as a visiting scientist in a dedicated bio-informatics
group formed as a result of a joint venture with an
American company. This assignment was aimed at
acquiring new technological expertise in the eld of
genomics that was missing in the European R&Dfacil-
ity and was needed to undertake gene-chip experiments
in the European site. Similarly, another senior scientist
went for 1 year to a US research facility to implement
a new high-throughput screening technology, which
had been developed in the home country laboratory.
The company believed that the relocation of the key
scientist involved in the development of the newequip-
ment would not only facilitate the transfer of technical
knowledge but would also avoid the not invented here
syndrome (Katz and Allen, 1982). In another company,
sabbatical visits were used for transferring expertise in
combinatorial chemistry and automation technologies
from one site to another.
1360 P. Criscuolo / Research Policy 34 (2005) 13501365
However, this socialisation mechanism is perceived
not only as a way for channelling transfers of knowl-
edge but also as a way of improving relations between
R&D locations, and thus it is used as a coordination
mechanism. This emerges very clearly from the fol-
lowing quote:
Historically, communication and cooperation between
European and US sites has been difcult due to the
different research philosophies. During the mid 1990s
there was more competition than cooperation with
them. During that time it was even difcult to exchange
knowledge. To overcome cultural differences and to
increase a common understanding among colleagues
fromdifferent continents staff exchange programs have
been established. Now, after the rst exchange of staff,
things are improving . . . People taking part in the
exchange programme should be the bridge between the
two sites. It is important to knowpeople, it is much eas-
ier to deal with them if you have worked with them
(emphasis added).
Similarly in another company a scientist said:
The R&Dfacility where I went was part of a company
we just acquired the year before and people there were
not very happy about this unfriendly take over. One
of the purposes of this exchange was to overcome the
initial reservation and try to build a collaboration that
would make us colleagues rather than competitors.
In all instances, this experience enabled researchers
to have better interaction and communication with the
people they met in the other facility. Interaction with
colleagues inside the company appeared to be the most
important source of learning, while very little interac-
tion seemed to have taken place between the seconded
researchers and other host country rms, universities,
or scientists. This nding conrms that the main pur-
pose of these assignments is network building among
researchers working in geographical distant R&D cen-
tres and increasing awareness of who knows what
in other centres. These relationships are in most cases
maintained once the scientist has returned to his/her
home laboratory. As stated by one researcher:
This experience made me knowing people to whom
I have been in touch later on my job, once back in
Europe. I got a better understanding of the technology
they were using in the US and I was also able to record
the contact people in the US subsidiary.
Researchers who had been on secondment seemed
to have also gained a better understanding of the R&D
organisation and management adopted in the US cen-
tre and this appears to be one of the primary goals
attributed to this type of socialisation mechanism by
the top management of the companies. According to
the head of a therapeutic area:
For me, in terms of delivering projects globally, it
is important that people [when sent to another R&D
centre] understand what it takes to deliver things in
a different country and how to do it most effectively.
When these people go back to their original site they
actually strengthen the site they came from because
they know how this organisation works globally rather
than knowing only how it works based on what they
have seen in one site . . . when you send someone
in another R&D site you have to make sure that [the
international assignment] is seen as a family thing not
an individual thing (emphasis added).
Thus, while scientists perceive the foreign expe-
rience as a chance to broaden their knowledge and
acquire new skills, top management see it as an oppor-
tunity to increase the productivity of the company by
getting the researchers acquainted with how the R&D
function is managed in other facilities.
However, even though rms acknowledge the cru-
cial role of scientists in the international diffusion of
knowledge and rely heavily on the ability and private
initiative of individual scientists to span the bound-
aries of their own area of specialisation and interact
with other scientists, the use of international assign-
ments is very limited. Across all rms in 2003 only
1% of the R&D staff employed was on secondment to
another unit in the same company for a period longer
than 6 months and secondments are generally organ-
ised only when needed. Only in one division of one
particular company is there an organised exchange
programme where a scientist from the European divi-
sion spends 1 year in the US facility while a scientist
from the US replaces her/him at the European site.
This programme has only been in place for the past
2 years and is an attempt at replacing the less popular
P. Criscuolo / Research Policy 34 (2005) 13501365 1361
sabbatical option, where a scientist goes to a foreign
R&D laboratory but is not replaced by another scien-
tist. However, this low degree of long-term mobility
is accompanied by a high degree of short-term mobil-
ity associated with the implementation of cross-border
projects and/or temporary assignments (from 1 week
to 6 months) for the transfer of specic expertise. As
observed by an R&D manager: because people travel
more often for weekly periods, there is less need to
send them for longer periods. Cross-border projects
are widely used because of the organisational structure
of the R&Dactivities, which often requires researchers
from different locations at different stages of the drug
discovery project. They also aim to replace long-term
secondments of researchers. As explained by an R&D
Resources are managed globally but people stay at the
site. For example imagine you have a diabetes project
in the European site where the expertise is and you need
chemical expertises. It might happen that the chemist is
working in the US facility. Typically you do not need
to move the chemist from the US to Europe because
you just include this particular chemist in the research
These international project teams allow individuals
from distant locations and from different cultural and
technological background to be brought together reg-
ularly. However, cross-border team projects are seen
more as an unavoidable consequence of the R&D
organisation than as a strategic means for increas-
ing knowledge diffusion inside the geographically dis-
persed R&D network. This clearly emerges from the
comments of an R&D manager:
The one-location teamis the preferable model because
it is the more efcient, but the realityof our organisation
is that most of our teams have members based in at least
two countries and some of them three. My personal
view is that if you can have one location team you are
going to be better off, if you can have all sitting in one
corridor is goingtoworkbetter. But this is the exception
to the rule.
Temporary assignments and short visits are also
quite often employed and they are easy to organize and
do not require senior manager authorization.
People will travel quite easily to have face-to-face
meetings. It is quite easy to bring someone over from
another facility. For example we had a resource short-
fall in the headquarters and we brought two people with
that particular skill from North America to work here
for three weeks. We do this quite often. It is resource
driven process, not part of individual personal devel-
This type of short-term assignment is not only
adopted when there is a shortfall of resources but is
also employed for transferring specic expertise as is
the case when a project passes fromthe discovery to the
development phase. At this stage of the drug discov-
ery process researchers from the development function
will work closely with the discovery team from 3 to 6
months before the compound has been identied.
Another way of bringing together people from
across the organisationwithsimilar expertise is through
seminars on specic scientic and technology areas.
In only one company is there an annual international
research meeting where all the researchers present their
work. This type of event does not take place in other
companies because of the size of the R&Dorganisation
and it has been replaced by smaller workshops focused
on a particular discipline. Similarly, companies pro-
mote and ofcially support communities of practice
(Wenger, 1998) and technology interest groups. These
communities are formed by people with expertise in a
target family, such as kinesis or protease, or in a dis-
cipline such as molecular biology, chemistry, pharma-
cology. The task of this group of experts is to integrate
and diffuse knowledge across different locations and
therapeutic areas.
The members of these communi-
ties interact regularly through intranet, meetings and
formal workshops. Technology councils are set up to
discuss problems faced in using particular technolo-
gies, such as those used in combinatorial chemistry or
high-throughput screening. These councils provide a
forum where individuals can share best practice and
some of the challenges they face in using a particular
technology. Another organisational measure adopted
by one company to gather people from different thera-
peutic areas and locations is cross-disciplinary project
proposal review boards.
They are also supposed to span outside the rms boundaries and
follow the developments of their scientic elds outside the rm.
1362 P. Criscuolo / Research Policy 34 (2005) 13501365
Thus, in line with what is found by B eret et al.
(2003) and with the general trend in international
labour mobility in MNEs, the adoption of an integrated
R&D network has lead to an increase in researchers
mobility but in the form of short-term visits which are
either connected to cross-border projects or organised
for transferring specic skills.
The general impression from the interviews with
R&Dmanagers is that the use of long-termmobility has
declined and is now very limited. Secondments abroad
were more often employed in the past when each centre
was organised as a completely self-sufcient R&Dunit
specialised in certain therapeutic areas where most of
the phases of the drug discovery and development pro-
cess were carried out. Because all the R&D staff from
research to development were in one place, there was
less short-term researcher mobility and cross-border
team projects were almost never implemented. Long-
term international assignments of researchers instead
were employed more often than in the current organi-
sation model as a mechanism for transferring complex
and location-specic knowledge.
There are two main factors explaining this low level
of long-term mobility of researchers in the form of
international assignments. First, it is very expensive
to send someone to another facility (according to an
R&D manager it doubles the cost of employment) and
the scientists themselves are reluctant, perhaps more
so than before, to leave lifestyles behind to take on
assignments abroad. Second, the R&D organisational
structure itself hampers long-term researcher mobil-
ity, a point emphasised also by B eret et al. (2003).
The high-degree of technological distance between
research sites implies that competences are not easily
transferable from one site to another, making mobility
difcult. As acknowledged by an R&Dmanager, scien-
tists do not move among different R&D units, because
the expertise they have is often on a specic therapeutic
area that cannot be applied in other research contexts.
This emerged very clearly from interviews with those
seconded researchers whose primary purpose was not
to transfer specic knowledge to the foreign R&D
When I was in the US I developed a completely new
class of drugs. I wanted to work more on that class of
compounds, but I did not, because it would have meant
implementing here a different program.
I was not able to apply much of what I have learnt in
the US because we use a different approach.
However, it is possible to identify a longer-term
potential for cross-fertilisation arising from exchanges
of personnel.
Coming from traditional medicinal chemistry it was
my rst chance to get in touch with automated chem-
istry. After staying three months in the automated
chemistry group in our US subsidiary I was aware of
the scope and limitations of the approach which helped
a lot in contributing to a global technology strategy.
While I was in the USI had started a collaboration with
a company expert in gene-chip technology. This expe-
rience improved my innovation activity because we are
trying to work more and more in my eld with these
micro-areas and this time in the US was a trigger for
me to think very hard on how I could use these tools.
7. Conclusion
This paper has analysed the roles of international
assignments and other forms of inter-unit researcher
mobility in six of the largest European pharmaceu-
tical MNEs as a coordination and technology trans-
fer mechanism. In these companies the innovative
efforts in the drug discovery phase are carried out
in an integrated network structure characterised by
inter-unit geographical, technological and organisa-
tional distance, which could hinder knowledge sharing
and coordination of the R&D activities. In this organ-
isational structure international assignments are used
to achieve three main goals. First, researchers are sec-
onded to other R&D units to transfer technical knowl-
edge from the headquarters or to acquire technical
knowledge fromother researchfacilities. Second, inter-
national assignments are used for individual leader-
ship development. Third, assigning researchers to other
R&D units is employed as a coordination mechanism,
i.e. as a means of reducing organisational distance
through social interaction among researchers. The sec-
ondment experience appears to increase researchers
overall understanding of R&D organisation and man-
agement in the other unit, helping themto identify who
knows what, and fostering the creation of personal
P. Criscuolo / Research Policy 34 (2005) 13501365 1363
relationships with other researchers. Thus, researcher
mobility seems to help overcoming some of the bar-
riers to knowledge sharing inside the R&D network
and fully exploiting the benets of this organisational
However, it was found that long-term international
assignments are not widely used even though com-
panies rely heavily on the initiative of individual
researchers to interact with other colleagues and to
enhance R&D synergies across projects. They have
been replaced by other, short-term forms of mobility
which are less costly and do not require the relocation
of the family. Researchers from different R&D centres
meet each other quite regularly either because there
are involved in cross-border team projects or because
they particip