The Interaction between Competition, Collaboration and Innovation in Knowledge Industries

by

Keyvan Vakili

A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy

Joseph L. Rotman School of Management University of Toronto

© Copyright by Keyvan Vakili 2013

ii

The Interaction between Competition, Collaboration and

Innovation in Knowledge Industries

Keyvan Vakili

Doctor of Philosophy

Joseph L. Rotman School of Management

University of Toronto

2013

Abstract

The three studies in this dissertation examine the relationship between the decision of market

participants to compete or collaborate on their innovation strategies and outcomes as well as the

broader industry structure and technological progress. The first study analyzes the impact of

modern patent pools on the innovative performance of firms outside the pool. Theories generally

predict that modern patent pools have a positive impact on innovation by reducing the cost of

access to the pool’s technology, but recent empirical research suggests that patent pools may

actually decrease the innovation rate of firms outside the pool. Using a difference-in-difference-

with-matching methodology, I find a substantial decline in outsiders’ patenting rate after the pool

formation. However I find that the observed reduction is mainly due to a shift in firms’

investment from additional patentable technological exploration toward implementing the pool

technology in their products. The results shed light on how the interaction between cooperation,

in the form of patent pooling, and competition shapes firms’ innovative strategies by enabling

opportunities for application development based on the pooled technologies.

In the second study, I examine the impact of restrictive stem cell policies introduced by George

W. Bush in 2001 on the U.S. scientists’ productivity and collaboration patterns. Employing a

iii

difference-in-differences methodology, I find that the 2001 Bush policy led to a decline in the

research productivity of U.S. scientists. However, the effect was short-lived as U.S. scientists

accessed non-federal funds within the United States and sought funds outside the United States

through their international ties. The results suggest that scientists may use international

collaborations as a strategic means to deal with uncertainties in their national policy

environment.

In the third study, I examine the effects of the fragmentation of patent rights on subsequent

investment in new inventions. Using a theoretical model and an empirical analysis of the

semiconductor industry, I seek to shed light on the contingency factors that shape the role of

technological fragmentation in explaining the investment decisions and appropriation strategies

of firms. The results provide a dynamic explanation of the interplay between firms’ R&D

investment, their patenting strategies, and technological fragmentation.

iv

Dedication

I dedicate this thesis to my parents, who have always been my source of inspiration, to my

brothers who have made me laugh through good times and bad times, and to Paria, without

whose love and support none of this would have been possible.

v

Acknowledgments

It would have been impossible to finish this dissertation without the constant support and

guidance of my advisor, mentor, and friend, Anita McGahan. I am extremely grateful to you for

encouraging me through every step of my experience at Rotman, for supporting me in every

possible means, and for allowing me to grow as a researcher over the past five years. I would

also like to thank my committee members – Sarah Kaplan, Ajay Agrawal, and Brian Silverman –

for your great support and brilliant comments and suggestions. You have been incredible

examples and mentors to me. My special thanks to Sarah Kaplan for helping me sharpen my

ideas and become a better researcher.

I would also like to thank everyone in the Rotman strategy group and Rotman community for

providing a friendly and stimulating research environment over the past five years. I am

particularly grateful to all the professors who have generously attended my practice talks and

helped me with their constructive feedback. I am also grateful to Alyson Colon for her

tremendous assistance.

I additionally would like thank all of my fellow graduate students for supporting me along the

way and making my experience at Rotman pleasant. Laurina Zhang, Elizabeth Lyons, Florenta

Teodoridis, Octavio Martinez, Sandra Barbosu, Jillian Chown, and Christian Catalini have been

great friends and wonderful colleagues. I am thankful to you for all the joyful and memorable

moments we have spent together.

vi

Table of Contents

Dedication ...................................................................................................................................... iv

Acknowledgments............................................................................................................................v

Table of Contents ........................................................................................................................... vi

List of Tables ................................................................................................................................. ix

List of Figures ................................................................................................................................ xi

List of Appendices ........................................................................................................................ xii

Introduction ................................................................................................................................1 1

Competitive Effects of Collaborative Arrangements: Impact of the MPEG-2 Pool on 2

Outsiders’ Innovative Performance...........................................................................................11

2.1 Introduction ........................................................................................................................11

2.2 Competitive Effects of Modern Patent Pools.....................................................................14

2.3 Theory ................................................................................................................................16

2.3.1 Why Outsiders? ......................................................................................................16

2.3.2 Theoretical Mechanisms ........................................................................................17

2.4 Institutional Background: MPEG-2 Patent Pool ................................................................20

2.5 Empirical Strategy .............................................................................................................21

2.5.1 Estimation Equation ...............................................................................................23

2.6 Data and Sampling .............................................................................................................25

2.6.1 Data Sources ..........................................................................................................25

2.6.2 Treatment Sample ..................................................................................................25

2.6.3 Control Samples .....................................................................................................26

2.7 Results ................................................................................................................................28

vii

2.7.1 Effect of Pool Formation on Technologically Proximate Outsiders ......................28

2.7.2 Alternative Explanations and Robustness Tests ....................................................29

2.7.3 Mechanisms ...........................................................................................................31

2.8 Conclusion .........................................................................................................................37

Scientists as Strategic Investors: Evidence from the Impact of the Bush hESC Policy on 3

U.S. Scientists’ Research Productivity......................................................................................53

3.1 Introduction ........................................................................................................................53

3.2 History of hESC Research and Policies .............................................................................56

3.3 Theory ................................................................................................................................57

3.3.1 Prior Research on the Impact of the Bush Policy ..................................................57

3.3.2 Theoretical Development .......................................................................................59

3.4 Empirical Methodology .....................................................................................................63

3.5 Data ....................................................................................................................................67

3.6 Empirical Analyses ............................................................................................................69

3.6.1 Descriptive Statistics ..............................................................................................69

3.6.2 Estimation results ...................................................................................................71

3.7 Discussion and Conclusion ................................................................................................75

The Impact of Technological Fragmentation on Firms’ R&D Investment and Patenting 4

Behavior ....................................................................................................................................88

4.1 Introduction ........................................................................................................................88

4.2 Prior Research on Technological Fragmentation ...............................................................91

4.3 The Model ..........................................................................................................................93

4.3.1 Boundary Conditions and Their Implications ......................................................101

4.4 Empirical Methodology ...................................................................................................103

4.4.1 Data ......................................................................................................................103

4.4.2 Variables ..............................................................................................................104

viii

4.4.3 Model Specification .............................................................................................107

4.5 Results ..............................................................................................................................108

4.6 Discussion and Conclusion ..............................................................................................109

References ....................................................................................................................................120

ix

List of Tables

2.1: Average patenting rate of firms in each pair of treatment and control samples pre-pool

formation ................................................................................................................................45

2.2: The effect of the formation of the MPEG-2 pool in 1997 on the weighted patenting rate

of firms in technological proximity to the pool using linear regression models. ..........................46

2.3: The effect of the formation of the MPEG-2 pool in 1997 on the weighted patenting rate

of firms in technological proximity to the pool using the Poisson model with robust standard

errors ................................................................................................................................47

2.4: The effect of MPEG-2 implementation on weighted patenting rate of firms after the

formation of the MPEG-2 pool. .....................................................................................................48

2.5: The effect of the MPEG-2 pool on the tendency of firms towards mentioning MPEG-2

implementation in annual 10-K reports after the pool formation. .................................................49

2.6: The effect of the formation of the MPEG-2 pool in 1997 on the weighted patenting rate

of licensees versus non-licensees. ..................................................................................................50

2.7: The effect of the formation of the MPEG-2 pool on subsequent inventions based on

initial pool patents ..........................................................................................................................51

2.8: The effect of the formation of the MPEG-2 pool on the financial performance of treated

firms ................................................................................................................................52

3.1: hESC research in the United States and in other developed countries with flexible

policies ................................................................................................................................82

3.2: Summary statistics ..................................................................................................................83

3.3: Yearly marginal difference between U.S. and non-U.S. scientists’ research output. .............84

3.4: Year-by-year marginal difference between U.S. and non-U.S. scientists’ hESC research

output for the subsamples of scientists with and without international collaboration

experience pre-2001 .......................................................................................................................85

3.5: Annual federal and state-level grants for hESC research. ......................................................86

3.6: Year-by-year marginal difference between U.S. and non-U.S. scientists’ cross-border

collaborations ................................................................................................................................87

4.1: Sample statistics ....................................................................................................................114

4.2: Main results: The impact of base and domain fragmentation levels on firms’ R&D

investment and patenting rate in semiconductors between 1980 and 1999 .................................115

x

4.3: Robustness checks ................................................................................................................116

xi

List of Figures

1.1: Broad research agenda . ..........................................................................................................10

2.1: Number of patent pools per decade. .......................................................................................40

2.2: The effect of the formation of the MPEG-2 pool in 1997 on the weighted patenting rate

of outsiders in technological proximity to the pool (treated firms) compared to other firms in

similar industries (control firms). ..................................................................................................41

2.3: The effect of the formation of the MPEG-2 pool in 1997 on the weighted patenting rate

of proximate outsiders in the matched treatment sample compared to firms in the matched

control sample ................................................................................................................................42

2.4: The yearly treatment effects of the MPEG-2 pool formation .................................................43

2.5: The number of patents with the term “MPEG-2” in their abstract per year before and

after the formation of the MPEG-2 pool in 1997. ..........................................................................44

3.1: Average number of hESC publications by hESC scientists in the United States versus

hESC scientists from other developed countries with flexible hESC policies. .............................78

3.2: Average number of ESC publications with collaborators from emerging countries with

flexible hESC policies by hESC scientists in the United States versus hESC scientists from

other developed countries with flexible hESC policies. ................................................................78

3.3: Estimated density function of Stem Cell publications by U.S. hESC scientists versus

hESC scientists in other developed countries with flexible hESC policies in 2000 ......................79

3.4: Impact of the 2001 Bush policy on the weighted hESC research output of U.S. scientists

compared to scientists in other developed countries with flexible hESC policies. .......................80

3.5: Difference between the yearly treatment effects of U.S. hESC scientists with

international collaboration experience pre-2001 and U.S. hESC scientists without such

experience ................................................................................................................................80

3.6: Impact of the 2001 Bush policy on the number of U.S. hESC scientis’ publications with

collaborators outside the United States in emerging countries with flexible hESC policies . .......81

3.7: Impact of the 2001 Bush policy on the research output of U.S. scientists’ collaborators

outside the United States . ..............................................................................................................81

4.1: The sequence of the two-stage model ...................................................................................111

4.2: The licensing-patenting decision ..........................................................................................112

4.3: Base and domain fragmentation levels based on patent citations .........................................113

xii

List of Appendices

Appendix A.1 ..............................................................................................................................117

Appendix A.2. ..............................................................................................................................117

Appendix A.3. ..............................................................................................................................117

Appendix A.4. ..............................................................................................................................118

Appendix A.5. ..............................................................................................................................118

Appendix A.6. ..............................................................................................................................119

Appendix A.7. ..............................................................................................................................119

Appendix A.8. ..............................................................................................................................119

1

Chapter 1

Introduction 1

Innovation plays a fundamental role in the development and growth of individuals, firms, and

countries (Agrawal & Goldfarb, 2006; Griliches, 1994; Jones, 2009a; Romer, 1994; Teece,

1986). As a result, a growing body of literature aims to unravel the sources of superior

innovative performance (Cockburn, Henderson, & Stern, 2000; Cohen & Levinthal, 1990; Dosi,

1988). An established regularity from this body of work is that changes in the levels of

competition and collaboration have profound impacts on the innovative performance of market

participants (Aghion, Bloom, Blundell, Griffith, & Howitt, 2005; Henderson & Clark, 1990;

Kaplan & Tripsas, 2008; McGahan & Silverman, 2006; Powell, Koput, & Smith-Doerr, 1996;

Teece, 1986). However, we know much less about the underlying mechanisms through which the

decisions of market participants to compete and collaborate shape their innovation strategies and

outcomes. The three studies in this dissertation attempt to address this gap. Moreover, these

studies together illustrate how such decisions by market participants would further translate into

broader changes in the industry structure and its underlying technological bases, which can in

turn influence the future levels of competition and collaboration in the market. Furthermore,

these studies illustrate how new policies, at the organizational, regional, or national levels, would

influence the path of technological progress by shaping the competitive and collaborative

relationships between market participants and their innovation strategies and outcomes.

Figure 1.1 depicts these research objectives in the frames of a broad research agenda. The first

link in the figure (top left) represents the potential mechanisms through which the decisions of

market participants to form or break their competitive or collaborative ties would influence their

innovation strategies and performance. In these studies, I focus on three particular mechanisms.

First, I investigate how these decisions change the opportunity structures in the related

technology and product markets and thus influence the direction and the level of investment in

follow-on technologies and products. Second, I explore how these decisions influence the

2

strategies that market participants pursue to appropriate the returns to their investments. Third, I

explore how they influence the risks associated with the availability of resources required for the

innovation process. The second link (top right) highlights the potential impacts of changes in the

market participants’ innovation strategies on the structure of the industries in which they operate

and the technology bases of these industries. Two aspects of the industry structure are of

particular interest: first, the size distribution of firms in the market and, second, the position of

firms in the value chain. Similarly, two aspects of the technology environment are of interest:

first, the distribution of patent rights in a particular technological domain and, second, the life-

cycle of the technologies. The last link (bottom) underlines the channels through which changes

in the related industry and technology structures can shape the decisions of market participants in

regard to forming or breaking their competitive or collaborative ties in the future. Each of the

three studies in this dissertation focus on one or more of these links.

Understanding these links and their underlying mechanisms is of significant importance from

theoretical, empirical, and practical perspectives. From a theoretical standpoint, these

mechanisms lay down the fundamental building blocks of any model of interplay between

competition, collaboration, and innovation. Any accurate explanation of the interaction between

these three elements requires a holistic picture of these mechanisms, the way they interact with

each other and the conditions under which they operate. Furthermore, these mechanisms work as

the roadmap for any empirical analysis of innovation and science policies; firms’ and

individuals’ innovation strategies and performance; and the competitive and collaborative

behavior of firms and individuals through the innovation process. They highlight the sources of

endogeneity. They show how changes in the innovation strategies of market participants can be

both the result of their decisions to compete or collaborate, and the cause of them. At the same

time, they point out the data that need to be collected and the dynamics of the phenomenon of

interest. From a policy point of view, awareness of these mechanisms can not only prevent

unintended policy consequences but also expand the range of policy levers that can be used to

effectively shape the innovation direction and performance of market participants.

3

However, at least two major issues create challenges for scholars studying these links and their

underlying mechanisms. The first arises from the co-evolution of the innovation strategies of

market participants, their competitive and collaborative interactions, and their surrounding

technological and institutional environment. A change in a firm’s competitive and collaborative

relationships can lead to changes in its innovation strategy and outcome. Similarly, shifts in the

innovative performance of an organization can trigger changes in its broader strategy thus

affecting its competitive and collaborative ties. Similarly, changes in the external institutional

and technological environments can be both the cause and the effect of shifts in market

participants’ innovation strategies and their competitive and collaborative ties. This co-evolution

raises endogeneity concerns that need to be addressed in any empirical examination of these

links.

The second issue is that the underlying mechanisms of interest may act as countervailing forces

and their individual effects may be obscured under various circumstances. This issue can in part

explain why studies implemented in different empirical contexts have reported different or even

inconsistent results. For example, while several theoretical and conceptual studies suggest that

patent pools – as one form of collaborative arrangements among competing firms – improve

innovative outcomes (Gilbert, 2004; Shapiro, 2001), some recent empirical studies have found

that patent pooling diminishes innovation (Joshi & Nerkar, 2011; Lampe & Moser, 2010).

Similarly, while Blundell, Griffith, & Van Reenen (1999) report a positive correlation between

competition intensity and innovation, Aghion, Harris, & Vickers (1997) find a negative

association between the two.

The three studies in this dissertation address these issues by using a mixture of theoretical

modeling and careful empirical designs. In order to address the first issue, in the first two studies,

I take advantage of natural experiments in the external policy or technological environments in

order to establish a causal link between market participants’ decisions to collaborate or compete

and their innovation strategies and outcomes. In the third study, I develop a theoretical model of

firms’ R&D investment and patenting behavior to explore the dynamic interplay between firms’

innovation strategies and the characteristics of their external technology environment. While

4

employing natural experiments in the first two studies helps me isolate and study single links

depicted in Figure 1.1 and eliminate the endogeneity issues associated with the co-evolution of

different moving elements in the picture, the model in the third study does in fact enable me to

explore this dynamic phenomenon. To address the second issue outlined above, in my empirical

analysis, I take advantage of the fact that different mechanisms have differential impacts on

various aspects of firms’ innovation strategies such as investment level, innovation direction and

type, and appropriation strategies. Analyzing these various aspects along with supplementary

analyses thus enables me to disentangle the effects of these mechanisms and study them

separately.

These three studies are set in different empirical settings: digital media standards, stem cell

research, and the semiconductors industry. Despite their differences, innovation plays a crucial

role in the overall performance of firms and individuals in all three settings (Bessen & Hunt,

2007; Hall & Ziedonis, 2001; Joshi & Nerkar, 2011; Vakili, McGahan, Rezaie, Mitchell, & Daar,

2013a). All three have experienced considerable technological advancements since the mid-

1990s and continue to grow rapidly. In addition, all three contexts are the subject of public

policy. Several developed and developing countries have made considerable investments in these

economically and technologically significant domains. These settings also host considerable

collaboration among organizations (Hall & Ziedonis, 2001; Joshi & Nerkar, 2011; Vakili et al.,

2013a). These features make these settings excellent candidates for studying the interplay

between collaboration, competition, and innovation. Employing these three different settings

have two important advantages. First, they enable me to focus on different links and different

mechanisms depicted in Figure 1.1. Second, each setting provides a unique opportunity to

explore different contingencies on which each mechanism of interest may or may not operate.

The first study (Chapter 2) focuses mainly on the first link in Figure 1.1 by exploring the impact

of an important modern patent pool on the innovation strategies and performance of firms

outside the pool. A patent pool is an agreement between multiple parties to license a group of

related patents on common terms to each other and to third parties. These pools are increasingly

used as a means for overcoming technological holdup that would otherwise arise from the

5

overlapping property rights of competing organizations. An important stream of prior studies on

patent pools and similar collaborative arrangements have focused on their total welfare effect

(Gilbert, 2004; Lerner, Strojwas, & Tirole, 2007; Lerner & Tirole, 2004; Shapiro, 2001), design

(Chiao, Lerner, & Tirole, 2007; Layne-Farrar & Lerner, 2011; Lerner et al., 2007; Lerner &

Tirole, 2004), and impact on constituting members or components (Joshi & Nerkar, 2011;

Rysman & Simcoe, 2008). Yet we know little about their broader impact on the innovative

performance of non-collaborating firms. And if they have any impact, how do the affected firms

respond strategically to the formation of these arrangements?

To answer these questions, I focus on the impact of the MPEG-2 pool – the pool formed around

the core patents on the MPEG-2 technology standard – on the innovation strategies and

performance of firms outside the pool. Theory predicts that, on the one hand, outsiders might

respond by shutting down investments in related innovation as the result of increased litigation

risks and competitive pressures from the pool members, but, on the other hand, the resolution of

uncertainty may spark investments in follow-on technologies and products based on the pooled

technology. Using a difference-in-differences with matching methodology, I find that the

formation of the MPEG-2 pool was associated with a marked reduction in patenting rates of

firms outside the pool. However, this does not reflect a reduction in innovation performance

overall but rather a shift from patenting to application development, resulting in sharp changes in

these firms’ positions in the product market. Furthermore, I find that the formation of the pool

actually boosted cumulative technological innovations based on the MPEG-2 patents in the pool.

More broadly, the results suggest that patent pools can shift the locus of investments in

innovation by firms and encourage cumulative innovation. The findings also highlight how

modern patent pools would shift the life cycle of pooled technologies by reducing uncertainties

around particular technology standards. Furthermore, the results suggest that these collaborative

arrangements can create fundamental shifts in the industry structure by shifting large incumbents

from the upstream technology market to the downstream product market and inviting a wave of

newcomers into the upstream level to seize the opened up technological opportunities (link 2).

6

In the second study (Chapter 3), part of a joint research program with Professor Anita McGahan,

Professor Will Mitchell, Professor Abdallah Daar, and Dr. Rahim Rezaie, I examine how

individual scientists use international collaboration as a strategic tool to deal with negative

shocks in their institutional environments as well as to seize emerging opportunities (link 3). I

also investigate how the choice of collaborators affected the research productivity of both U.S.

scientists and their non-U.S. collaborators (link 1). In order to deal with the potential

endogeneity issues, I take advantage of a policy shock in the United States in 2001 that

introduced restrictions on federal funding available for human embryonic stem cell (hESC)

research.

Employing a difference-in-differences methodology, I find that the 2001 Bush policy had a

negative impact on the research productivity of U.S. scientists compared to scientists in other

developed countries with flexible hESC policies. However, the effect of the Bush policy was

short-lived as U.S. scientists accessed non-federal funds within the United States and sought

funds outside the United States through their international ties. In particular, the results suggest

that international collaboration is used by scientists as a strategic means to manage the risks in

the local institutional environment that would negatively affect the resources required through

the research process. At a broad level, the findings suggest that scientists are not passive

recipients of new institutional conditions placed on their research activities. In contrast, they

actively engage in shaping their external institutional environment and collaboration is one of the

means through which they can achieve this. From a policy perspective, the results emphasize

how ignoring scientists’ strategic responses in the design of new science policies can lead to

unintended consequences such as unplanned cross-border scientific, technological, and capability

spillovers. Moreover, the results show how awareness of scientists’ strategic behavior can

provide valuable levers for policymakers who seek to attract international talent and to import

research and technological capabilities.

In the third study (Chapter 4), I explore how firms change their innovation strategies when their

inventions rely on multiple technological components patented by a fragmented set of firms in

the market. Scholars have previously identified incentives for firms to engage in patent

7

proliferation strategies when they face such a fragmented technology base (Hall & Ziedonis,

2001; Noel & Schankerman, 2013; Ziedonis, 2004). Prior research has also documented

countervailing disincentives related to high infringement costs associated with investing in

technologies with fragmented technological bases (Heller & Eisenberg, 1998; Murray & Stern,

2007; Shapiro, 2001). These two forces together create a theoretical puzzle where it is not clear

under what conditions the firms will continue investing in a fragmented technological area, how

they modify their innovation strategies in the face of technological fragmentation, and how their

innovation strategies shape the future technological progress of the field.

Using a formal theoretical model and an empirical analysis of the semiconductor industry, I

illustrate how these incentives and disincentives influence firms’ innovation and appropriation

strategies under different circumstances. Furthermore, through this study, I provide a dynamic

picture of the impact of firms’ appropriation strategies on their subsequent innovation

investments and patenting behavior (links 1, 2, and 3). The results suggest that, at lower levels of

technological fragmentation, as the number of patent holders on which a firm depends increases,

there is a higher likelihood that the firm engages in patent proliferation strategies to appropriate

its innovation outcomes. Also, the model depicts how pursuing such patent proliferation

strategies by all firms in a technological domain can lead to higher levels of technological

fragmentation in a technological domain. Subsequently, as the total level of fragmentation in a

technological domain grows, firms become less likely to invest in follow-on inventions thus

becoming less likely to come up with patentable inventions. Furthermore, the results also show

the differential impact of technological fragmentation on small versus large firms, suggesting

that as a technological domain becomes more fragmented, it becomes more dominated by larger,

established firms.

The findings from these three studies shed light on each of the links depicted in Figure 1.1. In

regard to the first link, the first study illustrates significant shifts in firms’ innovation direction

from technological exploration to downstream application development in response to the

formation of modern patent pools and the resulting decline in the uncertainty around the pooled

technology standards. Furthermore, the results from the second study on the impact of the 2001

8

Bush policy show how changes in scientists’ collaboration patterns lead to changes in their

research productivity and direction.

With respect to the second link, all three studies depict the fundamental impact of the firms’ and

individuals’ local innovation strategies on the broader industry structure and the future

technological and scientific progress. The first study in particular shows how the change in

firms’ innovation strategies in response to the formation of modern patent pools can lead to

major shifts in their positions in the technology and product markets. The second study illustrates

how scientists’ strategic reactions to external shocks generate long-term impacts on local and

global scientific progress through knowledge and capability spillovers. The last study highlights

the impact of firms’ short-term patenting strategies on the future technological progress through

altering the technological fragmentation level.

Finally, both the second study and the third study explore the third link in detail. The second

study on the impact of the 2001 Bush policy suggests that scientists expand their international

collaborative ties as a risk management mechanism when faced with uncertainties in the external

regulatory environment that controls their research inputs (link 3). The third study shows how

increased levels of fragmentation in a technological domain – as the result of firms’ patenting

strategies in the past – would gradually result in lower levels of investment in follow-on

technologies and products, causing potential technological stagnation.

At the broad level, these three studies attempt to expand our understanding of how firms and

individuals break the tradeoff between competition and collaboration through the innovation

process and how their short-term local strategies can have long-term structural consequences.

Focusing on different levels of analysis – firm level and scientist level –provides an opportunity

to understand the similarities and differences between the firms’ and individuals’ innovation

strategies. Prior literature has vastly conceptualized firms as strategic actors that actively engage

in shaping their external environment, while modeling individual scientists as passive recipients

of the changes in their external environment. However, the results from the second study on the

impact of the 2001 Bush policy on scientists’ research behavior demonstrate that individual

scientists act as strategically as firms with respect to deciding where to invest their time and

9

effort, shaping their external institutional environment, and protecting their investments using

risk management practices. At the same time, the results highlight important differences as well.

For example, while the first and the third studies suggest that firms craft their innovation

strategies with the goal of maximizing their profits, results from the second study suggest that

individual scientists have a more long-term view toward their investments and a stronger

preference for following the global scientific trends rather than seeking the highest short-term

payoff to their investments.

The results of these studies have important implications for policymakers and managers. The

result that firms and individuals change their innovation strategies in response to policy suggests

that the effects of such policies are complex. A simple implementation of policy that does not

account for strategic responses may lead to unanticipated outcomes. The range of such policies is

extensive: those that affect the inputs into the innovation process; policies that govern the

cooperative behavior amongst market participants; and policies that influence the value that

firms and individuals can potentially appropriate from their innovation investments.

Furthermore, insights from these studies equip managers with a more holistic picture of the

enduring impact of their short-term innovation strategies on the industry structure and

competition environment.

In the end, it is important to note that there are still several other mechanisms and questions that

remain unexplored in the broad research agenda presented in Figure 1.1. For example, while the

first study examines several mechanisms through which modern patent pools would affect the

innovation strategies of established firms outside the pool, it does not explore how these

arrangements affect the innovation strategies of new entrepreneurial ventures. Furthermore, one

may wonder how the findings apply to similar collaborative arrangements such as standard

setting organizations and R&D consortia. Similarly, while the second study highlights the role of

international collaboration as a risk reduction mechanism, it does not delve into other risk

management strategies that scientists may pursue to protect their research investments against the

uncertainties in their external environment. The question also remains as to how much the

findings hold or differ at the organizational level. Finally, in regard to the last study, several

10

questions remain unanswered: How can policymakers address the adverse impacts of strategic

patenting and potential technological stagnation in fragmented technological areas? How

effective are new organizational forms, such as patent pools and patent clearing houses, in terms

of resolving technological fragmentation? How does including other appropriation strategies,

such as secrecy and first-mover advantage, alter the predictions of the developed model? These

are just a few questions that naturally arise from these three studies. Answers to these questions

can further advance our understanding of the complex interaction between competition,

collaboration, and innovation.

Figure 1.1: Broad research agenda

11

Chapter 2

Competitive Effects of Collaborative Arrangements: 2Impact of the MPEG-2 Pool on Outsiders’ Innovative Performance

2.1 Introduction

This paper focuses on the impact of modern patent pools on the innovative performance of firms

outside the pools. While theory generally predicts that modern patent pools increase innovation

rates (Gilbert, 2004; Lerner & Tirole, 2004; Shapiro, 2001), two recent empirical studies suggest

that pools may have anti-innovative impacts (Joshi & Nerkar, 2011; Lampe & Moser, 2010).

These empirical findings raise new questions about the underlying mechanisms that link patent

pools and similar collaborative arrangements (including standard setting organizations and R&D

consortia) to the type, intensity and trajectory of firms’ investments in innovation. In this study, I

investigate four different mechanisms through which patent pools differentially affect the return

on innovative investments by firms outside the pool: access to licensed pool technology,

increased competition, increased litigation risk, and reshaped opportunities in product and

technology markets. By separately assessing the impact of these mechanisms, I seek to provide a

more nuanced picture of the ways in which collaborative arrangements influence the project

returns to the innovation of outsiders, and thus explain how, under certain specific conditions,

the implementation of a patent pool may create fundamental shifts in firms’ innovation

strategies.

This paper follows several previous studies that have explored the impact of patent pools and

similar collaborative arrangements on innovation (Joshi & Nerkar, 2011; Lampe & Moser, 2010;

Lerner et al., 2007; Lerner & Tirole, 2004; Rysman & Simcoe, 2008; Shapiro, 2001). The paper

most closely relates to the recent empirical works by Lampe and Moser (2010, 2012) and Joshi

12

and Nerkar (2011). All three studies report significant decline in the patenting rate of firms after

the formation of patent pools. Their results thus raise questions regarding whether modern patent

pools have detrimental impacts on innovation or not. The answer, though, depends on what

mechanisms are driving the observed impact on firms’ patenting rate. To the extent that this

decline is due to anticompetitive forces such as an unanticipated increase in litigation risk from

the pool members, it can reflect an undesirable anti-innovation impact of patent pools. In

contrast, to the extent that this decline is driven by a change in the innovation direction of firms

from more patentable innovations to less patentable ones, it does not support such a claim.

Therefore the focus of this paper is to shed more light on the mechanisms through which patent

pools influence the firms’ innovation strategies and performance.

Building on the previous literature, I first develop a theoretical framework to highlight four

potential mechanisms through which patent pools may affect outsiders’ innovation strategies and

performance: increasing litigation risk, increasing competition intensity, lowering the cost of

developing follow-on technological innovations, and establishing a dominant technology

standard. Subsequently, I verify the empirical findings of Joshi and Nerkar (2011) and Lampe

and Moser (2010) by examining the effect of the MPEG-2 pool on the patenting rate of firms

outside the pool. I use a difference-in-difference with matching methodology to compare the

change in the patenting rate of firms in technological proximity to the pool after the pool

formation. Several robustness tests are used to address the identification challenges in order to

establish a causal relationship. Next, using multiple empirical strategies and sources of

complementary data, I test the four potential mechanisms, examining the differential effects of

these mechanisms on firms’ financial performance, their R&D direction, and follow-on patents

based on the pool technology.

Similar to Lampe and Moser (2010) and Joshi and Nerkar (2011), I also find that, when

measured as the weighted patenting rate, the formation of the MPEG-2 pool caused a decline in

the patenting rate of outsiders in technological proximity of the pool. The estimation results

suggest that the average weighted patenting rate of these firms dropped by more than 35 percent

after the formation of the MPEG-2 pool relative to a closely matched control sample of firms.

13

However, analyzing the potential driving mechanisms, I find that the decline in the firms’

weighted patenting rate can be explained by a change in the direction of firms’ R&D investment

from further technological inventions to implementing the pool technology in their current and

future products. In other words, once the pool was formed, firms in its technological proximity

made a strategic shift from further (more patentable) technological exploration to (less

patentable) technological exploitation. I do not find evidence for the mechanisms associated with

an undesirable anticompetitive impact of patent pools. The findings suggest that although the

formation of modern patent pools may diminish the patenting rate of technologically proximate

outsiders, this decline should not be necessarily interpreted as a detrimental effect of patent pools

on innovation.

In fact, surprisingly, I find that the formation of the pool boosts further technological innovations

and patenting based on the pool technology. However, the extra patents are not introduced by the

pool members or the technologically proximate outsiders. These results together suggest that the

formation of modern pools can create fundamental shifts in the position of firms in the market. In

particular, I find that the formation of the MPEG-2 pool created incentives for technologically

proximate outsiders to redirect their investments into application development in the downstream

market. At the same time, it created incentives for other firms and organizations to invest in

developing follow-on innovations based on the pool technology. At the broader level, my

analysis shows that modern patent pools can in fact create fundamental changes in the life cycle

of related industries and technologies by creating a shift from exploring multiple alternative

technologies towards converging around a dominant technology standard.

The paper is organized as follows. First, I review the literature on the effect of patent pools on

innovation. Subsequently, I develop a theoretical framework that explains four mechanisms

through which patent pools can affect outsiders’ innovative performance. I then provide some

institutional background on the MPEG-2 pool. This is followed by the empirical strategy and the

base results. Finally, I present the results from the mechanism analyses and conclude.

14

2.2 Competitive Effects of Modern Patent Pools

The exponential increase in patent awards in the past two decades combined with the ambiguity

in the scope of patented inventions has led to fragmented technology fields with numerous firms

claiming overlapping property rights (Bessen & Hunt, 2007; Heller & Eisenberg, 1998;

Parchomovsky & Wagner, 2005; Ziedonis, 2004). Several theoretical and empirical works

suggest that this situation can lead to technological holdup due to high litigation risks, excessive

royalty rates and soaring transaction costs associated with cross-licensing negotiations (Heller &

Eisenberg, 1998; Lemley & Shapiro, 2007; Josh Lerner, 1995; Murray & Stern, 2007).1 One

solution to this problem is to put related patents into a common pool and consolidate all the

associated rights into a central entity (pool organizer) for the purpose of joint licensing (Merges,

1999; Shapiro, 2001). Patent pools are thus private arrangements between multiple parties to

license their interdependent patents to each other and to third parties usually based on defined

fixed terms.

Since the formation of the first modern pool in 1997, the MPEG-2 pool, patent pools have

consistently increased in number, economic significance and the extent of the technological

fields in which they are implemented (World Intellectual Property Report, 2011). A recent

estimate (Clarkson, 2003) shows that in 2001, the total sales of devices that were fully or

partially based on pooled technologies equaled more than $100 billion. The numbers continue to

increase (Figure 2.1).

Patent pools also have a very long history in the U.S., dating back to the formation of the Sewing

Machine Combination pool in 1856. At the same time, however, these arrangements raised

serious antitrust concerns (Carlson, 1999; Gilbert, 2004). With stronger enforcement of antitrust

laws in the early 20th century, traditional pools were increasingly recognized as anticompetitive

arrangements and many of them were eventually dismantled by antitrust authorities (Gilbert,

2004). No major pool arrangements had been established since the end of World War II until the

introduction of the MPEG-2 pool, the first instance of modern pools, in 1997.

1 See Chapter 4, “The Impact of Technological Fragmentation on Firms’ R&D Investment and Patenting Behavior”

for an extended analysis of technological fragmentation on firms’ R&D investment in follow-on technologies.

15

The MPEG-2 pool was originally formed around 27 patents that were essential to comply with

the MPEG-2 technology standard – one of the technologies developed for digital media

transmission. The structure of the MPEG-2 pool was later endorsed by the Department of Justice

for other parties interested in forming patent pools and gradually became the template for all

modern pools. The major difference between modern pools and their traditional counterparts is

that the former is only allowed to include essential and complementary patents. This policy is

intended to prevent opportunistic behavior by pool members to the extent possible (Shapiro,

2001). Because serious concerns about traditional pools (pools formed before 1997) rested on

their anticompetitive effect, most of the literature on modern pools has focused on their total

welfare effect (Gilbert, 2004; Lerner & Tirole, 2004; Shapiro, 2001) and their design (Chiao et

al., 2007; Layne-Farrar & Lerner, 2011; Lerner et al., 2007). The general prediction is that

modern pools should be welfare-enhancing based on the premise that they do not include

substitute or weak patents (Lerner & Tirole, 2004; Shapiro, 2001).

More recently, however, scholars have started to explore the impact of patent pools on

innovation at the firm level. Theoretical models of patent pools mainly predict that, unlike their

traditional counterparts, modern pools should encourage innovation by eliminating technological

holdups and decreasing the transaction and licensing costs associated with using the pool

technology (Carlson, 1999; Lerner & Tirole, 2004; Merges, 1999; Shapiro, 2001). Studying the

first traditional pool in U.S. history, the Sewing Machine Combination pool, Lampe and Moser

(2010) do, in fact, find a significant decline in the innovation rate of both pool members and

other firms in the sewing machine industry after the pool formation. In another working paper

Lampe and Moser (2012) report similar findings for the effect of traditional patent pools on the

patenting rate in 20 different industries in the 1930s. Their findings are consistent with

arguments regarding the anticompetitive and anti-innovation impact of traditional pools. In

another recent empirical study, Joshi and Nerkar (2011) investigate the impact of three modern

pools in the optical disk industry on the patenting rate of pool members and licensees.

Surprisingly, they also find a significant drop in the firms’ patenting rate after the formation of

these modern pools, concluding that modern patent pools may discourage innovation as well.

16

Their empirical findings raise serious concerns regarding the impact of modern patent pools on

innovation.

Interpreting these findings requires revisiting the underlying mechanisms that drive the negative

effect of patent pools on the innovation rate. While a drop in outsiders’ patenting rates due to

antitrust effects of patent pools or increased litigation risks can be interpreted as undesirable

effects, a decline in the innovation rate due to a change in firms’ R&D direction from

technological exploration to implementing the pool technology draws a more complex picture of

the effects of patent pools on innovation, which is not necessarily undesirable. The next section

describes these mechanisms and their effects on outsiders’ innovative performance in detail.

2.3 Theory

2.3.1 Why Outsiders?

This paper focuses on identifying and empirically examining the potential mechanisms through

which patent pools may impact the innovation strategies and performance of technologically

proximate outsiders. There are three reasons behind the focus on the outsiders rather than pool

members. First, despite evidence of the anticompetitive effects of their traditional counterparts,

the Department of Justice allowed modern patent pools to become established mainly because

they are believed to facilitate access to the pool technology for all firms in the market by

resolving fragmentation and technology holdups (Shapiro, 2001). As for the pool members, there

is little doubt that the formation of the pool should be to their benefit, or else they would have

not engaged in forming it in the first place. One should note that pool members are not legally

forced to form a pool.2 Hence, the question of whether or not modern pools harm innovation

mainly boils down to their impacts on outside firms that are working on technologies related to

the pool technologies.

Second, studying the impact of modern pools on outsiders can provide significant insights about

the impact of these arrangements on the industry structure, competition state, and the future of

2 It is theoretically possible that some firms join the pool formation process due to pressures from other participating

members. However, even in that case majority of pool members benefit.

17

technological progress in the field. A narrow focus on the handful of firms that formed the pool,

while important in its own right, is limited in terms of the insights it can provide about the

broader impacts of modern pools. Third, from a theoretical point of view, the mechanisms

through which these arrangements affect outsiders’ innovation strategies and performance would

be different from the mechanisms through which they affect those of the pool members. For

example, insiders may refuse to invest in developing follow-on innovations after the formation of

the pool in order to prevent cannibalization of their pooled technology. The same argument

however does not apply to the outsiders who do not have any share in the revenues generated by

the pool and thus do not financially gain from selling the pool technology itself. Considering that

exploring these mechanisms requires separate theoretical inquiry and empirical strategy,

focusing on the impact of modern pools on both the outsiders and the pool members into one

study is not practical.

Hence, this paper can be considered the first step towards unraveling some of the mechanisms

through which patent pools affect innovation by outsiders in technological proximity to the pool.

Future research can further advance insights from this study by exploring the mechanisms

through which modern pools influence innovation strategies of pool members, new ventures, and

other organizations such as universities and public research institutes.

2.3.2 Theoretical Mechanisms

Prior theoretical models of patent pools predict that modern pools can potentially eliminate

technological holdups and encourage innovation by lowering the negotiation, searching and

litigation costs associated with using the pool technology (Carlson, 1999; Lerner & Tirole, 2004;

Merges, 1999; Shapiro, 2001). According to the Antitrust Guidelines for the Licensing of

Intellectual Property, patent pools “may provide competitive benefits by integrating

complementary technologies, reducing transaction costs, clearing blocking positions, and

avoiding costly infringement litigation.” To the extent that patent pools successfully achieve this

goal, they can open up a window of opportunity for further technological developments based on

the pool technology. At the same time, the advantage of possessing a related established

technological asset suggests that outsiders in technological proximity to the pool would be

18

among the first to capitalize on their technological assets to seize this unblocked opportunity,

therefore experiencing a boost in their technological innovation rates based on the pool

technology. However, the recent empirical findings by Joshi and Nerkar (2011) on the negative

impact of modern pools on patenting rate challenge this prediction. Their results suggest that,

besides lowering the cost of developing follow-on technologies, there might be other

mechanisms through which modern patent pools affect innovation. Building on their work and

other theoretical and empirical studies on patent pools and similar arrangements, I argue that

there are at least three other mechanisms through which these arrangements may negatively

affect outsiders’ patenting rate.

First, patent pools may increase the litigation risk associated with developing new technologies

in the pool technological area for an outsider (Joshi & Nerkar, 2011; Lampe & Moser, 2010). It

is well known that firms with larger patent portfolios have a higher chance of winning

infringement lawsuits in court (Ziedonis, 2004). In fact, patent proliferation with the intention of

enlarging the patent portfolio is one of the main strategies used by firms in fragmented fields,

such as software and semiconductors, to protect their technological inventions (Bessen & Hunt,

2007; Noel & Schankerman, 2013; Ziedonis, 2004). Once a patent pool is formed, pool members

can act as a single party with a much bigger portfolio to enforce their property rights (Lampe &

Moser, 2010). As a result, because of the inherent vagueness of the scope of patented inventions

(Parchomovsky & Wagner, 2005; Shapiro, 2001), pool members can potentially use their

enhanced enforcement power to threaten any firm planning to develop a technology that will

potentially compete with the pool technology. Lampe and Moser (2010) in their analysis of the

Sewing Machine Combination pool, find evidence of an increase in litigation cases with

nonmembers as defendants after the formation of the pool. Furthermore, Simcoe, Graham and

Feldman (2009) show that firms, particularly small firms, are more likely to file litigation

lawsuits to enforce their rights once they disclose their patents to Standard Setting Organizations

(SSO). Considering that many of the current technology standards, including MPEG-2, are in

fact developed in these SSOs, their findings suggest that patent pools may indeed discourage

other firms from further technological development in technological proximity to the pool due to

increased litigation risk. Considering that firms cannot easily change their technological

19

capabilities in the short term, those with established technological assets in proximity to the pool

area may thus hold back or redirect their R&D investments to avoid any costly infringement

lawsuits. This shift in direction can not only lead to a decline in their technological innovation

and patenting rate, but can also harm their financial performance.

The second mechanism through which modern patent pools can adversely affect outsiders’

patenting rate is by negatively affecting their financial performance. A detrimental impact on

firms’ financial performance can directly translate into a negative impact on their R&D budget,

their innovation performance, and thus patenting rate. There are two possible scenarios in which

the formation of modern pools may cause a decline in outsiders’ financial performance. First, to

the extent that patent pools improve the financial performance of pool members by reducing their

litigation costs and boosting their licensing revenue, pool members would become stronger rivals

for proximate outsiders working on similar technologies. Second, lowering the cost of access to

the pool technology, patent pools may bring down the barriers to entry into the pool

technological area, thus increasing the rivalry by inviting a host of new entrants. An increase in

the number of rivals in the field may also “fish out” the technological opportunity pool, making

it more challenging for each single firm to innovate and thrive (Furman, Kyle, Cockburn, &

Henderson, 2005).

Finally, I argue that modern pools can lower outsiders’ patenting rate by reducing the uncertainty

in the downstream application development based on the pool technology, thus triggering a shift

in the outsider’s innovation strategies from investing in further patentable technological

innovations to less patentable product and process innovations. The majority of modern pools are

formed around some sort of technology standard whose different pieces are patented and owned

by different organizations in the market (Gilbert, 2004; Shapiro, 2001). This fragmentation in the

ownership rights is the source of high potential infringement costs and consequently high

uncertainty around the technology standard. Modern patent pools can reduce this uncertainty in

two ways. First, they transform a costly technology standard to an affordable one by suddenly

dropping the cost of access to the pool technology and streamlining the licensing process

(Merges, 1999; Shapiro, 2001). Second, the formation of the pool sends a strong and clear signal

20

that the main developers of the technology standard are committed to facilitating its adoption and

development over time. This signal is stronger when the pool members are considered to be the

pioneers in their respective technological domains. Thus, these two forces can help establish the

pool technology as the dominant technology standard in the industry. This mechanism is

particularly relevant when the pool is formed in a technological environment with several

competing standards and high uncertainty around which will become the dominant one in future.

Several works on the dominant design, standards, and general purpose technologies suggest that

the establishment of a dominant technological standard can lead to a shift in firms’ strategy from

further technological exploration towards technological exploitation, i.e., implementing the

standard in new products and markets (Besen & Farrell, 1994; David & Greenstein, 1990;

Gambardella & McGahan, 2010; Utterback, 1994). Such a strategic shift from technological

exploration to product implementation can translate into a decline in the outsiders’ patenting rate.

However, unlike the two previous mechanisms, it does not reflect a detrimental impact on

outsiders’ innovative performance, but rather a shift in their innovation direction and type.

These three mechanisms can all explain how modern patent pools induce a decline in the

outsiders’ patenting rate, however, they can operate simultaneously or separately, which makes

identification tricky. However each of these mechanisms has some unique features that help me

empirically identify them separately. I will elaborate on the identification strategies in the

following sections.

2.4 Institutional Background: MPEG-2 Patent Pool

On June 26, 1997, the Department of Justice (DOJ) sent a letter to the MPEG Licensing

Agreement (MPEG LA) approving the request to pool the patents “essential” to compliance with

the MPEG-2 standard. MPEG-2 is an international standard developed by the Moving Picture

Expert Group in the early ’90s and defined as “the generic coding of moving pictures and

associated audio information” (ISO/IEC 13818 MPEG2). The standard contains different coding

algorithms to compress and transfer digitalized video and audio signals. The MPEG-2 pool

initially contained 27 patents from nine different entities (eight corporations and Columbia

21

University). Thus, the formation of the MPEG-2 pool marked the first instance of modern patent

pools in history and soon became the template for all pools formed afterwards (Gilbert, 2004).

The approval letter from the DOJ clearly states that the portfolio only combines patents that are

complementary and essential.3 According to the MPEG-2 Patent Portfolio License, an essential

patent is “any patent claiming an apparatus and/or method necessary for compliance with the

MPEG-2 Standard under the laws of the country that issued or published the Patent.” The letter

also highlights that there is no technical substitute for any of the portfolio patents. Furthermore,

the pool agreement guarantees a non-discriminatory, non-exclusive international license with the

same terms and conditions to all would-be licensees.

The MPEG-2 pool grew rapidly; by the end of 1998, the number of patents in the pool was up to

226. By the end of 2011, the pool included more than 1,000 patents from 27 different entities and

was licensed to 1,518 different licensees. More importantly, the DOJ business review for the

MPEG-2 pool provided a template for pooling arrangements that would not violate antitrust

laws. In 1999, soon after the approval of the MPEG-2 pool, the DOJ approved a very similar

proposal for pooling the patents for Digital Versatile Disc (DVD) technology. Since then

numerous similar pool arrangements, mainly dealing with new technology standards, have been

approved. Many others are currently under construction (World Intellectual Property Report,

2011).

2.5 Empirical Strategy

The empirical strategy in this paper takes advantage of several features of the MPEG-2 pool

formation process in order to implement a difference-in-differences methodology (DID) and

establish the causal impact of the MPEG-2 pool on the innovative performance of the focal

outsiders. The goal is to compare the changes in the innovative performance of outsiders in

technological proximity to the pool (the treatment sample) after the formation of the pool

(treatment) to the changes in the innovative performance of a comparable group of firms (control

3 It is important to note that the Department of Justice did not assess the selected patents directly but rather

evaluated the lawfulness of the arrangement according to the claims of the License Administrator.

22

sample) that ideally were not affected by the formation of the pool. The main underlying idea is

that, under appropriate conditions, the outcome trend of the control sample after the pool

formation can be used as a proxy for the unobservable counterfactual trend for the treatment

sample.4 Thus, to the extent that the treated and control groups look similar pre-treatment, any

differences between their outcome trends post-treatment can be interpreted as the causal effect of

the treatment.

Two important assumptions are required for drawing a strong causal inference using the

difference-in-differences methodology (Imbens & Wooldridge, 2009). First, the treatment should

be exogenous to the treated sample; what Rubin (1990) labels as the “unconfounded

assignment.” There are multiple important features of the MPEG-2 pool that let me treat its

formation as a quasi-exogenous shock to the firms outside the pool. First, the letter from the DOJ

highlights that a group of independent patent experts was used to select the 27 essential patents

included in the pool. The experts reviewed around 8,000 U.S. patent abstracts and carefully

studied 800 patents belonging to 100 different assignees. Hence, nonparticipation in the pool was

not really a choice but rather the professional opinion of the hired experts that certain patents

should be excluded. Second, the formation process of the pool took place quickly. This made it

difficult for outside firms to predict the formation of the pool far in advance to be able to respond

accordingly by changing their R&D investments or innovation strategies before the pool’s

formation. Even for firms that may have had some early indication of the process, predicting the

approval of the DOJ would not have been easy considering both the DOJ’s negative view of

patent pools before 1997 as well as the absence of any pool formation for half a century. These

features suggest that the formation of the MPEG-2 pool can be considered as a quasi-exogenous

shock to the subjects of interest. It is important to note that the exogeneity claim here is only

relevant with respect to the firms that had no role in the formation of the pool and could not

predict it well in advance. By no means do I intend to claim that the formation of a patent pool is

a random event. Creating a patent pool is certainly a strategic decision by pool members and

other organizations involved in the formation process. However, to the extent that the formation

4 See Holland (1986) and Imbens & Woodbridge (2009) for a more detailed description of how to make causal

inferences using the difference-in-differences methodology.

23

of the MPEG-2 pool was not predictable by the outsiders of interest in this paper, the

endogeneity issue is not a concern. In the results section of the paper, I do some robustness tests

to assess this assumption more accurately. I further examine the possibility that the formation of

the MPEG-2 pool and the changes in outsiders’ innovation strategies may both be the result of a

change in the technological opportunity pool.

Second, a reliable causal inference requires treatment and control samples to be drawn from

similar distributions, or the “overlap assumption” (Imbens & Wooldridge, 2009). This

assumption implies that the control group should be similar enough to the treatment group in

order to be able to use its observable behavior as a valid proxy for the treatment groups’

unobservable counterfactual. Ignoring this condition may lead to biased results and

interpretations. In order to satisfy this condition, I use a matching method to construct a control

sample that exhibits similar behavior to the treated group before the formation of the pool.

2.5.1 Estimation Equation

In the first step, I measure the effect of the pool formation on the innovation rate of

technologically proximate outsiders in order to replicate the results reported by Joshi and Nerkar

(2011) and Lampe and Moser (2010, 2012). I use a straightforward difference-in-differences

estimator with firm and year fixed-effects to estimate the change in the innovation rate of firms

in the treatment sample relative to those in the control. Following previous works in the field, I

use firms’ citation-weighted patenting rate as the proxy for their innovation rate. Each patent is

weighted by the number of times it is cited by subsequent patents in the next four years since its

application date (Hall, Jaffe, & Trajtenberg, 2005).5 The results are robust to using a five- or six-

year window to count the number of forward citations. Several studies have shown a strong

correlation between the value of patents and their forward citation rate (Griliches, 1990; Hall et

al., 2005; Harhoff, Narin, Scherer, & Vopel, 1999). This analysis builds on recent similar works

using the patenting rate to investigate the impact of an institutional shock on a subsequent

5 The results are robust to using four- or six-year windows to count the number of forward citations.

24

innovation rate (Furman et al., 2005; Furman & Stern, 2011; Murray & Stern, 2007). The main

estimating equation is:

(1)

where stands for the weighted patenting rate of firm at year . I further use a log normal

transformation to account for the skewness of the data. is a dummy equal to

1 for firms in the treatment sample and 0 otherwise. is also a dummy equal to

1 for years after 1997, the year in which the MPEG-2 pool was formed, and 0 otherwise.

is a set of firm-specific time-varying controls including R&D intensity, capital intensity and

firm size. Since these data are mostly unavailable for private firms in the sample, they are only

used in estimations on the subsample of public firms. Following prior studies (Audretsch &

Feldman, 1996; Cohen, Levin, & Mowery, 1987), R&D intensity is calculated as the logarithm

of the amount of R&D spending divided by the total sales in each year. Similarly, capital

intensity is calculated as the log transformed ratio of a firm’s total assets to total sales. Firm size

is measured as the logarithm of the firms’ total sales per year. All three variables are lagged one

year to account for their delayed effect on a firm’s patenting rate.

and stand for firm and year fixed-effects respectively. Including firm fixed-effects helps

controlling for the effects of firms’ unobservable idiosyncratic characteristics. Furthermore, year

fixed-effects control for macroeconomic trends such as periods of technological expansion or

depression. is a firm-year error term uncorrelated with constructed dummies, fixed effects and

other control variables.

Coefficient captures the differential effect of the pool formation on the treatment sample

relative to the control sample, i.e., the treatment effect. A positive suggests that the treatment

has positively affected the innovation rate of firms in the treatment sample relative to those in the

control group and the opposite is true for a negative . For the sake of simplicity and robustness,

I use linear regression with robust standard errors. The sensitivity of results to other

specifications is tested in the robustness section. All the regressions are estimated for the time

25

period beginning three years before the formation of the pool in 1997 and ending three years

after (1994-2000).

In the second step I use several complementary analyses to investigate each of the mechanisms

that were described in the theoretical framework.

2.6 Data and Sampling

2.6.1 Data Sources

Several sources are used to collect the data for this study. The data on the MPEG-2 pool is

collected from the MPEG Licensing Agreement’s website6 and through direct contact with the

company. Patent data are extracted from the NBER and USPTO patent databases. I manually

matched the assignees from the NBER patent database with company names in the Compustat

Database to identify the public firms in the sample and collect their financial data including SIC

code, R&D investment, total assets and sales. Patent litigation data is collected from the Lex

Machina database,7 Internet searches and other secondary sources. Finally, the data on public

firms’ annual 10-K reports are collected through SEC’s EDGAR database.

2.6.2 Treatment Sample

I use two separate search techniques to identify the group of outsiders in the technological

proximity of the pool. In the first step, I identify all the firms owning patents that cited patents in

the MPEG-2 pool before the formation of the pool, taking advantage of a well-known feature of

patents: if patent A cites patent B, it indicates that the idea in patent A is partially built upon the

idea developed in patent B. Patent citations thus have been frequently used as a means to

measure knowledge flows between firms and the distance between their technological

capabilities (Jaffe, Trajtenberg, & Henderson, 1993; Mowery, Oxley, & Silverman, 1998). Using

this approach, I am able to identify all the firms that were directly working on technologies based

on the patents in the MPEG-2 standard before the pool formation.

6 See http://www.mpegla.com/main/default.aspx

7 See https://www.lexmachina.com.

26

In the second step, I identify all the firms owning any patent applied before 1997 with any of the

terms “MPEG-1,” “MPEG-2,” “H.261,” and/or “H.263” mentioned in their specifications. The

MPEG-2 standard is basically a major improvement over the MPEG-1 standard. Also, according

to technology reports and articles, H.261 and its antecedent, H.263, were the main competitors

for the MPEG-2 technology around the time when the MPEG-2 pool was assembled. The

considerable technical similarities between these technologies assure that any firms familiar with

one would also be familiar with the other. The union of both collections resulted in 239

organizations. I further excluded pool members and organizations other than for-profit firms. The

final treatment sample includes 175 firms.

These 175 firms have reported 33 different 4-digit SIC codes as their principal products

manufactured or major services provided. The most represented industries are communications

services (including telecommunication services, telephone communication, radio and television

broadcasting and cable and other pay television services), electronics (including semiconductors,

computer equipment and electronic components) and software (including packaged software and

motion picture & video software). The less represented industries include measuring and

controlling devices, photographic equipment and office machines. Most of these industries were

experiencing a boom cycle with increasing sales and number of active firms. Also, the sample is

mainly comprised of established firms. Large firms and public firms are over-represented in the

sample compared to the general distribution of firms in the corresponding industries.

2.6.3 Control Samples

The validity and reliability of difference-in-differences estimates depend heavily on the extent to

which the observable outcomes for the control sample can represent the unobservable

counterfactual for the treatment sample. The goal of this study is not to compare the effect of the

pool on insiders (members) and outsiders (nonmembers), but rather to identify how modern

pools affect those outsiders that are technologically close to the pool area compared to other

outsiders that are not ideally affected by the formation of the pool. Thus, the control group is

comprised of firms outside the pool that are not affected directly by its formation due to their

longer technological distance. In order to acquire robust results, I construct two control samples.

27

For the first control group, I extracted all the representative industries in the treatment sample

using the SIC codes of the public firms in the sample. Next, I collected all the public firms in

those industries excluding the pool members and the firms selected into the treatment sample. I

further limited the collected firms to those that had at least one patent before the formation of the

pool to assure that the control firms indeed engaged in some sort of technological innovation

before the formation of the pool. It resulted in a group of 2,166 firms in the first control sample.

Columns 1 and 2 in Panel A of Table 2.1 show the weighted patenting rate for firms in the

treatment and first control sample for the three years before the formation of the pool (1994 to

1996). Column 3 reports the t-test stats for the difference in means between the two samples.

Firms in the treatment sample have substantially more patents throughout the whole period,

suggesting a lack of sufficient overlap between the distributions of firms in the two samples.

While firms in the first constructed control sample serve as a good initial point of reference, they

cannot be considered as a reliable control sample. Hence, when I compare firms in the treatment

sample with firms in this control group, I simply intend to compare the treated firms to “average”

firms in the same industries rather than necessarily to a comparable group of firms with a similar

innovation track prior to the formation of the pool.

In order to resolve the overlap issue, in the second stage I match each firm in the treatment

sample with another firm from the control sample constructed in the previous step based on the

weighted patenting rate in each of the three years prior to the pool formation and the average

growth rate in the weighted patenting rate over the three-year period. Any treated firm with no

matched control is excluded from the matched treatment sample. This reduces the size of the

treatment and control samples to 67 firms in each group. Panel B in Table 2.1 shows the

weighted patenting rate for the matched treatment and control samples and the difference in their

means for each of the three years prior to the formation of the pool. As can be seen, firms in the

second control sample look much similar to those in the matched treatment group in terms of

pre-treatment innovation rate. The differences between the mean innovation rates are

insignificant for all three years.

28

One should note that using a matching method in this context produces more conservative

results. It is simply impossible to ensure that none of the selected firms in the matched control

sample were affected by the pool formation. There may be firms in technological proximity to

the pool that never cited any of the MPEG-2 patents and never mentioned any of the searched

terms (“MPEG-1,” “MPEG-2,” “H.261,” or “H.263”) in their patent specifications prior to the

pool formation. With more stringent matching criteria, the possibility of matching an

unidentified firm that should have been ideally selected into the treatment sample with an

identified firm in the treatment sample is more likely. This can potentially lead to attenuated

results as some of the treated and matched control pairs may in fact both similarly be affected by

the formation of the pool. However, since this issue works against finding any significant

treatment effect, I would be able to establish the robustness of findings as long as I get similar

results using both control samples.

2.7 Results

2.7.1 Effect of Pool Formation on Technologically Proximate Outsiders

The first set of results show the net effect of the MPEG-2 pool formation on the innovative

performance of firms in the treatment sample. Figure 2.2 depicts the trend in the innovation rate

of firms in the treatment sample compared to firms in the broad control sample three years before

and after the pool formation. The graph shows an obvious downward trend in the innovation rate

of treated firms compared to “average” firms in the broad control sample starting right after the

pool formation. Figure 2.3 shows the same trends for the matched treatment and control samples:

a negative impact on the average weighted patenting rate of the treatment sample relative to the

matched control sample. To analyze the effect formally, I now turn to the regression results.

Table 2.2 reports the difference-in-differences estimation results. The first column shows the

results for the regression with the broad control group (all other firms in representative

industries) using a linear regression model with robust standard errors. The interaction

coefficient ( ) shows a substantial negative effect

significant at the 1 percent level, implying that the formation of the pool resulted in a more than

29

60 percent fall in the weighted patenting rate of firms in technological proximity to the pool

compared to “average” firms in the same industries.

Column 2 shows the results for estimation with the matched control sample. Similar to the

previous case, the interaction coefficient is negative and significant, suggesting a 35 percent drop

in the innovation rate of firms in the treatment sample compared to those in the matched control

sample. The lower coefficient relative to the previous case can be partially due to the more

conservative nature of the estimates with the matched control sample as explained earlier. Figure

2.4 depicts the estimated yearly treatment effect using a flexible model that captures the

difference between the patenting rates of firms in the treated sample relative to those in the

control sample. The graph shows a substantial and significant decline in the patenting rate of the

treatment group relative to the matched control sample.

In model 3, I add two placebo treatments for the two years prior to the formation of the pool to

test for any pre-treatment decline in the innovation rate of firms or any divergence in the

behavior of the treatment and control groups prior to the formation of the pool. The results show

no significant decline in the innovation rate of either group, or any pattern of divergence between

the two, before the pool was assembled. The main coefficient of interest, , still suggests a

substantial decline in the innovation rate of treated firms post-treatment.

Finally, in model 4 I run the regression on the subsample of public firms controlling for one year

lagged R&D intensity, capital intensity and sales. The main result remains the same although less

significant, probably due to the smaller sample size. Overall, the results suggest that the

formation of the MPEG-2 pool led to a substantial drop in the weighted patenting rate of firms in

its technological proximity compared to other firms in similar industries but more

technologically distant from the pool.

2.7.2 Alternative Explanations and Robustness Tests

One may be concerned that the formation of the MPEG-2 pool and the decline in the patenting

rates of treated firms may both be caused by diminishing technological opportunities in the

market. In other words, both are strategic responses by two different groups of firms – pool

30

creators and firms in technological proximity to the pool – to receding technological

opportunities in the external market. To examine this possibility, I examine the general patenting

trends in the pool technological area before and after the pool formation. In the case of

diminishing technological opportunities, I expect to see an overall decline in the number of

patents on the MPEG-2 technology or other technologies dealing with digital media compression

and transmission. To examine this hypothesis, I first collected all patents with the term “MPEG-

2” in their abstract. Figure 2.5 depicts the number of such patents pre- and post-pool formation.

The graph shows a steady growth in the number of patents that are directly related to the MPEG-

2 technology. Expanding the set of patents based on a wider set of keywords (“MPEG-1,”

“MPEG-2,” “H.261,” and “H.263”) generates similar patterns. These trends are not consistent

with the depleting technological opportunity story. Furthermore, a gradual decline in the

technological opportunity pool should arguably lead to a gradual decline in firms’ technological

innovations. This is not consistent with the observed sudden drop in the outsiders’ patenting rate

after a period of considerable growth in patenting rate pre-pool formation. These trends thus

reject a diminishing technological opportunity argument. However, it is important to note that

the causal impact of the pool on the outsiders’ patenting rate can indeed occur through changing

the technological opportunity structure for outsiders. I will explore this idea more thoroughly in

the next section.

Another source of concern is that the MPEG LA experts might have selected firms that later

decided not to join the pool. The absence of any data on these firms makes it hard to predict how

they react to the formation of the pool. However, to the extent that they may redirect their

strategy away from engaging in any further patenting (to avoid engaging in potential IP

infringements with pool members) and towards taking advantage of their strategic position in

licensing negotiations with pool customers, including them in the treatment sample may lead to

overestimating the pool’s real effect on the innovation rate of other outsiders. To address this

issue, I identified all firms with any track of IP disclosure to the MPEG working group before the

formation of the pool. These firms comprise the main candidates for selection into the pool. In an

unreported analysis, I removed these firms from the sample and repeated the main regressions.

The exclusion has modest effects on the significance of coefficients particularly in regressions

31

with matched samples mainly due to smaller sample sizes, but for the most part the results are

similar.

Furthermore, to assure that the results are not mainly driven only by those firms working on

competing technologies (H.261 and H.263), I broke down the treatment sample into two sub-

samples: firms with patents mentioning “H.261” or “H.263” in their specifications and others. I

repeated the analysis on each sub-sample separately (unreported here). For both samples, I find

similar results to those reported in the previous section. Furthermore, there is no significant

difference between the innovation rates of firms in the two sub-samples.

Finally, I tested the sensitivity of results to other functional forms. Table 2.3 reports the

regression results for the panel Poisson model with robust standard errors. The results are

essentially similar in terms of size and significance. The interaction coefficient suggests a

sharper decline in the innovation rate of treatment firms compared to those estimated in the

linear models. Results are also robust to Negative Binomial specification. Furthermore, using a

five- or six-year window to measure the forward citation rate of patents does not change the

results.

2.7.3 Mechanisms

2.7.3.1 Test of Shift in Innovation Strategy

The first mechanism I examine here is the change in R&D investment direction of firms as a

strategic response to the pool formation. As explained in the theory section, a switch from

investing in developing new technologies to implementing the pool technology in current and

new products may indeed explain the decline in the technologically proximate outsiders’

patenting rate.

A direct examination of such shifts in the investment direction of firms requires detailed data on

how firms spend their R&D money. In the absence of such detailed data, I rely on the

information public firms disclose in their annual 10-K reports to investors. Each report contains

activities during the past year and plans for the future. These reports typically have

comprehensive sections on business and product strategies of firms and their current and future

32

products. Considering at that time MPEG-2 was an increasingly important technology in

industries such as broadcasting, software and semiconductors, mentioning any activities towards

implementing it in current or future products could potentially give firms a positive edge in

investors’ evaluations. This lets me identify any investment towards implementing the MPEG-2

technology in the annual reports of collected firms.

In particular, I collect the annual reports of all public firms in the treatment and control samples

between 1994 and 2001 in the top six representative industries.8 I further exclude the firms that

had no annual reports disclosed before 1997 or after 1998 to assure coverage before and after the

formation of the pool. Subsequently, I identify the first report in which each firm mentions a

product compatible with the MPEG-2 technology or a plan to implement it in current or new

products. In doing so, I am able to identify all the firms that have reported their attempt towards

implementing the MPEG-2 technology between 1994 and 2001. This results in identifying 14

firms in the pool of treated firms and 26 firms in the associated constructed control sample.

In order to test this mechanism, I use a two-step approach. In the first step, I verify whether

mentioning any strategic intention towards implementing the MPEG-2 technology subsequently

has any association with a lower patenting rate. I expect the firms that mentioned any effort

towards implementing MPEG-2 technology to experience a decline in their innovation rate

relative to other firms. In the second step, I test whether firms in technological proximity to the

pool were more likely to mention MPEG-2 implementation in their annual reports compared to

other firms in the control samples.

For the first step, I use the following estimation equation:

(2)

where is equal to 1 for firms mentioning MPEG-2 implementation in at

least one of their annual reports between 1997 and 2000 and 0 otherwise. Everything else is

8 Selected industries include Radio and TV Broadcasting, Telephone Communications, Cable and Other Pay

Television Services, Communication Services, Services-Packaged Software, and Semiconductors.

33

similar to the main estimating equation. Column 1 of Table 2.4 reports the estimation results for

the full sample of eligible firms. The negative and significant indicates that firms mentioning

MPEG-2 implementation indeed experienced a decline in their weighted patenting rate after the

pool was formed. In order to assure that the results are not solely driven by firms in the treatment

sample, I repeat the estimation focusing only on firms in the control sample. The new

estimations are reported in the second column of Table 2.4. The interaction coefficient is still

negative, significant, and of the same magnitude. Taken together, the results suggest that any

attempt towards implementing MPEG-2 technology is indeed associated with a lower patenting

rate post-pool formation.

In the next step, I compare the rate at which firms in technological proximity to the pool shifted

their R&D investments towards implementing the MPEG-2 technology relative to their

counterparts in the constructed control sample using the following difference-in-differences

estimation model:

(3)

where is equal to 1 if firm has already mentioned MPEG-2 implementation in any of its

annual reports in years or before. All the other variables are similar to what was explained in

the base estimation model. The interaction coefficient, , measures the difference in the response

rates of treatment and control samples. If firms in technological proximity to the pool are, in fact,

more likely to shift their investments towards implementing the MPEG-2 technology, I expect

to be positive and significant. I only focus on the broad control sample since all five firms in the

matched treatment and control samples that mentioned MPEG-2 implementation in their 10-K

reports solely belong to the treatment sample.

The regression results are depicted in Table 2.5. The positive and significant interaction

coefficient indicates that firms in the treatment sample had a higher rate of investment

redirection towards implementing the MPEG-2 technology relative to control firms in the time

period of analysis. The size of the coefficient suggests that treated firms were 16 percent more

likely to mention MPEG-2 implementation in their annual reports in the four years following the

34

formation of the pool relative to their counterparts in the control samples. Results from an

unreported hazard rate analysis also shows that firms in the treatment sample are on average

eight times more likely to mention MPEG-2 implementation in their annual reports than those in

the broad control sample.

The results from this two-step analysis strongly support the idea that the decline in the patenting

rate of technologically proximate outsiders is due to a shift in their innovation strategies from

further technological exploration to implementing the pool technology in their current and future

products. Moreover, unreported supplementary analyses show a continuous increase in the R&D

investment and R&D intensity of these firms after the pool formation, providing further evidence

for this mechanism.

2.7.3.2 Test of Increased Litigation Risk

Several studies have used the number of litigation cases to identify the level of litigation risk

(Agarwal, Ganco, & Ziedonis, 2009; Simcoe, Graham, & Feldman, 2009). However, in this case,

using the change in the number of litigation cases involving pool patents after the pool formation

may result in misleading conclusions. On the one hand, any increase in the number of litigation

lawsuits can be interpreted as an increase in the litigiousness of pool members and, consequently,

an increase in the litigation risk associated with developing technologies based on the pool

technology (Agarwal et al., 2009; Simcoe et al., 2009). On the other hand, any decrease in the

number of such lawsuits may indeed be a signal that outsiders have retreated from the pool

technological area altogether due to the increased perceived litigation risk. In other words, any

change in the number of litigation cases (either an increase or decrease) can potentially lead to

the same conclusion: an increase in litigation risk. In fact, between 1994 and 2000 there has been

only one litigation lawsuit involving one of the MPEG-2 pool patents, filed by France Telecom

against Compaq Computer Corporation.

Considering the difficulty involved with using litigation data, instead I use the pool licensing

data to test whether litigation risk had any role in explaining the decline in the innovation rate of

outsiders. Assuming that acquiring the pool license eliminates potential litigation threats to a

35

great extent, a comparison between innovation rates of licensees versus non-licensees in the

treatment sample provides a unique opportunity to measure the effect of the litigation risk.

Hence, I first identify all the firms that acquired a license to the pool technology by the end of

1997 (the year in which the pool was formed). Subsequently, I use a difference-in-differences

estimation to measure the difference in the innovation rate of licensees versus others in each of

the treatment samples separately. It is important to note that, in this case, I measure the

difference in the patenting rates of two sub-samples of the treatment sample. This is similar to

the approach used by Simcoe, Graham and Feldman (2009) to measure the difference between

behaviors of small firms versus large firms after disclosing patents to SSOs. The following

estimation equation is used:

(4)

where is equal to 1 if the treated firm is a licensee by the end of 1997 and 0 otherwise.

Everything else is similar to the main estimation model explained earlier. The results are

presented in Table 2.6. Columns 1 and 2 report the results for the linear and Poisson models

respectively. The interaction term is insignificant in both models, indicating no difference in the

patenting rate of licensees and non-licensees after the pool formation. Furthermore, the

coefficients of year dummies (not reported here) show a similar decline in the innovation rate of

both groups after the pool formation. The size of the decline in each model is, in fact, very

similar to the previously measured decline in the base estimations. The results are also robust to

the choice of cutoff year to identify licensees (the end of 1998 or 1999).

Furthermore, any increase in litigation risk after the formation of the pool should have a

depressing impact on the follow-on innovations based on the pool patents. To test the effect of

the pool on subsequent innovations, building on the method proposed by Mehta, Rysman and

Simcoe (2010) and Rysman and Simcoe (2008), I compare the forward citation patterns of pool

patents to a closely matched sample of patents. Each pool patent is matched with another patent

from a similar technological class, the same application year, and a similar pre-pool citation

trend. Their method makes use of the variance in processing time of patents from the same

application-year cohort to identify the age effects. Using their method with little modification, I

36

estimate the effect of the pool formation on the yearly citation rates of pool patents relative to

their matched counterparts after the pool formation. I use the following difference-in-differences

model:

(5)

where stands for the log transformed number of citations received by patent in year (as

measured by the application year of the citing patent), and are the fixed effects for the

application year and citing year respectively, is the set of citing age effects, and is a

patent-year error term uncorrelated with the constructed dummies. Coefficient captures the

differential effect of the pool on pool patents compared to matched patents. I limit the sample of

pool patents to those that were initially selected into the pool in 1997 due to potential differences

between the initial selection process and the process used later for adding subsequent patents to

the pool.

Column 1 in Table 2.7 presents the result for the linear regression with robust standard errors.

The interaction coefficient suggests that the pool formation led to an average 40 percent increase

in the yearly citation rate of pool patents relative to the matched controls. In model 2, I replace

the matched control sample with a sample of patents on the competing technologies (“H.261”

and “H.263”). The results are essentially the same, suggesting that the pool formation not only

had no depressing impact on subsequent inventions based on pool patents but, in fact, also

boosted such inventions. The results are robust to other specifications such as Poisson and

Negative Binomial. Overall the results demonstrate that an increased litigation risk does not play

a role in explaining the observed decline in the weighted patenting rate of treated firms.

2.7.3.3 Test of Diminishing Financial Performance

Patent pools can potentially harm outsiders’ financial performance, and thus innovation

performance due to stronger competition from pool members and potential new entrants into the

pool technological area. In fact, the increased number of patents with the term “MPEG-2” in

their abstract along with the decline in the patenting rate of the firms in the treatment sample

already suggests that more firms started to invest into the pool technological area. In order to

37

examine the role of competition as a potential driving mechanism, I analyze the change in the

profit margin of firms in the treatment sample post-pool formation. I use the same difference-in-

differences estimation model as presented in equation 1, replacing the innovation rate with

different measures of financial performance as the dependent variable. I further limit the analysis

to the public firms in the sample due to data availability.

Column 1 in Table 2.8 shows the results for the effect of the pool on the net profit margin of

firms in the treatment sample compared to those in the broadly constructed control sample. The

interaction coefficient ( ) is insignificant indicating that the formation of the pool had no direct

impact on the profit margin of firms in the treatment sample relative to those in the control

sample. Columns 2 and 3 report the results for two other measures of financial performance:

ROA and Tobin’s Q, respectively. Columns 4 to 6 report similar estimations for the matched

samples. The results are essentially the same. Except for model 3 (with broad control sample and

Tobin’s Q as the dependent variable), which shows a positive impact, the other results indicate

no significant change in the financial performance of firms after the pool formation. Overall, the

results do not provide support for a decline in the financial performance of technologically

proximate outsiders. However, one should note that the results do not reject the increased

competition intensity post-treatment. In fact, anecdotal evidence and an observed increase in

total technological innovation in the field suggest the opposite. However, the results show no

support for a decline in treated firms’ patenting rate due to diminishing financial performance.

2.8 Conclusion

Despite their growing and widespread use in advanced fields of technology and their significant

economic importance, research has investigated the competitive effects of modern patent pools.

More importantly, we know little about how firms that are affected by these organizations

strategically respond to their formation. In this paper, I first investigate the effect of the

formation of the MPEG-2 pool on the innovation rate of outsiders in technological proximity of

the pool. Subsequently, I examine three different mechanisms through which the pool formation

may have differentially influenced the innovative performance of these firms.

38

The results suggest that the formation of the MPEG-2 pool had a substantial negative impact on

proximate outsiders’ innovation rate, measured by their weighted patenting rate. The net negative

effect on innovation rate is consistent with results reported by Lampe and Moser (2010) and

Joshi and Nerkar (2011). The analyses of potential mechanisms further suggest that the decline

is, in fact, a manifestation of a shift in outsiders’ R&D strategy from further technological

exploration to implementing the pool technology in their products. Considering that this

mechanism should apply similarly to both pool members and technologically proximate

outsiders, one should expect a similar shift in pool members’ innovation strategies leading to a

decline in their patenting rate. In an unreported analysis, I find that pool members also

experienced a significant decline in their patenting rate post-pool formation. In addition, I find

that all the pool members introduced new products based on the MPEG-2 technology in the years

following the pool formation. Furthermore, I find that increased litigation risk and diminished

financial performance cannot explain the observed decline in outsiders’ patenting rate.

Interestingly, the results suggest that the formation of the pool did actually encourage further

technological innovations based on the MPEG-2 technology. An analysis of the patents that have

cited the pool patents after its formation reveals that less than 20 percent of these follow-on

innovations are indeed assigned to the pool members or firms in technological proximity to the

pool. These figures suggest that the MPEG-2 pool formation created different opportunities for

different market participants. From the pool members and technologically proximate outsiders,

the real opportunity was in developing downstream applications based on the pool technology.

For some other market participants, the real opportunity was to develop follow-on technologies.

Investigating the determinants of this differential opportunity structure is an interesting avenue

for future research. In particular, examining the impact of modern patent pools on the business

plans of new ventures can shed more light on the entrepreneurial effect of these arrangements.

At the broad level, the findings provide a more nuanced understanding of how patent pools affect

the competitive arena through changing the investment incentives of firms outside the pool

across several industries and markets. These changes not only can introduce fundamental

structural shifts in a broad range of industries, but can also affect the technological trajectories of

39

pooled technologies in the long term. The results suggest that patent pools can create

fundamental shifts in the life cycle of related technologies by establishing dominant technology

standards and changing the opportunity structure for further technological development. This

study further contributes to the growing literature on the effect of institutional innovations on the

follow-on innovation (Furman et al., 2005; Furman & Stern, 2011; Murray & Stern, 2007).

Finally, the results of this paper call for more attention to studying how hybrid collaborative

organizations such as Standard Setting Organizations and R&D consortia may affect outside

organizations.

Finally, one should note that this paper focuses only on one modern patent pool, the MPEG-2

pool. Although, the fact that the MPEG-2 pool has been the template for almost all other modern

pools enhances the external validity of current findings, however, further research on other

modern pools can shed more light on the contextual factors that mediate the impact of these

arrangements on innovation. In particular, the supported mechanism in this study – the shift from

technological exploration to product development – relies on specific circumstances, most

notably the presence of high uncertainty around the dominant technology standard. Future

research can shed more light on how different levels of uncertainty in the technology market

would influence the response of market participants to the pool formation.

40

Figure 2.1: Number of patent pools per decade

Note: Each bar shows the number of patent pools formed in each decade of the 20th

century. As

can be seen, the number peaked in the ‘30s and declined subsequently due to the rise and

enformcement of antitrust laws. No major pool was formed after the end of World War II until

the formation of the MPEG-2 pool, the first instance of modern pools, in 1997. The numbers

continue to increase. (Source: World Intellectual Property Report 2011, WPO Economics &

Statistics Series)

0

5

10

15

20

25

30

35

1910s 1920s 1930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s

nu

mb

er

of

pat

en

t p

oo

ls

41

Figure 2.2: The effect of the formation of the MPEG-2 pool in 1997 on the weighted patenting

rate of outsiders in technological proximity to the pool (treated firms) compared to other firms in

the same industries (control firms)

Note: The weighted patenting rate of control firms is plotted on the right axis.

5

7

9

11

13

15

17

19

21

23

25

300

350

400

450

500

550

600

650

700

1994 1995 1996 1997 1998 1999 2000

we

igh

ted

pat

en

tin

g ra

te

we

igh

ted

pat

en

tin

g ra

te

incumbents in technological proximity of the pool (treatment sample)

other firms in similar industris (broad control sample) - scale on the right axis

After pool formation Before pool formation

outsiders in technological proximity to the pool (treatment sample)

Other firms in the same industries (broad control sample) – scale on the right axis

42

Figure 2.3: The effect of the formation of the MPEG-2 pool in 1997 on the average weighted

patenting rate of proximate outsiders in the matched treatment sample compared to that of firms

in the matched control sample.

60

80

100

120

140

160

180

200

1994 1995 1996 1997 1998 1999 2000

we

igh

ted

pat

en

tin

g ra

te

incumbents in technological proximity of the pool (matched treatment sample)

matched set of firms (matched control sample)

outsiders in technological proximity to the pool (matched treatment sample)

After pool fromation Before pool formation

43

Figure 2.4: The yearly treatment effects of the MPEG-2 pool formation

Note: The solid line shows the differential effects of the MPEG-2 pool formation on the log

normalized weighted patenting rate of firms in the treatment sample relative to those in the

matched sample. The gray dotted lines show the 95% confidence interval.

-1.4

-1.2

-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

1995 1996 1997 1998 1999 2000

After pool formation Before pool formation

44

Figure 2.5: The number of patents with the term “MPEG-2” in their abstract per year before and

after the formation of the MPEG-2 pool in 1997

0

10

20

30

40

50

60

70

80

90

1994 1995 1996 1997 1998 1999 2000

nu

mb

er

of

pat

en

ts

After pool formation Before pool formation

45

Table 2.1: Average patenting rate of firms in each pair of treatment and control samples pre-pool

formation

Treatment Control Difference

Panel A: treatment sample vs. broad control sample

Weighted patenting rate one year prior to the pool

formation (year=1996)

578.886

[175]

18.682

[2166]

560.204**

(0.000)

Weighted patenting rate two years prior to the pool

formation (year=1995)

484.703

[172]

15.180

[2101]

469.523**

(0.000)

Weighted patenting rate three years prior to the pool

formation (year=1994)

422.9438

[160]

12.580

[1976]

410.363**

(0.000)

Panel B: matched treatment sample vs. matched control

sample

Weighted patenting rate one year prior to the pool

formation (year=1996)

163.030

[67]

156.075

[67]

6.955

(0.890)

Weighted patenting rate two years prior to the pool

formation (year=1995)

120.328

[67]

106.106

[67]

14.222

(0.704)

Weighted patenting rate three years prior to the pool

formation (year=1994)

92.016

[62]

80.277

[65]

11.739

(0.684)

Note: In brackets in each cell is the number of observations. Column 3 also reports the p-value for testing

the null hypothesis that the average weighted patenting rates are equal across treatment and control

samples. ** p<0.01, * p<0.05, + p<0.1

46

Table 2.2: The effect of the formation of the MPEG-2 pool in 1997 on the weighted patenting

rate of firms in technological proximity to the pool using linear regression models

(1) (2) (3) (4)

Dependent variable:

ln(weighted patenting rate)

broad control

sample

matched

sample

matched sample,

w/ pre-treatment

placebos

matched sample,

public firms

Treatment * After Pool Formation

(1997)

-0.951**

(0.415)

-0.428*

(0.198)

-0.488*

(0.205)

-0.589+

(0.343)

Treatment * After 1996

0.126

(0.262)

0.435

(0.421)

Treatment * After 1995

-0.068

(0.261)

-0.069

(0.381)

Constant 0.995**

(0.022)

3.353**

(0.105)

3.353**

(0.105)

3.858**

(0.882)

Controls for lagged R&D

intensity, lagged capital intensity,

and lagged sales

No No No Yes

Firm fixed effects Yes Yes Yes Yes

Year fixed effects Yes Yes Yes Yes

Observations 15817 903 903 302

Number of firms 2778 134 134 46

Note: All estimates are from panel ordinary-least-squares (OLS) models with firm fixed effects.

Dependent variable is the ln(weighted patenting rate) for all models. Treatment (0/1) = 1 if firm is in

technological proximity of the pool, 0 otherwise. Columns 3 and 4 include two placebo treatments to

control for pre-pool formation trends. Robust standard errors are shown in parentheses. ** p<0.01, *

p<0.05, + p<0.1

47

Table 2.3: The effect of the formation of the MPEG-2 pool in 1997 on the weighted patenting

rate of firms in technological proximity to the pool using the Poisson model with robust standard

errors

(1) (2) (3) (4)

Dependent variable: weighted

patenting rate

broad control

sample

matched

sample

matched sample,

pre-treatment

placebos

matched

sample, public

firms

Treatment * After Pool Formation

(1997)

[0.677]**

-0.390

(0.096)

[0.650]*

-0.430

(0.170)

[0.642]**

-0.443

(0.153)

[0.543]**

-0.610

(0.169)

Treatment * After 1996

[1.062]

0.060

(0.111)

[1.241]

0.215

(0.148)

Treatment * After 1995

[0.928]

-0.074

(0.149)

[0.884]

-0.123

(0.199)

Controls for lagged R&D

intensity, lagged capital intensity,

and lagged sales

No No No Yes

Firm fixed effects Yes Yes Yes Yes

Year fixed effects Yes Yes Yes Yes

Observations 11383 903 903 302

Number of firms 1921 134 134 46

Note: All estimates are from panel Poisson models with robust standard errors and firm fixed effects.

Incident-rate ratios reported in square brackets. Estimated coefficients are reported in the second line.

Estimations are done using xtpqml in Stata 11 (Simcoe, 2007). Dependent variable is the “weighted

patenting rate” for all models. Treatment (0/1) = 1 if the firm is in technological proximity to the pool, 0

otherwise. Columns 3 and 4 include two placebo treatments to control for pre-pool formation trends.

Robust standard errors are shown in parentheses. ** p<0.01, * p<0.05, + p<0.1

48

Table 2.4: The effect of MPEG-2 implementation on weighted patenting rate of firms after the

formation of the MPEG-2 pool

(1) (2) (3) (4)

Pooled sample of firms in both

treatment and broad control

samples

Only firms in the broad control

sample

Linear regression

DV: ln(weighted

patenting rate)

Poisson

regression

DV: weighted

patenting rate

Linear regression

DV: ln(weighted

patenting rate)

Poisson

regression

DV: weighted

patenting rate

MPEG-2 implementation *

After Pool Formation (1997)

-0.323+

(0.194)

[0 .698]**

-0.359

(0.140)

-0.390+

(0.230)

[0 .537]**

-0.621

(0 .155)

Constant 1.031**

(0.059)

0.920

(0.063)

Firm fixed effects Yes Yes Yes Yes

Year fixed effects Yes Yes Yes Yes

Observations 2463 2007 2003 1563

Number of firms 325 263 299 232

Note: Treatment is equal to 1 if the firm mentioned MPEG-2 implementation in at least one of its annual

10-k reports, 0 otherwise. Estimates in columns 1 and 3 are from panel ordinary-least-squares (OLS)

models with firm fixed effects. Estimates in models 2 and 4 are from panel Poisson models with robust

standard errors and firm fixed effects. In columns 2 and 4, incident-rate ratios are reported in square

brackets. Estimated coefficients are reported in second line. Poisson estimations are done using xtpqml in

Stata 11 (Simcoe, 2007). Dependent variable is ln(weighted patenting rate) for models 1 and 3 and

“weighted patenting rate” for models 2 and 4. MPEG-2 implementation (0/1) = 1 if firm has mentioned

MPEG-2 implementation in its annual 10-K reports to investors, 0 otherwise. Robust standard errors are

shown in parentheses. ** p<0.01, * p<0.05, + p<0.1

49

Table 2.5: The effect of the MPEG-2 pool on the tendency of firms toward mentioning MPEG-2

implementation in annual 10-K reports after the pool formation

(1)

DV: mentioning MPEG-2 implementation in annual 10-K report

(0/1)

broad control samples

Treatment * After Pool Formation 0 .164**

(0.059)

Constant -0.002

(0.011)

Firm fixed effects Yes

Year fixed effects Yes

Observations 2984

Number of firms 325

Note: Estimates are from panel ordinary-least-squares (OLS) models with firm fixed effects. Dependent

variable is a dummy variable equal to 1 if a firm has mentioned MPEG-2 implementation in its annual 10-

K reports after the formation of the MPEG-2 pool, 0 otherwise. Treatment (0/1) = 1 if firm is in

technological proximity to the pool, 0 otherwise. Robust standard errors are shown in parentheses.

** p<0.01, * p<0.05, + p<0.1

50

Table 2.6: The effect of the formation of the MPEG-2 pool in 1997 on the weighted patenting

rate of licensees versus non-licensees

(1) (2)

Linear regression

DV: ln(weighted

patenting rate)

Poisson regression

DV: weighted

patenting rate

Licensee * After Pool Formation (1997) 0.169

(0.359)

[1.093]

0.089

(0.096)

Constant 3.874**

(0.114)

Firm fixed effects Yes Yes

Year fixed effects Yes Yes

Observations 1101 1052

Number of firms 191 171

Note: Treatment is equal to 1 if the firm has acquired a license from the pool, 0 otherwise. Estimates in

column 1 are from panel ordinary-least-squares (OLS) models with firm fixed effects. Estimates in model

2 are from panel Poisson models with robust standard errors and firm fixed effects. In column 2, incident-

rate ratios are reported in square brackets. Estimated coefficients are reported in second line. Poisson

estimations are done using xtpqml in Stata 11 (Simcoe, 2007). Dependent variable is ln(weighted

patenting rate) for model 1 and “weighted patenting rate” for model 2. Licensee (0/1) = 1 if firm has

acquired a pool license by the end of 1997, 0 otherwise. Robust standard errors are shown in parentheses.

** p<0.01, * p<0.05, + p<0.1

51

Table 2.7: The effect of the formation of the MPEG-2 pool on subsequent inventions based on

initial pool patents

(1) (2)

Dependent variable:

ln(citation rate)

Pool patents vs. matched

patents

Pool patents vs. patents on

competing technologies

Pool Patent * After Pool Formation (1997) 0.345+

(0.173)

0.420**

(0.117)

Constant 0.345*

(0.158)

0.164

(0.164)

Patent fixed effects Yes Yes

Age since grant dummies Yes Yes

Application year dummies Yes Yes

Citing year dummies Yes Yes

Observations 232 2631

Number of patents 34 476

Note: All estimates are from panel ordinary-least-squares (OLS) models with patent fixed effects.

Dependent variable is the ln(citation rate) for both models. Pool patent (0/1) = 1 if patent was initially

included in the MPEG-2 pool, 0 otherwise. Model 1 compares the forward citation rate of pool patents

with that of a set of matched patents. Model 2 compares the forward citation pattern of pool patents with

that of patents on main MPEG-2 competitors (H.261 and H.263). Robust standard errors are shown in

parentheses. ** p<0.01, * p<0.05, + p<0.1

52

Table 2.8: The effect of the formation of the MPEG-2 pool on the financial performance of

treated firms

(1) (2) (3) (4) (5) (6)

Treatment and broad control sample Matched treatment and control samples

DV:

net profit

margin

DV:

ROA

DV:

ln(Tobin’s

Q)

DV:

net profit

margin

DV:

ROA

DV:

ln(Tobin’s

Q)

Treatment * After

Pool Formation

36.699

(101.122)

7.870

(8.060)

0.198**

(0.052)

45.029

(37.080)

-5.681

(5.962)

-0.118

(0.134)

Constant 419.970

(902.540)

-52.300

(62.801)

1.930**

(.114)

-1195.191+

(708.617)

40.061**

(13.557)

0.227

(0. 518)

Controls for lagged

R&D intensity,

lagged capital

intensity, and

lagged sales

Yes Yes Yes Yes Yes Yes

Firm fixed effects Yes Yes Yes Yes Yes Yes

Year fixed effects Yes Yes Yes Yes Yes Yes

Observations 10039 10053 9798 296 296 293

Number of firms 2070 2074 2049 46 46 46

Note: All estimates are from panel ordinary-least-squares (OLS) models with firm fixed effects.

Treatment (0/1) = 1 if firm is in technological proximity of the pool, 0 otherwise. Each model has a set of

control dummies for 1 year lagged R&D intensity, capital intensity and sales. Robust standard errors are

shown in parentheses. ** p<0.01, * p<0.05, + p<0.1

53

Chapter 3

Scientists as Strategic Investors: Evidence from the 3Impact of the Bush hESC Policy on U.S. Scientists’ Research Productivity9

3.1 Introduction

Policymakers at organizational, regional, and national levels design numerous policies every

year to shape the future of scientific and technological progress in the territories under their

control. An important question is how effectively these policies influence scientists’ research

productivity and direction? In order to answer this question, prior research, particularly empirical

research, has largely modeled scientists as knowledge production functions whose input and

production technology are shaped by the external regulatory, societal and technological

environment (Acs, Anselin, & Varga, 2002; Feldman, 1999; Fritsch, 2002; Ponds, Van Oort, &

Frenken, 2010). Such models mainly rely on the assumption that scientists passively comply

with new institutional conditions placed on their research activity. This is in sharp contrast to an

alternative view where scientists actively and strategically change the external conditions on

which they depend (Furman, Murray, & Stern, 2012; Murray, 2010). From this perspective, the

efficiency of new policies depends heavily on how accurately they take into account such

strategic reactions by scientists. Despite its importance, previous research has devoted little

attention to investigating such strategic behavior by scientists. This study attempts to address this

9 This study is the third paper from an ongoing research program with Professor Anita McGahan, Professor Will

Mitchell, Professor Abdallah Daar, and Dr. Rahim Rezaie. The two first papers, both established as working papers

(and under consideration at journals), at the time of submission of this dissertation, explore changes across countries

in the publication of journal articles among independent and collaborating scientists in regenerative medicine. This

study is distinct from these two prior works in focusing on scientist choice and on changes in their research

productivity given shifts in the external policy environment. I have obtained an agreement from my co-authors in the

prior two studies to write this chapter of my dissertation as a sole-authored piece. I would like to thank my

colleagues in this research program for their support.

54

gap by exploring how scientists respond strategically to shocks and uncertainties in their external

environment.

For that purpose, I study a policy shock that limited the research funds, an important research

input, available to U.S. scientists for human embryonic stem cell (hESC) research. In 2001, the

Bush administration placed severe restrictions on federal funding of hESC research due to ethical

issues and instead increased funds for other subfields of embryonic stem cell (ESC) research

(NIH, 1999). This policy shock provides an excellent opportunity to investigate how U.S.

scientists reacted to the increased restrictions and increased uncertainty in their national policy

conditions. A passive view of scientists suggests that U.S. researchers would redirect their

research away from hESC towards other areas of ESC research where funding was more

abundant. This was indeed a serious concern shared by both politicians and academics in

response to the policy (Fletcher, 2001; Holden, 2005; S. Holland, Lebacqz, & Zoloth, 2001;

Holm, 2002; Levine, 2004; Levine, 2006; McCormick, Owen-Smith, & Scott, 2009). In this

paper I argue that U.S. scientists could circumvent the introduced restriction by actively seeking

new funding sources to reduce their reliance on federal funding. In this study, I highlight two

possible strategic responses that U.S. scientists could follow: first, they could switch to the non-

federal funds in the United States that emerged a few years after the introduction of the 2001

policy. While reaching out to non-federal financial sources could reduce reliance on federal ones

to some extent, it could not completely address the broad uncertainties in the national policy

environment governing hESC research as well as a lack of supplementary research materials

such as repositories for new hESC lines. As a more effective mechanism, I argue that U.S.

researchers could further seek more favorable and reliable hESC policies, financial resources and

other research materials outside the United States by expanding their international collaborative

ties. With this perspective, international collaboration can be interpreted as a diversification

mechanism to reduce risks that may arise due to shocks in the national institutional environment.

In order to test these arguments, I use longitudinal data on the research activity of all the

scientists in the Scopus database – the most extensive dataset on scientific publications – who

have published at least one article in the field of Stem Cell (SC) research in the 15-year period

55

from 1995 to 2010. Furthermore, I employ a difference-in-differences methodology to establish

the causal impact of the 2001 policy on the research output and collaboration patterns of U.S.

scientists compared to those of their counterparts in other developed countries with more

permissive hESC policies.

My analysis provides a number of core findings. First, I find that the restrictions on federal

funding of hESC research by the Bush administration in 2001 had a significant negative impact

on the research output of U.S. scientists in both SC and hESC areas compared to that of scientists

in other developed countries with more permissive policies towards hESC research. Second, I

find that these declines were quite short-lived and, in fact, by 2005 U.S. scientists experienced an

increase in their research output. The results provide support for two strategic responses by U.S.

scientists to the Bush policy. First, I find qualitative evidence that U.S. scientists switched to

non-federal funding sources. Furthermore, I find an increase in the U.S. scientists’ cross-border

collaborations post-2001 in other countries with flexible hESC policies. The findings also show

that those U.S. scientists who already had some cross-border collaborators prior to 2001

experienced little decline in their research output after the implementation of the 2001 Bush

policy. In addition, I find a boost in the hESC research output of U.S. scientists’ collaborators

after the 2001 Bush policy.

The findings contribute to the growing literature on science policy by delving deeper into how

scientists respond strategically to policies that affect their external institutional conditions

(Furman & Stern, 2006; Jaffe, 2008; Lane, 2009; Lane & Bertuzzi, 2011; Murray, 2010).

Considering the growing globalized nature of science, the results underline how ignoring

scientists’ strategic responses can lead to unintended international knowledge and capability

spillovers through cross-country collaborative ties (Vakili, McGahan, Rezaie, Mitchell, & Daar,

2013b). In addition, the paper contributes to the broad literature on scientific collaborations and

provides further insights into the role of international collaboration as a risk aversion mechanism.

The paper is organized as follows: The next section summarizes the policy history of human

embryonic stem cell research in the United States. Section 3 reviews the relevant prior research

56

on this topic and presents the set of hypothesized predictions. Section 4 describes the empirical

methodology to test them. Results and discussion follow.

3.2 History of hESC Research and Policies

Stem cells are undifferentiated biological cells that are capable of dividing and differentiating

into specialized cell types such as skin cells, nerve cells, or muscle cells. There are two broad

types of stem cells: adult stem cells, which can be found in any type of body tissues; and

embryonic stem cells (ESC), which can only be derived from the inner cell mass of early-stage

embryos. Adult stem cells are already successfully used in treating several severe conditions.

However their big limitation is that they are lineage-restricted, meaning that they can only

develop into particular forms of cells. A blood stem cell can differentiate into several types of

blood cells, however, it cannot develop into nerve or brain cells. In contrast, embryonic stem

cells are known to be pluripotent, i.e., they can develop into any type of cell. Furthermore, they

have the ability to self-propagate indefinitely under certain conditions.

Embryonic stem cells were first derived from mouse embryos in 1981 by two independent

research teams. It took another 17 years before James Thomson and his research team at the

University of Wisconsin-Madison made their breakthrough and developed a technique to grow

and isolate human embryonic stem cells (Thomson et al., 1998). Because of their pluripotency

and the fact that they are derived from human embryos, hESCs are acknowledged to have the

highest potential among all types of stem cells to advance current clinical treatments and

introduce novel therapies (Vogel, 1999). Yet, despite their scientific and economic importance,

there have been ongoing political debates over hESC research (and, more broadly, ESC research)

due to ethical concerns.

In the United States, the first laws prohibiting research on fetuses and embryos date back to

1973. However, these laws have not been rigorously enforced. In 1995, just a short while after

the Clinton administration approved federal funding for research on leftover embryos created

through in vitro fertility treatments, Congress passed the Dickey Amendment bill that prohibited

any federal funds for use in research that could potentially result in the creation or destruction of

57

a human embryo. It was during this period when Thomson made his breakthrough discovery of

human embryonic stem cells in 1998 using private funding. Following Thomson’s breakthrough

and in recognition of the massive opportunities it opened up, in 1999, the Clinton administration

loosened the policies governing federal funds available for embryonic research. Just a few

months later, on August 9, 2001, the newly appointed President George W. Bush announced his

administration’s stem cell policy. Despite scientific community hoping for growing federal funds

available for hESC research, the announced policy banned any federal funding for research on

new hESC lines, while approving federal funding for existing hESC lines developed in the

United States or outside.

3.3 Theory

3.3.1 Prior Research on the Impact of the Bush Policy

Bush’s stem cell policy act set off waves of concern in the scientific and regulatory communities

(Fletcher, 2001; Holden, 2004; S. Holland et al., 2001; Holm, 2002; Johnson & Williams, 2007;

Vogel, 2001). In April 2004, 206 members of Congress released a letter to President Bush urging

him to relax embryonic stem cell research restrictions (Holden, 2005). Among other points, the

letter highlights that “this promising field of research is moving overseas” and that “leadership in

this area of research has shifted to the United Kingdom.” Furthermore, the letter states that “it is

increasingly difficult to attract new scientists to this area of research because of concerns that

funding restrictions will keep this research from being successful.” At a hearing by a Senate

appropriations subcommittee in 2005, the then NIH Director, Elias Zerhouni, cited “mounting

evidence” that numerous problems have been uncovered due to genetic instability of the aging 22

approved cell lines. In the same session, the Chair of the subcommittee, Arlen Specter, released

several letters drafted by various institute directors warning that the NIH would fall behind other

local and international institutions in the field due to restrictions (Holden, 2005).

In the wake of these concerns, several studies have attempted to document the impact of the

Bush policy on the follow-on hESC research in the United States relative to other countries

(Furman et al., 2012; Levine, 2004; Owen-Smith & McCormick, 2006; Scott, McCormick, &

Owen-Smith, 2009; Vakili et al., 2013a). Using five less controversial biomedical research areas

58

as the baseline, Levin (2004) reports that the share of hESC publications credited to U.S.

scientists dropped considerably in 2003 and remained at this lower level in 2004. Studying the

same time period, Owen-Smith and McCormic (2006) also report a decline in the relative share

of hESC publications by U.S. scientists compared to scientists elsewhere in the world. While

both studies report a decline in the share of U.S. hESC research, it is not clear whether this share

is due to a decline in the relative productivity of U.S. scientists or rather an increase in the total

number of stem cell scientists in other regions of the world, particularly given the evidence of the

growth in stem cell research in emerging markets such as China, South Korea, and Singapore

(Vakili et al, 2013). Furthermore, both studies only examine the short-term impact of the Bush

policy in the two years following the policy change due to data constraints. Looking at a much

extended timespan, Vakili et al. (2013) find that the United States share of hESC publications

stopped declining after 2003, increased slightly in 2004, and then remained consistently steady at

about 33 percent until 2010. Similarly, using a difference-in-differences methodology to analyze

the causal impact of the 2001 policy, Furman, Stern and Murray (2012) also find that the United

States’ cumulative production of hESC research declined between 2001 and 2003 and rebounded

in the subsequent years.

While these studies provide important insight into the impact of the Bush policy on the total

share of the United States in hESC research, they do not report how these policies affected

individual scientists’ research productivity. Moreover, these studies provide little explanation of

how individual scientists responded strategically to the Bush policy intervention and how the

United States’ share of hESC research recovered so quickly despite the continuous enforcement

of federal funding restrictions. Both Furman et al. (2012) and Vakili et al. (2013b) provide

evidence of an increase in the aggregate level of international collaborations involving U.S.

scientists, suggesting that the quick recovery might be related to the increased level of

international collaboration. Yet their results are not fine-grained enough to establish a direct link

between the quick recovery and increased amount of international collaboration.

59

3.3.2 Theoretical Development

Faced with the 2001 Bush policy, U.S. researchers who had already invested or were planning to

invest in hESC research suddenly found themselves in a new institutional setting with very

limited funding available for research on new hESC lines. A passive view of scientists suggests

that U.S. researchers who were mainly dependent on federal funding would shift their research

direction to either approved hESC lines or non-human ESC areas. This was indeed the intended

purpose of the policy (Vogel, 2001).

However, there are several reasons to believe that such a shift in research direction was not in

line with U.S. scientists’ preferences and could thus potentially trigger their reactions to avoid

this shift as much as possible. First, hESCs were considered to have the highest potential to cure

severe human diseases (Vogel, 1999). Prior to 2001, when the Bush administration enacted their

policy, hESC research had been growing faster than any other subfield of stem cell research.

Furthermore, while U.S. scientists had the option to work on the approved hESC lines under the

Bush policy, these lines had limited range of genetic diversity and many of them were

contaminated with mouse embryonic feeder cultures by 2004 (Holden, 2004, 2005). Moreover,

many new hESC lines with higher potential for therapeutic purposes were developed after the

announcement of the Bush policy (Holden, 2004, 2005). Also, a more limited number of hESC

lines mean higher competition among U.S. researchers to come up with new ideas and

discoveries.

In contrast to the passive view of scientists, one can alternatively think of scientists as investors

who try to maximize the return on their investments subject to given constraints. From this

perspective, each research project can be seen as an investment project that requires some input

and that, with some uncertainty, leads to some output. Inputs include research funds, research

technologies and materials, and knowledge. Outputs include scientific publications, credit and

reputation in academia, usually some monetary benefits, and more knowledge to be used in

future research. The investment also involves a value adding process that transforms inputs into

outputs, using methods such as brainstorming, knowledge recombination, exploration and

experimentation.

60

This model of scientists’ investment in research projects resembles closely the way scholars have

modelled firms’ R&D investment in different innovation projects (Amit & Livnat, 1988;

Huchzermeier & Loch, 2001; Mikkola, 2001). However, despite the vast literature on how firms

strategically manage their R&D investments and risks associated with it, little effort has been

devoted to explore how individual scientists decide about their level of investments and how they

deal with external shocks and risks associated with their research projects. There are, in fact,

various risks associated with research investments. The sources of these risks can be internal to

the research process. There is no guarantee that a research project leads to desired outcomes or

any outcomes at all. This kind of risk is inherent in any research investment and can only be

mitigated to a certain extent. Careful research design, using simulations, as well as iteration are

all examples of procedures that are commonly used by researchers to minimize risks associated

with the research process itself. Risks can also be due to uncertainties in the external

environment. Examples include a sudden advancement in a particular research technology, or

sudden changes in the regulatory environment that dictates what kind of experimentation can be

done or what type of outputs can or cannot be produced.

To the extent that scientists are aware of these risks and behave strategically, I expect them to

respond to shocks that affect their research inputs and engage in some risk management activities

to protect their investments. In this light, I would argue that U.S. scientists could follow at least

two strategies to deal with the sudden drop of federal funds available for hESC research in the

United States. First, they could replace federal funds with funds from non-federal agencies in the

United States. This solution, of course, depends on the availability of such financial sources.

About four years after the Bush policy announcement in 2001, some universities and private

organizations started funding hESC research. For example, in 2005, Harvard University

launched a fundraising campaign for hESC research focused on special therapeutic purposes.

Several corporations such as ViaCell, Inc., Geron, and BioTime decided to invest in hESC

research with financial interests in mind. Furthermore, the federal law did not impose any

restrictions on whether states could invest in hESC research. While a few states such as

Louisiana, Michigan, and Minnesota decided to re-enforce the restrictive policies further at the

state level, some other states took a completely opposite approach. The state of New Jersey

61

became the first state to allocate funds for hESC research in 2005. Also, the State of California

took out $3 billion in bond loans to fund hESC research in 2004, planning to invest $300 million

per year for 10 years. To the extent that U.S. scientists were keen on continuing their research in

hESC, they could seek funding from these non-federal sources as a strategic response to the Bush

policy shock. There are reports indicating that the six states of California, Connecticut, Illinois,

Maryland, New Jersey, and New York funded more hESC research than the federal government

(Fossett, 2007).

One problem with this solution was that these alternative sources of funding were not abundantly

available until four years after the introduction of the policy. Another major issue with this

strategy was that, while it could resolve the lack of financial resources, it could not address the

lack of other supplementary resources required for hESC research such as repositories for new

hESC lines that were not abundantly available within the United States. More importantly, this

strategy could not resolve the uncertainties in the national policy environment that could

potentially influence the resources required for hESC research or its conduct directly within the

U.S. national territories. One could imagine that in such an uncertain and unfavourable policy

environment for the hESC research, government could negatively affect the availability of non-

federal funds. These issues reduce the efficiency of any strategic response that relies on

alternative sources of funding and supplementary resources inside the United States.

However, non-federal agencies were not the only sources from which U.S. scientists could seek

financial resources for their hESC research. A second channel through which U.S. scientists

could obtain both funds as well as other supplementary resources for their hESC research was by

collaborating with funded scientists in countries with more flexible policies. Compared to the

previous strategy, international collaboration is, on average, a more costly option. In a series of

works, Cummings and his colleagues argue that the institutional, geographic, and temporal

distances between collaborators in scientific or non-scientific environments leads to lower levels

of productivity and output quality due to higher costs of coordination (Cummings, 2011;

Cummings, Espinosa, & Pickering, 2009; Cummings & Kiesler, 2005, 2007). In the case of

hESC research, these cross-border collaborations were even more challenging due to different

62

international policies governing the hESC research and the need for an appropriate division of

labor that could allow such collaborations to take place (Furman et al., 2012).

Yet despite these costs, international collaboration had two important benefits over a switch to

non-federal sources of funding within the United States. First, unlike the private- and state-level

funds within the United States that emerged primarily with considerable delay, there were plenty

of funds available outside the United States at the same time that the Bush policy was

announced. In contrast to the United States, several countries, including the United Kingdom,

Japan, Sweden, Canada, China, Israel, and South Korea increased their legal support and funding

for hESC research to gain a competitive advantage in this promising field. By expanding their

collaborations overseas, U.S. scientists could potentially gain access to these funds immediately.

In addition, international ties would not only help U.S. scientists access new financial sources,

they would also offer a channel through which U.S. scientists could obtain other required inputs

for their research, such as repositories for new hESC lines.

Furthermore, by crossing U.S. national boundaries, international collaboration can work as a

diversification strategy through which U.S. scientists reduce their reliance on resources within

the U.S. territory. Numerous studies on risk management have regarded diversification as an

important strategic means to deal with uncertainties in the external environment (McNeil, Frey,

& Embrechts, 2010). The same idea can be applied to scientists and the strategies they would

pursue to deal with uncertain external environments. Hence, the high uncertainty in the U.S.

policy environment regarding hESC research can arguably justify the higher cost associated with

diversification through international collaborations. This line of reasoning also suggests that

those U.S. scientists who already had access to a more diverse set of funding sources through

their international ties pre-2001 would experience less impact on their research productivity as a

result of the Bush policy intervention.

It is also important to note that while cross-border ties could help U.S. researchers mitigate the

negative impact of the Bush policy, scientists in countries with permissive hESC policies could

also gain from collaborating with U.S. researchers. A large body of literature has established the

significant role of collaboration as a means to transfer knowledge, particularly tacit knowledge

63

between individuals and firms (Jones, 2009b; Mowery, Oxley, & Silverman, 1996; Powell, 1998;

Singh, 2005; Singh & Fleming, 2010). Hence, collaborating with U.S. scientists could potentially

work as a means to access knowledge and capabilities developed in the United States. This is

particularly important for scientists in emerging countries such as China, South Korea, and

Singapore who lacked the necessary capabilities in hESC research pre-2001 and hence could

potentially benefit a lot from such collaborations (Fox, 2007; Greenwood et al., 2006; Hwang,

2005).

In summary, modelling U.S. scientists as strategic investors, I expect them to respond to the

Bush policy intervention in at least two ways. First, I expect an increase in U.S. scientists’

projects funded through non-federal agencies within the United States post-2001. I also expect an

increase in their international collaborations with scientists in countries with flexible hESC

policies post-2001. One should note that both of these responses naturally take some time. This

means that in the short run, U.S. scientists may experience some decline in their research

productivity in hESC and, more broadly, in SC. However, with the growth in the level of

international collaborations and also the rise of state-level and private funds, I expect to see a

bounce back in the U.S. scientists’ research output. This also implies that those scientists who

already had some cross-border collaborative ties pre-2001 could respond faster to the Bush

policy shock by relying more heavily on the international sources and thus experienced less

impact due to the policy shock. Finally, assuming that the 2001 Bush policy pushed U.S.

scientists to spend more time and effort on their internationally funded projects, I expect a boost

in the research output of their international collaborators post-2001.

3.4 Empirical Methodology

My empirical strategy includes four steps. The first step involves estimating the impact of the

Bush policy shock on the subsequent research output of U.S. scientists in the areas of hESC and

SC. In particular, I am interested in measuring the year-by-year change in U.S. scientists’

research output to be able to capture the dynamic impact of the policy and the subsequent

strategic responses of U.S. scientists. The main challenge to estimating the causal effects of

these events is that we cannot observe what would have happened to U.S. scientists’ research

64

output should the Bush policy and the strategic responses it potentially triggered had not

happened, i.e., we cannot observe the counterfactual. A change in the variables of interest can be

driven by these events, or some other unobservable factors. For example, a possible decline in

the hESC research output of U.S. scientists after the 2001 Bush policy could be driven by a

general drop in hESC research at the global level. Alternatively, it could be driven by an increase

in the global competition level in the field due to a sudden surge of scientists interested in this

area.

In order to deal with this issue, I compare the changes in the research output of U.S. scientists

(treated sample) after the Bush policy intervention (treatment) to those of another similar group

of scientists (control sample) using a difference-in-differences (DID) methodology. The main

underlying idea is that, under appropriate conditions, the trends for the control sample provides a

proxy for the unobservable counterfactual trend for the treated sample (Holland, 1986; Imbens &

Wooldridge, 2009). Thus, assuming that control and treated samples are in fact comparable and

similar, any difference between their outcome trends after the treatment can be interpreted as the

causal impact of the treatment on the treated sample.

The reliability of a DID estimator depends on two necessary conditions. First, the treatment

should be exogenous to both the treated and control samples (Rubin, 1990). In the case of this

study, this means that the 2001 Bush policy was not predictable by the hESC research

community well in advance. There is significant evidence suggesting that the 2001 Bush policy

was not fully predictable by the U.S. research community. The policy was announced just a few

months after the 2000 presidential election, the outcome of which was very hard to predict well

in advance. Also, the conversation surrounding the hESC research in the period prior to the

policy announcement highlights considerable uncertainty in the policy environment. Ultimately,

considering the very strong support for hESC research within the NIH through the final years of

the Clinton administration, the 2001 policy announcement met with surprise in the media and

amongst scientists (Fletcher, 2001; Furman et al., 2012). Nevertheless, to the extent that the Bush

policy was not really a shock and was predictable by U.S. scientists well in advance, I expect to

observe changes in U.S. scientists’ research behavior prior to the policy announcement. In the

65

empirical section of the paper, I examine this possibility by exploring the US scientists’

publication and collaboration trends pre-2001.

Second, for the control sample to provide a reliable proxy for the unobservable treated sample’s

counterfactual, control and treated samples need to be drawn from similar distributions. In other

words, the treated and control samples should look similar enough prior to the treatment. In this

study, for each group of U.S. researchers in a particular research area, I use the non-U.S.

researchers in developed countries with flexible hESC policies in the same research area as the

control group. The assumption is that prior to the 2001 policy, these two groups of scientists

were similar in terms of their research output and behavior. In the results section, I provide

descriptive statistics to support this assumption.

For this difference-in-differences estimation, I use the following linear equation:

(1) ∑ ( )

where stands for the research output measure in hESC or SC for scientist in

year , is one if scientist is a US-based researcher and is 0 otherwise, is

1 for scientist in year and is 0 otherwise, is a researcher fixed effect, and is a year fixed

effect. is the main coefficient of interest and captures the marginal difference between U.S.

researchers and researchers in the control sample in year . This flexible specification allows me

to capture the dynamic year-by-year effect of the Bush policy on U.S. scientists’ research

outcome. Including researcher fixed effects control for idiosyncratic differences between

different researchers. Also, including year fixed effects control for the macro time trends.

In the second step, my goal is to examine the two hypothesized strategic responses by U.S.

scientists after the Bush policy in 2001. For the change in the amount of U.S. scientists’

international collaborations in response to the policy, I use a similar difference-in-differences

estimation to that in the previous step:

(2) ∑ ( )

66

where is the number of publications by scientist in year that

involves at least one international collaborator from an emerging market with permissive hESC

policies such as China, South Korea, or Singapore. The reason for focusing on collaborative ties

with scientists in emerging markets is twofold: First, it is technically infeasible to include the

publications with collaborators from the developed countries with flexible hESC policies. The

reason is that such collaborative publications will be counted for both U.S. scientists in the

treated group and for their collaborators in the control group and thus they will be automatically

cancelled out through the DID estimation. Second, theoretically, I expect scientists in emerging

markets to be more eager to establish collaborative ties with the U.S. scientists to access

knowledge and capabilities developed in the United States.

To test whether U.S. scientists switched to non-federal sources of funding, ideally I would like to

compare how frequently U.S. scientists switched to non-federal funds relative to a comparable

group of non-U.S. scientists. However, this is not feasible due to data constraints and

methodological issues. The first issue is that I do not have data on the sources of funding for

hESC research outside the United States. Thus, I cannot create a comparable control group to

empirically test this idea. The more severe issue is that the emergence of non-federal funds and

the switch of U.S. scientists to these funds are simultaneously observed. Hence I cannot

separately estimate the impact of the Bush policy on the switching rate of U.S. scientists to non-

federal funds. Considering these difficulties, I only compare the total amount of state-level

funding of hESC research awarded to U.S. scientists in each year to the total amount of federal

funds allocated to hESC research in the same year. This is far from ideal. However, it gives us an

idea of the aggregate reliance of U.S. scientists on non-federal funds relative to federal funds.

In the third step, the goal is to compare the impact of the Bush policy on two different groups of

U.S. scientists: those who already had some collaborative ties with scientists in countries with

flexible hESC policies and those who did not. The idea is to examine whether those U.S.

scientists who already had more diversified sources of funding through international

collaboration were less affected by the 2001 policy shock or not. To test this idea, I repeat the

estimation strategy in the first step for these two groups of U.S. scientists separately. For each

67

group, I construct a comparable group of scientists outside the United States in developed

countries with flexible hESC policies as the control sample. The estimating equation is

essentially the same as equation (1).

Finally, in the last step, in order to compare the research output of those non-U.S. researchers

who had collaboration with U.S. scientists prior to 2001 to other non-U.S. researchers, I use a

similar difference-in-differences estimation as the one used in the first and second steps with

some minor changes:

(3) ∑ ( )

where is 1 for non-U.S. researchers who had at least one

collaboration with a U.S. scientist in the five years prior to the 2001 Bush policy and is 0 for

other non-U.S. researchers. All the other variables remain the same. This model is only estimated

for the sample of non-U.S. researchers and basically estimates the year-by-year difference in the

research output of non-U.S. researchers who had U.S. collaborators pre-2001 and other non-U.S.

researchers.

3.5 Data

The data for this study is collected by searching the Scopus on-line database from 1980 to 2012

for all articles that mention keywords relevant to Stem Cell in their title or abstract. With more

than 47 million records from about 18,500 peer-reviewed scientific journals, Scopus is currently

the most comprehensive and encompassing database of scientific publications. More than 20

percent of the articles recorded in Scopus are in languages other than English and more than half

of them are assigned to scientists outside the United States.10

For the purpose of this study, I

excluded all the conference proceedings, reviews, and reports as well as all publications before

1995 and after 2010. This resulted in 65,759 articles in SC. Next, all the articles were categorized

into different subclasses based on particular word combinations (such as “human embryonic

stem cell,” “hESC,” “stem cell,” etc.) in their title and abstract. This resulted in 1,658 articles on

10

Every article in Scopus has an English abstract.

68

hESC, 3,896 articles on ESC (excluding hESC), and 60,205 on SC (excluding ESC). In order to

assess the reliability of the classification, an expert in the field reviewed 200 randomly selected

articles in detail and manually categorized them. Comparing the automatic categorization based

on keywords to results from the manual categorization by the expert revealed more than 90

percent accuracy.11

In the next step, for each article I identified the list of authors and their affiliations. Scopus

assigns a unique identifier to each author in the database. Taking advantage of these unique

identifiers, I created longitudinal yearly records of all the publications in each research category

for each unique scientist in the sample. In 59,359 cases (less than 5 percent of the total sample),

some author data was missing or could not be extracted successfully and, hence, was dropped.

Furthermore, Scopus reports the number of citations that each publication has received from

subsequent works published until the time of data collection. Following prior research in this

stream, I created a yearly citation-weighted publication count in each research category for each

scientist in the sample. I use this weighted publication count as the measure of scientists’

research output. The issue with the simple publication count is that it does not account for the

heterogeneity in the quality of different publications. Assuming that the number of citations that

a publication receives from subsequent research is a reliable proxy of its quality, this weighting

scheme can therefore account for quality differences in the publications sample (Furman et al.,

2012; Furman & Stern, 2006; Harhoff et al., 1999; Trajtenberg, 1990). Furthermore, in order to

deal with the skewness in the distribution of this measure, following prior studies, I first added

one to the calculated weighted counts and then log normalized it.

Subsequently, I identified the country of affiliation for each author on each article by analyzing

the text of the affiliation. In cases that multiple affiliations were assigned to an author on a

particular article (about 19 percent of the time), only the first affiliation was used. Next, using

the country data, I determined the location of each author in each year from 1995 to 2010. Each

country in the sample is then categorized as either “constrained” or “flexible” in its policy

regime with regards to hESC research based on an analysis of the public records on regulations,

11 Both the false positive rate and false negative rate were below 10

percent.

69

laws, and policies implemented in each country between 1995 and 2010. Countries classified as

“constrained” include the United States, Austria, Colombia, France, Germany, Italy, Norway,

Poland, Slovakia, and Tunisia. Countries in the “flexible” category are Argentina, Australia,

Belgium, Brazil, Canada, Chile, China, the Czech Republic, Denmark, Finland, Greece, Hong

Kong, Hungary, Iceland, India, Iran, Israel, Mexico, the Netherlands, New Zealand, Portugal,

Russia, Singapore, South Africa, South Korea, Spain, Sweden, Switzerland, Taiwan, Turkey, and

the United Kingdom.

Also, each country is further labeled as “developed market” or “emerging market” according to

its status in the UN database. Countries marked as “developed market” include the United States,

Canada, Japan, countries of Western Europe (the United Kingdom, Germany, the Netherlands,

Sweden, France, Italy, Spain, Austria, and others), and the Antipodes (Australia and New

Zealand). Countries in the “emerging market” category with active RM scholars are those in

Asia (China, South Korea, Singapore, Taiwan, India, Malaysia, Hong Kong, Thailand, and

others), Latin America (Brazil, Mexico, Argentina, and others), the Middle East (Israel, Turkey,

Iran, Saudi Arabia, and others), Eastern and Central Europe (Russia, Czech Republic, Hungary,

Poland, Romania, and others), and Africa (Egypt, Tunisia, South Africa, and others).

Using the above categorizations for the countries in the sample and also the co-authorship data

on each article, for each author I calculated the yearly average number of international co-

authors as well as the yearly number of publications that involves at least one cross-border

collaborator from emerging countries with flexible hESC policies.

3.6 Empirical Analyses

3.6.1 Descriptive Statistics

Table 3.1 illustrates the yearly number of hESC scientists, their yearly average weighted

publication count, and their yearly average number of international collaborators between 1999

and 2010, both inside the United States and in other developed countries with flexible hESC

policies. New hESC scientists enter into the sample as soon as they have their first hESC

publication. The average number of international collaborators for each scientist is calculated

70

using all the ESC publications by that scientist each year. The justification behind using all the

ESC publications is that hESC projects often involve non-human ESC experiments as well and,

hence, lead to both human and non-human ESC publications.

Figures 3.1 and 3.2 illustrate the trends for the “average number of hESC publications” and the

“average number of ESC publications with collaborators from developed countries with flexible

hESC policies” for the treated and control samples respectively. The trends in Figure 3.1 suggest

that, on average, U.S. scientists underperformed in hESC research relative to non-U.S. scientists

in the control sample in the three years following the Bush policy shock. The graph also shows a

bounce back for U.S. scientists in 2005. The trends in Figure 3.2 suggest a similar story. U.S.

scientists seem to lag slightly behind scientists in the control sample in terms of level of

international collaboration until 2003. In contrast, from 2004 until 2010, U.S. scientists

persistently published more articles in collaboration with scientists from emerging countries with

flexible hESC policies. Both trends are consistent with the predictions and previous findings at

the country level (Furman et al., 2012; Levine, 2004; McCormick et al., 2009; Vakili et al.,

2013a). However, an important issue with these graphs is that they do not control for scientists’

fixed effects. Hence, it is not obvious whether the observed trends are due to changes in an

individual scientist’s productivity level, or rather due to changes in the distribution of scientists

in the treated and control samples. As a result, the increase in U.S. scientists’ research output in

2004 can be due to the entrance of new scientists in the United States rather than an increase in

the productivity of those scientists who were already active in the field. In the next section, I

present the results from the DID estimations to deal with these issues.

As mentioned in the empirical methodology section, one of the conditions for a reliable DID

estimation is the similarity between treated and control samples pre-treatment. Figure 3.3 shows

the estimated density function of publication counts in SC for the treated and control samples in

2000. The graphs suggest that the treated and control samples were quite similar in terms of

research productivity in SC. I further performed a Kolmogorov-Smirnov equality-of-distributions

test to compare the distribution of SC and ESC researchers within the United States to those in

other developed countries with flexible policies in 2000. All three tests rejected a significant

71

difference between the distribution of U.S. and non-U.S. researchers in terms of research output

in these areas of research in 2000.

Table 3.2 presents some summary statistics for the research outputs and collaboration patterns of

hESC scientists in the treated and control samples in 2000. Both samples include those scientists

who were active in SC in 2000 and later started publishing in hESC. The difference between the

two samples and the calculated t-stats are reported in the third columns. There is no significant

difference in the mean publication counts, mean weighted publication counts, and mean number

of co-authors between the treated and control samples in the year prior to the policy change.

Overall, the equality-of-distributions tests and summary statistics suggest that treated and control

samples were quite similar before the Bush policy shock and thus satisfy the second assumption

for the DID estimation.

3.6.2 Estimation results

The results from estimating equation 1 are presented in Table 3.3. All models are estimated using

a panel OLS regression with scientist and year fixed effects. Model 1 estimates the year-by-year

change in the research output of U.S. hESC scientists relative to that of other hESC scientists in

developed countries with flexible hESC policies. The model only includes scientists who have

published in hESC. Each scientist enters into the sample on the year of her first hESC

publication. The sample structure and the inclusion of scientist fixed effects guarantee that the

estimated coefficients solely reflect the changes in each scientist’s research output. Figure 3.4

presents the estimated yearly treatment effects. The estimates show a significant decline in U.S.

hESC scientists’ output in 2003 and 2004 and a subsequent recovery and growth in their hESC

publications. The results support the hypothesized predictions. They are also consistent with

reported findings in previous studies (Furman et al., 2012; Levine, 2004; McCormick et al.,

2009; Vakili et al., 2013a).

One issue with model 1 is that it includes those U.S. hESC scientists who started their research

activity after 2001. To the extent that this group of scientists were different from those who were

active pre-2001 in terms of quality, their inclusion can lead to biased results. To address this

72

issue, model 2 only includes those scientists who had some research activity in the broader field

of SC both pre-2001 and post-2004. The estimated coefficients confirm both the relative decline

in research output of U.S. hESC scientists after the policy shock and also their recovery

subsequently in 2005. However, unlike model 1, estimations from model 2 show no significant

difference between the research output of U.S. and non-U.S. researchers after 2005. Comparing

the results from the two models suggests that the disproportional growth in the research output of

U.S. hESC scientists after 2005, as suggested by model 1, is likely driven by a higher

productivity of those U.S. scientists who entered post- 2005. One possible explanation is that

only high quality U.S. scientists who already had secured non-federal funds for their research

entered into the hESC research area after 2001. The results are robust if I replace the post-2004

condition in model 2 with post-2005 or post-2006.

In order to test a potential short-term switch from human to non-human ESC research post-2001

as a short-term strategy to deal with the policy shock, in model 3, I change the sample to ESC

researchers and estimate the relative change in the U.S. scientists’ research output in non-human

ESC research. An ESC scientist is defined as a scientist who has at least one publication in ESC.

Similar to the previous models, each ESC scientist enters into the sample after her first ESC

publication. Also, similar to model 2, the sample only includes those researchers who have been

active in SC both pre-2001 and post-2004. The results show an increase in U.S. scientists’ non-

human ESC publications in 2003 and 2004. Considering the decline in U.S. scientists’ hESC

research in the same two years, these results suggest that U.S. scientists actually changed their

research direction to some extent from human to non-human ESC research following the 2001

policy change. However, the estimates show no sign of switch back after the recovery in 2005.

One interpretation is that the growth of non-federal and international funds helped both hESC

and non-human ESC research alike and thus there was no need to make such a trade-off once the

funds were available.

Model 4 repeats the estimation in model 3 for the broader SC research area. The model includes

all the scientists who had some publication activity in SC pre-2001 and post-2004. The estimates

from model 4 shows a significant decline in U.S. scientists’ research output in the broad field of

73

SC starting in 2004 and lasting for three years until it finally recovered. This more prolonged

decline in the broader research productivity of U.S. scientists might be due to the extra costs

associated with switching research directions and seeking new sources of funding. Also, focusing

on non-human ESC research might have attracted less attention amongst SC researchers globally

and thus generated lower follow-on citations from future work, costing U.S. scientists lower

weighted publication counts post-policy change. These results together suggest that the 2001

Bush policy indeed had a short-term negative impact on the U.S. scientists’ research output both

in hESC and in the broader field of SC research. Furthermore, the results suggest that, consistent

with the prediction, U.S. scientists managed to recover their hESC research output within a short

while after the policy shock, despite the fact that the policy was still in effect.

Models 1 through 3 in Table 3.4 present the impact of the Bush policy on two subgroups of U.S.

hESC scientists: those who had some international collaborative ties in the broader field of SC

research pre-2001 and those who did not. All models are estimated using equation 1. Model 1

presents the yearly treatment effects for the first group. The control sample includes those non-

U.S. researchers in developed countries with flexible hESC policies who also had international

ties in the SC area pre-2001. Estimated coefficients show a slight decline in the hESC research

output of U.S. scientists in 2004 followed by a fast and significant growth in 2005. Model 2

repeats the same estimations for the sample of hESC researchers with no international

collaboration experience pre-2001. The estimated coefficients show significant decline in the

weighted hESC publications in 2003 and 2004 and remain negative for all the subsequent years.

Nevertheless the coefficients are insignificant or barely significant post-2005 suggesting some

recovery. Figure 3.5 illustrates the estimations from models 1 and 2 next to each other. The

results suggest that U.S. researchers with some prior international ties were minimally affected

by the 2001 Bush policy. Others experienced a more prolonged negative impact on their research

output post-2001.

One might be concerned that the difference between the estimates in models 1 and 2 in Table 3.4

might be driven by the difference between the quality of scientists in the subgroup of U.S.

scientists who are included in these models. In other words, it might be possible that the research

74

of those U.S. scientists who had some international ties pre-2001 was of higher quality and thus

they had an easier time locating new sources of funds for their hESC research post-2001.

Therefore, the difference between the estimates in models 1 and 2 is driven by quality

differences rather than differences in access to international sources of funding. In fact,

analyzing the publication records of the U.S. scientists in these two subgroups in SC in 2000

shows significant differences in their average research output. U.S. scientists with international

ties in SC pre-2001 produced, on average, 280 SC publications (weighted) in 2001, whereas

other U.S. scientists with no cross-border ties produced, on average, 102 SC publications

(weighted) in that same year. The difference between the two groups is significant at the 1

percent level.

In order to address this issue, I constructed another subgroup of U.S. scientists who had no

international ties pre-2001 but were similar to their counterparts in the United States with pre-

2001 cross-border links in terms of their publication record in SC before the Bush policy. I

further created a matched control sample of scientists from developed countries with no pre-2001

international tie. Model 3 in Table 3.4 repeats model 2 using these new treated and control

samples. The estimates are essentially the same as those in model 2. It is important to mention

that more than 60 percent of the U.S. scientists who had some international ties pre-2001 also

had publications with no international collaborators in the five years before the Bush policy. This

basically suggests that U.S. scientists with pre-2001 cross-border ties were at least partially

dependent on funds within the United States for some of the projects in which they were

involved. However, consistent with the predictions, their international links helped them

significantly circumvent the negative impact of the Bush policy.

Table 3.5 compares the yearly total amount of federal and state-level funds allocated to hESC

research. While the first state-level funds did not materialize until four years after the Bush

policy, they quickly surpassed the federal funds in just two years. An analysis of the research

history of the scientists who were awarded funds sponsored by one of the states reveals that

about 25 percent of them had at least one publication in SC before their first state-level grant.

75

These figures basically suggest that U.S. scientists were actively seeking state-level funds in the

face of a limited federal budget for hESC, which is consistent with the expectations.

Table 3.6 illustrates the results for the impact of the 2001 policy on the collaboration patterns of

U.S. researchers and the research productivity of their collaborators. Model 1 shows the year-by-

year impact of the policy on the number of U.S. scientists’ hESC publications with at least one

cross-border collaborator from an emerging country with flexible hESC policies using equation

2. The estimated coefficients are depicted in Figure 3.6. Consistent with the prediction, the

results suggest that U.S. scientists significantly increased their cross-border projects involving

scientists from emerging countries after the Bush policy. Model 2 presents the estimation results

for the impact of the 2001 policy on the output of non-U.S. hESC researchers who had some

U.S. collaborators pre-2001 compared to those who had no such experience. Yearly treatment

effects are illustrated in Figure 3.7. The estimates show that the hESC research output of non-

U.S. scientists who had U.S. collaborators pre-2001 increased significantly between 2001 and

2005 relative to other non-U.S. scientists. The results also suggest that the difference between the

two groups fade away after 2005. There are two possible explanations. First it is possible that

with the emergence of private and state-level funds, U.S. scientists drew most of their attention

back to the projects they could fund within the United States. Another potential explanation is

that those non-U.S. researchers who had no pre-2001 U.S. collaborators managed to catch up

with other non-U.S. researchers who had pre-2001 U.S. ties, either by learning from them or by

developing U.S. ties post-2001. Nevertheless, the results suggest that non-U.S. researchers

benefited significantly from the Bush policy intervention through increased collaborative

activities with U.S. scientists.

3.7 Discussion and Conclusion

This paper provides a detailed analysis of the impact of the restrictions put on the federal funding

of hESC research in the United States in 2001 on U.S. scientists’ research output and behavior in

hESC. The results show that U.S. hESC researchers experienced a decline in their research

output for a short while after the 2001 policy, but recovered quickly. The paper also presents an

in-depth analysis of how U.S. scientists responded strategically to the changes in the policy

76

environment by switching to non-federal funding sources and expanding their international

collaborative ties post-2001 policy. The results contribute to the literature on scientific

collaboration by highlighting how scientists engage in strategic behavior in order to mitigate

negative policy shocks that affect the resources they require for their research. The results

suggest that scientists do not easily conform to the restrictive policies that are imposed on them,

rather that they actively seek strategies to maintain their research agenda in the global

community (Bozeman & Corley, 2004; Murray, 2010). The results particularly call into question

the efficiency of restrictive policies governing the funds allocated to the stem cell research in

shaping scientists’ research direction and portfolio. The findings illustrate how scientists’

strategic choices in response to the introduced policies can undermine the intended purpose of

these policies and lead further to unintended consequences.

This issue is particularly important when we consider the current globalized nature of science. A

new policy that limits access to a particular resource in one country can easily become

ineffective due to an unintended increase in cross-country collaborations that facilitates access to

that resource in other countries. This can potentially lead to knowledge spillovers and persuade

scientists to engage in collaborations that would otherwise not be planned or desired. On the

flipside, facilitating access to a particular resource for a group of scientists can attract new

collaborators seeking to take advantage of new opportunities. Considering the longevity of

collaborative ties, encouraging international collaborations can lead to long-term positive

impacts on national scientific progress and technological advancement. There is large evidence

that collaboration is an important channel through which tacit knowledge and research

capabilities can be transferred (Inkpen & Pien, 2006; Jones, 2009b; Mowery et al., 1996; Powell,

1998; Singh, 2005). In the case of current study, for example, the results suggest that scientists

outside the United States who had collaborative ties with U.S. scientists prior to the 2001 policy

experienced a significant boost in their research output after the 2001 Bush policy. Hence,

understanding how scientists actually shape and alter their collaborative ties not only helps better

anticipate the impacts of policies, but also provides new levers for policymakers to advance local

scientific and technological capabilities through attracting international collaborators. There is

anecdotal and qualitative evidence that several emerging markets such as China, South Korea,

77

and Singapore have boosted their scientific and technological progress by providing supportive

institutional environments for international collaboration and complementing that with a

considerable amount of resources for research in several areas of science including stem cell,

nanotechnology, and biotechnology (Fox, 2007; Greenwood et al., 2006; Hwang, 2005; Xin,

2007).

The paper also provides a new potential explanation for why scientists engage in international

collaborations. Previous studies have mentioned incentives such as sharing ideas, exchanging

data, tackling global societal challenges, and improving national competitiveness as possible

drivers of cross-border collaborations. The results from this study suggest that scientists may

expand their international ties in order to diversify their research resources and shield themselves

against uncertainties in their national legal environment.

The findings in this paper also point to several opportunities for future research. First, this study

looks at only one of the U.S. scientists’ strategic responses to the introduced hESC policy in

2001. Studying other potential reactions, for instance, moving to states or countries with more

supportive policies or quitting institutions that are more heavily dependent on federal funding

(such as public universities) and joining more independent private institutions, provide further

insight into scientists’ strategic behavior. Furthermore, this study mainly focuses on the impact

of the 2001 Bush policy on the productivity and collaboration patterns of U.S. scientists. As the

results suggest, a considerable amount of cross-border policy spillovers have occurred through

international collaborative ties. Studying these policy spillovers in a broader context provides

another avenue to better understand the impact of national policies on the global scientific

progress. Finally, the growth of cross-border collaborative ties with scientists outside the United

States, particularly in emerging markets, as a response to the 2001 U.S. policy provides an

excellent context to study how internationalization of knowledge production impacts the

knowledge commercialization and use further down the value chain.

78

Figure 3.1: Average number of hESC publications by hESC scientists in the United States

versus hESC scientists from other developed countries with flexible hESC policies

Figure 3.2: Average number of ESC publications with collaborators from emerging countries

with flexible hESC policies by hESC scientists in the United States versus hESC scientists from

other developed countries with flexible hESC policies

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

US scientists

Scientists from other developed countries with flexible hESC policies

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

1998 2000 2002 2004 2006 2008 2010 2012

average number of ESC publications with collaborators from emerging countries with flexible hESC policies

US scientists

Scientists from other developed countries with flexible hESC policies

79

Figure 3.3: Distribution of the number of Stem Cell publications by U.S. hESC scientists versus

hESC scientists in other developed countries with flexible hESC policies in 2000

0

0.1

0.2

0.3

Density

0 5 10 15

Stem Cell publication count

US scientists

Scientists in developed countries with flexible hESC policies

Distribution of scientists’ publication count in SC research in 2000

80

Figure 3.4: Impact of the 2001 Bush policy on the weighted hESC research output of U.S.

scientists compared to scientists in other developed countries with flexible hESC policies

Figure 3.5: Difference between the yearly treatment effects of U.S. hESC scientists with

international collaboration experience pre-2001 and U.S. hESC scientists without such

experience

-4

-3

-2

-1

0

1

2

3

4

5

1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

Emergence and growth of state-level and private funds

Introduction of the Bush hESC policy

-4

-3

-2

-1

0

1

2

3

4

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

With internation collaboration pre-2001

Without internation collaboration pre-2001

Emergence and growth of state-

Introduction of the Bush hESC

81

Figure 3.6: Impact of the 2001 Bush policy on the number of U.S. hESC scientis’ publications

with collaborators outside the United States in emerging countries with flexible hESC policies

Figure 3.7: Impact of the 2001 Bush policy on the research output of U.S. scientists’

collaborators outside the United States

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

1999 2001 2003 2005 2007 2009

emergence and growth of state-level and private funds

Introduction of the Bush hESC policy

-4

-3

-2

-1

0

1

2

3

4

5

1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

emergence and growth of state-level and private funds

Introduction of Bush hESC policy

82

Table 3.1: hESC research in the United States and in other developed countries with flexible

policies

Year

Number of

scientists

Average publication count Average number of publications with collaborators

in emerging countries with flexible hESC policies

US Other US Other Difference US Other Difference

1999 1 1 1.000

(.)

1.000

(.)

0.000

(.)

0.000

(0.000)

0.000

(0.000)

0.000

(.)

2000 11 4 0.909

(0.301)

0.750

(0.500)

0.159

(p=0.459)

0.125

(0.334)

0.061

(0.242)

0.064

(p=0.284)

2001 36 8 1.083

(0.649)

1.250

(0.886)

0.167

(p=0.542)

0.000

(0.000)

0.064

(0.307)

-0.064+

(p=0.079)

2002 32 11 0.281

(0.457)

0.000

(0.000)

0.281*

(p=0.049)

0.000

(0.000)

0.000

(0.000)

0.000

(.)

2003 43 32 0.558

(0.665)

0.781

(0.553)

-0.223

(p=0.128)

0.000

(0.000)

0.009

(0.096)

-0.009

(p=0.307)

2004 61 95 0.492

(0.722)

0.880

(0.531)

-0.392**

(p=0.000)

0.057

(0.232)

0.043

(0.204)

0.013

(p=0.557)

2005 144 128 0.812

(0.579)

0.766

(0.768)

0.047

(p=0.568)

0.077

(0.266)

0.007

(0.083)

0.070**

(p=0.000)

2006 298 288 0.832

(0.690)

0.906

(0.664)

-0.074

(p=0.186)

0.027

(0.161)

0.002

(0.045)

0.025**

(p=0.001)

2007 720 626 0.836

(0.668)

0.906

(0.743)

-0.070+

(p=0.070)

0.098

(0.305)

0.076

(0.306)

0.017

(p=0.157)

2008 997 931 0.671

(0.701)

0.741

(0.840)

-0.070*

(p=0.046)

0.076

(0.306)

0.059

(0.270)

0.017

(p=0.157)

2009 1364 1321 0.674

(0.685)

0.559

(0.705)

0.114**

(0.000)

0.063

(0.264)

0.015

(.132)

0.048**

(p=0.000)

2010 1615 2749 0.637

(0.740)

0.640

(0.743)

-0.003

(0.895)

0.097

(0.344)

0.080

(0.300)

0.017+

(p=0.073)

** p<0.01, * p<0.05, + p<0.1

83

Table 3.2: Summary statistics in 2000

Variable US hESC Scientists

hESC Scientists in

developed countries

with flexible hESC

policies Difference

Number of scientists (in 2000) 315 258 57

SC publication count (in 2000) 0.860

(1.539)

0.705

(1.377)

0.155

(p= 0.210)

Weighted SC publication count (in 2000)

113.492

(341.578)

[n= 315]

95.604

(273.900)

[n= 258]

-17.887

(p= 0.496)

ESC publication count (in 2000) 0.063

(0.244)

0.039

(0.213)

-0.024

(p=0.202)

Weighted ESC publication count (in 2000) 17.062

(151.684)

16.797

(102.729)

16.916

(p= 0.980)

Number of SC publications with international

collaborators (in 2000)

0.146

(0.442)

0.190

(0.490)

-0.044

(p=0.260)

Number of SC publications with international

collaborators from emerging countries with

flexible policies (in 2000)

0.032

(0.013)

0.027

(0.185)

0.005

(p=0.791)

** p<0.01, * p<0.05, + p<0.1

84

Table 3.3: Yearly marginal difference between U.S. and non-U.S. scientists’ research output

Model: (1) (2) (3) (4)

Sample: hESC scientists hESC scientists with

pre-2001 and post-2004

activity

ESC scientists with

pre-2001 and post-2004

activity

SC scientists with pre-

2001 and post-2004

activity

DV: Ln(weighted

hESC pub count)

Ln(weighted hESC pub

count)

Ln(weighted ESC

,excluding hESC, pub

count)

Ln(weighted SC pub

count)

Regression: Panel OLS with

fixed effects

Panel OLS with fixed

effects

Panel OLS with fixed

effects

Panel OLS with fixed

effects

US Scientist ×

2000

0.507

(1.595)

0.490

(1.858)

0.370*

(0.153)

0.142**

(0.055)

US Scientist ×

2001

2.621**

(0.804)

2.350**

(0.763)

0.038

(0.163)

0.058

(0.052)

US Scientist ×

2002

0.819

(0.623)

0.608

(0.685)

0.033

(0.166)

0.005

(0.051)

US Scientist ×

2003

-1.772**

(0.604)

-0.217

(0.734)

0.640**

(0.147)

-0.004

(0.052)

US Scientist ×

2004

-0.725+

(0.414)

-1.335**

(0.435)

0.490**

(0.162)

-0.092+

(0.052)

US Scientist ×

2005

1.344**

(0.388)

1.744**

(0.510)

0.176

(0.148)

-0.155**

(0.053)

US Scientist ×

2006

0.452

(0.360)

0.682

(0.455)

0.222

(0.167)

-0.110*

(0.053)

US Scientist ×

2007

0.703*

(0.343)

-0.115

(0.398)

0.399*

(0.165)

-0.045

(0.054)

US Scientist ×

2008

0.712*

(0.332)

-0.287

(0.383)

0.409*

(0.170)

-0.023

(0.054)

US Scientist ×

2009

0.975**

(0.332)

0.424

(0.372)

0.399*

(0.161)

-0.029

(0.054)

US Scientist ×

2010

0.980**

(0.331)

0.245

(0.371)

0.453**

(0.160)

-0.116*

(0.054)

Scientist fixed

effects Yes Yes Yes Yes

Year fixed

effects Yes Yes Yes Yes

Constant 6.696**

(0.208)

5.518**

(0.141)

0.779**

(0.067)

1.383**

(0.020)

Observations 9,292 1,895 10,059 116,527

N. of Scientists 4,342 588 2,126 12,186

R2 0.281 0.250 0.035 0.008

Robust standard errors are shown in parentheses. ** p<0.01, * p<0.05, + p<0.1

85

Table 3.4: Year-by-year marginal difference between U.S. and non-U.S. scientists’ hESC

research output for the subsamples of scientists with and without international collaboration

experience pre-2001

Model: (1) (2) (3)

Sample: Scientists with international

collaborations pre-2001

Scientists without

international

collaborations pre-2001

Matched sample of scientists

without international

collaborations pre-2001

DV: Ln(weighted hESC pub

count)

Ln(weighted hESC pub

count)

Ln(weighted hESC pub

count)

Regression: Panel OLS with fixed

effects

Panel OLS with fixed

effects

Panel OLS with fixed effects

US Scientist × 2000 - - -

US Scientist × 2001 2.919

(1.817)

1.063

(1.317)

-1.768*

(0.836)

US Scientist × 2002 1.134

(1.968) -

-

US Scientist × 2003 2.103

(1.977)

-3.147*

(1.323)

-3.550*

(1.498)

US Scientist × 2004 -0.280

(0.719)

-2.984*

(1.184)

-2.990*

(1.457)

US Scientist × 2005 3.234+

(1.817)

-0.004

(1.232)

-0.215

(1.470)

US Scientist × 2006 1.285

(1.760)

-1.303

(1.207)

-1.305

(1.370)

US Scientist × 2007 1.074

(1.769)

-1.955+

(1.090)

-1.202

(1.067)

US Scientist × 2008 0.705

(1.743)

-2.085+

(1.137)

-1.921

(1.182)

US Scientist × 2009 1.990

(1.700)

-1.583

(1.158)

-1.963+

(1.180)

US Scientist × 2010 1.492

(1.721)

-1.833

(1.151)

-2.510*

(1.183)

Scientist fixed effects Yes Yes Yes

Year fixed effects Yes Yes Yes

Constant 6.055**

(0.756)

3.695**

(0.376)

5.975**

(0.422)

Observations 583 1,220 395

N. of Scientists 167 395 117

R2 0.171 0.266 0.250

Robust standard errors are shown in parentheses. ** p<0.01, * p<0.05, + p<0.1

86

Table 3.5: Annual federal and state-level grants for hESC research

year Total federal grants for hESC research Total state-level grants for hESC research

1999 0 0

2000 0 0

2001 0 0

2002 10.1 m$ 0

2003 20.3 m$ 0

2004 24.3 m$ 0

2005 39.6 m$ 5.1 m$

2006 37.8 m$ 73.2 m$

2007 88.1 m$ 246.2 m$

2008 142.6 m$ 426.1 m$

2009 165.2 m$ 519.5 m$

2010 123.0 m$ 214.3 m$

87

Table 3.6: Year-by-year marginal difference between U.S. and non-U.S. scientists’ cross-border

collaborations

Model: (2) (3)

Sample: hESC scientists hESC scientists

DV: Num. of publications with collaborators

from emerging countries with flexible

hESC policies

ln(weighted hESC pub count)

Regression: Panel OLS with fixed effects Panel OLS with fixed effects

US Scientist × 2000 0.399

(0.261)

-0.635

(1.105)

US Scientist × 2001 -0.095

(0.135)

3.032**

(0.597)

US Scientist × 2002 0.090

(0.074)

0.863+

(0.443)

US Scientist × 2003 0.162+

(0.093)

1.112*

(0.533)

US Scientist × 2004 0.123+

0.063)

0.880*

(0.356)

US Scientist × 2005 0.174**

(0.053)

1.053**

(0.323)

US Scientist × 2006 0.088+

(.050)

0.291

(0.258)

US Scientist × 2007 -0.001

(0.064)

-0.235

(0.245)

US Scientist × 2008 -0.016

(0.064)

-0.289

(0.231)

US Scientist × 2009 0.058

(0.053)

-0.202

(0.237)

US Scientist × 2010 0.012

(0.063)

-0.192

(0.234)

Scientist fixed effects Yes Yes

Year fixed effects Yes Yes

Constant 0.065**

(0.020)

5.967**

(0.120)

Observations 1895 8,898

N. of Scientists 588 3,987

R2 0.0302 0.265

Robust standard errors are shown in parentheses. ** p<0.01, * p<0.05, + p<0.1

88

Chapter 4

The Impact of Technological Fragmentation on Firms’ 4R&D Investment and Patenting Behavior

4.1 Introduction

Patent rights are intended to encourage R&D investment in new technologies by helping firms

and individuals protect the returns on their inventions (Arora, Ceccagnoli, & Cohen, 2008). The

belief that patents do actually induce more innovative activity has underpinned a trend towards

strengthening patent protection and expanding the range of patentable inventions over the past

three decades. However, several studies have raised concerns over whether patents serve their

intended purpose particularly in technological domains where a new technology is usually built

upon several existing technological components patented by a fragmented set of firms (Bessen &

Maskin, 2009; Heller & Eisenberg, 1998; Shapiro, 2001). The current patent law gives each of

the patents owners the power to potentially halt any sales of the new technology unless its

inventor secures a licensing deal with the respective patent holder. Several studies argue that

such intellectual property (IP) ownership fragmentation can disproportionally increase

negotiation and infringement costs associated with investments in follow-on inventions (Lemley

& Shapiro, 2007; Shapiro, 2001), an argument commonly referred to as the tragedy of the

anticommons in fragmented technological areas. Such fragmentation can thus lead to lower

investment levels and lower inventive and patenting activity (Heller & Eisenberg, 1998; Murray

& Stern, 2007).

However, despite the average growth in the technological fragmentation level in several

technological areas such as semiconductors, ICT, telecommunication and software over the past

three decades, the total R&D investment and the number of applied and granted patents in these

areas have exponentially grown, rather than declined. To explain this observation, several studies

89

posit that when firms face a fragmented technology base on which they have to build their new

technologies, they tend to pursue aggressive patenting behavior and increase their patent

portfolio size (Blind, Cremers, & Mueller, 2009; Hall & Ziedonis, 2001; Noel & Schankerman,

2013; Ziedonis, 2004). This seemingly stands in contrast to the tragedy of the anticommons

argument, which suggests that increased fragmentation lowers inventive activity and patenting

rates. To address this issue, this paper attempts to shed light on the contingencies that shape how

firms change their investment and patenting strategies in response to higher or lower levels of

technological fragmentation. In addition, considering that technological fragmentation is

endogenously shaped through firms’ patenting behavior, this study attempts to provide a

dynamic synthesis of the interplay between firms’ R&D investment, their patenting strategies,

and technological fragmentation. The results also illustrate how technological fragmentation can

explain the heterogeneity in the size distribution of firms, their innovation investment levels and

their appropriation strategies in different technological areas.

To do so, I first develop a two-stage model of a firm’s R&D investment and patenting behavior

in a fragmented technological area. In the first stage, the focal firm chooses its level of R&D

investment and in the second stage it captures the inventive outcomes of its investment and

decides on the optimal strategy mix of licensing and patenting that will allow it to appropriate the

returns on its invention. The model differentiates between two types of fragmentation: the base

fragmentation level and the domain fragmentation level. The base fragmentation level refers to

the number of firms whose patented technological components are used as the base in the focal

firm’s new inventions. This fragmentation level is only revealed in the second stage of the model

once the focal firm realizes the inventive outputs to its R&D investment. The domain

fragmentation level refers to the total fragmentation level that the focal firm can observe in the

technological domain in which it is planning to invest in the first stage.

There are two main tensions that drive the model’s behavior. First, as the level of base

fragmentation increases, the benefit to cost ratio of a defensive patent proliferation strategy

grows faster than that of a strategy that entails acquiring licenses from all the parties with patents

on which the focal firm relies. Hence, at higher levels of base fragmentation, one should expect

90

firms to rely more on patent proliferation strategies than licensing. Second, while a higher level

of domain fragmentation would shift the appropriation strategy balance towards more patenting

and less licensing in the second stage, it also reduces the overall expected profit from investment

in that technological domain in the first stage for any firm.

The model makes a few core predictions. First, firms will patent more aggressively when they

face a higher level of base fragmentation in the second stage. Second, a higher domain

fragmentation level negatively impacts the focal firm’s R&D investment in the first stage, thus

lowering its patenting rate in the second stage. Third, the model suggests that as more firms

engage in aggressive patenting strategies, the domain fragmentation level rises which in turn,

increases the costs associated with appropriating the returns on investment in follow-on

technologies. The intuition here is that, as the domain fragmentation level increases, each single

firm is more likely to engage in a patent proliferation strategy to appropriate the returns on its

invention, thus increasing the total fragmentation level for all the other firms, and decreasing

their expected profit from future investments in follow-on technologies. In the empirical section

of the paper, I provide qualitative evidence for these predictions using data on firms’ patenting

behavior in the semiconductor industry between 1980 and 1999.

This study extends the current literature on firms’ innovation and patenting strategies by

providing a more holistic and dynamic image of how fragmentation in a technological area

shapes firms’ R&D investment and appropriation strategies. The developed model reconciles the

seemingly inconsistent predictions regarding the impact of fragmentation on firms’ patenting

rates by differentiating between two types of fragmentation and the different channels through

which they impact firms’ patenting strategies. The model also presents a dynamic synthesis of

how the local short-term strategies that firms pursue in order to maximize their profits would

impact their collective expected profit in the long-term. Also, assuming that the increase in the

cost of appropriation strategies differentially impacts the incumbents and startups, the model

further suggests that as a technological domain becomes more fragmented, it becomes more

dominated by large incumbents. It also suggests that in such fragmented technological domains,

91

startups would rely more on investing in niche technologies with the goal of being acquired by

current incumbents.

On a broad level, this study illustrates how policymakers can use the legal costs associated with

patenting and infringement lawsuits as a lever to influence firms’ R&D investment and patenting

behaviors. The results also speak to the current policy discussions on what can be patented and

what the minimum scope of a patent should be (Bessen & Meurer, 2008; Burk & Lemley, 2009;

Lerner, 1994; Merges & Nelson, 1990). The results suggest that more lenient patenting

requirements may lead to undesirable technological stagnation in the long-term under particular

circumstances which are explored in this study. An analysis of the model’s boundary conditions

also highlights the importance of new organizational arrangements such as patent pools and

patent clearinghouses in alleviating the issues associated with technological fragmentation and

thus encouraging cumulative innovation in fragmented technological domains.

4.2 Prior Research on Technological Fragmentation

There is no unique definition for technological fragmentation in the literature. It generally refers

to the situation where multiple firms claim patent rights on the technological base of a new

technology or product (Bessen, 2003; Heller & Eisenberg, 1998; Shapiro, 2001; Ziedonis, 2004).

DVD technology is a good example of a technology with a highly fragmented technology base. It

is developed based on several different technological components patented by different entities.

According to patent laws, any firm that uses this technology in the products or services that it

sells, or plans to make a profit by advancing the DVD technology needs to acquire licenses to all

of these patents. Drugs and chemicals on the other hand are examples of technologies/products

with low fragmentation in their technology base.

Prior research on the impact of technological fragmentation on firms’ inventive activities can be

categorized into two broad streams. The first stream, mainly built upon the notion of the tragedy

of anticommons, predicts that technological fragmentation stifles follow-on inventions (Heller &

Eisenberg, 1998; Murray & Stern, 2007). The tragedy of the anticommons refers to a situation

where the existence of multiple gatekeepers for a common resource can lead to underutilization

92

of that resource (Buchanan & Yoon, 2000; Heller, 1998). Heller and Eisenberg (1998) point out

the case of biotechnology patents and argue that when firms need to negotiate with large

numbers of patent holders who own IP rights on a particular technology, they may face excessive

costs associated with multiple licensing negotiations which can potentially diminish their

incentives to invest in follow-on research. Several other conceptual and theoretical works have

also argued that a fragmented ownership of IP rights on a technology can lead to over-valued

licensing royalty rates and substantial infringement costs (Lemley & Shapiro, 2007; Shapiro,

2001). A few empirical studies support these arguments. Lerner (1995) shows that firms with

high litigation costs are indeed more likely to avoid highly fragmented technology classes with

too many awards granted to other firms, particularly to firms with low litigation costs . Focusing

on third-generation cellular telephones and Wi-Fi systems, Lemley and Shapiro (2007) show that

excessive royalty rates can, in fact, cause serious disincentives for investment in follow-on

technologies.

The second category of studies, mainly under the notion of strategic patenting, predicts that when

firms face a fragmented technology base they will pursue aggressive patenting strategies. The

main idea is that by increasing the size of their relevant patent portfolio, they can increase their

bargaining power in licensing negotiations or potential infringement disputes (Cohen, Nelson, &

Walsh, 2000; Hall & Ziedonis, 2001; Ziedonis, 2004). Using theoretical modeling, Bessen

(2003) and Hunt (2006) demonstrate that whenever a new technology or product reads on patents

owned by many market participants and the cost of patenting is sufficiently low, firms will

engage in aggressive patenting behavior. Based on the results from an extensive survey of R&D

managers in different industries, Cohen et al. (2000) report that, especially in industries that

involve fragmented technologies such as semiconductors and software, the two main motives for

extensive patenting, besides prevention of copying, are that patents are useful negotiation tools

and that they also help prevent lawsuits. Studying the firms in semiconductors, Hall and Ziedonis

(2001) and later Ziedonis (2004) show that the strengthening of U.S. patent rights during the

1980s has created patent amassing among capital-intensive firms aimed at achieving a stronger

position in licensing negotiations. Similarly, Bessen and Hunt (2007) find evidence of strategic

patenting in software patents.

93

Both the tragedy of the anticommons and the strategic patenting arguments do in fact rely on

technological fragmentation as the driving force behind their predictions. However, while the

tragedy of the anticommons argues that increased technological fragmentation would lead to

lower investment in follow-on technologies and thus implicitly lower patenting activity based on

fragmented technologies, the strategic patenting argument posits that higher technological

fragmentation would trigger firms to engage in aggressive patenting strategies. In order to

integrate the ideas behind these two arguments, in the next section I develop a theoretical model

that captures the main features of both, exploring how these two arguments interact with each

other in a dynamic fashion, and discovering the contingencies that influence the impact of

technological fragmentation on firms’ R&D investment and patenting activities.

4.3 The Model

In this section I develop a two-stage model of a firm’s R&D investment and patenting behavior

in the face of technological fragmentation. In the first stage, the firm decides the level of R&D it

wants to invest in a particular technology with the goal of maximizing the expected returns on its

investment. In the second stage, the technological invention is realized and the firm decides on

the number of patents it wants to file and the number of licenses it wants to acquire to optimally

appropriate the returns on its invention. Figure 4.1 shows these two stages with the decision that

the firm has to make and the information to which it has access in each stage. I start with finding

the optimal strategy for the firm in the second stage once it realizes the inventive outcome of its

investment. Next I move to the first stage and calculate the optimal investment amount with

respect to the firm’s expectation of what it can optimally appropriate in the second stage. In the

whole model, I assume that the firm is profit maximizing.

An important feature of the model is that both the level of R&D investment and the number of

patents are endogenously determined in the model. Furthermore, the model incorporates two

main elements of the current patent system that are believed to fuel technological fragmentation.

First, lower patenting costs and requirements are argued to induce more aggressive patenting

strategies (Lemley & Shapiro, 2007; Shapiro, 2001). Second, several scholars claim that the

costs associated with infringement lawsuits (hereafter referred to as court costs) play a

94

significant role in shaping firms’ patenting strategies (Burk & Lemley, 2009). Higher court costs

can potentially push firms to avoid infringement disputes to the greatest extent possible and rely

more on licensing agreements as the primary appropriation strategy. Yet, larger patent portfolios

can also increase firms’ bargaining power in licensing negotiations and, thus, it is not clear

whether increasing court costs can tame the unwanted patent races. The model includes both of

these costs and explores their roles.

I start by analyzing the firm’s behavior in the second stage. In this stage, the firm has already

committed its R&D investment and obtained its inventive outcome. The firm’s goal is to

appropriate the maximum returns possible on its invention. In the case that the inventive

outcome does not depend on any previously patented technologies, the firm can simply file one

patent to protect its invention and guarantee monopolistic profits, , from its invention over the

course of the patent life. In this case, the only cost that the firm faces is the cost of obtaining one

patent. Now imagine that the invention is developed based on patented technological

components owned by other firms. In this case, the firm has two options to protect its profits

against each of these patent holders who can potentially threaten the focal firm with a

permanent injunction on any sales from its invention: negotiating with the patent holder and

securing a license (or an alternative contractual agreement) to use its patents, or proceeding

without permission from the patent holder and relying on defensive actions in case the patent

holder files an infringement lawsuit to obtain an injunction. The focal firm’s strategy is

hypothetically a mixture of these two strategies: acquiring licenses from patent holders and

defending itself against firms in potential infringement lawsuits. The optimum depends

on the costs and benefits associated with each strategy.

What is the cost of each licensing agreement for the focal firm? Note that the focal firm has to

make separate licensing deals with each of the patent holders. From the perspective of each

patent holder, the focal firm has two choices: either get a license or defend itself in court. If the

focal firm chooses to acquire a license, its expected net profit would be equal to where

stands for the total royalty it should pay to the patent holder (indexed with where ).

Here I assume that the cost of negotiation is negligible in respect to the royalty rate. In the case

95

where the negotiation cost exceeds the royalty rate, patent holder would optimally forego the

royalty benefits considering the larger costs associated with the licensing negotiation itself. If the

focal firm decides to rely on a defensive strategy in court, its expected net profit would be equal

to its chance of winning in court ( ) multiplied by monopoly profits it can secure from its

invention ( ), minus the chance of losing in court ( ) multiplied by the court costs ( ).

Figure 4.2 shows the expected outcomes of these two decisions.

The equilibrium royalty rate is thus equal to the rate at which the expected profits from these two

decisions will be equal for the focal firm. In the case of a lower royalty rate, patent holder

would negotiate for a higher royalty rate since from its perspective the focal firm would still

strongly prefer the licensing choice than defending itself in court. For any royalty rate above this

amount, the focal firm would strongly prefer the choice of defending itself in court, which leaves

worse off compared to the equilibrium royalty rate. Hence, the equilibrium royalty rate between

the focal firm and each patent holder would be equal to:

(1) ( )( )

where is the probability that the focal firm wins its case against patent holder in a potential

infringement lawsuit. Following previous works, I assume that this probability increases with the

number of relevant patents that the focal firm holds on its invention. For now, I assume that all

the patent holders have a similar number of patents relevant to the focal firm’s invention. This

means that is equal for all s. Hence, for the rest of this section, I drop the index and use as

the probability that the focal firm would win in each potential infringement lawsuit against each

of the patent holders. The results hold if I relax this assumption. With this assumption, the

would be the same for all the patent holders and thus the focal firm faces the following total

royalty rate for licensing deals:

(2) ( )( )

Note that the total royalty rate to use the patents that belong to these firms is larger than the

royalty rate that the focal firm would have paid if it had made only one licensing deal with a

96

single firm that owned all the relevant patents (proof in the Appendix A.1.). This highlights the

over-valued royalty rates (also known as royalty stacking) associated with technological

fragmentation (Lemley & Shapiro, 2007; Shapiro, 2001).

If the focal firm chooses any , it has to rely on defensive strategies against the remaining

patent holders. For the focal firm to secure its profit against these patent holders, it needs

to win in all the potential infringement lawsuits it may face once it starts sales based on its

invention. Losing in any lawsuit would arguably result in zero profits as the claimant can

threaten the focal firm with a permanent injunction and simply ask for any profits it made. Also,

each loss in an infringement lawsuit means that the focal firm would incur the court costs

( ) as well. Given the total royalty rate for licensing deals and the expected benefits and

costs associated with relying on defensive strategies against patent holders, the focal firm

faces the following expected profit with respect to and the number of patents, , it needs to file

to increase its chances in potential infringement disputes and to achieve better licensing

terms.

(3) ( ) ( )( ) ( )( )

where is the expected profit that the firm can secure if it manages to win in

potential infringement disputes, ( )( ) is the total cost that the focal firm

incurs if it loses in one or more disputes (proof in the Appendix A.2.), ( )( ) is

the total royalty rate it has to pay to patent holders, is the cost of obtaining patents, and

is the cost associated with an R&D investment equal to . Following prior research (Bessen,

2003; Ziedonis, 2004), I assume that a larger portfolio of relevant patents increases the focal

firms’ chance of winning in an infringement dispute ( ). This means that, without putting any

restrictions on the functional form of , I assume that it has the following standard properties:

(4) ( )

97

where is the number of patents that the focal firm would optimally obtain in the second stage

to protect its invention. The

condition implies that has a diminishing marginal rate

with respect to . In other words, adding one more patent to an already large patent portfolio has

a small impact on the chance the firm would win in the potential infringement disputes.

Moreover, without assuming any particular functional form, I assume that has the following

properties with respect to :

(5) ( )

The last condition assures that for any particular invention, the next patent on the same invention

will be more costly. This puts a limit on the number of patents that a firm would file for any

given invention. The cost is also determined exogenously by the nature of the invented

technology as well as the patent laws. For example, it would be less costly for the focal firm to

obtain multiple patents on a modular invention in a technological area such as semiconductors

versus on a non-modular invention such as a new drug. It also depends on the definition of

patentable invention. Stricter rules on what is patentable and what is not can also increase the

marginal costs of extra patenting.

The focal firm thus solves the following maximization problem with respect to , the number of

licenses it would acquire, and , the number of patents it would obtain on its invention.

(6) ( )

( )

For now, I assume that is zero. The results hold as long as is much larger than .

The first order conditions are:

(7) ( )

( ) ( )

(8) ( )

( )

98

Proposition 1: if , we have ( )

,and hence .

(Proof in the Appendix A.3.)

This proposition suggests that, if the is negligible relative to , for , the focal firm

would forego any licensing negotiations and rely completely on defensive strategies.

Proposition 2: if and ( ) , there is a unique which maximizes

( ).

(Proof in the Appendix A.4.)

Propositions 1 and 2 suggest that when an invention draws from multiple patented technologies

and the court and patenting costs are negligible relative to the monopoly profits that the focal

firm can make based on its invention, a defensive strategy based on patent proliferation is always

preferred to a strategy that involves licensing negotiations. Note that very high patenting costs

relative to expected monopoly profits can create a situation in which the focal firm foregoes both

licensing and patenting strategies. Proposition 2 also suggests that technological areas in which

each technology is comprised of multiple patentable modules are more prone to such patent

proliferation strategies due to lower costs associated with extra patenting. Both propositions also

highlight the role of legal systems in shaping firms’ appropriation strategies. Lowering both the

patenting and court costs encourages defensive strategies based on patent proliferation. On the

other hand, increasing these costs can push firm towards more collaborative licensing

arrangements.

Proposition 3: For a given R&D investment level and obtained invention, an increase in

fragmentation level will lead to an increase in the number of patents that the firm obtains to

protect its invention, i.e.

.

(Proof in the Appendix A.5.)

Proposition 3 is at the core of the strategic patenting argument. It suggests that, once the focal

firm commits to a certain level of R&D investment and obtains its inventive output, it would file

a larger number of patents if it has to deal with a larger number of firms whose patents shape the

99

base for its invention. Note that this proposition assumes that some R&D investment is already

made and an inventive outcome is already realized.

Next, I move to the first-stage and calculate the focal firm’s optimal R&D investment strategy

subject to its expectation from the second-stage. The important difference between the two stages

is that in the first stage, the firm does not know the value of its inventive outcome ( ) as it is not

realized yet. The firm also does not know on how many patents its inventive outcome will

potentially infringe in the next stage. Therefore, its decision is based on its expectation of and

in the next stage. I will use and notations to show the expected values rather than

realized values. The following shows the focal firm’s expected profit in the first-stage:

(9) ( ) ( )( ) ( )( )

Similar to the second stage, I assume that the monopoly profit that the firm expects to gain from

its investment far exceeds the costs associated with a single patent to protect its invention and

court costs ( and ( )). Given the results from the second stage, the

first order condition for the maximization problem in the first stage is:

(10) ( )

.

Proposition 4: There exists a unique that maximizes the focal firm’s expected profit in the first

stage.

(Proof in the Appendix A.6.)

This proposition suggests that as long as the normal conditions hold (that the expected benefits

exceed the potential costs of defending the inventive outcome), there is a unique optimal

investment level that maximizes the focal firm’s expected returns on investment. The following

propositions can be further derived from proposition 10:

Proposition 5: a higher leads to lower investment level in the first stage, i.e.

.

(Proof in the Appendix A.7.)

100

Proposition 6: a higher leads to lower patenting rate in the second stage, i.e.

.

(Proof in the Appendix A.8.)

These two propositions are in line with the tragedy of the anticommons argument. They suggest

that higher levels of observable fragmentation in a particular technological domain can translate

into lower levels of R&D investment in the first stage and, hence, a lower patenting rate in the

second stage. The effect is shaped by how the focal firm sets its expectation of the number of

firms that it needs to deal with (either in licensing negotiations or infringement disputes) once it

makes its investment and obtains its inventive output. Note that while the impact of realized

fragmentation in the focal firms’ technology base ( ) on the optimal number of patents is

positive (proposition 4), the impact of the observable fragmentation level in the target

technological domain ( ) on the patenting rate is negative (proposition 6). These two

propositions underline where the strategic patenting and the tragedy of the anticommons

arguments differ. In the former, the fragmentation refers to the number of firms with whom the

focal firm should potentially engage in licensing negotiations. In the latter, however, the

fragmentation refers to the observable fragmentation level in a particular technological domain

before the firm makes any investment. For the rest of the paper, I will refer to the first

fragmentation as the “base fragmentation level” and the second fragmentation as the “domain

fragmentation level.” From an empirical perspective, the base fragmentation level can be directly

calculated based on the number of firms that the focal firm should potentially deal with to be

able to appropriate the returns on its inventions. The domain fragmentation level, however,

depends on the distribution of patent rights among the market participants in a particular

technological domain.

The interaction between these two effects also highlights an important dynamics. For lower

levels of base fragmentation level, the optimal strategy for the focal firm would be to make the

investment in the first stage and then pursue an aggressive defensive patenting in the second

stage. This defensive patenting strategy increases the domain fragmentation level for all the other

firms active in the same technological domain. If other firms pursue the same strategy as well,

the technological domain becomes more and more fragmented over time. To justify the higher

101

costs associated with the defensive patenting strategy, the higher domain fragmentation level

demands a larger R&D investment in the first stage in order to assure a higher level of monopoly

profits in the second stage. Assuming that startups are usually more limited in terms of the

amount of R&D investment, they are most likely to be the first to avoid long-term investments in

a fragmented technological domain and leave the field for the established incumbents who have

already garnered cross-licensing agreements among themselves. This is consistent with previous

findings by Lerner (1995) and Lemley and Shapiro (2007). In the lack of long-term investment

opportunities, an alternative strategy for startups is to invest in niche technological opportunities

and rely on acquisition by an incumbent as their goal; a phenomenon that is very common in the

ITC and semiconductor industries.

4.3.1 Boundary Conditions and Their Implications

The model developed in the previous section rests on some assumptions that are worth exploring.

First, the model focuses mainly on licensing and defensive patenting as the two alternative

appropriation strategies that firms can choose from. However, there are other methods through

which firms can protect their inventions, such as secrecy, trade-secrets, and other forms of IP

protection such as copyright and trademarks. The choice of appropriation strategy relies heavily

on the type of innovation and the industry in which a firm operates. For example, while patent

protection is a viable and commonly used appropriation strategy in the semiconductor industry,

copyright protection is much less useful since the type of inventions in this industry generally

cannot be copyrighted. In contrast, in the software industry, both patents and copyrights are

considered to be viable appropriation strategies and thus might be used as alternative options. In

publishing, patenting is no longer a viable appropriation strategy. Thus, while the tensions

examined in this paper are relevant to technological areas that have traditionally been prone to

high levels of patenting and licensing, they might be less useful in technological areas where

patents occupy second place after secrecy, copyrights or trademarks, or technological areas

where licensing contracts are not feasible or effectively enforceable.

As another consideration, one should also note that through this model, I am excluding extreme

strategic choices that firms may pursue such as opening up their patents to the public or licensing

102

out their patents at a zero or almost zero fee. While these strategic choices may have important

implications for theory, they have yet to become a common strategy in practice. This of course

does not reduce the importance of exploring these strategies in future research12

. The model also

excludes the possibility that firms might be able to find alternative technological paths to achieve

similar inventive outcomes13

.

Furthermore, the current model does not explore the possibility that firms would form

cooperative organizations such as patent pools to potentially alleviate the technological

fragmentation issue and the supra-normal licensing costs associated with it. Patent pools –

collaborative arrangements amongst multiple organizations in which participants agree to license

the patents they own on a particular technology to each other and to third parties – have

increasingly been employed to resolve technological fragmentation in the telecommunications

and ICT industries in the past decade. However, while the model excludes the possibility of these

organizational arrangements, it does provide some insights regarding their impacts on firms’

innovation strategies and future technological progress. According to the model, any change that

brings down the level of technological fragmentation without affecting other elements in the

model can potentially trigger higher investments in follow-on technologies and tip the firms’

optimized appropriation strategy mix towards higher levels of licensing and lower levels of

defensive patenting. Hence, to the extent that patent pools and similar organizational

arrangements manage to reduce the costs associated with technological fragmentation without

creating any anti-competitive impacts, they would be able to encourage higher investment in

follow-on technologies and products based on the pooled technologies.

12

See Schlam (1997) and Rosen (2004) for discussions of these strategies.

13 To the extent that such alternative technological paths exist, I suspect that fragmentation in any path would

gradually spillover to other paths. The idea is that once firms redirect their investments from one path to another, the

same dynamics would gradually occur in the alternative path.

103

4.4 Empirical Methodology

In this section I provide empirical evidence for the testable theoretical predictions outlined in the

previous section by using an extensive dataset of firms’ patenting behavior in the semiconductor

industry. Due to the nature of the data in hand, I face some limitations worth highlighting. First,

due to a lack of detailed data on firms’ licensing activities I am not able to test the impact of

fragmentation on firms’ licensing behavior as predicted in proposition 1. Furthermore, as I

cannot observe how firms allocate their R&D investment to different technological opportunities,

I can only test the model predictions at the aggregated firm level. The idea is that if the

predictions hold at the firm-technology level, they should also hold at the aggregated firm level.

With these limitations in mind, I focus on the following predictions that are empirically testable:

First, a higher base fragmentation level leads to more aggressive patenting (proposition 3). Thus,

I expect a positive association between firms’ patenting rates and the base fragmentation level

they face. Second, a higher domain fragmentation level results in lower R&D investments

(proposition 5) and lower patenting rates (proposition 6). Therefore, I expect a negative

association between firms’ R&D investment and patenting rates and the domain fragmentation

level they observe.

I should emphasize that through the following empirical exercise, I do not claim any causal

inference. The employed empirical methodology in this paper can only provide evidence for

positive or negative associations between the firms’ R&D investment and patenting rates, and the

independent variables of interest, base, and domain fragmentation levels. Establishing a causal

link between these variables is an important and interesting avenue for future research.

4.4.1 Data

I use a longitudinal dataset of the patenting behavior of public firms in the semiconductor

industry from 1980 to 1999. The semiconductor industry is known as a research-intensive

industry with a large degree of complementarity between technologies developed by different

players in the market (Cohen et al., 2000; Grindley & Teece, 1997; Hall & Ziedonis, 2001;

Ziedonis, 2004). Moreover, new technologies and products in the semiconductor industry are

104

highly dependent on several prior inventions developed both internally and externally (Grindley

& Teece, 1997; Hall & Ziedonis, 2001; Ziedonis, 2004). In fact, several previous studies have

found evidence for the extensive use of patent proliferation strategies especially among capital-

intensive firms in this industry (Hall & Ziedonis, 2001; Ziedonis, 2004). All of these

characteristics make the semiconductor industry an appropriate benchmark for my empirical

exercise.

In order to construct the dataset, a longitudinal sample of all publicly-traded U.S. firms reporting

their main line of business to be in semiconductors (SIC3674) was compiled from

COMPUSTAT.14

The resulting dataset is an unbalanced panel with 201 firms. 55 firms with less

than three years of data were subsequently removed from the sample. To identify the patents

assigned to the sampled firms, I used the dynamic matching table between NBER dataset and

Compustat provided by NBER.15

The sampled firms collectively were granted 28,395 U.S.

patents between 1980 and 1999. However, in 683 firm-year observations certain firms do not

have any successful patent applications, resulting in missing values for the main independent

variables. These observations are further removed from the sample.

4.4.2 Variables

In this section I describe the variables used in my empirical analysis.

Dependent Variable: The main outcome variables are the total amount of R&D

investment and the number of successful patent applications made by each firm in a given year.

Following prior research (Griliches, 1990; Hall, Jaffe, & Trajtenberg, 2001), I use a log normal

transformation to deal with the skewness of the R&D data. The level of observation is firm-year.

Independent Variables: In order to test the predictions of the theoretical model, I need

to derive the measures of base fragmentation and domain fragmentation levels.

14

Prior to 1980, truncation bias in the constructed variables could potentially distort the results.

15 See Hall, Jaffe, and Trajtenberg (2001) for details.

105

To construct the base fragmentation level, I use the fragmentation index proposed by Ziedonis

(2004). A number of studies have used this measure previously especially in industries such as

semiconductors and ICT (Noel & Schankerman, 2013; Von Graevenitz, Wagner, & Harhoff,

2011; Ziedonis, 2004). The index is calculated using the patent citation data and is developed

based on the legal principle that each citation from one of firm A’s patents to one of firm B’s

patents gives B the power to obtain a permanent injunction on A’s sales from its patent unless A

acquires a license to use B’s patent. This means that as the number of parties whose patents are

cited increases, the number of required licensing deals and potential infringement lawsuits also

increase. The index is based on the Herfindahl index of citation concentration:

∑(

)

where refers to the number of citations made in firm ’s patents to firm ’s patents

at year and is the total number of citations listed on firm ’s patents at year . Self-

citations and citations to non-patented materials are excluded. The measure takes the value 0 if

all the citations of firm are pointed to only one other entity. As the dispersion of citations

increases, so does the fragmentation index. A firm with 10 citations all pointed to one entity has

a local fragmentation index of 0, whereas a firm with 10 citations each pointed to a different

entity will end up with a local fragmentation index of 0.9.

In order to derive a measure of domain fragmentation level, again I take advantage of the patent

citation data. According to the model, the measure should reflect the observable fragmentation

level in the target technological domains in which the focal firm is planning to invest before it

obtains its inventive output and the base fragmentation level it faces. From the focal firm’s

perspective, the patent citation network among the firms that are active in the target

technological domains can provide an idea of what it should expect in terms of base

fragmentation level once it makes its investment. Very few patent citations among these firms

suggest that they face low base fragmentation level. On the other hand, a highly dense network

of patent citations among these firms suggests that the focal firm should probably expect a high

106

base fragmentation level once it makes its investment. Based on this idea, I use the total number

of inter-firm patent citations in a firm’s patent citation network (i.e., the firms whose patents are

cited in the focal firm’s patent portfolio) divided by the maximum potential number of such

citations as a measure for the total fragmentation level.

Figures 4.3a and 4.3b give a better illustration of these two fragmentation measures and their

differences. Each circle represents a firm in the figures and each directional link from one firm to

another represents one or more patent citations from the patents assigned to the former to one or

more patents assigned to the latter. In both figures, the focal firm (red circle) cites patents owned

by the same number of firms. Thus in both figures the focal firm faces the same level of base

fragmentation level. However, in Figure 4.3a, there is no patent citation between the cited

entities; whereas in Figure 4.3b, there is a dense web of patent citations between them. The large

number of inter-firm citations in Figure 4.3b suggests that any new patent filed in this

technological domain is very likely to read on patents assigned to several other firms in that

domain.

I use a five year window to construct the sample of firms whose patents are cited by the focal

firm. I expect that an inter-firm citation more than five years old is likely to be irrelevant to the

current technological stock of the focal firm. Considering the pace of technological development

in the semiconductor industry, five years seems to be a reasonable time frame. I further divide

the observed number of inter-firm citations by the maximum potential number of such inter-firm

citations to have a normalized measure of domain fragmentation level ranging from 0 to 1. If

firm cites different firms in a five-year window, the maximum number of potential

directional patent citations between these n firms is ( ). Thus the domain fragmentation

index is constructed as follows:

( )

Note that all the citations from firm A’s patents to firm B’s patents will be counted as one

directional citation from firm A to firm B.

107

Control Variables: To identify the firms’ R&D investment and patenting rate beyond or

below what is otherwise predicted, similar to Ziedonis (2004), I construct a baseline estimate

using the key determinants of R&D investment and patenting rate identified in the previous

research (Hall & Ziedonis, 2001; Kortum & Lerner, 2000; Pakes & Griliches, 1980). The

variables used in the baseline estimate will be used as controls in all the other estimations. The

key variables in the baseline model are as follows.

R&D Intensity: the one-year lagged logarithm of the amount of R&D spending (in $M 1984)

divided by the number of employees in the same year.

Firm’s size: the one-year lagged logarithm of the number of employees.

Capital-intensity: the one-year lagged logarithm of the deflated value of the firm’s total assets (in

$M 1984) divided by the number of employees.

Time Dummies: a dummy variable for each year from 1980 to 2000 to control for the

macroeconomic trends such as periods of technological recession or ferment.

Table 4.1 shows the summary statistics for the described variables. The reported values show

that an average firm in the data makes approximately 40 successful patents each year. It spends

$20.95 million (1984) dollars per 1,000 employees on R&D. It has a local fragmentation index of

0.88, a total fragmentation index of 0.48, and a prior portfolio strength index of 0.86. The stats

are similar to those reported in Ziedonis (2004).

4.4.3 Model Specification

I have two dependent variables. For the estimations with the log normalized R&D intensity as

the dependent variable, I use a linear OLS regression with firm and year fixed effect. Firm fixed

effects control for any time-invariant characteristic of firms. Some firms might be simply better

at assimilating and using external knowledge and developing new technologies based on them.

Using firm fixed effects, I can control for such unobserved time-invariant heterogeneity among

the firms in the sample. For estimations with the number of successful patent applications as the

dependent variable, following previous literature (Hausman, Hall, & Griliches, 1984), I use a

negative binomial estimation method with firm fixed effects and robust standard errors. Previous

studies (Cameron & Trivedi, 1998) show that negative binomial estimation is a better alternative

108

to a pure Poisson model when the sample variance exceeds the sample mean. However, in the

robustness checks section I also report the results of the OLS and Poisson estimations with firm

fixed effects and robust standard errors. I further re-estimate the models with conditional quasi-

maximum likelihood (QML) method based on the fixed-effect Poisson model developed by

Hausman, Hall, & Griliches (1984).

4.5 Results

Table 4.2 presents the main results. Models 1 and 2 show the baseline estimates for the effect of

control variables on the R&D intensity and patenting rate, respectively. Model 1 is estimated

using an OLS regression and model 2 is estimated using a negative binomial regression. The

coefficients of all three control variables, R&D intensity, size, and capital intensity are positive

and significant in both regressions with the exception of the coefficient for R&D intensity in

model 2. The estimates in the first model suggest that, not surprisingly, larger firms with higher

levels of capital and R&D intensities in any given year are more likely to have higher levels of

R&D investment in the following year. Furthermore, consistent with previous findings (Ziedonis,

2004), the results show that larger firms with higher capital intensity have a higher propensity to

patent as well.

To test the impact of the domain fragmentation level on R&D investment, I add the total

fragmentation index in model 3. The coefficients for the three control variables stay positive and

significant. The coefficient for the total fragmentation index is negative and significant. The

estimated coefficient suggests that a standard deviation increase in the total fragmentation index

is associated with about a three percent decrease in the level of firms’ R&D investment relative

to the baseline levels depicted in column 1. The results show strong support for proposition 6.

In model 4, I test the impact of domain fragmentation level on firms’ patenting rate by adding the

domain fragmentation index to model 2. As predicted by the model, the estimated coefficient of

the domain fragmentation index is negative and significant. The results suggest that one standard

deviation increase in the domain fragmentation index is correlated with about 20 percent decline

in firms’ patenting rate compared to the baseline levels presented in model 2.

109

Finally, in model 5, I add the base fragmentation level to model 4 to test the relationship between

the local fragmentation level and firms’ patenting rate. Results strongly support the model

predictions. The positive and significant estimated coefficient of the base fragmentation level

suggests that one standard deviation increase in the base fragmentation index is associated with

more than a 100 percent increase in firms’ patenting rate. Both the size and the sign of the base

fragmentation index are consistent with those reported in Ziedonis (2004). Note that the

coefficient of the domain fragmentation index remains negative and significant. The results show

support for both the strategic patenting and the tragedy of the anticommons arguments and

highlight the importance of differentiating between the two fragmentation indexes as they impact

firms’ patenting rate through two different channels.

Table 4.3 shows the results for two robustness checks. The first model repeats model 5 in the

previous step using a conditional quasi-maximum likelihood (QML) Poisson regression with

fixed effects and robust standard errors. The results remain essentially the same. Model 2 repeats

the same estimation using linear OLS regression with log transformed patenting count as the

dependent variable. Again, the results are quite similar. Overall, the results are consistent with

model predictions.

4.6 Discussion and Conclusion

This study provides a first attempt to reconcile the seemingly inconsistent predictions in the

literature regarding the impact of the technological fragmentation on firms’ R&D investment and

patenting strategies. In particular, the model makes a distinction between the base fragmentation

level, which reflects the fragmentation in the set of firms that provide the technological base for

a new invention, and the domain fragmentation level, which reflects the overall fragmentation

level in a technological domain that is observable before a firm makes any investment in that

domain. The model suggests that while the base fragmentation level can lead to aggressive

patenting strategies, as suggested by the strategic patenting argument, the domain fragmentation

level can have a negative impact on firms’ patenting rate by lowering their R&D investment, as

proposed by the tragedy of the anticommons argument. The developed model also depicts a

110

dynamic synthesis of how the short-term local strategies that firms pursue to appropriate the

returns on their R&D investments can create a dense web of overlapping patent rights in a

technological domain in the long-term, thus substantially increasing the costs associated with

investing in follow-on innovations in that domain substantially. The model further illustrates

how increased technological fragmentation can lead to the dominance of large incumbent firms

in a fragmented technological domain.

On a broad level the results suggest that, if not handled properly, the current patent laws

combined with the modular nature of many technologies developed day-to-day in the

semiconductor, ICT, and electronics industries may cause serious technological stagnation. The

developed model also renders patenting costs and requirements as major policy levers that can be

used to address some of the issues raised by technological fragmentation. At the firm level, the

model suggests that defensive patent proliferation strategies may not be sustainable in the long

term as suggested by other scholars previously (Shapiro, 2001; Ziedonis, 2004).

There are several opportunities to further advance the results from this study. First, the model

developed in the paper focuses on the defensive nature of patenting. Previous studies suggest that

firms may patent for several other reasons. For instance, they may patent to block their

competitors (Cohen et al., 2000; Parchomovsky & Wagner, 2005) or to obtain financing (Hall,

Jaffe, & Trajtenberg, 2005). Future research could provide a more accurate picture of firms’

patenting behavior by including such motives. Second, the model does not include alternative

appropriation strategies such as secrecy, trade-secrets, and the development of complementary

assets. Analyzing how firms choose one appropriation strategy over another is another avenue

for future research. Finally, the empirical analysis is based on the data from a single industry:

semiconductors. Further empirical analysis in other sectors and industries is required to test the

generalizability of the predictions.

111

Figure 4.1: The sequence of the two-stage model

Stage 1

Decision: R&D investment ( )

Available Information: domain

fragmentation level ( ),

expected monopoly returns on

investment in the second stage

( )

Stage 2

Decision: number of licenses ( )

and number of patents ( )

Available Information:

potential monopoly returns on

investment ( ), base

fragmentation level ( )

112

Figure 4.2: The licensing-patenting decision

The expected outcome of each of the focal firm’s decistions from the perspective of the

patent holder

License Defend in court

( ) ( ) ( )

113

Figure 4.3: Base and domain fragmentation levels based on patent citations

A B: A citation from a patent assigned to A to a patent assigned to B

: The focal firm

: A cited firm

Figure 4.3a: a technological domain

with domain fragmentation index of 0

Figure 4.3b: a technological domain

with domain fragmentation index of 1

114

Table 4.1: Sample statistics

Mean SD Correlations

R&D investment ($M

1984) 75.98 254.16

Patent Count 39.68 152.65 0.60**

Base Fragmentation

Index 0.88 0.17 0.10** 0.14**

Domain Fragmentation

Index 0.48 0.22 0.11** 0.06 0.28**

Capital Intensity

($M 1984/1000

employ)

162.14 133.91 0.07+ 0.12** 0.10** 0.10*

R&D Intensity

($M 1984/1000

employ)

20.95 18.79 -0.01 -0.02 0.07+ 0.02 0.50**

Firm Size (1000

employ) 4.41 11.86 0.70** 0.33** 0.01 0.12** -0.12** -0.18**

** p<0.01, * p<0.05, + p<0.1

115

Table 4.2: Main results: The impact of base and domain fragmentation levels on firms’ R&D

investment and patenting rate in semiconductors between 1980 and 1999

Model (1) (2) (3) (4) (5)

Dependent

Variable

Ln(R&D

investment) Patent count

Ln(R&D

investment) Patent count Patent count

Estimation Method OLS Negative

Binomial OLS

Negative

Binomial

Negative

Binomial

Base fragmentation

1.981**

(0.610)

Domain

fragmentation

-0.154*

(0.074)

-2.125**

(0.216)

-1.856**

(0.233)

Ln(Capital

intensity)

0.432**

(0.058)

0.352**

(0.099)

0.421**

(0.57)

0.396**

(0.090)

0.399**

(0.089)

Ln(R&D intensity) 0.239**

(0.067)

-0.043

(0.081)

0.243**

(0.066)

0.101

(0.078)

0.078

(0.078)

Ln(Firm size) 0.686**

(0.042)

0.150**

(0.050)

0.678**

(0.041)

0.163**

(0.054)

0.078

(0.056)

Constant 0.337

(0.216)

-0.092

(0.448)

0.020

(0.233)

0.439

(0.407)

-1.470*

(0.713)

Year fixed effects Yes Yes Yes Yes Yes

Firm fixed effects Yes Yes Yes Yes Yes

Observations 610 587 610 587 587

Number of unique

firms 116 93 116 93 93

R-Squared 0.911 - 0.913 - -

Log-likelihood - -1567.8839 - -1519.1348 -1513.4055

Chi-squared

(p-value)

140.41**

(0.000)

221.02**

(0.000)

135.56**

(0.000)

416.84**

(0.000)

390.15**

(0.000)

** p<0.01, * p<0.05, + p<0.1

116

Table 4.3: Robustness checks

Model (1) (2)

Dependent Variable Patent count Ln(Patent count)

Estimation Method QML Fixed Effect Poisson Model OLS

Base fragmentation 2.476+

(1.362)

2.765**

(0.640)

3yr moving average of base fragmentation

Domain fragmentation -3.895**

(0.545)

-1.551**

(0.312)

Ln(Capital intensity) 0.729**

(0.182)

0.640**

(0.158)

Ln(R&D intensity) 0.172

(0.148)

0.222

(0.142)

Ln(Firm size) 0.729**

(0.183)

0.580**

(0.131)

Constant - -3.189**

(0.674)

Year fixed effects Yes Yes

Firm fixed effects Yes Yes

Observations 587 611

Number of unique firms 93 117

R-Squared 0.630

Log-likelihood -2124.6104 -

Chi-squared

(p-value)

16057.02**

(0.000)

12.56**

(0.000)

** p<0.01, * p<0.05, + p<0.1

117

Appendices

Appendix A.1.

Consider a single firm that owns all the patents on which the focal firm’s new technology

relies ( ). Based on the line of reasoning behind equation 1, the equilibrium royalty that

the focal firm should pay to firm A would be equal to ( )( ), where is

the probability that the focal firm wins its case in the court against , or alternatively, ( ) is

the probability that wins its case against the focal firm. Considering that

, we have

( ), thus, multiplying both sides by ( ), we have

.

Appendix A.2.

The probability that the focal firm loses in out of ( ) cases and wins in the other (

) ones is ( ) . The expected cost associated with losses is thus equal to

( ) . Considering that when faced with ( ) potential infringement

cases, the focal firm may lose in any number of cases between 1 and ( ), the total expected

cost that the focal firm would incur is equal to

( )

∑ (

) ( )

Given that , using Pascal’s rule and the induction method, we have:

( ) ( )( )

Appendix A.3.

Differentiating equation (3) with respect to (FOC), we have

118

( )

( ) ( )

Lemma: ( ) for where ( ) ( )

Proof: Note that ( ) , ( ) and ( ) for . Thus ( ) for

.

Since and , we have . Hence:

( ) ( ) ( )

Appendix A.4.

Given , the first order condition in equation (8) becomes:

( )

Hence, the optimal number of patents, , is given by:

Considering that both terms are positive and the left term is decreasing in for and the

term on the right is increasing in , the Intermediate Value Theorem asserts that there is one

unique that solves this equation.

Appendix A.5.

Taking partial derivatives from equation (8) (where ) with respect to , we have:

119

[

(

)

( ) ( )

]

( )

Since the coefficient of

on the left hand side is negative, and the right hand side is positive,

we have

for any .

Appendix A.6.

The first order condition in equation (10) suggests:

Considering that both terms are positive and the term on the left is decreasing in and the term

on the right is increasing in , the Intermediate Value Theorem asserts that there is one unique

that solves this equation.

Appendix A.7.

Taking partial derivatives from equation (10) with respect to gives:

( )

( )

( )

Appendix A.8.

Note that

. Given that

and

, we have

.

120

References

Acs, Z. J., Anselin, L., & Varga, A. (2002). Patents and innovation counts as measures of

regional production of new knowledge. Research policy, 31(7), 1069-1085.

Agarwal, R., Ganco, M., & Ziedonis, R. H. (2009). Reputations for toughness in patent

enforcement: implications for knowledge spillovers via inventor mobility. Strategic

Management Journal, 30(13), 1349-1374.

Aghion, P., Bloom, N., Blundell, R., Griffith, R., & Howitt, P. (2005). Competition and

innovation: An inverted-U relationship. The Quarterly Journal of Economics, 120(2),

701-728.

Aghion, P., Harris, C., & Vickers, J. (1997). Competition and growth with step-by-step

innovation: An example. European Economic Review, 41(3): 771-782.

Agrawal, A. K., & Goldfarb, A. (2006). Restructuring research: Communication costs and the

democratization of university innovation: National Bureau of Economic Research.

Amit, R., & Livnat, J. (1988). RESEARCH NOTES. DIVERSIFICATION AND THE RISK-

RETURN TRADE-OFF. Academy of Management Journal, 31(1), 154-166.

Arora, A., Ceccagnoli, M., & Cohen, W. M. (2008). R&D and the patent premium. International

Journal of Industrial Organization, 26(5), 1153-1179.

Audretsch, D. B., & Feldman, M. P. (1996). R&D spillovers and the geography of innovation

and production. The American Economic Review, 86(3), 630-640.

Besen, S. M., & Farrell, J. (1994). Choosing how to compete: Strategies and tactics in

standardization. The Journal of Economic Perspectives, 8(2), 117-131.

Bessen, J. (2003). Patent thickets: Strategic patenting of complex technologies. Available at

SSRN 327760.

Bessen, J., & Hunt, R. M. (2007). An empirical look at software patents. Journal of Economics

& Management Strategy, 16(1), 157-189.

Bessen, J., & Maskin, E. (2009). Sequential innovation, patents, and imitation. The RAND

Journal of Economics, 40(4), 611-635.

Bessen, J., & Meurer, M. J. (2008). Patent failure: How judges, bureaucrats, and lawyers put

innovators at risk: Princeton University Press.

Blind, K., Cremers, K., & Mueller, E. (2009). The influence of strategic patenting on companies’

patent portfolios. Research policy, 38(2), 428-436.

121

Blundell, R., Griffith, R., & Van Reenen, J. (1999). Market share, market value and innovation

in a panel of British manufacturing firms. The Review of Economic Studies, 66(3): 529-

554.

Bozeman, B., & Corley, E. (2004). Scientists’ collaboration strategies: implications for scientific

and technical human capital. Research policy, 33(4), 599-616.

Buchanan, J. M., & Yoon, Y. J. (2000). Symmetric tragedies: Commons and anticommons. JL &

Econ., 43, 1.

Burk, D. L., & Lemley, M. A. (2009). The patent crisis and how the courts can solve it:

University of Chicago Press.

Cameron, A. C., & Trivedi, P. K. (1998). Regression analysis of count data: Cambridge

University Press.

Carlson, S. C. (1999). Patent pools and the antitrust dilemma. Yale Journal on Regulation, 16,

359.

Chiao, B., Lerner, J., & Tirole, J. (2007). The rules of standard‐setting organizations: an

empirical analysis. The RAND Journal of Economics, 38(4), 905-930.

Cockburn, I., Henderson, R., & Stern, S. (2000). Untangling the origins of competitive

advantage.

Cohen, W. M., Levin, R. C., & Mowery, D. C. (1987). Firm Size and R&D Intensity: A Re-

examination. Journal of Industrial Economics, 35(4), 543-565.

Cohen, W. M., & Levinthal, D. A. (1990). Absorptive capacity: a new perspective on learning

and innovation. Administrative science quarterly, 128-152.

Cohen, W. M., Nelson, R. R., & Walsh, J. P. (2000). Protecting their intellectual assets:

Appropriability conditions and why US manufacturing firms patent (or not): National

Bureau of Economic Research.

Cummings, J. N. (2011). Geography is alive and well in virtual teams. Communications of the

ACM, 54(8), 24-26.

Cummings, J. N., Espinosa, J. A., & Pickering, C. K. (2009). Crossing spatial and temporal

boundaries in globally distributed projects: A relational model of coordination delay.

Information Systems Research, 20(3), 420-439.

Cummings, J. N., & Kiesler, S. (2005). Collaborative research across disciplinary and

organizational boundaries. Social studies of Science, 35(5), 703-722.

122

Cummings, J. N., & Kiesler, S. (2007). Coordination costs and project outcomes in multi-

university collaborations. Research policy, 36(10), 1620-1634.

David, P. A., & Greenstein, S. (1990). The Economics Of Compatibility Standards: An

Introduction To Recent Research 1. Economics of innovation and new technology, 1(1-2),

3-41.

Dosi, G. (1988). Sources, procedures, and microeconomic effects of innovation. Journal of

Economic Literature, 1120-1171.

Feldman, M. P. (1999). The new economics of innovation, spillovers and agglomeration:

Areview of empirical studies. Economics of innovation and new technology, 8(1-2), 5-25.

Fletcher, L. (2001). US stem cell policy comes under fire. Nature Biotechnology, 19(10), 893-

894.

Fossett, J. W. (2007). Federalism by necessity: state and private support for human embryonic

stem cell research: Rockefeller Institute Policy Brief.

Fox, C. (2007). Cell of cells: the global race to capture and control the stem cell: WW Norton &

Company.

Fritsch, M. (2002). Measuring the quality of regional innovation systems: A knowledge

production function approach. International Regional Science Review, 25(1), 86-101.

Furman, J., Murray, F., & Stern, S. (2012). Growing Stem Cells: The Impact of U.S. Policy on

the Organization of Scientific Research. forthcoming at Journal of Policy Analysis &

Management.

Furman, J. L., Kyle, M. K., Cockburn, I., & Henderson, R. M. (2005). Public & Private

Spillovers, Location and the Productivity of Pharmaceutical Research. Annales

d'Économie et de Statistique, 79/80, 165-188.

Furman, J. L., & Stern, S. (2006). Climbing atop the shoulders of giants: The impact of

institutions on cumulative research: National Bureau of Economic Research.

Furman, J. L., & Stern, S. (2011). Climbing atop the Shoulders of Giants: The Impact of

Institutions on Cumulative Research. The American Economic Review, 101(5), 1933-

1963.

Gambardella, A., & McGahan, A. M. (2010). Business-model innovation: General purpose

technologies and their implications for industry structure. Long Range Planning, 43(2),

262-271.

Gilbert, R. J. (2004). Antitrust for patent pools: A century of policy evolution. Stanford

Technology Law Review, 2004.

123

Greenwood, H. L., Thorsteinsdóttir, H., Perry, G., Renihan, J., Singer, P., & Daar, A. (2006).

Regenerative medicine: new opportunities for developing countries. International journal

of biotechnology, 8(1), 60-77.

Griliches, Z. (1990). Patent Statistics as Economic Indicators: A Survey. Journal of Economic

Literature, 28(4), 1661-1707.

Griliches, Z. (1990). Patent Statistics as Economic Indicators: A Survey. journal of Economie

Literature, 28, 1661-1707.

Griliches, Z. (1994). Productivity, R&D, and the data constraint. The American Economic

Review, 84(1), 1-23.

Grindley, P., & Teece, D. (1997). Managing intellectual capital: Licensing and cross-licensing in

semiconductors and electronics. California management review, 39(2), 8-41.

Hall, B. H., Jaffe, A., & Trajtenberg, M. (2005). Market value and patent citations. RAND

Journal of Economics, 16-38.

Hall, B. H., Jaffe, A., & Trajtenberg, M. (2005). Market value and patent citations. The RAND

Journal of Economics, 36(1), 16-38.

Hall, B. H., Jaffe, A. B., & Trajtenberg, M. (2001). The NBER patent citation data file: Lessons,

insights and methodological tools: National Bureau of Economic Research.

Hall, B. H., & Ziedonis, R. H. (2001). The patent paradox revisited: an empirical study of

patenting in the US semiconductor industry, 1979-1995. RAND Journal of Economics,

101-128.

Harhoff, D., Narin, F., Scherer, F. M., & Vopel, K. (1999). Citation frequency and the value of

patented inventions. Review of Economics and statistics, 81(3), 511-515.

Hausman, J. A., Hall, B. H., & Griliches, Z. (1984). Econometric models for count data with an

application to the patents-R&D relationship: National Bureau of Economic Research

Cambridge, Mass., USA.

Heller, M. A. (1998). The tragedy of the anticommons: property in the transition from Marx to

markets. Harvard law review, 621-688.

Heller, M. A., & Eisenberg, R. S. (1998). Can patents deter innovation? The anticommons in

biomedical research. Science, 280(5364), 698-701.

Henderson, R. M., & Clark, K. B. (1990). Architectural innovation: the reconfiguration of

existing product technologies and the failure of established firms. Administrative science

quarterly, 9-30.

124

Holden, C. (2004). NIH report fuels push to alter US stem cell policy. Science.

Holden, C. (2005). Restiveness Grows at NIH Over Bush Research Restrictions. Science,

308(5720), 334-335.

Holland, P. W. (1986). Statistics and causal inference. Journal of the American Statistical

Association, 81(396), 945-960.

Holland, S., Lebacqz, K., & Zoloth, L. (2001). The human embryonic stem cell debate: Science,

ethics, and public policy: MIT Press.

Holm, S. (2002). Going to the roots of the stem cell controversy. Bioethics, 16(6), 493-507.

Huchzermeier, A., & Loch, C. H. (2001). Project Management Under Risk: Using the Real

Options Approach to Evaluate Flexibility in R… D. Management Science, 47(1), 85-101.

Hunt, R. M. (2006). When Do More Patents Reduce RD? The American Economic Review,

96(2): 87-91.

Hwang, S. W. S. (2005). Asia jockeys for stem cell lead.

Imbens, G. W., & Wooldridge, J. M. (2009). Recent Developments in the Econometrics of

Program Evaluation. Journal of Economic Literature, 47(1), 5-86.

Inkpen, A. C., & Pien, W. (2006). An examination of collaboration and knowledge transfer:

China–Singapore Suzhou Industrial Park. Journal of Management Studies, 43(4), 779-

811.

Jaffe, A. B. (2008). The “Science of Science Policy”: reflections on the important questions and

the challenges they present. The Journal of Technology Transfer, 33(2), 131-139.

Jaffe, A. B., Trajtenberg, M., & Henderson, R. (1993). Geographic localization of knowledge

spillovers as evidenced by patent citations. The Quarterly Journal of Economics, 108(3),

577.

Johnson, J. A., & Williams, E. D. (2007). Stem cell research: Federal research funding and

oversight: Congressional Research Service.

Jones, B. F. (2009a). The burden of knowledge and the “Death of the Renaissance Man”: Is

innovation getting harder? The Review of Economic Studies, 76(1), 283-317.

Jones, B. F. (2009b). The Burden of Knowledge and the “Death of the Renaissance Man”: Is

Innovation Getting Harder? Review of Economic Studies, 76(1), 283-317.

125

Joshi, A. M., & Nerkar, A. (2011). When do strategic alliances inhibit innovation by firms?

Evidence from patent pools in the global optical disc industry. Strategic Management

Journal, 32(11), 1139-1160.

Kaplan, S., & Tripsas, M. (2008). Thinking about technology: Applying a cognitive lens to

technical change. Research policy, 37(5), 790-805.

Kortum, S., & Lerner, J. (2000). Assessing the contribution of venture capital to innovation.

RAND Journal of Economics, 674-692.

Lampe, R., & Moser, P. (2010). Do Patent Pools Encourage Innovation? Evidence from the

Nineteenth-Century Sewing Machine Industry. The Journal of Economic History, 70(04),

898-920.

Lampe, R. L. and P. Moser (2012). Do patent pools encourage innovation? Evidence from 20 US

Industries under the New Deal, National Bureau of Economic Research.

Lane, J. (2009). Assessing the impact of science funding. Science, 324(5932), 1273-1275.

Lane, J., & Bertuzzi, S. (2011). Measuring the results of science investments. Science,

331(6018), 678-680.

Layne-Farrar, A., & Lerner, J. (2011). To join or not to join: examining patent pool participation

and rent sharing rules. International Journal of Industrial Organization, 29(2), 294-303.

Lemley, M. A., & Shapiro, C. (2007). Patent Holdup and Royalty Stacking. Texas Law Review,

85.

Lerner, J. (1994). The importance of patent scope: an empirical analysis. The RAND Journal of

Economics, 319-333.

Lerner, J. (1995). Patenting in the Shadow of Competitors. JL & Econ., 38, 463.

Lerner, J., Strojwas, M., & Tirole, J. (2007). The design of patent pools: The determinants of

licensing rules. The RAND Journal of Economics, 38(3), 610-625.

Lerner, J., & Tirole, J. (2004). Efficient Patent Pools. The American Economic Review, 94(3),

691-711.

Levine, A. (2004). Trends in the geographic distribution of human embryonic stem-cell research.

Politics and the Life Sciences, 23(2), 40-45.

Levine, A. D. (2006). Research policy and the mobility of US stem cell scientists. Nature

Biotechnology, 24(7), 865-866.

126

McCormick, J. B., Owen-Smith, J., & Scott, C. T. (2009). Distribution of human embryonic stem

cell lines: who, when, and where. Cell Stem Cell, 4(2), 107.

McGahan, A. M., & Silverman, B. S. (2006). Profiting from technological innovation by others:

The effect of competitor patenting on firm value. Research policy, 35(8), 1222-1242.

McNeil, A. J., Frey, R., & Embrechts, P. (2010). Quantitative risk management: concepts,

techniques, and tools: Princeton university press.

Mehta, A., Rysman, M. & Simcoe, T. (2010). Identifying the age profile of patent citations: New

estimates of knowledge diffusion. Journal of Applied Econometrics, 25(7): 1179-1204.

Merges, R. P. (1999). Institutions for intellectual property transactions: the case of patent pools.

University of California at Berkeley Working Paper.

Merges, R. P., & Nelson, R. R. (1990). On the complex economics of patent scope. Columbia

Law Review, 90(4), 839-916.

Mikkola, J. H. (2001). Portfolio management of R&D projects: implications for innovation

management. Technovation, 21(7), 423-435.

Mowery, D. C., Oxley, J. E., & Silverman, B. S. (1996). Strategic alliances and interfirm

knowledge transfer. Strategic Management Journal, 17, 77-91.

Mowery, D. C., Oxley, J. E., & Silverman, B. S. (1998). Technological overlap and interfirm

cooperation: implications for the resource-based view of the firm. Research policy, 27(5),

507-523.

Murray, F. (2010). The Oncomouse That Roared: Hybrid Exchange Strategies as a Source of

Distinction at the Boundary of Overlapping Institutions1. American Journal of Sociology,

116(2), 341-388.

Murray, F., & Stern, S. (2007). Do formal intellectual property rights hinder the free flow of

scientific knowledge?: An empirical test of the anti-commons hypothesis. Journal of

Economic Behavior & Organization, 63(4), 648-687.

Murray, F., & Stern, S. (2007). Do formal intellectual property rights hinder the free flow of

scientific knowledge?:: An empirical test of the anti-commons hypothesis. Journal of

Economic Behavior & Organization, 63(4), 648-687.

NIH. (1999). Fact Sheet on Stem Cell Research. Bethesda, MD: Office of the NIH Director.

Noel, M. D., & Schankerman, M. (2013). Strategic patenting and software innovation. Journal of

Industrial Economics, forthcoming.

127

World Intellectual Property (2011). World Intellectual Property Report: The Changing Face of

Innovatio WIPO Economics & Statistics Serie.

Owen-Smith, J., & McCormick, J. (2006). An international gap in human ES cell research.

Nature Biotechnology, 24(4), 391-392.

Pakes, A., & Griliches, Z. (1980). Patents and R&D at the firm level: A first report. Economics

Letters, 5(4), 377-381.

Parchomovsky, G., & Wagner, R. P. (2005). Patent portfolios. University of Pennsylvania Law

Review, 154(1), 1-77.

Ponds, R., Van Oort, F., & Frenken, K. (2010). Innovation, spillovers and university–industry

collaboration: an extended knowledge production function approach. Journal of

Economic Geography, 10(2), 231-255.

Powell, W. W. (1998). Learning from collaboration. California management review, 40(3), 228-

240.

Powell, W. W., Koput, K. W., & Smith-Doerr, L. (1996). Interorganizational collaboration and

the locus of innovation: Networks of learning in biotechnology. Administrative science

quarterly, 116-145.

Romer, P. M. (1994). The origins of endogenous growth. The Journal of Economic Perspectives,

8(1), 3-22.

Rubin, D. B. (1990). Formal models of statistical inference for causal effects. Journal of

Statistical Planning and Inference, 25(3), 279-292.

Rysman, M., & Simcoe, T. (2008). Patents and the Performance of Voluntary Standard-Setting

Organizations. Management Science, 54(11), 1920-1934. doi: DOI

10.1287/mnsc.1080.0919

Scott, C. T., McCormick, J. B., & Owen-Smith, J. (2009). And then there were two: use of hESC

lines. Nature Biotechnology, 27(8), 696-697.

Shapiro, C. (2001). Navigating the Patent Thicket: Cross Licenses, Patent Pools and Standard

Setting. In A. Jaffe, J. Lerner & S. Stern (Eds.), Innovation Policy and the Economy (Vol.

1, pp. 1190-1250): National Bureau of Economic Research.

Shapiro, C. (2001). Navigating the patent thicket: Cross licenses, patent pools, and standard

setting Innovation Policy and the Economy, Volume 1 (pp. 119-150): MIT Press.

Simcoe, T. S., Graham, S. J. H., & Feldman, M. P. (2009). Competing on standards?

Entrepreneurship, intellectual property, and platform technologies. Journal of Economics

& Management Strategy, 18(3), 775-816.

128

Singh, J. (2005). Collaborative networks as determinants of knowledge diffusion patterns.

Management Science, 51(5), 756-770.

Singh, J., & Fleming, L. (2010). Lone inventors as sources of breakthroughs: Myth or reality?

Management Science, 56(1), 41-56.

Teece, D. J. (1986). Profiting from technological innovation: Implications for integration,

collaboration, licensing and public policy. Research policy, 15(6), 285-305.

Thomson, J. A., Itskovitz-Eldor, J., Shapiro, S. S., Waknitz, M. A., Swiergiel, J. J., Marshall, V.

S., & Jones, J. M. (1998). Embryonic stem cell lines derived from human blastocysts.

Science, 282(5391), 1145-1147.

Trajtenberg, M. (1990). A penny for your quotes: patent citations and the value of innovations.

The RAND Journal of Economics, 172-187.

Utterback, J. M. (1994). Mastering the Dynamics of Innovation: How Companies Can Seize

Opportunities in the Face of Technological ChangeHarvard Business School Press.

Boston, MA.

Vakili, K., McGahan, A. M., Rezaie, R., Mitchell, W., & Daar, A. S. (2013a). Did Restricted US

Funding for Research on Human Embryonic Stem Cells Actually Restrict Research?

working paper.

Vakili, K., McGahan, A. M., Rezaie, R., Mitchell, W., & Daar, A. S. (2013b). Regenerative

Medicine Research is Global Science. working paper.

Vogel, G. (1999). Capturing the promise of youth. Science, 286(5448), 2238-2239.

Vogel, G. (2001). Bush squeezes between the lines on stem cells. Science, 293(5533), 1242-

1245.

Von Graevenitz, G., Wagner, S., & Harhoff, D. (2011). How to measure patent thickets—A

novel approach. Economics Letters, 111(1), 6-9.

Xin, H. (2007). Stem Cell Research, China Style. Science, 317(5837), 439-439.

Ziedonis, R. H. (2004). Don't fence me in: fragmented markets for technology and the patent

acquisition strategies of firms. Management Science, 50(6), 804-820.