The Interaction between Competition, … The Interaction between Competition, Collaboration and...
Transcript of The Interaction between Competition, … The Interaction between Competition, Collaboration and...
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.