J Evol Econ (2012) 22:785–810
DOI 10.1007/s00191-012-0287-z
REGULAR ARTICLE
Schumpeterian patterns of innovation and the sources
of breakthrough inventions: evidence from a data-set
of R&D awards
Roberto Fontana · Alessandro Nuvolari ·
Hiroshi Shimizu · Andrea Vezzulli
Published online: 15 July 2012
© Springer-Verlag 2012
Abstract This paper examines the relationship between Schumpeterian pat-
terns of innovation and the generation of breakthrough inventions. Our data
source for breakthrough inventions is the “R&D 100 awards” competition
organized each year by the magazine Research & Development. Since 1963, this
magazine has been awarding this prize to 100 most technologically significant
new products available for sale or licensing in the year preceding the judgment.
We use USPTO patent data to measure the relevant dimensions of the
technological regime prevailing in each sector and, on this basis, we provide a
characterization of each sector in terms of the Schumpeter Mark I/Schumpeter
Mark II archetypes. Our main finding is that breakthrough inventions are more
likely to emerge in ‘turbulent’ Schumpeter Mark I type of contexts.
R. Fontana
University of Pavia, Via Ferrata 1, 27100 Pavia, Italy
R. Fontana (B)
KITeS – Bocconi University, Via Sarfatti 25, 20139 Milano, Italy
e-mail: roberto.fontana@unibocconi.it
A. Nuvolari
LEM – Sant’Anna School of Advanced Studies,
Piazza Martiri della Liberta’ 33, 56172 Pisa, Italy
e-mail: alessandro.nuvolari@sssup.it
H. Shimizu
Institute of Innovation Research, Hitotsubashi University, Tokyo, Japan
e-mail: shimizu@iir.hit-u.ac.jp
A. Vezzulli
UECE-ISEG, Universitade Técnica de Lisboa,
Rua Miguel Lupi, 20, 1249-078 Lisboa, Portugal
e-mail: andreav@iseg.utl.pt
786 R. Fontana et al.
Keywords Innovation patterns · Radical innovations ·
Schumpeter Mark I and Mark II
JEL Classification O31 · O33
1 Introduction
One of the robust findings emerging from the rich body of empirical research
on innovation carried out over the last thirty years is that innovative activities
differ across industries along several dimensions, such as the knowledge
base underlying innovation processes, the type of actors and institutions
involved in innovative activities, the characteristics and the economic effects
of innovations (Malerba 2005). These differences have been highlighted both
by detailed case studies of individual sectors (see, for example, the essays
collected in Mowery and Nelson 1999) and by empirical contributions that
have systematically compared quantitative measures of innovation with other
economic characteristics of sectors (Cohen 2010).
In the evolutionary economics literature these differences in patterns of
innovative activities across sectors have been captured by means of taxo-
nomic exercises. The aim of these exercises was to identify in the welter of
the empirical evidence some archetypical configurations capturing the key-
dimensions in which the structure of innovative activities differs systematically
across sectors. Within this approach, one of the most common distinctions
proposed to summarize the inter-sectoral differences in patterns of innovation
is the characterization of industries in terms of Schumpeter Mark I and
Schumpeter Mark II patterns. Schumpeter Mark I industries are characterized
by turbulent environments with relatively low entry barriers, where innova-
tions are (mostly) generated and developed by new ‘entrepreneurial’ firms.
Accordingly, technological competition among firms in Schumpeter Mark I
industries assumes the form of “creative destruction”, with successful innovat-
ing entrants replacing incumbents. In contrast, Schumpeter Mark II industries
are characterized by stable environments with relatively high entry barriers in
which innovations are generated and developed by large established firms. In
Schumpeter Mark II industries, technological competition assumes the form
of “creative accumulation”, with incumbent firms introducing innovations by
mean of a process of consolidation of their technological capabilities along
well established technological trajectories (Malerba 2005: 382). The terms
Schumpeter Mark I and Mark II refer to the well-known distinction between
the early view of innovation that Schumpeter advanced in The Theory of
Economic Development (1911) (“Schumpeter Mark I”) and the later view pro-
posed by Schumpeter in Capitalism, Socialism and Democracy (“Schumpeter
Mark II”).
A substantial empirical literature has shown the existence of these two
patterns of innovation as characteristic of many industrial sectors in different
countries using data such as patents (Malerba and Orsenigo 1995, 1996)
Schumpeterian patterns of innovation and the sources of breakthrough inventions 787
or responses to innovation surveys (Castellacci 2007). One relatively ro-
bust empirical finding is that Schumpeterian patterns of innovation are, by
and large, technology-specific. More specifically, in different countries, the
same industries display similar patterns of innovation (Malerba and Orsenigo
1996). Following this cue, most research efforts have tried to relate the two
Schumpeterian patterns to a number of specific technological dimensions
summarized by the concept of technological regime. A technological regime,
as defined by Malerba and Orsenigo (1995, 1996, 1997), Breschi et al. (2000) is
a synthetic description of the “framework conditions” (Castellacci 2007: 1111)
in which innovative activities take place. These conditions shape the processes
of variety generation and selection among the firms in the sector and, through
this channel, they affect both the organization of innovative activities and the
market structure of the industry. (Malerba and Orsenigo 1996, 1997) have
proposed that the key dimensions of a technological regime are the level of
technological opportunities, the degree of appropriability of innovations, the
cumulativeness of technological advances and the characteristics of the knowl-
edge base underlying innovative activities. In general, the evidence suggests
that Schumpeter Mark I patterns of innovation emerge in the presence of high
technological opportunities, low appropriability, and low cumulativeness. By
contrast, high appropriability and high cumulativeness are conducive to the
emergence of Schumpeter Mark II patterns.1
While most of the contributions in this field have studied the precise
relationships between the dimensions of technological regimes and the sectoral
patterns of innovative activities, the overall connection between technological
regimes and the innovation performance of sectors have received much less
attention. A notable exception is the recent contribution of Castellacci (2007)
investigating the relationship between technological regimes and productivity
growth.
In this paper, we focus on the relation between sectoral patterns of innova-
tion and a more specific dimension of innovative performance, the generation
of breakthrough inventions. This approach is somewhat reminiscent of the
debate on the ‘sources of invention’ triggered by the contribution of Jewkes
et al. (1958) who, on the basis of 70 case studies of breakthrough inventions,
argued that, notwithstanding the emergence of corporate research laborato-
ries, the most important inventions of the first half of the twentieth century had
been actually generated by individual inventors and small companies. In other
words, the ultimate source of truly significant inventions was outside the walls
of the corporate research and development laboratories. For our purposes, we
consider as breakthrough inventions the inventions that have won a prestigious
competition organized by one of the leading magazines for R&D practitioners.
In comparison to other measures of innovative performance such as patents or
productivity, this type of indicator seems to represent a more ‘direct’ measure
1Schumpeter Mark II patterns are, in principle, consistent both with low and high degrees of
technological opportunities (Breschi et al. 2000: 395).
788 R. Fontana et al.
of innovative performance. Furthermore, since in this paper we shall follow the
common practice to use patent data to measure the relevant dimensions of the
technological regimes, it seems useful to have a direct indicator of innovative
performance at the sectoral level that is not also constructed using patents.
The paper is structured as follows. Section 2 reviews the relevant background
literature. Section 3 introduces our database. Section 4 presents the empirical
results, and Section 5 concludes.
2 Background literature
2.1 Technological regimes and Schumpeterian patterns of innovation
In retrospect, some of the modern research on sectoral patterns of innovation
emerged out of a feeling of dissatisfaction towards the ‘mixed’ evidence pro-
duced by the testing of the so-called ‘Schumpeterian hypothesis’ postulating a
positive effect of firm size and market concentration on innovation. Following
a suggestion of Nelson and Winter (1982), Malerba and Orsenigo (1995,
1996, 1997) argued that the inconclusive results of the literature studying
the relationship between market structure and rates of innovation were due
to a failure properly to acknowledge the specific conditions of technological
opportunities and appropriability prevailing in each sector and, relatedly, to
recognize that both innovation and market structure ought to be regarded
as endogenous variables jointly determined by the nature of the prevailing
technological regimes.
Malerba and Orsenigo’s approach to this issue was to examine system-
atically sectoral patterns of innovation across countries using patent data.
In general, they found that it was possible to use the Schumpeter Mark I–
Schumpeter Mark II distinction to characterize sectoral patterns of innovative
activities in all the major industrialized countries. In particular, Malerba and
Orsenigo (1995) examined patterns of innovation in different technology
classes using USPTO patents over the period 1969–1986 for four European
countries (Germany, France, UK and Italy), while Malerba and Orsenigo
(1996) carried out a similar exercise using EPO patents over the period 1978–
1991 for six major industrialized countries (USA, Germany, UK, France, Italy
and Japan). The dimensions considered in the assessment of the patterns of
innovation were: i) concentration and asymmetries among innovating firms
in each sector (measured, respectively, by the C4 concentration ratio and
the Herfindahl index computed using the shares of patents hold by different
firms); ii) size of the innovating firms (measured as the total share of patents
in the technology class belonging to firms with more than 500 employees);
iii) changes over time in the hierarchy of innovators (measured using the
Spearman correlation coefficient of the patents owned between the innovating
firms in different periods); iv) relevance of the entry of new innovators
(measured as the share of patents of firms applying for the first time in a
specific technology class).
Schumpeterian patterns of innovation and the sources of breakthrough inventions 789
Malerba and Orsenigo’s findings showed that technology classes with low
concentration and reduced asymmetries among innovating firms were char-
acterized by the relatively small size of innovating firms, changes in the
hierarchy of innovators and considerable innovators’ entry, pointing towards
a Schumpeter Mark I pattern. By contrast, technology classes with high
concentration and asymmetries among innovating firms were characterized by
the large size of innovators, a relative stability in the hierarchy of innovators,
and limited entry, pointing towards a Schumpeter Mark II pattern. These
results were further corroborated by a principal component analysis on the
variables mentioned above. In all countries, the principal component analysis
produced one dominant factor (explaining in all cases more than 50 % of
the variance) the loadings of which are fully consistent with the Schumpeter
Mark I/Schumpeter Mark II distinction. The overall conclusion of these inves-
tigations was the recognition of systematic differences across industries in the
patterns of innovation (differences that it is possible to characterize in terms of
the Schumpeter Mark I and Schumpeter Mark II dichotomy) and of similarities
across countries in sectoral patterns of innovation for a specific technology
(Malerba and Orsenigo 1997: 94).
Malerba and Orsenigo’s interpretive hypothesis of this finding is that
the emergence of these two sectoral patterns of innovation is accounted
for by different ‘technological regimes’ that shape and constrain innovative
processes in different sectors. In their definition, a technological regime is
a synthetic description of the technological environment in which firms act.
More specifically, a technological regime is a specific combination of some
basic characteristics of technologies: opportunity conditions, appropriability
conditions, cumulativeness of technical progress, and the nature of the knowl-
edge base (Malerba and Orsenigo 1997: 94). The hypothesis is that Schumpeter
Mark I patterns of innovation emerge in contexts characterized by high tech-
nological opportunities, low appropriability and low cumulativeness, whereas
Schumpeter Mark II pattern emerge in contexts of high appropriability and
cumulativeness (technological opportunities can be both high or low). Breschi
et al. (2000) provided a first (successful) test of these hypotheses concerning
the relationship between technological regimes and sectoral patterns of inno-
vation using data from the PACE innovation survey to measure the relevant
dimensions of the technological regimes and EPO patents to measure the
sectoral patterns of innovation.
Further contributions have confirmed the merits of introducing the
Schumpeter Mark I/ Schumpeter Mark II distinction.2 Van Dijk (2000) studied
the industrial structure and dynamics in Dutch manufacturing and found con-
sistent differences in the patterns of industrial dynamics between Schumpeter
2Other contributions have, however, argued that the Schumpeter Mark I–Schumpeter Mark II
distinction may be too narrow and does not map adequately the large empirical variety of inter-
sectoral patterns of innovative activities. Therefore more articulated taxonomies of innovation
patterns have been proposed. The most famous example is the Pavitt’s taxonomy (Pavitt 1984).
For a comprehensive discussion, see Marsili and Verspagen (2002).
790 R. Fontana et al.
Mark I and Schumpeter Mark II industries. The distinction between
Schumpeter Mark I and Schumpeter Mark II seems also useful to study
patterns of innovation with broad technological fields. For example, Corrocher
et al. (2007) have shown the existence of Schumpeter Mark I and Schumpeter
Mark II patterns of innovation examining patents taken in different sub-
segments of ICT applications.
More recently, the focus of the empirical investigations has shifted towards
the connection between technological regimes and innovation performance.
Castellacci (2007) studied the relationship the relationship between differences
in sectoral productivity growth and technological regimes in nine European
countries (Germany, France, Italy, Netherlands, Norway, Portugal, Sweden,
UK and Austria) in the period 1996–2001. Technological regimes are defined
in terms of technological opportunities, appropriability and cumulativeness,
and the measurement of the different dimensions of technological regimes
is based on responses to the CIS surveys. Castellacci finds that Schumpeter
Mark II sectors are characterized by higher rates of productivity growth.
Furthermore, the relationship between the different characteristics of the
technological regimes and productivity is different in the two Schumpeterian
patterns.
2.2 Schumpeterian patterns of innovation and breakthrough inventions
Another critical dimension of technological performance is the emergence of
breakthrough inventions. The recent emphasis on the key role of breakthrough
inventions is related to the growing appreciation of the highly skewed nature of
innovation size distributions (Silverberg and Verspagen 2007). Clearly, if the
majority of innovations yield only modest returns and most economic value is
actually generated by relatively few breakthrough inventions situated in the
tail of the value distribution, the search for the possible determinants of these
breakthrough inventions becomes a fundamental research issue (Scherer and
Harhoff 2000).
Existing approaches to the study of the role of breakthrough invention
can be classified into two camps. On the one hand, there are historians of
technology and economic historians who have frequently acknowledged that
serendipity plays a large role in the generation of breakthrough inventions.
Mokyr (1990: 13) is possibly summarizing what is the conventional wisdom
on this issue when he writes: “macro-inventions [. . . ] do not seem to obey
obvious laws, do not necessarily respond to incentives and defy most attempts
to relate them to exogenous economic variables. Many of them resulted from
strokes of genius, luck or serendipity. Technological history, therefore, retains
an unexplained component that defies explanation in purely economic terms.
In other words, luck and inspiration mattered, and thus individuals made a
difference”. Still, some economic historians have been able to unravel some
significant relationship between breakthrough inventions and economic and
social variables (Khan and Sokoloff 1993).
Schumpeterian patterns of innovation and the sources of breakthrough inventions 791
On the other hand, there is the recent literature in management. Ahuia and
Lampert (2001) assess the relationship between breakthrough inventions and
R&D strategies at firm level. Their findings suggest that established firms tap-
ping new technologies are more likely to introduce breakthrough inventions.
Chandy and Tellis (2000) look at the role of incumbent firms in the generation
of radical innovations. They find that, despite their inertia, established firms
can be an important source of radical innovations. Finally, Schoenmakers
and Duysters (2010) analyze the connection between breakthrough inventions
and different types of knowledge. They find that radical inventions are to a
higher degree based on existing knowledge rather than incremental inventions.
They also find that inter-firms collaborations play an important role in the
development of radical inventions, as highlighted also by Singh and Fleming
(2010) at the individual inventor level.
This paper contributes to this emerging literature on the sources of break-
through inventions by examining this issue from the perspective of the litera-
ture on Schumpeterian patterns of innovation. More specifically, we shall not
deal directly with the issue of the possible economic and social determinants
of major macro-inventions, but we shall limit ourselves to study the possible
role played by different Schumpeterian patterns of innovation in the genera-
tion of breakthrough inventions. If it turns out that Schumpeterian patterns
affect the generation of breakthrough inventions, it is important that future
contributions devoted to study of the sources of breakthrough inventions at
micro level will try to control explicitly for the dimensions of the technological
regime prevailing in the industries under consideration. A similar exercise was
carried out by Granstrand and Alange (1995) for the Swedish case using a
sample of 100 ‘significant’ inventions that occurred in the period 1945–1980,
although their focus was not so much on the impact of the technological
regimes but on the relative contribution of different organizational structures
(individual inventors, small firms, large firms) to the generation of inventive
breakthroughs. Their findings were mixed. They found that large firms were
responsible for 80 % of the inventions in their sample, but still a sizable share
of breakthrough inventions (i.e. the remaining 20 %) could be ascribed to
individual inventors and small firms.
3 The “R&D 100” awards database
Our source of data is the ‘R&D 100 Awards’ competition organized by the
magazine Research and Development (previously called Industrial Research).
The magazine, founded in 1959, is one of the most authoritative regular publi-
cations for R&D practitioners.3 The ‘R&D 100 Awards’ competition has been
3The information reported here on R&D magazine and the R&D competition was retrieved from
http://www.rdmag.com, last accessed on 7/7/2011.
792 R. Fontana et al.
running continuously since 1963. Each year the magazine awards with a prize
the 100 most technologically significant products available for sale or licensing
in the year preceding the judgment.
Throughout the years, breakthroughs inventions such as Polacolor film
(1963), the flashcube (1965), the automated teller machine (1973), the halo-
gen lamp (1974), the fax machine (1975), the liquid crystal display (1980),
the printer (1986), the Kodak Photo CD (1991), the Nicoderm antismoking
patch (1992) and Taxol anticancer drug (1993) have received the prize. In
order to apply for the prize, the inventors or their companies must fill out
an application form providing a detailed description of the invention. The
prize consists of a plaque which is presented at a special ceremony. There is
no monetary prize. The prize is awarded by a jury composed of university
professors, industrial researchers and consultants with a certified level of
competence in the areas they are called upon to asses. Jury members are
selected by the editor of the magazine and inventions are assessed according
to two criteria: i) technological significance (i.e., whether the product can
be considered a major breakthrough from a technical point of view) and ii)
competitive significance (i.e., how the performance of the product compares
to rival solutions available on the market). R&D 100 awards are accolades
comparable to the Oscars for the motion picture industry as “they carry
considerable prestige within the community of R&D professionals” (Block and
Keller 2009: 464).
There are a number of characteristics of the R&D 100 awards competition
that, prima facie, appear particularly promising for the study of inventive
breakthroughs. First, the R&D 100 awards competition represents a good
opportunity for companies, and government laboratories to showcase their
inventions. Second, R&D 100 awards are granted to inventions that, at least in
principle, should embody a clearly documented improvement of the state-of-
the-art (i.e. a technological breakthrough). Third, the selection of the awards
is made by a competent, authoritative jury of experts. Fourth, R&D awards
may be assigned both to patented and not-patented inventions. Finally, there
seems to be limited space for strategic behaviors and attempts to conditioning
the jury, because the nature of the prize is simply honorific.
Given these properties, it is surprising that economists of innovation have
so far paid scant attention to this type of data. To the best of our knowledge,
the R&D 100 awards data have been so far only used in three contributions:
Carpenter et al. (1981), Scherer (1989) and, more recently, Block and Keller
(2009). Carpenter et al. (1981) used these data to study differences in citations
received between patents covering awarded inventions and a random sample
of patents, providing an important corroboration for the use of forward
citations as an indicator of patent quality. Scherer (1989) used information
on the mean and maximum R&D costs of the awarded inventions to study
the distribution of R&D investments. Finally, Block and Keller (2009) used
the R&D 100 awards to gain insights on the growing importance of public
institutions in the US innovation system over the period 1971–2006. From our
perspective, it is reassuring that three authoritative contributions in the field of
Schumpeterian patterns of innovation and the sources of breakthrough inventions 793
innovation studies have employed the data to study the nature of breakthrough
inventions.
4 Empirical analysis
Retrieving the information from different issues the magazine, we have con-
structed a dataset with all the R&D 100 awards granted from 1963 to 2005. In
this section we use the dataset to study the impact of different Schumpeterian
regimes on the generation of breakthrough inventions. We proceed in two
steps. First, we introduce some preliminary descriptive statistics of the dataset
to check the reliability of the source. Second, we carry out an econometric
study of the probability of the occurrence of breakthrough invention as a
function of the Schumpeterian regime prevailing at the sectoral level.
4.1 Descriptive statistics
Figure 1 displays the share of awards granted to US applicants for the prize.
The nationality of the applicants has been assigned using the organization,
rather than by looking at the nationality of the inventors. Over the period
1963–2002, the share of US awards declined indicating that other countries
closed the gap with the US in terms of technological performance. Interestingly
enough, during the period 2003–2005, the US recovered their edge, but, of
course, this is a too short span of time for detecting clear trends.
Figure 2 displays the share of awards received by applicants from different
countries by sub-periods excluding the US that, as one would have expected
given the nature of the competition and the place of publication of the
magazine, dominate the sample. The figures clearly indicate that Japan and
Germany are the two most prominent contenders of US technological leader-
ship. Figure 2 shows how this effort of closing the gap evolved over time, with
Fig. 1 Share of “R&D
100”awards received by
US applicants
.
82
.
84
.
86
.
88
.
9
1963-1972 1973-1982 1983-1992 1993-2002 2003-2005
794 R. Fontana et al.
0
.
01
.
02
.
03
.
04
.
05
.
06
.
07
1963-1972 1973-1982 1983-1992 1993-2002
Canada Finland France Germany
Israel Italy Japan Netherland
Sweden Switzerland UK
Fig. 2 Share of “R&D 100” awards received by applicants of different countries
Japan and Germany progressively overtaking two older established players,
France, and UK.
Figure 3 shows the shares of awards granted to different type of organi-
zations. The trends here are consistent with the literature that has recently
0
.
2
.
4
.
6
.
8
1
1960 1970 1980 1990 2000 2010
years
Academic_Institution Company Government_Organization
Fig. 3 Shares of “R&D 100” awards granted to different type of organization
Schumpeterian patterns of innovation and the sources of breakthrough inventions 795
0
10
20
30
40
Co
lla
bo
ra
tiv
e
_
in
ve
n
tio
ns
1963 1968 1973 1978 1983 1988 1993 1998 2003
years
Fig. 4 Number of collaborative inventions receiving an “R&D 100” award
pointed out the increasing involvement in inventive activities of a number of
new actors such as government laboratories and universities. Whereas in the
early 1960s, corporations were the primary source of inventions, in the most
recent years this has clearly not been the case.
Figure 4 displays the number of awards that are the outcome of collabora-
tive activities. The figure shows an increasing trend which is fully consistent
with the emphasis that has been put on the growing role of cooperation and
networking in the field of innovative activities (Freeman 1991).
To carry out our analysis at the sectoral level, we classified each awarded
invention according to a technology-oriented classification of 30 different
sectors based on the co-occurrence of the International Patent Classification
(IPC) codes proposed by the Observatoire des Sciences et des Techniques
(OST).4 We assigned each R&D 100 invention to only one of the 30 OST
sectors. These sectors were further aggregated into 5 ‘macro’ technological
classes (called ‘OST5’ henceforth) defined according to the ISI-INIPI-OST
patent classification based on the EPO IPC technological classes, as reported
in Table 1.5
Figure 5 contains histograms showing the distribution of the awarded
inventions across the 30 OST sectors.
4See Hinze et al. (1997).
5Technology-oriented classification system jointly elaborated by the German Fraunhofer Institute
of Systems and Innovation Research (ISI), the French Patent Office (INIPI) and the Observatoire
des Science and des Techniques (OST).
796 R. Fontana et al.
Table 1 Aggregation of the 30 ISI-INPI-OST sectors in 5 macro-classes
MacroISI-INIPI-OST ISI-INIPI-OST Technological class
1 1, 2, 3, 4, 5 Electrical engineering
2 6, 7, 8, 27 Instruments
3 9, 10, 11, 12, 14, 15 Chemistry & pharmaceuticals
4 13, 16, 17, 18, 20, 24, 25 Process engineering
5 19, 21, 22, 23, 26, 28, 29, 30 Mechanical engineering
As one would have expected, there is some distortion towards ‘high-tech’
sectors such as instruments, biotechnology, information and communication
technologies, optics (lasers), and semiconductors. The predominant sector is
instrumentation (control instruments). On the one hand, this may be clearly
explained by the interests of the editors and the readership of the magazine,
given that instrumentation plays a central role in the majority of modern R&D
processes. On the other hand, this may be the consequence of the fact that
it is easier for inventions in these categories to prove that they are superior
to the state of the art by means of quantitative assessment of technological
performance. All in all, these results confirm that the R&D 100 awards tend to
cover, as one would have expected, a high-tech R&D intensive segment of the
economy.
Finally, we check whether the R&D 100 inventions that were patented
(more specifically, those for which we were able to match with one USPTO
patent) receive more citations than an analogous random sample of patents.
Accordingly, for each R&D invention with a USPTO patent we construct a
0
20
0
40
0
60
0
Fr
e
qu
en
cy
0
El
ec
tri
ca
l E
ng
in
e
er
in
g
Au
di
o
vis
u
al
TL
C IC
T
Se
m
ic
on
du
ct
o
rs
O
pt
ic
s
Co
n
tro
l I
n
st
ru
m
e
n
ts
M
ed
ica
l
O
rg
an
ic 
Ch
em
is
try
Po
ly
m
er
s
Ph
ar
m
a
Bi
o
te
ch
M
at
er
ia
ls Fo
od
M
a
te
ria
ls
 
ch
em
is
try
Ch
em
ic
al
 
En
gi
n
e
er
in
g
Su
rfa
ce
M
a
te
ria
ls
 
pr
oc
e
ss
in
g
Th
er
m
a
l P
ro
ce
ss
es
En
vi
ro
n
m
en
t
M
ac
hi
ne
 to
ol
s
En
gi
n
es
M
ec
ha
n
ica
l e
le
m
e
n
ts
Ha
n
dl
in
g
Fo
od
 
pr
oc
e
ss
in
g
Tr
a
n
sp
or
t
Nu
cle
ar
Sp
ac
e
Co
n
su
m
er
 g
oo
ds
OST sectors
Fig. 5 Distribution of “R&D 100” awards across technology classes, 1963–2005
Schumpeterian patterns of innovation and the sources of breakthrough inventions 797
Table 2 Patent citations received by R&D 100 inventions and a random sample of patents
(matched by granted year and technology class)
Number Mean Median Standard Min Max
deviation
R&D 100 patents 535 12.88037 7 16.17822 0 137
Random Sample 5331 8.483024 4 14.11133 0 329
Mann–Whitney test rejects the null hypothesis of equal populations
‘matched random’ sample of ten patents from the same grant year and from
the same IPC class. The results of this test are reported in Table 2.6 The non
parametric Mann–Whitney test confirms that the median number of citations
of patents associated with a R&D 100 invention is significantly higher than
the median of the random matched sample. These results confirm the early
findings of Carpenter et al. (1981) obtained for the two years 1969–1970 of
awards and provides an important corroboration for our use of the R&D 100
data set as an indicator of breakthrough inventions.
4.2 The econometric exercise
In this section, we carry out our econometric exercise. Our main explanatory
variables are constituted by a set of time-varying indicators constructed using
patent based data for each of the five macro-classes mentioned above. These
indicators aim at capturing different patterns of innovative activities across
classes and over time.7 Following the contributions of Breschi et al. (2000),
Hall et al. (2001) and Corrocher et al. (2007), we computed the indicators as
follows (where j = 1, .., 5 for each OST5 sector and t = 1976, . . ., 2006 is the
year of granting of each patent):
1) PATGROWTHjt = patjt−patjt−1patjt−1
where patjt is the total number of patents granted in OST5 class j in year t.
2) Entryjt = newpatjtpatjt
where newpatjt is the total number of patents granted in OST5 class j in
year t by new innovators (i.e. by firms patenting for the first time in class j).
6The random matched sample includes 5331 patents and not 5350 because, for some specific years
in some technology classes, it was not possible to collect enough patents to create the match.
7Our main source of information is the NBER Patent Data Project which collects a very
comprehensive set of information on USPTO patents for the 1976–2006 period (e.g. dates of
application and grant, inventors and applicant’s name, number of claims, technological classes,
forward and backward citations, etc.). The reclassification of all USPTO patents according to the
2008 IPC classification system is available on the NBER Patent Data Project website and it has
been performed on the basis of the International Patent Classification Eighth Edition available at:
http://www.uspto.gov/go/classification/uspc002/us002toipc8.htm. For a comprehensive description
of the database, see Hall et al. (2001).
798 R. Fontana et al.
-
2
-
1
0
1
2
SC
HU
M
P
1977 1982 1987 1992 1997 2002
year
Electrical_engineering Instruments
Chemistry_Pharmaceuticals Process_engineering
Mechanical_engineering
Fig. 6 The dynamics of SCHUMP for each OST5 macro sector (1977–2005)
3) C4jt representing the concentration ratio of the top four patenting firms
(in terms of number of patents granted in a given year t and class j).
4) Stabilityjt is the Spearman rank correlation coefficient between hierarchies
(in term of number of patents granted) of firms patenting in year t and
firms patenting in year t − 1 in class j.
Following Breschi et al. (2000), the last three indicators (Entry, C4 and
Stability) are consolidated in a unique indicator called Schump jt by means of
principal component analysis. Schump jt is our main variable of interest and
represents the prediction obtained using the scoring coefficients of the first
component and the standardized values of the original variables.8 It provides
an indication of the type of Schumpeterian pattern of innovation prevailing
in a given class i in year t. High values of Schump jt indicate a Schumpeter
Mark II type regime (i.e., a “deepening” pattern of innovative activities with
a concentrated and stable population of innovators). Low values of Schump jt
indicate a Schumpeter Mark I type regime (i.e., a “widening” pattern with a
large and turbulent population of innovators) (Breschi et al. 2000). Figure 6
depicts the different trend of Schump jt across the OST5 macro sectors within
our time window.
Two sectors (Electrical Engineering and Chemistry & Pharmaceuticals)
are consistently close to a Schumpeter Mark II type of pattern, while two
8The extracted principal component accounts for about 70 % of the total variance. The correla-
tions between the principal component and our three original indicators C4, Entry, and Stability
are 0.37, −0.67 and 0.64, respectively.
Schumpeterian patterns of innovation and the sources of breakthrough inventions 799
other sectors (Mechanical and Process Engineering) are close to a Schumpeter
Mark I type of pattern and one sector (Instruments) displays an intermediate
pattern between these two.
5) Herfsourcestechjt is an index of the relative variety of knowledge sources
across technological classes and is calculated in a similar way as in
Corrocher et al. (2007). Let ajht = cjhtcjt be the share of backward citations
from patents granted in year t and belonging to OST5 class j to previous
patents in IPC class h (defined at 4 digit level), where cjht is the total
number of patents belonging to IPC class h and cited by patents granted
in year t and belonging to OST5 class j and c jt = ∑h cjht.
Let then vjht = pjhtpjt be the share of patents (for each granting year t)
in OST5 class j belonging to IPC class h. Let Herf techjt and Herfcittechjt
be the corrected Herfindahl indexes (Hall et al. 2001) calculated using,
respectively, the shares cjht and vjht and indicating how much each OST5
class j and its knowledge sources are concentrated (in term of number of
patents granted and number of backward citations made) across different
IPC 4 digit sub-classes in a given year t. The resulting relative index of
concentration of knowledge sources across IPC technological classes is
given by the ratio of the previous two indexes: Herfsourcestechjt =
Herfcittechjt
Herf techjt
.
6) Herfsourcesf irmjt =
Herfcitf irmjt
Herf f irmjt
. This is an index of the relative variety of
knowledge sources across firms and is calculated (for each granting year t)
in a similar way as Herfsourcestechjt . Here the Herfindahl index at the
numerator is calculated using the shares of backward citations from
patents in class j to patents applied by firm z: b jzt = d jztd jt , where djzt is
the total number of cited patents from OST5 class j applied by firm z
(excluding self citations) and d jt = ∑z djzt. The Herfindahl index at the
denominator measures the degree of concentration across firms in a given
class j calculated with respect to the number of patents granted in a given
year t.
7) Selfsourcesjt = scjtcjt is an index of intensity of internal knowledge sources
and is defined for each OST5 class j and granting year t as the ratio
between the total number of self-citations (i.e. backward citations to
patents applied by the same firm z) over the total number of backward
citations.
In addition to these indicators we also include ‘applicant level’ variables and
further controls. Our final reference period of analysis ranges from 1977 to
2005 with a total of 2802 inventions awarded.9 Table 3 gives a comprehensive
overview of the variables used in the econometric exercise.
9We dropped the first (1976) and last (2005) year of reference to avoid possible inconsistencies
when calculating our time-varying industry indicators based on patent data.
800 R. Fontana et al.
Table 3 Description of the variables
Description Type
Dependent variable
OST5 Invention-type classification according 5 categories:
to OST5 (see Table 1) j = 1, 2, 3, 4, 5.
Independent variables
Sector-level characteristics j = category of the invention (OST5);
t = year of award
PAT_GROWTHjt Patent growth rate Continuous
SCHUMPjt Schumpeterian pattern of innovative Continuous
activities index
HERFSOURCES_TECHjt Variety of knowledge sources across Continuous
technological classes index
HERFSOURCES_FIRMjt Variety of knowledge sources across Continuous
firms index
SELFSOURCESjt Intensity of internal knowledge Continuous
sources index
Invention-level characteristics
MAPPL = 1 for multiple applicant organizations, Dummy
= 0 otherwise
NINV Number of inventors Count
USA = 1 if at least one applicant is a U.S. Dummy
organization, = 0 otherwise
GOV = 1 if at least one applicant is a Dummy
governmental organization,
= 0 otherwise
ACAD = 1 if at least one applicant is an academic Dummy
organization, = 0 otherwise
Other controls
dum1986_1995 = 1 the invention has been awarded in the Dummy
1986–1995 decade, = 0 otherwise
dum1996_2005 = 1 the invention has been awarded in the Dummy
1996–2005 decade, = 0 otherwise
Tables 4 and 5 report the main descriptive statistics of the variables used in
the analysis, as well as the distribution of the awarded inventions across sectors
and over time.
Table 4 Descriptive statistics
Variable Obs Mean Std. dev. Min Max
OST5 2802 2.514 1.322 1 5
PAT_GROWTHjt 2802 0.049 0.126 −0.290 0.478
SCHUMPjt 2802 0.261 0.733 −1.412 1.602
HERFSOURCES_TECHjt 2802 0.521 0.103 0.273 0.910
HERFSOURCES_FIRMjt 2802 0.841 0.156 0.565 1.382
SELFSOURCESjt 2802 0.142 0.048 0.085 0.448
MAPPL 2802 0.256 0.437 0 1
NINV 2802 1.665 0.902 1 5
USA 2802 0.877 0.329 0 1
GOV 2802 0.320 0.467 0 1
ACAD 2802 0.074 0.262 0 1
dum1986_1995 2802 0 0 0 1
dum1996_2005 2802 0.322 0.467 0 1
Schumpeterian patterns of innovation and the sources of breakthrough inventions 801
T
ab
le
5
D
is
tr
ib
ut
io
n
of
‘R
&
D
10
0’
aw
ar
ds
ac
ro
ss
se
ct
or
s
Y
ea
r
E
le
ct
ri
ca
le
ng
.
In
st
ru
m
en
ts
C
he
m
is
tr
y
&
ph
ar
m
a
P
ro
ce
ss
en
g.
M
ec
ha
ni
ca
le
ng
.
A
ll
se
ct
or
s
19
77
20
(2
0.
2
%
)
38
(3
8.
38
%
)
14
(1
4.
14
%
)
18
(1
8.
18
%
)
9
(9
.0
9
%
)
99
(1
00
%
)
19
78
24
(2
4.
24
%
)
37
(3
7.
37
%
)
17
(1
7.
17
%
)
14
(1
4.
14
%
)
7
(7
.0
7
%
)
99
(1
00
%
)
19
79
33
(3
2.
35
%
)
32
(3
1.
37
%
)
18
(1
7.
65
%
)
12
(1
1.
76
%
)
7
(6
.8
6
%
)
10
2
(1
00
%
)
19
80
35
(3
2.
11
%
)
32
(2
9.
36
%
)
8
(7
.3
4
%
)
30
(2
7.
52
%
)
4
(3
.6
7
%
)
10
9
(1
00
%
)
19
81
24
(2
4.
74
%
)
47
(4
8.
45
%
)
7
(7
.2
2
%
)
13
(1
3.
4
%
)
6
(6
.1
9
%
)
97
(1
00
%
)
19
82
25
(2
5.
25
%
)
40
(4
0.
4
%
)
7
(7
.0
7
%
)
17
(1
7.
17
%
)
10
(1
0.
1
%
)
99
(1
00
%
)
19
83
20
(2
0.
2
%
)
38
(3
8.
38
%
)
6
(6
.0
6
%
)
19
(1
9.
19
%
)
16
(1
6.
16
%
)
99
(1
00
%
)
19
84
24
(2
4.
24
%
)
44
(4
4.
44
%
)
0
(0
%
)
21
(2
1.
21
%
)
10
(1
0.
1
%
)
99
(1
00
%
)
19
85
36
(3
6.
36
%
)
39
(3
9.
39
%
)
1
(1
.0
1
%
)
19
(1
9.
19
%
)
4
(4
.0
4
%
)
99
(1
00
%
)
19
86
34
(3
4.
34
%
)
37
(3
7.
37
%
)
0
(0
%
)
23
(2
3.
23
%
)
5
(5
.0
5
%
)
99
(1
00
%
)
19
87
25
(2
5
%
)
50
(5
0
%
)
0
(0
%
)
20
(2
0
%
)
5
(5
%
)
10
0
(1
00
%
)
19
88
15
(1
5
%
)
60
(6
0
%
)
0
(0
%
)
25
(2
5
%
)
0
(0
%
)
10
0
(1
00
%
)
19
89
22
(2
2.
22
%
)
49
(4
9.
49
%
)
0
(0
%
)
21
(2
1.
21
%
)
7
(7
.0
7
%
)
99
(1
00
%
)
19
90
23
(2
3
%
)
46
(4
6
%
)
0
(0
%
)
25
(2
5
%
)
6
(6
%
)
10
0
(1
00
%
)
19
91
22
(2
2
%
)
35
(3
5
%
)
5
(5
%
)
30
(3
0
%
)
8
(8
%
)
10
0
(1
00
%
)
19
92
21
(2
1
%
)
32
(3
2
%
)
8
(8
%
)
24
(2
4
%
)
15
(1
5
%
)
10
0
(1
00
%
)
19
93
29
(2
9
%
)
29
(2
9
%
)
8
(8
%
)
22
(2
2
%
)
12
(1
2
%
)
10
0
(1
00
%
)
19
94
26
(2
6
%
)
35
(3
5
%
)
5
(5
%
)
22
(2
2
%
)
12
(1
2
%
)
10
0
(1
00
%
)
19
95
18
(1
7.
82
%
)
29
(2
8.
71
%
)
6
(5
.9
4
%
)
27
(2
6.
73
%
)
21
(2
0.
79
%
)
10
1
(1
00
%
)
19
96
31
(3
0.
69
%
)
29
(2
8.
71
%
)
8
(7
.9
2
%
)
28
(2
7.
72
%
)
5
(4
.9
5
%
)
10
1
(1
00
%
)
19
97
27
(2
7
%
)
26
(2
6
%
)
12
(1
2
%
)
23
(2
3
%
)
12
(1
2
%
)
10
0
(1
00
%
)
19
98
26
(2
6
%
)
33
(3
3
%
)
1
(1
%
)
30
(3
0
%
)
10
(1
0
%
)
10
0
(1
00
%
)
19
99
28
(2
8
%
)
32
(3
2
%
)
1
(1
%
)
26
(2
6
%
)
13
(1
3
%
)
10
0
(1
00
%
)
20
00
26
(2
6
%
)
29
(2
9
%
)
7
(7
%
)
33
(3
3
%
)
5
(5
%
)
10
0
(1
00
%
)
20
01
26
(2
6
%
)
35
(3
5
%
)
4
(4
%
)
24
(2
4
%
)
11
(1
1
%
)
10
0
(1
00
%
)
20
02
32
(3
2
%
)
26
(2
6
%
)
11
(1
1
%
)
23
(2
3
%
)
8
(8
%
)
10
0
(1
00
%
)
20
03
31
(3
1
%
)
40
(4
0
%
)
6
(6
%
)
12
(1
2
%
)
11
(1
1
%
)
10
0
(1
00
%
)
20
04
25
(2
5
%
)
28
(2
8
%
)
16
(1
6
%
)
21
(2
1
%
)
10
(1
0
%
)
10
0
(1
00
%
)
T
ot
al
72
8
(2
5.
98
%
)
1,
02
7
(3
6.
65
%
)
17
6
(6
.2
8
%
)
62
2
(2
2.
2
%
)
24
9
(8
.8
9
%
)
2,
80
2
(1
00
%
)
802 R. Fontana et al.
In our first model, we analyze the factors affecting the probability of
observing a breakthrough invention in each OST5 sector by considering both
industry-level technological regimes and invention specific characteristics.
Even though in our setting this probability does not obviously reflect directly
the specific choice made by an individual amongst a fixed set of alternatives
maximizing a latent utility function, we can assume that the observed distribu-
tion of prizes across sectors (as resulting by the yearly decision of the awarding
board) would mimic quite closely how ‘nature’ chooses in which sectors a
breakthrough invention is more likely to occur.
We, therefore, rely on the estimation of a Conditional Multinomial Logit
(CML) model with both alternative-varying and individual-varying covariates.
In this setting, the probability of observing a breakthrough invention i in a
given macro-sector j is defined as:
prij =
exp
(
Xijβ + Ziγi
)
m∑
i=1
exp
(
Xijβ + Ziγi
)
(1)
where Xij are a set of alternative-specific and Zi are a set of case-specific
covariates, respectively. Table 6 reports the estimated coefficients for the
model.
The marginal effects for individual-specific covariates are computed as
follows:
∂prij
∂ Zi
= prij
(
γ j − γ¯t
)
(2)
where γ¯t is a probability weighted average of the estimated coefficients. The
marginal effect for a given alternative-specific covariate xrik (i.e. the value of
the covariate xr for individual i and alternative k) is computed as:
∂prij
∂xrik
=
{
prij
(
1 − prij
)
βr for j = k
−prij prikβr for j = k
}
. (3)
Thus the own-marginal effect (for j = k) has the same sign of the estimated
coefficient, whereas the cross-marginal effect (for j = k) has the opposite sign.
In Table 7 below, we report only individual-specific and own alternative-
specific marginal effects. For each alternative, they are computed at the
average value of each covariate.
Collaboration (i.e. having a multiple applicant) (MAPPL) decreases the
probability of observing a breakthrough invention in the sector of Instruments
(−0.073), whereas it increases the probability of observing a breakthrough
invention in the sector of Mechanical Engineering (+0.087). Breakthrough
inventions with at least one U.S applicant organization are more likely to occur
in the Chemistry & Pharmaceuticals and Process Engineering sectors, whereas
they are less likely to occur in the Electrical Engineering sector. The presence
of at least one governmental applicant decreases the probability of observing a
Schumpeterian patterns of innovation and the sources of breakthrough inventions 803
Table 6 Conditional Multinomial Logit regressions
Variables (1) (2) (3) (4) (5)
All Instruments Chemistry Process Mechanical
sectors pharma eng. eng.
MAPPL −0.237* −0.349 0.0453 0.605***
(0.130) (0.242) (0.138) (0.174)
NINV 0.0682 0.185** 0.132** 0.0366
(0.0583) (0.0910) (0.0627) (0.0829)
USA 0.380*** 0.918*** 0.848*** 0.254
(0.145) (0.294) (0.186) (0.215)
GOV −0.124 −0.567*** −0.0606 −0.388**
(0.113) (0.212) (0.124) (0.172)
ACAD 0.486** 0.319 −0.455* −0.846**
(0.201) (0.333) (0.247) (0.355)
dum1986_1995 −0.0995 −0.595** 0.196 −0.114
(0.152) (0.246) (0.164) (0.226)
dum1996_2005 −0.416*** 0.340 −0.0359 −0.348
(0.146) (0.311) (0.181) (0.260)
PAT_GROWTH 0.603
(0.509)
SCHUMP −0.481***
(0.178)
HERFSOURCES_TECH −0.676
(1.084)
HERFSOURCES_FIRM −1.106***
(0.325)
SELFSOURCES 7.326***
(2.311)
Constant −0.508** −3.060*** −2.146*** −1.919***
(0.234) (0.481) (0.435) (0.433)
Observations 14010 14010 14010 14010 14010
Robust standard errors in parentheses *** p < 0.01, ** p < 0.05, * p < 0.1
breakthrough in the Chemistry & Pharmaceuticals and Mechanical Engineer-
ing sectors, whereas it increases the probability of observing an invention in
the Electrical Engineering sector. Finally, a breakthrough invention with at
least one academic applicant is less likely to occur in the Process Engineering
and Mechanical Engineering sectors, whereas it is more likely to occur in the
Instruments sector.
Turning our attention to the impact of alternative-specific covariates,
SCHUMP, which is our main variable of interest, has a negative and significant
marginal effect. This result suggests that breakthrough inventions are more
likely to occur in sectors characterized by a Schumpeter Mark I type of
innovation patterns than in Schumpeter Mark II. This result appears both in
Tables 6 and 7. This finding is of particular interest also because it is likely
that our measure of breakthrough invention will probably be biased towards
inventions emerging from the corporate R&D segment of the economy.
Interestingly enough, concerning the variety of knowledge source across
firms indicator (HERFSOURCES_FIRM), we find that the more the amount
of relevant knowledge in a sector is concentrated across firms, the lower is
804 R. Fontana et al.
Table 7 Conditional Multinomial Logit regressions—marginal effects
Variables (1) (2) (3) (4) (5)
Electrical Instruments Chemistry Process Mechanical
eng. pharma eng. eng.
Pr(OST5 = j | 1 selected) 0.264 0.372 0.056 0.221 0.087
MAPPL 0.008 −0.073*** −0.017 0.017 0.065***
(0.021) (0.023) (0.010) (0.020) (0.016)
NINV −0.018* 0.001 0.007 0.014 −0.003
(0.010) (0.011) (0.004) (0.009) (0.006)
USA −0.110*** 0.006 0.025** 0.089*** −0.010
(0.029) (0.029) (0.010) (0.022) (0.017)
GOV 0.033* −0.001 −0.023** 0.014 −0.022*
(0.020) (0.022) (0.009) (0.018) (0.011)
ACAD −0.026 0.167*** 0.013 −0.098*** −0.057***
(0.034) (0.039) (0.019) (0.024) (0.012)
dum1986_1995 0.009 −0.025 −0.029*** 0.052** −0.007
(0.026) (0.027) (0.010) (0.022) (0.017)
dum1996_2005 0.045 −0.089*** 0.031* 0.029 −0.015
(0.028) (0.027) (0.018) (0.026) (0.017)
PAT_GROWTH 0.117 0.141 0.032 0.104 0.048
(0.099) (0.119) (0.027) (0.088) (0.041)
SCHUMP −0.093*** −0.112*** −0.025*** −0.083*** −0.038***
(0.035) (0.042) (0.010) (0.031) (0.014)
HERFSOURCES_TECH −0.131 −0.158 −0.036 −0.116 −0.054
(0.210) (0.253) (0.057) (0.533) (0.086)
HERFSOURCES_FIRM −0.215*** −0.258*** −0.059*** −0.190*** −0.088***
(0.063) (0.076) (0.001) (0.056) (0.026)
SELFSOURCES 1.423*** 1.712*** 0.388*** 1.261*** 0.582***
(0.449) (0.541) (0.125) (0.399) (0.187)
Observations 14010 14010 14010 14010 14010
Robust standard errors in parentheses *** p < 0.01, ** p < 0.05, * p < 0.1
the probability of observing a breakthrough invention in that sector. At the
same time, however, the probability of observing a breakthrough increases
with the degree of knowledge ‘cumulativeness’ in a given sector as captured by
the relative degree at which each firm exploits its internal source of knowledge
(SELFSOURCES).
4.3 Robustness checks and sensitivity analysis
The CML model estimated above relies on the Independence of Irrelevant
Alternatives (IIA) assumption which states that the relative odds between two
alternatives considered (e.g. the probability of awarding an invention in the
Instruments vs. Electrical Engineering macro-sectors) is not affected by adding
another alternative (e.g. by adding another macro-sector not considered in
our analysis) or by changing the characteristics of a third alternative (e.g. by
splitting in two the Chemistry & Pharmaceutical macro-sectors). Although
this assumption seems plausible in our setting, since we have classified ex-
post the awarded inventions in the OST sectors with respect to the decision
Schumpeterian patterns of innovation and the sources of breakthrough inventions 805
of the awarding board,10 we report in this sub-section (as a robustness check
exercise) the estimates of an alternative econometric model which relaxes the
IIA assumption. The Alternative-Specific Multinomial Probit (ASMNP) re-
gression model (Drukker and Gates 2006) assumes a multinomial distribution
for the error terms εij in each j-alternative latent variable equation pr∗ij with a
user-specified correlation structure :
pr∗ij = Xijβ + Ziγi + εij and
ε′j = (εi1, ..., εiJ) ∼ MVN (0,) , for j = 1, . . . , J and i = 1, . . . , N. (4)
The simulated maximum likelihood estimator for the ASMNP is computed us-
ing the command asmprobit on STATA 11—SE version which implements the
GHK algorithm (Geweke 1989; Hajivassiliou and McFadden 1998; Keane and
Wolpin 1994) to approximate the multivariate distribution function. Tables 8
and 9 report respectively the estimated coefficients and marginal effects of the
ASMNP model.11 In most of the cases, the sign, the statistical significance and
the magnitude of the estimates are similar to the CML estimates.
Moreover, for those sectors in which the alternative-specific regressors have
the most significant estimated impact (Instruments, Chemistry & Pharmaceu-
ticals, and Mechanical Engineering), Fig. 7 shows the degree of sensitivity of
the marginal effects with respect to different levels of the alternative specific
regressors considered in different sectors.
Interestingly enough, the estimated impact of the Schumpeterian regime
indicator (SCHUMP), although being always negative, shows a different be-
havior with respect to the sector considered. In the sector Instruments, the
estimated negative marginal effect tends to become stronger the more the
Schumpeterian regime gets closer to a Mark II type, whereas in Mechanical
Engineering, the negative impact tends to become weaker. For Chemistry
& Pharmaceuticals, although on average the estimated marginal effect of
SCHUMP is negative, we observe a U-shaped pattern with a rate of change
in the simulated probability of getting an invention awarded which decreases
(i.e. the estimated negative impact becomes stronger) when moving from an
highly ‘turbulent’ Schumpeterian Mark I type to an ‘intermediate’ type, and
then increases when moving from an ‘intermediate’ type to an highly ‘stable’
Mark II type regime.
A similar non-monotonic pattern is found when considering the effect of
HERFSOURCES_FIRM in the Instruments sector. The rate of change in the
simulated probability of observing a breakthrough invention in this sector de-
creases when moving from a low concentrated (in terms of relevant knowledge
10As we already mentioned, the R&D 100 awarding board was not faced with a real choice
amongst macro-sectors alternatives when deciding which invention deserved the prize (i.e. there
were no ’pre-determined‘ shares of awards reserved for each sector).
11The marginal effects are computed considering the mean value for continuous variables and a
discrete change 0–1 for binary variables.
806 R. Fontana et al.
Table 8 Alternative Specific Multinomial Probit regression
Variables (1) (2) (3) (4) (5)
All Instruments Chemistry Process Mechanical
sectors pharma eng. eng.
MAPPL −0.134*** −0.142*** 0.871 −0.0289
(0.0375) (0.0368) (1.231) (0.0271)
NINV 0.0820*** 0.0878*** 0.980** 0.0590***
(0.0183) (0.0174) (0.406) (0.0141)
USA 0.431*** 0.468*** 5.941 0.329***
(0.0409) (0.0395) (3.730) (0.0289)
GOV −0.148*** −0.161*** 0.721 −0.131***
(0.0385) (0.0362) (1.004) (0.0295)
ACAD 0.211 0.206 −6.023** 0.0583
(0.157) (0.162) (2.661) (0.127)
dum1986_1995 0.0403 0.0669 3.687*** 0.0177
(0.0443) (0.0422) (1.152) (0.0351)
dum1996_2005 −0.133*** −0.0753* 3.857*** −0.145***
(0.0435) (0.0416) (1.177) (0.0336)
PAT_GROWTH 0.0905***
(0.0231)
SCHUMP −0.0971***
(0.00389)
HERFSOURCES_TECH −0.151***
(0.0177)
HERFSOURCES_FIRM −0.249***
(0.00691)
SELFSOURCES 0.712***
(0.0962)
Constant 0.162*** 0.0459 −2.03*** 0.171***
(0.0612) (0.0656) (0.1109) (0.0470)
Observations 14010 14010 14010 14010 14010
Robust standard errors in parentheses *** p < 0.01, ** p < 0.05, * p < 0.1
Table 9 Alternative Specific Multinomial Probit regression—marginal effects (alternative specific
regressors)
Variables (1) (2) (3) (4) (5)
Electrical Instruments Chemistry Process Mechanical
eng. pharma eng. eng.
Pr(OST5 = j | 1 selected) 0. 258 0. 378 0.055 0.218 0.086
PAT_GROWTH 0.0015 0.192** 0.123*** 0.0005 0.076*
(0.001) (0.080) (0.046) (0.001) (0.044)
SCHUMP −0.0014* −0.190*** −0.121*** −0.0005 −0.075**
(0.0008) (0.041) (0.027) (0.001) (0.030)
HERFSOURCES_TECH −0.003 −0.355* −0.227* −0.001 −0.140
(0.002) (0.209) (0.091) (0.005) (0.097)
HERFSOURCES_FIRM −0.004** −0.472*** −0.301*** −0.001 −0.186***
(0.002) (0.078) (0.075) (0.005) (0.026)
SELFSOURCES 0.011* 1.473*** 0.939*** 0.004 0.581**
(0.006) (0.360) (0.215) (0.009) (0.258)
Observations 14010 14010 14010 14010 14010
Robust standard errors in parentheses *** p < 0.01, ** p < 0.05, * p < 0.1
Schumpeterian patterns of innovation and the sources of breakthrough inventions 807
Fig. 7 Estimated marginal effects (red line) for different values of the covariates in different
sectors (95 % confidence interval is the grey area)
owned by firms) to an ‘average’ concentrated scenario, and then increases
when moving to an highly concentrated one. In the other two sectors con-
sidered (Chemistry & Pharmaceuticals and Mechanical Engineering), the
estimated negative marginal effects monotonically decreases with the de-
gree of concentration. Finally, concerning the estimated positive impact of
the relevance of the internal sources of knowledge (SELFSOURCES), we
can see that its intensity tends to decrease with the degree of knowledge
808 R. Fontana et al.
‘cumulativeness’ in the Instruments sector whereas the pattern is inverted-
U-shaped for the Chemistry & Pharmaceuticals sector and constant for the
Mechanical Engineering sector.
5 Concluding remarks
Economists of innovation have been aware for a long time that patterns
of innovative activities differ across industries. So far, most research efforts
have been devoted to the construction of taxonomies that could be fruitfully
employed to interpret the variety of sectoral innovation patterns. In this
respect, the Schumpeter Mark I/ Schumpeter Mark II distinction has been,
together with the Pavitt (1984) taxonomy, the interpretative approach that has
gained the widest currency. In fact, the characterization of sectoral patterns of
innovation in terms of the Schumpeter Mark I/ Schumpeter Mark II distinction
has consistently emerged in different countries using different type of data to
measure innovative activities (e.g., USPTO patents, EPO patents and national
Innovation Surveys responses).
In this paper, we have expanded on this line of research by examining the
relationship between different sectoral patterns of innovation (characterized in
terms of technological regimes and Schumpeter Mark I/ Schumpeter Mark II
patterns) and the generation of breakthrough inventions. To address this issue,
we have used two different sources of data. We have used USPTO patents to
capture the relevant dimensions of the technological regime prevailing in each
sector and to construct an indicator of the degree in which each sector can be
identified as either a Schumpeter Mark I or Schumpeter Mark II. We have used
a new data set of awarded inventions to measure the number of breakthrough
inventions generated by each sectors. Our findings indicate that, in general,
a Schumpeter Mark I ‘turbulent’ environment rather than a more ‘stable’
Schumpeter Mark II is conducive to a higher probability of the occurrence
of breakthrough inventions.
Though preliminary and in need of further corroboration, we think that our
results bear some important implications for the existing literature on innova-
tion. First, they extend the analysis of the relationship between Schumpeterian
pattern of innovation and economic performance to the case of breakthrough
inventions. In this respect, our findings appear somewhat consistent with those
of Castellacci (2007) on the relationship between productivity growth and
sectoral patterns of innovation. Also in that case he found that the relationship
between productivity growth and the dimensions of the technological regime
was articulated in a different way in Schumpeter Mark I and Schumpeter
Mark II patterns. Second, our results complement the evidence provided by re-
cent studies in the management tradition that look at the sources of innovative
breakthroughs mainly at the individual level. While the probability of achiev-
ing a breakthrough may be related to inventors’ past experience and ability
(Conti et al. 2010) and/or to the organizational setting in which the research
activity takes place (Jeppesen and Lakhani 2010), industry characteristics seem
Schumpeterian patterns of innovation and the sources of breakthrough inventions 809
also to play an important role and ought to be considered when carrying out
firm level studies. Finally, our results bear also some policy implications. If
an entrepreneurial regime is an environment relatively more conducive to
breakthrough inventions, then it is clear that intelligent innovation policies
would better follow the advice of Jewkes et al. (1958) and pay attention to
the role of small firms and/or individual entrepreneurs rather than focusing
exclusively inside the walls of the research and development facilities of large
corporations.
Acknowledgements We thank the editors of the special issue and an anonymous reviewer
for helpful suggestions. We have also benefited from comments of seminar participants at the
International Schumpeter Society Conference 2010, Workshop on ‘Knowledge Integration and
Innovation’, EPIP conference 2009, EMAEE Conference 2009. Fontana and Nuvolari acknowl-
edge financial support from European Union, project FP6 – 043345, NEST-2005-Path-CUL:
“CID-Cultural Innovation Dynamics. Explaining the uneven evolution of human knowledge”.
References
Ahuia G, Lampert C (2001) Entrepreneurship in the large corporation: a longitudinal study of
how established firms create breakthrough inventions. Strateg Manag J 22:521–543
Block F, Keller M (2009) Where do innovations come from? Transformations in the US economy.
Socio-Econ Rev 7:459–483
Breschi S, Malerba F, Orsenigo L (2000) Technological regimes and schumpeterian patterns of
innovation. Econ J 110:388–410
Carpenter MP, Narin F, Woolf P (1981) Citation rates to technologically important patents. World
Pat Inf 3:160–163
Castellacci F (2007) Technological regimes and sectoral differences in productivity growth. Ind
Corp Change 16:1105–1145
Chandy R, Tellis G (2000) The incumbent’s curse? Incumbency, size and radical product innova-
tion. J Market 64:1–17
Cohen WM (2010) Fifty years of empirical studies of innovative activity and performance. In:
Hall B, Rosenberg N (eds) The handbook of economics of innovation. Elsevier, Amsterdam,
pp 129–213
Conti R, Gambardella A, Mariani M (2010) Learning to be Edison? Individual inventive experi-
ence and breakthrough inventions. Paper presented at DRUID Academy conference 2010
Corrocher N, Malerba F, Montobbio F (2007) Schumpeterian patterns of innovation in the ICT
field. Res Policy 36:418–432
Drukker DM, Gates R (2006) Generating Halton sequences using Mata. Stata J 6:278–294
Freeman C (1991) Networks of innovators. A synthesis of research issues. Res Policy 20:499–514
Geweke J (1989) Bayesian inference in econometric models using Monte Carlo integration.
Econometrica 57:1317–1339
Granstrand O, Alange S (1995) The evolution of corporate entrepreneurship in Swedish
industry—was Schumpeter wrong? J Evol Econ 5:133–156
Hajivassiliou VA, McFadden DL (1998) The method of simulated scores for the estimation of
LDV models. Econometrica 66:863–896
Hall B, Jaffe A, Trajtenberg M (2001) The NBER patent citations data file: lessons, insights,
methodological tools. NBER Working Paper n. 849w8
Hinze S, Reiss T, Schmoch U (1997) Statistical analysis on the distance between fields of technol-
ogy. Report for the European Commission, TSER project
Jeppesen LB, Lakhani KR (2010) Marginality and problem solving effectiveness in broadcast
search. Org Sci 21(5):1016–1033
Jewkes J, Sawers D, Stillerman R (1958) The sources of invention. MacMillan, London (rev. edn.
1969)
810 R. Fontana et al.
Keane MP, Wolpin KI (1994) The solution and estimation of discrete choice dynamic program-
ming models by simulation and intermpolation: Monte Carlo evidence. Rev Econ Stat 76:648–
672
Khan Z, Sokoloff K (1993) “Schemes of practical utility”: entrepreneurship and innovation among
“great inventors” in the United States, 1790–1865. J Econ Hist 53:289–307
Malerba F (2005) Sectoral systems: how and why innovation differs across sectors. In: Fagerberg J,
Mowery DC, Nelson RR (eds) The Oxford handbook of innovation. Oxford University Press,
Oxford, pp 380–406
Malerba F, Orsenigo L (1995) Schumpeterian patterns of innovation. Camb J Econ 19:47–65
Malerba F, Orsenigo L (1996) Schumpeterian patterns of innovation are technology-specific. Res
Policy 25:451–478
Malerba F, Orsenigo L (1997) Technological regimes and sectoral patterns of innovative activities.
Ind Corp Change 6:83–117
Marsili O, Verspagen B (2002) Technology and the dynamics of industrial structures: an empirical
mapping of Dutch manufacturing. Ind Corp Change 11:791–815
Mokyr J (1990) The lever of riches. Oxford University Press, Oxford
Mowery DC, Nelson RR (eds) (1999) Sources of industrial leadership. Cambridge University
Press, Cambridge
Nelson RR, Winter S (1982) An evolutionary theory of economic change. Harvard University
Press, Cambridge
Pavitt K (1984) Patterns of technical change: towards a taxonomy and a theory. Res Policy 13:343–
373
Scherer FM (1989) “Comments” on Z. Griliches, “Patents: recent trends and puzzles”. Brook Pap
Econ Act 9:291–330
Scherer FM, Harhoff D (2000) Technology policy for a world of skewed distributed outcomes. Res
Policy 29:559–566
Schoenmakers W, Duysters G (2010) The technological origins of radical inventions. Res Policy
39:1051–1059
Silverberg G, Verspagen B (2007) The size distribution of innovations revisited: an application of
extreme value statistics to citation and value measure of patent significance. J Econ 139:318–
319
Singh J, Fleming L (2010) Lone inventors as sources of breakthroughs: myth or reality? Manag Sci
56:41–56
Van Dijk M (2000) Technological regimes and industrial dynamics: the evidence from Dutch
manufacturing. Ind Corp Change 9:173–194