is this novel technology going to be a hit? antecedents ...mouse” created in 1996 by seo jeon sun,...
TRANSCRIPT
Is ThIs Novel TechNology goINg To be a hIT? aNTecedeNTs PredIcTINg TechNologIcal NovelTy dIffusIoN
Documents de travail GREDEG GREDEG Working Papers Series
Michele PezzoniReinhilde VeugelersFabiana Visentin
GREDEG WP No. 2018-22https://ideas.repec.org/s/gre/wpaper.html
Les opinions exprimées dans la série des Documents de travail GREDEG sont celles des auteurs et ne reflèlent pas nécessairement celles de l’institution. Les documents n’ont pas été soumis à un rapport formel et sont donc inclus dans cette série pour obtenir des commentaires et encourager la discussion. Les droits sur les documents appartiennent aux auteurs.
The views expressed in the GREDEG Working Paper Series are those of the author(s) and do not necessarily reflect those of the institution. The Working Papers have not undergone formal review and approval. Such papers are included in this series to elicit feedback and to encourage debate. Copyright belongs to the author(s).
1
Is This Novel Technology Going to be a Hit?
Antecedents Predicting Technological Novelty Diffusion
Michele Pezzoni1, Reinhilde Veugelers2, Fabiana Visentin3
GREDEG Working Paper No. 2018-22
Abstract: Despite the high interest of scholars in identifying breakthrough inventions, little
attention has been devoted to investigating how the technological content of those breakthrough
inventions is re-used over time. We overcome this limitation by focusing on the dynamics of
diffusion of the novel technologies incorporated in inventions. Specifically, we consider the factors
affecting the time needed for a technology to be legitimated as well as its technological potential.
We find that the dynamics of diffusion of a novel technology are affected by the characteristics of
its building blocks, i.e. the technological components that combined together for the first time
generate a novel technology. Combining similar technological components, components familiar
to the inventors’ community, and components with a high level of appropriability generates a
technology that requires a short time to be legitimated but with a low technological potential.
Combining technological components with a science-based nature generates technologies with a
longer legitimation time but also higher technological potential. Finally, when large firms are the
main innovative actors, novel technologies show a higher technological potential.
JEL code: O33
Keywords: Technological novelty, diffusion, combinatorial process, initial characteristics, patent
data
1 Université Côte d’Azur, CNRS, GREDEG, France; ICRIOS, Bocconi University, Milan, Italy; BRICK, Collegio
Carlo Alberto, Torino, Italy; email: [email protected] 2 KULeuven; email: [email protected] 3 School of Business and Economics Maastricht University and UNU-MERIT, Maastricht, Netherlands
2
1. Introduction
While there is a rising interest in identifying individual breakthrough inventions, little work has
investigated how the novel technologies embodied in these inventions diffuse generating follow-
on inventions (Arts et al., 2013). We consider novel technologies as the result of a combination of
existing components (Schumpeter 1939; Nelson & Winter 1982; Strumsky and Lobo 2015) where
each component represents by itself a technology (Arthur 2009). We analyze the diffusion curve
of a novel technology by looking at all the patented inventions that build-up on that technology.
Looking at the diffusion curve of a novel technology, we are particularly interested in identifying
the time that a novel technology needs to be legitimated within the inventors’ community and its
technological potential defined as the number of follow-on inventions that the novel technology
can generate. In our analysis, we investigate how the characteristics of the combined components
influence both novel technology legitimation time and technological potential. By investigating
how the characteristics of the combined components impact on legitimation and technological
potential, we aim to reply to a series of what-if questions such as, what would have been the
diffusion of a novel technology if the combined components were more established (or less) within
the inventors’ community? What would have been the diffusion of a novel technology if the two
components were more science-based? Or more similar to each other? More generally, we ask,
what are the antecedents predicting a “hit” technology?
The discussion of a renowned invention and of the diffusion of its embodied novel technology
helps to illustrate the main idea of our paper. The “onco-mouse’’, a mouse largely used in
laboratory cancer research, is an example of a well-recognized breakthrough invention embodying
a novel technology. In the mid-80s, Philip Lader and Timothy Steward at the Harvard laboratories
had the revolutionary idea of isolating cancer-related genes and injecting them into a mouse eggs
3
transplanted into a female mouse generating the first transgenic mammal likely to develop a
specific disease (Murray, 2010). The onco-mouse was both patented and published. The novel
transgenic mammal technology to mimic diseases was embodied in the onco-mouse invention and
was the result of an unprecedented combination of two existing technological components: “Gene
isolation” and “Injection of material into animals”. The novel technology generated a variety of
follow-on inventions represented by other transgenic mammals used in labs to experiment
treatments for a variety of diseases. Geneticists patented at least 110 transgenic mammals designed
to develop diseases from Alzheimer to cystic fibrosis (Murray, 2010). For instance, the “diabetic
mouse” created in 1996 by Seo Jeon Sun, a professor from Seoul National University is a follow-
on patented invention building up on the novel transgenic mammal technology. We look at the
patents’ technological content and we reconstruct the diffusion curve of the transgenic mammal
technology to mimic diseases, by tracing the reuse of the novel combination “Gene isolation” and
“Injection of material into animals” in follow-on patents4.
In this work we assume that the diffusion of the novel technology in the follow-up inventions
depends from the characteristics of the two technological components combined to generate the
novel technology. The two combined components at the base of the transgenic mammal
technology have each its own characteristics and so has their resulting combination. The “Gene
isolation” component was rather new within the technological community, as well as the “Injection
of material into animals” component at the time when they were recombined for the first time in
the onco-mouse. Furthermore, the two components belong to two different technological domains
4 Interestingly, we observed that patent citations do not connect all the patents belonging to the diffusion curve of the
transgenic mammal technology. Indeed, citations refer to the specific prior art used and not necessarily to the
technological content of an invention. For example, the onco-mouse patent does not appear in the backward citation
list of the diabetic-mouse.
4
and they were both strongly science-based. Finally, the two components were characterized by a
low concentration of the actors conducting the innovative activities.
In this paper, we conduct an empirical study that traces the diffusion of 11,009 novel technologies
over the period 1985-2015. We consider the existing technological components as the
technological classes used in the European Patent Office classification, and novel technologies as
an unprecedented combination of technological classes in the history of patented inventions
(Strumsky and Lobo, 2015).
The contribution of our study is twofold. First, we provide insights to the economics of innovation
literature that is interested in identifying the key drivers of the evolution of a “technological
trajectory” (Dosi, 1985)5. Second, we provide managerial implications that support the decisional
process of practitioners asked to predict the success of a novel technology.
The remaining of the paper is organized as follow. Section 2 illustrates our novel technology
definition and develops the hypothesis on how the characteristics of the technological components
recombined affect the diffusion of the resulting technology. Section 3 illustrates the method used
in the analysis. Section 4 presents the results that are discussed in Section 5. Section 6 concludes.
2. Novel technology diffusion and antecedents
5 A neo-Schumpeterian scholar would define a novel technology as a new technological paradigm and the diffusion
of a novel technology as a technological trajectory. Over time, radical inventions introduce new technological
paradigms. Technology advancement is cumulative and new paradigms open new venues for further technological
developments shaping new technological trajectories. Technological trajectories can be traced over time by following
the patent paper trail related to the technological paradigm (Achilladelis, 1993).
5
While a large number of studies has investigated the diffusion of an innovation looking at its use
by the relevant population of potential adopters (Hall, 2004), a limited number of studies has
investigated the diffusion path of a novel technology in the follow-on pool of inventions. Extant
works rely on case studies developed for specific technologies. For instance, Achilladelis (1993)
follows the diffusion of the novel technology related to Sulpholamide antibacterial drugs by tracing
all the patented inventions based on Sulpholamide appeared over 70 years. Differently from the
extant literature, our study adopts a systematic quantitative approach that allows us (i) to identify
the introduction of a novel technology, (ii) to trace its diffusion, and (iii) to assess the antecedents
impacting on its diffusion.
Identifying a novel technology
The first step in our analysis is to define a novel technology. In his seminal work on the origins of
innovation, Schumpeter claimed that “innovation combines components in a new way, or that it
consists in carrying out New combination” (1939, pp. 88, Schumpeter 1939). Fleming (2001)
elaborated on the concept of invention as a process of “recombinant search” (pp. 118) by arguing
that inventors search among existing technological components and recombine them to realize
something new. In the same vein, Arthur (2009) states that novel technologies are the result of the
combination of existing components and that these existing components are themselves
technologies. According to this “recombinant search” approach, we consider a novel technology
as the result of pre-existing technological components that are combined for the first time
(Verhoeven et al., 2016).
Not all the novel technologies diffuse. The aim of our study is to analyze the diffusion of successful
technologies. To do that, we consider those novel technologies that recombine existing
technological components and reach a minimum level of diffusion. Therefore, following Amabile
6
(1996) and Fleming and co-authors (2007), we require to the result of creative efforts to be both
novel and useful. Novel technologies that do not show a minimum number of industrial
applications are considered not useful and are excluded from our analysis.
Tracing novel technology diffusion
In our study we assume that the diffusion of a novel technology is characterized by two distinct
phases: a legitimation phase and a phase of convergence towards the technological potential.
Initially a technology diffuses slowly since it needs time to gain legitimation within the inventors’
community. Inventors using the novel technology in follow-on inventions need time to learn and
familiarize with the novel technology and to abandon the established competing technologies.
Once legitimated, the diffusion of the novel technology accelerates since it becomes the dominant
technology. The second phase of asymptotical convergence to a ceiling level is determined by the
potential of a technology. Two different trends can lead a technology to reach its ceiling. First, all
the possible applications of the technology have been implemented exhausting the inventive
opportunities. Second, the technology loses its appeal in favor of emerging alternative
technologies.
An insightful example of technology experiencing a first phase of legitimation followed by a
second phase of convergence to its technological potential, is represented by the Sulpholamide
drugs. The Sulpholamide is a class of synthetic antibacterial drugs introduced in 1935. The first
Sulpholamide-based drug, Prontosil, was developed and launched by Bayer. In the next 10 years
following the Prontosil introduction, other companies and public research laboratories gradually
developed and marketed more than 5,000 innovative Sulpholamide-based drugs. The success of
Sulpholamide-based drugs was exhausted by the mid- ‘50s for two reasons. First, additional
improvements of the therapeutic benefits of this class of drugs become more difficult reducing the
7
opportunities to generate other inventions. Second, new antibiotics substituting the Sulpholamide-
based drugs appeared (for details on the case, see Achilladelis, 1993).
In reconstructing the two phases of the diffusion of a novel technology, we assume that a diffusion
process follows a sigmoid-curve (Griliches, 1957). The values of the length of the legitimation
period combined with the level of technological potential define the shape of the S-curve of the
corresponding novel technology. Figure 1 compares the diffusion curves of a technology having a
long legitimation phase and a high technological potential with another technology having a short
legitimation phase but a low technological potential.
8
Figure 1: Comparison between diffusion curves
Technological, cognitive and institutional antecedents
There is general consensus among innovation scholars that novel technologies spur from an
unprecedented combination of existing components (Fleming, 2001). Recent studies look at the
antecedents of the probability of observing those new combinations (Curran, 2013; Caviggioli,
2016). Nevertheless, none of the previous studies consider the antecedents as predictors of
diffusion of the novel technology. In this paper we assume that the characteristics of the combined
components affect the resulting novel technology characteristics and therefore its diffusion
(Fleming, 2001). We identify technological, cognitive and institutional characteristics of the
combined components as relevant for the diffusion of the novel technology. Technological
characteristics are those related exclusively to the technological aspects of the combined
components. Cognitive characteristics are those related to the understanding of the combined
components by the inventors’ community. Finally, institutional characteristics are those related to
the level of appropriability of the combined components and to the characteristics of the industrial
sector.
9
As technological characteristics we consider both the technological similarity and science-based
nature of the combined components. Novel technologies can result from the combination of
components that differ substantially or that are similar between each other. When combining
similar components, a large number of inventors might have already the competences to use the
resulting technology. On the contrary, combining dissimilar components requires cross-field
competences and inventors need to team-up to be able to use the novel technology or to expand
their competences. Since teaming-up and acquiring new competences are time consuming
activities, we expect that a novel technology resulting from the combination of similar components
is legitimated earlier than a novel technology resulting from the combination of dissimilar
components. At the same time, we expect that the combination of components that are based on
similar technological principles generates a novel technology that only marginally differs from the
existing components, harming its technological potential. On the contrary, combining dissimilar
components generates a novel technology with a greater original content (Verhoeven et al., 2016),
enhancing its technological potential.
We expect that:
Hypothesis 1 (combining similar components): A novel technology resulting from the
combination of two similar components (a) requires a shorter time to be legitimated and (b)
has a lower technological potential.
Novel technologies can result from the combination of science-based or applied components.
Science-based components are those components where the knowledge used has abstract content,
while applied components are those components where specialized knowledge is used for specific
applications (Breschi et al., 2000). For science-based components, the scientific knowledge
advancements need time to be incorporated in the population of potential inventors, requiring the
10
education and training of cohorts of inventors employed in industrial R&D (Klevorick et al., 1995).
On the contrary, for applied components the inventors’ activities rely on stable set of knowledge
that is immediately available and does not need scientific updates. For these reasons, we expected
that a novel technology resulting from the combination of science-based components takes more
time to be legitimated in the inventors’ community. At the same time, the generality of the
scientific knowledge used in science-based components is expected to open up a broader set of
technological opportunities (Klevorick et al., 1995). On the contrary, novel technologies relying
on applied components are expected to have a limited set of possible technological applications.
For these reasons, we expect that a novel technology resulting from the combination of science-
based components has a higher technological potential while the combination of applied
components has lower technological potential.
Hypothesis 2 (combining science-based components): A novel technology resulting from the
combination of science-based components (a) requires a longer time to be legitimated and
(b) has a higher technological potential.
Cognitive characteristics are relevant since components are recombined by individuals who need
to receive and interpret information before being able to use it in a creative way (Cohen and
Levinthal, 1990). As cognitive characteristics, we consider the familiarity of the inventors’
community with the components. A novel technology can result from the combination of
components that have been frequently or rarely used within the inventors’ community. A
component frequently used is well-known and can be exploited, while a component rarely used is
unknown and has to be explored (March, 2001). Following Fleming’s argument (2001), we assume
that there is a greater chance that inventors build-up their inventions on components they are
familiar with. Exploiting familiar components reduces the risks for unexpected results and
11
facilitates the legitimation of the resulting novel technology. However, familiar components have
been already largely exploited in the past so the novel technology resulting from their combination
is expected to have a lower technological potential.
We formulate the following hypothesis:
Hypothesis 3 (combining familiar components): A novel technology resulting from the
combination of familiar components (a) requires a shorter time to be legitimated and (b) has
a lower technological potential.
As institutional aspects we consider appropriability, defined as the level of protection from
imitation guaranteed to inventions spurring from the components. The level of protection largely
depends on extant intellectual property legislation. A broad scope of protection provides an
incentive to incumbent inventors and firms to invest in R&D activities. Higher investments in
R&D are expected to shorten the legitimation period. At the same time, a broad scope of protection
prevents outsider inventors from benefiting from the latest technological advancements generated
by the incumbent inventors (Breschi et al., 2000; Klevorick et al., 1995). Specifically, when the
scope of the protection is board, outsiders are deterred to start new R&D activities that build-up
on incumbent inventors’ results since the risk of infringing incumbents’ intellectual property rights
is high (Merges and Nelson, 1994). As a result of high entry barriers for new incumbents, the
community of inventors remains small. Due to its small size the community of inventors is not
able to exploit all the potential of the novel technology leaving several inventive opportunities
unexplored. As a consequence, we expect that the combination of components with high
appropriability leads to a novel technology with lower technological potential.
12
Hypothesis 4 (combining components with high appropriability): A novel technology
resulting from the combination of components with high appropriability (a) requires less time
to be legitimated and (b) reaches lower technological potential.
Finally, we include as an additional institutional aspect the concentration of the innovative
activities of the firms characterizing the two combined technological components. Specifically,
for the two components, we consider if the innovative activity is conducted by a handful group of
large firms (the so-called Schumpeter Mark 1 model) or if the innovative activity is spread over a
large number of small firms (the so-called Schumpeter Mark 2 model) (Breschi et al. 2000). We
expect that when innovative activities are concentrated the legitimation time is reduced due to the
presence of an organized and formalized R&D activity within large firms. On the contrary, when
the innovative activity is spread over a large number of small firms, the R&D is less formalized
and conducted less systematically. The later approach to the R&D activity increases the
legitimation time. Moreover, concentrated environments are more stable and we expect firms
benefiting from the cumulativeness of their R&D efforts. R&D cumulativeness leads to a faster
legitimation of the novel technology. On the contrary, fragmented environments are unstable due
to a high firms’ turnover. Firms are not able to retain their R&D experience and this slows down
the legitimation process. Concerning the technological potential, we expect that large innovative
firms develop a greater number of industrial applications of the novel technologies for two main
reasons. First, large firms in a concentrated environment have established research labs and it is
easier for them to develop a stable flow of industrial applications of the novel technology (Malerba
et al., 1997). Second, large firms tend to patent for strategic reasons. They might aim to cover all
the industrial applications of their novel technology preventing the entry of other inventive firms.
13
In a concentrated environment, both the availability of large R&D labs and strategic patenting,
lead to higher technological potentials.
Hypothesis 5 (combining components with high concentration): A novel technology resulting
from the combination of components with high concentration (a) requires shorter time to be
legitimated and (b) reaches higher technological potential.
3. Methods
Our analysis relies on a sample including all the patents filed to the European Patent Office (EPO)
in the period 1985-2015. Following Fleming and his coauthors (2007), we consider the technology
classes in which a patent is classified (International Patent Classification -IPC- codes) as the
technological components on which the patent build-up. We mark the appearance of a novel
technology as the first time ever6 appearance of a combination of IPC codes in a patent. Limiting
our definition of novel technology to the first time ever appearance of a combination, we might
consider as novel technologies, pairs of IPC codes that appears for the first time but that are never
re-used after their appearance7. Following Amabile (1996) and Fleming and co-authors (2007), we
require to the novel technology to be useful. Therefore, we restrict our novel technology definition
only to those novel combinations re-used in at least 20 patents in the 20 years following the novel
technology appearance8. We end up with a study sample that includes 11,009 successful novel
6 To flag a technology as novel, we would need the complete history of its component to be able to evaluate when a
pair of component appear together for the first time. We use the first available data period at EPO, 1978-1984, as a
buffer period to capture the history of our components and we track novel combinations starting from 1985. 7 The generation of pairs of IPC codes that are not re-used might be also the also be the result of the mechanical merge
of all IPC codes included in all the EPO patent applications. 8 Note that, as suggested by Jaffe (2002), “such selection does not create any selectivity bias” (pp. 25). We are
estimating the effects of our main variables of interest for a specific group of novel technologies, successful novel
technologies.
14
technologies which generated 250,979 distinct follow-up patents9. In Appendix A we validate our
novel technology measure by applying a topic modelling algorithm to assess the coherence of the
patents included in the novel technology diffusion curves.
To represent the diffusion curves of each novel technology we proceed in two steps. First, we
count the yearly number of patents that embodied the novel technology (yt). Doing so, we construct
the actual patent cumulated distribution over a window-period of 20 years (𝑌𝑡 = ∑ 𝑦𝑝𝑡𝑝=1 )
following the appearance of the novel technology10. Second, we use the actual cumulated
distribution of each novel technology to fit the corresponding trend function. The trend function
represents an algebraic approximation of the diffusion curve of the technology. Following the
arguments detailed in Section 2, we opt for an S-curve (or sigmoid curve) characterized by a slow
initial growth and by an asymptotic convergence to a ceiling level (Geroski, 2000). The use of an
estimated diffusion curve, instead of actual data, allows us to measure the technological potential
and the legitimation period also for those technologies that do not exhaust their innovative
potential during the observation period of 20 years. In fact, technologies with particularly slow
legitimation might not show any asymptotic convergence to a ceiling level after the 20 years
covered by the observed data. One example of a technology with a particularly long diffusion
process is the light bulb that was introduced in 1909 and that continued to generate additional
patented inventions until 1955 (Abernathy and Utterback, 1978). Sixty-eight percent of the novel
technologies included in our sample reaches the estimated ceiling level11 after 20 years.
9 58% of the generated 250,979 patents contain one novel technology, 18% two novel technologies, 8% three novel
technologies, and 4% 4 novel technologies. 10 We consider all the novel technologies until the patent cohort of 1996 to leave a 20-year window forward to this
last cohort. 11 We consider a novel technology reaching the ceiling level if the difference between the actual cumulated number
of patents after 20 years and the estimated value of the ceiling is less than 10% of the ceiling value.
15
The technological diffusion S-curve is identified by three parameters, i.e, Midpoint, Alpha and
Ceiling and can be expressed by the following equation:
�̂�𝑡 =𝐶𝑒𝑖𝑙𝑖𝑛𝑔
1 + 𝑒(−
(𝑡−𝑀𝑖𝑑𝑝𝑜𝑖𝑛𝑡)𝐴𝑙𝑝ℎ𝑎
)
(Equation 1)
where �̂�𝑡 is the cumulated number of patents predicted at time t, t is the number of years elapsed
since the appearance of the novel technology, the parameter Ceiling is defined as the upper
asymptote of the S-curve, Midpoint is the required time to reach the fifty percent of the ceiling,
and Alpha is the inverse of the curve slope at the Midpoint. An increase of the value of Alpha leads
to flatter diffusion curves while a decrease of the value of Alpha leads to stepper diffusion curves.
We use the three parameters describing the S-curve as proxies of the concept of Legitimation and
Technological potential illustrated in Section 2. We assume that a technology is legitimated when
it is accepted by the community and we fix the “acceptance” point as the moment when the
technology reaches the ten percent of its ceiling (Griliches, 1957). Using the formula in Equation
1, legitimation is calculated as a linear combination of Midpoint and Alpha, Midpoint-ln(9)*Alpha
(See Appendix B for the details concerning the mathematical formulation) and is expressed in
number of years since the appearance of the novel technology until its legitimation (Griliches,
1957). The potential of a technology equals to ceiling and indicates the maximum number of
inventions that the novel technology is expected to generate.
In this paper, we conduct a set of regression exercises aiming to estimate the impact of
technological, cognitive and institutional characteristics of the components combined on the
16
legitimation and technological potential of a novel technology. Specifically, we estimate with an
Ordinary Least Squares (OLS) the following two equations:
Legitimation =β0+ Component characteristics*β1 + Inventors’ characteristics*β2 +
Applicants’ characteristics*β3 + Other controls* β4 + ε
(Equation 2)
Technological potential =β0+ Component characteristics*β1 + Inventors’
characteristics*β2 + Applicants’ characteristics*β3 + Other controls*β4 +ε
(Equation 3)
where Component characteristics is a vector of variables that includes the characteristics of the
components that combined to generate the novel technology namely, Similarity, Science-based
content, Familiarity, Appropriability, and Concentration of the inventive activity. The other three
vectors of variables represent our controls. Inventors’ characteristics is a vector including the
inventors’ characteristics. Applicants’ characteristics is a vector accounting for applicants’
characteristics. As other controls we include a set of time-dummies representing the calendar year
when the novel technology appeared (Technology entry year) and a set of technology class
dummies (Dummy Technology class).
To measure Similarity between the components combined, we exploit the hierarchical structure of
the IPC code classification where each additional digit denotes a higher degree of refinement of
the technological classification. Precisely, we define two components being similar when they
have the first three digits of their IPC codes in common. To construct the variable Science-based
content, we consider the patent applications including the two combined components over a rolling
time window from t-1 to t-4. For each component we compute the average number of references
17
to the non-patent literature per patent application (Meyer-Krahmer & Schmoch, 1998). Finally, we
calculate the Science-based content variable as the average number of references per patent
between the two components. To measure Familiarity, we refer to Fleming’s measure (2001). To
this end, we count the number of patent applications in a four-year rolling window for each of the
two combined components. Then, we calculate the average number of patents between the two
components. Similarly, to construct the variable Appropriability, we calculate the average number
of claims of the patents in a four-year rolling window for each of the two combined components
(Merges and Nelson, 1994; Tong and Frame, 1994)12. Then, we calculate the average between the
two components. We measure the Concentration of the inventive activity by calculating the
Herfindahl–Hirschman Index (HHI) of the patents owned by the applicants using the combined
components. For each component (limited at the first four digits) in the time window from t-1 to
t-4, we calculate the share of patents owned by each applicant. Based on these shares we calculate
the HHI of each component and we calculate the average of the HHI of the two components.
In our regression model we consider the logarithm transformation of the variables Science-based
content, familiarity, Appropriability, and Concentration to interpret the estimated effects as semi-
elasticities.
Several time-invariant unobserved characteristics of the components might bias our estimations of
the technological, cognitive and institutional characteristics such as, the unmeasurable propensity
of the component to generate additional inventions and its technological complexity. To take into
12 Our interpretation is that the larger the number of claims the higher the appropriability of the invention. However
other interpretations exist. An alternative interpretation is that a larger number of claims correspond to a lower
appropriability of the invention since the large number of claims protects very specific industrial applications. On the
contrary, a patent with a limited number of claims protects a less specific industrial application with a higher
appropriability.
18
account these issues, we include as explanatory variables a set of 118 three-digit technological
class dummies referring to the combined components. A technology class dummy equals one if
the first three digits of the IPC codes of at least one of the two combined components equal to the
three-digit identifying the technological class dummy, zero otherwise. A complete list with the
detailed description of the 118 technological class dummies is reported in Appendix C.
As inventors’ characteristics we consider inventors’ experience and inventors’ team size. To build
the inventors’ experience variable, we proceed in three steps. First, we extract the inventors listed
in the patents filed during the first year when the novel technology appears. Second, we reconstruct
the patent history of each inventor and count the number of patents she filed. Finally, we define a
dummy that equals one if at least one of the inventors has a previous patented invention. The team
size is the number of inventors appearing in the patents filed during the first year when the novel
technology appears.
As applicants’ characteristics during the first year when the novel technology appears, we consider
their experience, type (university or public research center versus private company), being a single
applicant, and their countries. The applicants’ experience is computed similarly as for inventors.
However, instead of generating a dummy variable, we compute a count variable that records
applicants’ previous patented inventions. University applicant is a dummy that equals one if at
least one of the applicant is a university or a research center, zero otherwise. More than one
applicant is a dummy that equals one if there is more than one applicant, zero otherwise.
Applicant’s country is a set of dummies, one for each applicant’s country reported on the patent
documents, that identifies the geographical location of the applicant(s). As for the inventors’
characteristics, the set of variables that refer to applicants is computed on the patents filed during
the first year when the novel technology appears.
19
Table 1 reports the descriptive statistics of all our dependent and independent variables.
Table 1: Descriptive statistics.
Obs. Mean Sd Min Max
Dependent variables
Technological potential [# patents] 11,009 65.97 93.84 16.32 998.29
Legitimation (10%) [# years] 11,009 5.86 3.13 0.00 16.78
Other sigmoid parameters
Midpoint (50%) [# years] 11,009 12.37 3.32 1.53 24.89
Alpha 11,009 2.96 1.16 0.21 7.47
Component characteristics
Similarity 3 digits (dummy) 11,009 0.33 0.47 0.00 1.00
Science-based content 11,009 3.50 7.41 0.02 283.99
Familiarity 11,009 346.24 427.84 0.50 5490.50
Appropriability 11,009 10.91 2.09 1.50 28.75
Concentration (HHI) 11,009 0.03 0.05 0.00 0.89
Inventor’s characterisics
Inventors' experience (dummy) 11,009 0.47 0.50 0.00 1.00
Inventors' team size 11,009 3.36 2.89 1.00 55.00
Applicant characteristics
Applicants' experience 11,009 637.19 1503.72 0.00 17279.00
University applicant (dummy) 11,009 0.05 0.22 0.00 1.00
More than one applicant (dummy) 11,009 0.25 0.43 0.00 1.00
The transgenic mammal technology used to mimic diseases provides an illustrative example of
how we identify the diffusion curve of a novel technology and of how we calculate the
characteristics of the combined components. We mark as transgenic mammal novel technology
the combination of the IPC codes “Introducing […] material into […] the body of animals” (IPC
code A01K67) and “[…] DNA or RNA concerning genetic engineering […]” (IPC code C07H21)
that appeared for the first time in onco-mouse patent. We observe the novel transgenic mammal
technology for a 20-year window. Figure 2 shows the actual cumulated distribution (Yt) of the user
patents embodying the combinations of the IPC codes A01K67-C07H21 (dotted line). Referring
to the patent distribution, we estimate the three parameters of the corresponding S-curve by using
20
a maximum likelihood estimation methodology. The solid line in Figure 2 represents the fitted S-
curve curve (�̂�𝑡). This curve is identified by an estimated ceiling equals to 267.38 patents, a
Midpoint of 15.49 years, and an Alpha of 3.45. In terms of legitimation and technological potential,
it results that the legitimation of the technology is reached after 7.91 years since the technology
appearance while 267.38 patents correspond to the technological potential reachable by the
technology. The use of a specific functional form allows us to predict the diffusion of a novel
technology at any point in time by means of the three estimated parameters. In our example, by
substituting the estimated parameters in Equation 1, we can predict that, after 10 years since the
transgenic mammal technology appearance (�̂�10), the cumulated number of user patents
embodying the transgenic mammal technology equals 45.24 (267.38
1+𝑒(−
(10−15.49)3.45
)=45.24). In the same
way, we can predict that after 30 years from its appearance (�̂�30), the cumulated number of user
patents embodying the transgenic mammal technology equals to 263.45 (267.38
1+𝑒(−
(30−15.49)3.45
)=263.45).
Figure 3 plots all the estimated diffusion curves of the novel technologies in our sample. The black
line highlights the transgenic mammal technology diffusion curve.
Concerning the characteristics of the combined components generating the transgenic mammal
technology, we find that they are not similar (Similarity = 0 [sample average = 0.33]), they are
science based (Science-based content = 9.51 [sample average = 3.50]), the inventors’ community
is not familiar with the components (Familiarity=160.5 [sample average = 346.24]), and, that their
Appropriability is slightly higher than the average of the other components (Appropriability=13.11
[sample average = 10.91]). Finally, the Concentration of the innovative activities is lower than the
average (Concentration = 0.014 [sample average = 0.03]).
21
Figure 2: Diffusion curve of the novel onco-mouse technology
Figure 3: Novel technologies’ diffusion curves. The back curve identifies the onco-mouse technology
22
4. Results
Table 2 shows the results of our econometric exercise. Columns 1 and 2 report the estimation of a
baseline model including only the component characteristics and technology entry year dummies.
Columns 3 and 4 add as controls the inventors’ and applicants’ characteristics. Columns 5 and 6
add the technological class dummies. According to our most complete model specification,
represented in Equation 2 and 3 and estimated in columns 5 and 6, we find that combining two
similar components reduces significantly the technology legitimation time. Specifically, if the
novel technology results from the combination of two similar components the legitimation time is
15.12 months13 shorter than a technology resulting from the combination of two dissimilar
components. Component similarity significantly affects also the technological potential. Having
similar component decreases the technological potential by 14.8 patents. Then, hypothesis one,
combining similar components, is confirmed.
Combing components with a higher science-based content generates a novel technology that
requires more time to be legitimated but has a higher technological potential. Precisely 50% more
of science-based content of the technology increases the legitimation time by 1.38 months, while
it increases the technological potential by 9.1 patents. Therefore, hypothesis two, combining
science-based components, is confirmed.
A novel technology resulting from the combination of two familiar components requires a shorter
time to be legitimated. Increasing the level of familiarity by 50% decreases the time needed for a
novel technology to be legitimated by 1 month. The same technology has lower technological
13 15.12 is obtained multiplying the coefficient of Similarity in the regression with Legitimation as dependent variable
(-1.26) by the number of months in a year (12).
23
potential by 3.5 patents. These results are in line with hypothesis three, combining familiar
components.
Combining components with a high appropriability generates a novel technology that requires
shorter time to be legitimated and has lower technological potential. Precisely 50% more of
appropriability decreases the legitimation time by 5.94 months and decreases the technological
potential by 13.65 patents. Therefore, hypothesis four, combining components with high
appropriability, is confirmed.
Combining components with a higher concentration of the innovative activity generates a novel
technology with a higher technological potential. Precisely, 50% more of concentration increases
the technological potential by 5.5 patents. Legitimation time is not affected. Therefore, hypothesis
five, combining components with high concentration, is partially confirmed.
Concerning the controls, we find that the inventors’ experience reduces the time needed for a
technology to be legitimated while the inventors’ team size increases the technological potential.
The experience of the applicants, having multiple applicants, and the presence of a university
among the applicants promoting the novel technology lead to a technology with a higher potential.
The legitimation time is negatively affected by the presence of multiple applicants. The applicant
concentration in the sector fosters the technological potential.
24
Table 2: Regression results. OLS estimations for Equation 1 and 2.
(1) (2) (3) (4) (5) (6)
VARIABLES
Legitimation
(10%)
Technological
potential
Legitimation
(10%)
Technological
potential
Legitimation
(10%)
Technological
potential
Component characteristics
Similarity 3 digits (dummy) 0.039 18.9*** 0.077 18.1*** -1.26*** -14.8***
log(Science-based content) 0.46*** 23.7*** 0.49*** 22.3*** 0.23*** 18.2***
log(Familiarity) -0.10*** -4.90*** -0.092*** -4.96*** -0.17*** -6.91***
log(Appropriability) -2.71*** -52.6*** -2.51*** -54.4*** -0.99*** -27.3***
log(Concentration) 0.27*** 9.56*** 0.26*** 9.81*** 0.042 11.0***
Inventor’s characterisics
Inventors' experience (dummy) -0.30*** -3.08 -0.20*** -1.18
log(Inventors' team size) -0.17*** 4.03*** -0.063 5.55***
Applicant characteristics
log(Applicants' experience) -0.027** 0.14 -0.010 0.60*
University applicant (dummy) 0.10 7.62* 0.17 10.3***
More than one applicant (dummy) -0.77*** 7.76*** -0.86*** 5.71**
Dummy Technology class (3 digits) No No No No Yes Yes
Dummy Applicant’s country No No Yes Yes Yes Yes
Dummy Technology entry year Yes Yes Yes Yes Yes Yes
Constant 14.0*** 251*** 14.3*** 238*** 12.9*** 248***
Observations 11,009 11,009 11,009 11,009 11,009 11,009
R-squared 0.095 0.080 0.128 0.096 0.243 0.155
In Table 2 we present a set of regressions with legitimation and technological potential as
dependent variables. While the later variable has a direct correspondence in one of the parameter
of a sigmoid curve representing the diffusion curve, legitimation is a linear combination of the
remaining two parameters, Midpoint and Alpha. Table 3 reports a set of regressions with Midpoint
and Alpha as dependent variables in order to disentangle the determinants of the two parameters
that linearly combined define Legitimation. Note that the coefficients estimated for each variable
in Table 2, columns 5 and 6, can be calculated as follow:
�̂�𝐿𝑒𝑔𝑖𝑡𝑖𝑚𝑎𝑡𝑖𝑜𝑛 = �̂�𝑀𝑖𝑑𝑝𝑜𝑖𝑛𝑡 − �̂�𝐴𝑙𝑝ℎ𝑎 ∗ 2.2
25
(Equation 4)
Table 3: OLS estimation using as dependent variables Midpoint and Alpha and adopting the same
specifications reported in Equation 1 and Equation 2.
(1) (2)
VARIABLES Midpoint (50%) Alpha
Technological component characteristics
Similarity 3 digits (dummy) -0.48*** 0.35***
log(Science-based content) -0.20*** -0.20***
log(Familiarity) -0.12*** 0.022**
log(Appropriability) -0.83*** 0.074
log(Concentration) -0.069 -0.051***
Inventor’s characterisics
Inventors' experience (dummy) -0.13** 0.031
log(Inventors' team size) -0.078* -0.0072
Applicant characteristics
log(Applicants' experience) -0.028*** -0.0081**
University applicant (dummy) -0.30** -0.21***
More than one applicant (dummy) -0.27*** 0.27***
Dummy Applicant’s country yes yes
Dummy technology entry year yes yes
Dummy Technology Class (3 digits) yes yes
Constant 18.1*** 2.37***
Observations 11,009 11,009
R-squared 0.339 0.259
The results concerning Legitimation (time to reach 10%) and Midpoint (time to reach 50%), appear
consistent (see Table 2 column 5 versus Table 3 column 2). Indeed, Similarity, Familiarity,
Appropriability, and Concentration show qualitatively the same impact on Legitimation and
Midpoint. The level of Science-based content of the combined components represents the only
exception. While Science-based content increases the legitimation time, it decreases the time
needed to reach the Midpoint. The parameter Alpha explains this finding. A high Science-based
content of the combined components leads to a diffusion curve that grows slowly in its very first
26
part (that reflects in a positive and significant coefficient of the Science-based variable in the
Legitimation regression) but that catches up quickly due to its steepness (at the base of a negative
and significant sign of the coefficient of the Science-based variable in the Alpha regression) and
reaches faster the midpoint (with a negative and significant sign of the coefficient of the Science-
based variable in the Midpoint regression).
To test the reliability of our results we implemented a set of robustness checks that are reported in
Appendixes from D to I.
In Appendix D we check the robustness of the main results reported in Table 2. Specifically, we
change the functional form by using the logarithmic transformation of our dependent variables,
legitimation time and technological potential. In this regression exercise, the estimated coefficients
for the component characteristics can be interpreted as elasticities. Results are consistent across
model specifications.
The arbitrariness of the assumption that the diffusion process follows an S-curve might raise a
concern. In Appendix E, to reply to this concern, we report a variant of the econometric exercise
of Table 2. Instead of using as dependent variables the estimated legitimation time and the
estimated technological potential, we use the actual values of the two parameters. Specifically, we
calculate the legitimation time as the time required to reach the ten percent of the actual ceiling.
We define the actual ceiling, i.e. technological potential, as the cumulated number of patented
inventions using the novel technology 20 years after its appearance. Results are consistent when
using actual legitimation time and actual technological potential.
In the same vein of Appendix E, we test our assumption on the validity of a S-curve as
representative diffusion curve by conducting two separated econometric exercises on two samples.
27
The first sample includes novel technologies reaching the technological potential within the 20
years observed, while the second sample includes the technologies not reaching the technological
potential. We adopt the same specification of the models estimated in columns 5 and 6 of Table 2.
Results are reported in Appendix F and are in line with those obtained in Table 2 for both samples.
Another possible concern regards the presence of multiple novel technologies within the same
patent. As reported in section 3 (Methods), 58% of the follow-up patents contains only one of the
11,009 novel technologies, while the 42% contains multiple novel technologies. The fact that a
patented invention embeds more than one novel technology, and therefore contributes to multiple
diffusion curves, is not in conflict with the hypotheses at the base of our analysis. However, in
order to test the robustness of our results, we restrict our sample to the novel technologies that
have diffusion curve where the large majority of the patents (at least the 90%) is mono-technology.
Then, we replicate the same econometric exercise as the one reported in Table 2, columns 5 and
6. Results are reported in Appendix G and are consistent with results in Table 2.
In Appendix H we conduct a similar robustness check as the one conducted in Appendix G.
Specifically, we limit the sample to those novel technologies having only one patent the first year
when they appear. Results are in line with those obtained in Table 2, columns 5 and 6.
In Appendix I, we consider a less stringent definition of novel technology avoiding imposing a
minimum number of reuse after its appearance and present a 2-step Heckman estimation strategy
that correct for possible selection biases. Results are coherent with those reported in Table 2,
columns 5 and 6.
5. Discussion
28
Novel technology diffusion and component characteristics
Ideally an attractive technology has a short legitimation period and a high technological potential.
To visualize the economic impact of the component characteristics on the novel technology
diffusion, we consider five scenarios where the novel technology is obtained by combining (1)
similar components (compared to dissimilar components), (2) components with a science-based
content increased by 50%, (3) components with familiarity increased by 50%, (4) components with
appropriability increased by 50%, and (5) components with concentration of the innovative
activities increased by 50%. The coefficients considered are those estimated in the econometric
models presented in Table 2, columns 5 and 6. Figure 4 and 5 show the increase/reduction of
legitimation time and technological potential obtained in the five scenarios, while Figure 6 reports
the complete diffusion curve corresponding to each scenario. We observe a window period of
twenty years after the introduction of the novel technology14.
The five scenarios show there is a trade-off when changing the values of each component
characteristics (Similarity, Science-based content, Familiarity, Appropriability, and
Concentration) to shorten the legitimation or increase the technological potential. Looking at
Figures 4 and 5, we observe that by augmenting the value of the variables Science-based content
and Concentration, the corresponding technology has higher technological potential. However, it
requires a longer time to be legitimated. On the contrary, augmenting the value of Familiarity, the
novel technology decreases the time needed to be legitimated and, as a drawback, reduces its
14 Although our model allows to predict technological potential for any time span, we selected twenty years because
it seems a reasonable period over which both managers and policy makers might be requested to predict the
technological potential of a novel technology.
29
technological potential. Higher Similarity as well as higher Appropriability decreases the
legitimation time and the technological potential.
Although each component characteristic shows a trade-off, the magnitude of these trade-offs
differs significantly according to the characteristic considered. Augmenting the Science-based
content and the Concentration increases to a considerable extent the technological potential,
augmenting only slightly the legitimation time. Familiarity exhibits limited effects both on
legitimation and technological potential. Appropriability and Similarity are characterized by the
largest trade-offs, namely shortening the legitimation time is obtained at the cost of significantly
lower technological potential and vice versa. We conclude that Familiarity has a limited effect,
Science-based content and Concentration increase significantly the technological potential with a
limited drawback in terms of legitimation time, Appropriability and Similarity show relevant trade-
offs between legitimation and technological potential.
Novel technology diffusion across different technological domains
The results discussed regard the average effects of the combined component characteristics in
different technological domains. However, the effects might differ across technological domains.
In order to further investigate the technological domain heterogeneity, we conduct two separated
regression exercises on the most representative sectors in our dataset: the Bio-pharmaceutical
sector and the Digital technology sector, respectively. Table 4 replicates the econometric exercise
as in Table 3, columns 5 and 6, for the bio-pharmaceutical and for the digital technology sectors15.
15 We identify bio-pharmaceutical and for the digital technology sectors according to Schmoch’s classification (2008)
30
Table 4: Regression results. OLS estimations for Equation 1 and 2 in the Bio-pharmaceutical sector and
Digital technology sector
Bio-pharmaceutical sector Digital technology sector
(1) (2) (3) (4)
Legitimation (10%) Technological potential Legitimation (10%) Technological potential
Technological component characteristics
Similarity 3 digits (dummy) -1.70*** -24.1*** -0.73** 23.8*
log(Science-based content) 0.13 24.7*** 0.083 3.86
log(Familiarity) -0.23*** -5.69*** -0.21*** -12.8***
log(Appropriability) -0.20 -8.66 -1.61*** -56.6***
log(Concentration) -0.83*** -22.6*** 0.14 35.9***
Inventor’s characterisics
Inventors' experience (dummy) -0.13 -1.74 -0.27** -1.06
log(Inventors' team size) -0.046 12.3*** 0.11 3.55
Applicant characteristics
log(Applicants' experience) 0.0062 1.61* -0.041** 0.21
University applicant (dummy) 0.21 4.62 -0.98*** -10.3
More than one applicant (dummy) -0.73*** 2.10 -0.82*** 9.52
Dummy Applicant’s country Yes Yes Yes Yes
Dummy technology entry year Yes Yes Yes Yes
Dummy Technology Class (3 digits) Yes Yes Yes Yes
Constant 8.19*** 58.9 15.8*** 389***
Observations 3,303 3,303 3,252 3,252
R-squared 0.295 0.151 0.321 0.244
We find substantial differences between the two sectors in the impact of the variables Similarity,
Science-based, Appropriability, and Concentration, respectively. In the bio-pharmaceutical sector,
consistently with our main results (Table 2), Similarity has a negative impact on the technological
potential. On the contrary, in the digital technology sector, the Similarity of the combined
components increases the technological potential of the novel technology by 23.8 patented
inventions. The different level of complexity across the two sectors might explain these different
results. In the digital sector, inventions are complex since they often result from assembling
31
numerous parts, while inventions in the bio-pharmaceutical sector are simpler being often
represented by a unique molecule or a chemical compound. In highly complex sectors, combining
similar components does not penalize the technological potential, on the contrary increase the
probability that a larger number of inventors engages in developing industrial applications that use
the novel technology. The Science-based content does not affect novel technologies diffusion in
the digital sector, while it increases significantly the technological potential of a bio-
pharmaceutical novel technology. This result might be explained by the strong connection between
laboratory scientific research and industrial applications in the bio-pharmaceutical sector. Greater
research on a molecule or chemical compound leads to a better knowledge of its functioning
biological mechanisms and to a larger number of applications. On the contrary in the digital sector,
laboratory scientific research and industrial applications are less connected and having a greater
number of scientific paper citations does not affect the potential of the resulting technology.
Appropriability does not show any effect on the diffusion of novel technologies in the bio-
pharmaceutical sector, while it is statistically significant and negative both for legitimation and
technological potential in the digital technology sector. In the bio-pharmaceutical sector, a patent
guarantees the protection of an invention regardless its form, i.e. the number of claims, since
property rights over molecules and chemical compounds are easily defined. Contrary to our
hypotheses, Concentration of the firms’ inventive activity reduces the technological potential in
the pharmaceutical sector. This result might be explained by the fact that few large firms can easily
exclude potential entrants and a limited number of actors in the sector limits the number of
industrial applications developed.
32
Strategical implications
Our general results can be used for practical purposes when managers or policy makers are asked
to decide the long terms R&D strategies of a company or a region. Table 5 reports an example of
a hypothetical situation. A firm is competing in a market where two emerging novel technologies
co-exist, A and B. The Chief Technological Officer (CTO) of the firm is asked to select the
technology that is more promising in terms of technological potential in order to plan the long run
R&D investment strategy of her firm. Moreover, she is asked to estimate if there is a significant
difference between the two novel technologies in terms of the time needed to generate a
“reasonable” number of industrial applications (legitimation time). Novel technology A is
characterized by a combination of dissimilar components with an average value of Familiarity,
Science-based content, Appropriability, and Concentration. Differently, novel technology B, is
characterized by the combination of similar components and has a significantly lower Science-
based content. Familiarity, Appropriability and Concentration of technology A equal to those of
technology B. Table 3 allows the CTO to compare the two technologies and to take a decision. It
shows that, according to the components characteristics, the novel technology A has higher
technological potential than B (+26.08 patents) although reaching a reasonable number of
industrial applications will require more than technology B (+1.4 years).
Table 5: Legitimation and technological potential of the novel technology A vs. B
A B A-B
∆Legitimation [years]
∆Technological Potential [patents]
Similarity 0 1 -1 +1.26 +14.8
log(Familiarity) 5.26 5.26 0 0 0
log(Science-Based content) 0.74 0.12 0.62 +0.14 +11.28
log(Appropriability) 2.37 2.37 0 0 0
log(Concentration) -3.91 -3.91 0 0 0
Total: +1.4 +26.08
33
Figure 4: Novel technologies’ legitimation time increase/reduction obtained by changing the values of the
technological component characteristics. The average legitimation time in the study sample equals 5.86 years
34
Figure 5: Novel technologies’ technological potential obtained by changing the values of the technological
component characteristics
35
Figure 6: Novel technologies’ diffusion curves obtained by changing the values of the technological
component characteristics
36
6. Conclusion
In this paper we investigate if the conditions at the time of the appearance of a novel technology
(technological antecedents) impact on its subsequent diffusion curve. To do that, we conduct an
empirical study that traces the diffusion of 11,009 successful novel technologies over the period
1985-2015. Using patent data, we identify a novel technology as an unprecedented combination
of existing technological components, as represented by the technological classes where a patent
is classified. Then, we trace all the patents incorporating that novel technology and we estimate
the corresponding diffusion curve. Under the assumption that a diffusion curve is S-shaped, we
look at the technological, cognitive and institutional characteristics of the combined components
affecting the duration of the legitimation phase (first part of the curve) and the potential of a
technology (ceiling of the curve).
We find that a high degree of similarity between the combined components, familiarity of
inventors with the components, and high level of appropriability shorten the legitimation time and
reduce the technological potential. The science-based nature of the combined components plays
an opposite role and increases the legitimation time, while positively affects the technological
potential of a novel technology. Concentration of innovative activities increases the technological
potential. By looking at the economic impact of our results, we find that increasing the science-
based nature of the combined components as well as the concentration of the innovative activities
significantly augment the technological potential with a limited increase in the legitimation time.
Familiarity has a limited effect both on technological potential and legitimation. Finally, similarity
and appropriability show a trade-off of a relevant extent: decreasing the legitimation period has a
significant cost in terms of reduction of technological potential (and vice versa).
37
Our results have strategical implications in predicting how the legitimation period and the
technological potential vary according to a set of characteristics of the novel technology observed
at its appearance. These predictions might be used in the private sector, when managers face the
choice between two emerging technologies for their businesses, as well as in the public sector,
when policy makers are asked to choose, for instance, the technological orientation of a region.
38
References
Abernathy, W.J. and Utterback, J.M. 1978. Patterns of Industrial Innovation. Technology
Review, 80: 40–47.
Achilladelis, B. 1993. The dynamics of technological innovation: The sector of antibacterial
medicines. Research Policy, 22: 279–308.
Amabile, T. M. 1996. Creativity in Context. Boulder: Westview Press.
Arrow, K. 1962. Economic welfare and the allocation of resources for invention. In The rate and
direction of inventive activity: Economic and social factors. New Jersey: Princeton
University Press.
Arthur, W. B. 2009. The Nature of Technology: What It Is and How It Evolves. Simon and
Schuster.
Arts, S., Appio, F. P., & Van Looy, B. 2013. Inventions shaping technological trajectories: do
existing patent indicators provide a comprehensive picture? Scientometrics, 97: 397–419.
Breschi, S., Malerba, F., & Orsenigo, L. (2000). Technological Regimes and Schumpeterian
Patterns of Innovation. The Economic Journal, 110(463), 388–410.
Cohen, W. M., & Levinthal, D. A. 1990. Absorptive capacity: a new perspective on learning and
innovation. Administrative Science Quarterly, 128–152.
Curran, C.-S. (2013). The Anticipation of Converging Industries. London: Springer London.
Caviggioli, F. (2016). Technology fusion: Identification and analysis of the drivers of technology
convergence using patent data. Technovation, 55–56.
Dosi, G. (1982). Technological paradigms and technological trajectories: a suggested
interpretation of the determinants and directions of technical change. Research Policy,
11:147–162.
39
Fleming, L. (2001). Recombinant uncertainty in technological search. Management Science, 47:
117–132.
Fleming, L., Mingo, S., & Chen, D. (2007). Collaborative brokerage, generative creativity, and
creative success. Administrative Science Quarterly, 52: 443–475.
Geroski, P. A. 2000. Models of technology diffusion. Research Policy, 29: 603–625.
Griliches, Z. 1957. Hybrid Corn: An Exploration in the Economics of Technological Change.
Econometrica, 25: 501.
Hall, B. H. 2004. Innovation and diffusion. Cambridge, Mass.: National Bureau of Economic
Research.
Adam B. J. 2002. Building Programme Evaluation into the Design of Public Research‐ Support
Programmes, Oxford Review of Economic Policy, 18 (1): 22–34.
Malerba, F., Orsenigo, L., & Peretto, P. (1997). Persistence of innovative activities, sectoral
patterns of innovation and international technological specialization. International Journal
of Industrial Organization, 15(6), 801–826.
March, J. 1991. Exploration and exploitation in organizational learning. Organization Science, 2
71-87.
March, J. G. 2013. Handbook of Organizations. U.K.: Routledge.
Meyer-Krahmer, F., & Schmoch, U. 1998. Science-based technologies: university–industry
interactions in four fields. Research Policy, 27 835–851.
Merges, R. P., & Nelson, R. R. 1994. On limiting or encouraging rivalry in technical progress:
The effect of patent scope decisions. Journal of Economic Behavior & Organization, 25: 1–
24.
40
Murray, F. (2010). The onco-mouse that roared: Hybrid exchange strategies as a source of
distinction at the boundary of overlapping institutions. American Journal of Sociology,
116(2), 341–388.
Nelson, R. R., & Winter, S. G. 2004. An evolutionary theory of economic change. Cambridge,
Mass.: Harvard Univ. Press.
Schumpeter, J. A. 1939. Business cycles. New York: McGraw-Hill.
Schmoch, U. 2008. Concept of a Technology Classification for Country Comparisons. Final
Report to the World Intellectual Property Organization (WIPO), Fraunhofer Institute for
Systems and Innovation Research, Karlsruhe.
Strumsky, D., & Lobo, J. 2015. Identifying the sources of technological novelty in the process of
invention. Research Policy, 44: 1445–1461.
Verhoeven, D., Bakker, J., & Veugelers, R. 2016. Measuring technological novelty with patent-
based indicators. Research Policy, 45: 707–723.
41
Appendix A: Validation of the novel technology measure with topic modelling
A possible concern with the identification of a novel technology as the combination of two
different components (represented by the combination of two IPC codes) is that there is no proof
that the resulting combination is meaningful. It could be that the novel combination is a mere
artifact of our way of identifying novel technologies without any meaningful content. To
investigate the existence of a meaningful content, we implement a topic modelling analysis that
verifies the coherence of the content of the patents embedding the same novel technology. If the
patents associated to the same novel technology are coherent in terms of content, we can assume
that they embed a meaningful common content, i.e. the novel technology.
In detail, we proceed as follow. First, we pulled together the titles of all the patents included
in our study sample (250,979 patents) and we inferred the topics of each patent by using the Lantent
Dirichler Allocation (LDA) method. The idea of LDA is that each document is the results of a set
of (latent) topics, and topics generate words with a probability distribution (Steyvers and Griffiths,
2007). For example, if a document includes a unique topic, e.g., transgenic modification methods,
the words that are likely to appear in the document will be “protein”, “human”, “dna”,
“transgenic”. We used the Gibbs’s algorithm to estimate the LDA parameters and to classify each
patent according to 20 topics. We chose the number of topics according to the optimization method
proposed by Steyvers and Griffiths (2007) that aims at finding the number of topics that maximizes
the log-likelihood value of the Gibb’s estimation of the LDA parameters. The results of this
optimization method are reported in Figure A1. Figure A1 shows that the maximum value of log-
likelihood is obtained for 20 topics. Figure A2 illustrates the most likely four words generated by
each of the 20 topics.
42
Figure A1. Optimal topic selection number of topics identifying the patent titles text in our sample.
Figure A2: The most likely four words generated by each topic.
Once the 20 topics have been identified, we redefined each patent as a combination of topic shares.
Figure A3 shows an illustration of how topics contribute to the patent titled “Method and device
for preparing fibrous plant bodies.”. The patent is characterized by topic 5 and 12 that have the
highest shares.
Figure A3: Share contribution of topics in patent titles.
Ttitle 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Method and
device for
preparing
fibrous
plant
bodies.
5% 5% 5% 5% 8% 5% 5% 5% 5% 5% 5% 8% 5% 5% 5% 5% 5% 5% 5% 5%
43
To prove that each novel technology has a meaningful content, we need to show that patents
belonging to the same group are homogenous in terms of topic content. To do that, we grouped
the patents belonging to the same novel technology according to our definition reported in section
2: a patent is assigned to a technology if it has a specific pair of IPC codes in its IPC classification
list. Then, we evaluated the level of homogeneity of each topic within the group of patents
embedding the same novel technology. To do so, we calculate the standard deviation of each topic
share (see Figure A4). Then, we calculate the average of those standard deviations by group (novel
technology) (see Figure A5).
Figure A4: Standard deviation of each topic share by novel technology. A patent is assigned to a technology if
it has a specific pair of IPC codes in its IPC classification list.
IPC class IPC class 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 #patents
A01D45 A01D43 2.8% 0.8% 0.6% 0.7% 0.7% 1.0% 0.6% 0.8% 0.5% 0.5% 0.3% 1.1% 0.1% 0.5% 0.7% 0.1% 0.7% 0.4% 0.4% 0.4% 26
Figure A5: Average standard deviation of topics shares by novel technology. A patent is assigned to a
technology if it has a specific pair of IPC codes in its IPC classification list.
PC class IPC class Average
standard devation
#patents
A01D45 A01D43 0.69% 26
To assess if topics are homogeneous or not within each group, as represented by the average
standard deviation calculate in Figure A5, we need to have a comparison value. We constructed a
44
comparison sample where each patent is assigned randomly to a group (and not grouped according
to its IPC codes). Each group maintains the same number of patents as in our original classification.
We computed the average standard deviation of topics shares of each of those groups where patents
are randomly assigned (see Figure A6).
Figure A6: Average standard deviation of topics shares by novel technology. A patent is assigned
randomly to a technology.
PC class IPC class Average standard
devation
#patents
A01D45 A01D43 1.04% 26
Finally, we test if the average standard deviations in the two samples, the original sample and the
one with patent randomly assigned to technologies, are statistically different. We find that the
average standard deviation of the topics share in the original sample is significantly lower than in
the sample with randomly assigned patents (0.008 vs. 0.012, P-value 0.000). This result shows that
topics are more homogeneous in a sample where patents are assigned to the novel technology
according to the IPC code combinations rather than when they are assigned randomly. In other
words, the method applied to identify a novel technology aggregates patents with coherent
contents.
45
Appendix B: Legitimation as a linear combination of midpoint and alpha of an S-curve
technology diffusion curve
In this appendix details the mathematical formulation relating legitimation (defined as the time
needed for diffusion curve to reach the 10% of its ceiling point) to the midpoint and alpha. Starting
from Equation 1, we extract t as follow:
�̂�𝑡 =𝐶𝑒𝑖𝑙𝑖𝑛𝑔
1 + 𝑒(−
(𝑡−𝑀𝑖𝑑𝑝𝑜𝑖𝑛𝑡)𝐴𝑙𝑝ℎ𝑎
)
1 + 𝑒(−
(𝑡−𝑀𝑖𝑑𝑝𝑜𝑖𝑛𝑡)𝐴𝑙𝑝ℎ𝑎
)=𝐶𝑒𝑖𝑙𝑖𝑛𝑔
�̂�𝑡
−(𝑡 − 𝑀𝑖𝑑𝑝𝑜𝑖𝑛𝑡)
𝐴𝑙𝑝ℎ𝑎= ln(
𝐶𝑒𝑖𝑙𝑖𝑛𝑔
�̂�𝑡− 1)
−𝑡 +𝑀𝑖𝑑𝑝𝑜𝑖𝑛𝑡 = 𝐴𝑙𝑝ℎ𝑎 ∗ ln(𝐶𝑒𝑖𝑙𝑖𝑛𝑔
�̂�𝑡− 1)
𝑡 = 𝑀𝑖𝑑𝑝𝑜𝑖𝑛𝑡 − 𝐴𝑙𝑝ℎ𝑎 ∗ ln (𝐶𝑒𝑖𝑙𝑖𝑛𝑔
�̂�𝑡− 1)
Then we set the 10% of the ceiling point, �̂�𝑡
𝐶𝑒𝑖𝑙𝑖𝑛𝑔=
1
10 and compute the corresponding t
𝑡10% = 𝑀𝑖𝑑𝑝𝑜𝑖𝑛𝑡 − 𝐴𝑙𝑝ℎ𝑎 ∗ ln(10 − 1)
𝑡10% = 𝑀𝑖𝑑𝑝𝑜𝑖𝑛𝑡 − 𝐴𝑙𝑝ℎ𝑎 ∗ 2.2
46
Appendix C: List of the technological class dummies of our sample (three digits)
Tech class Number of patents (whole Patstat) Legitimation Potential Class name
C12 226231 5.40 56.74 biochemistry; beer; spirits; wine; vinegar; microbiology; enzymology; mutation or
genetic engineering
C40 4502 6.28 74.89 combinatorial technology
B82 5427 6.19 62.41 nanotechnology
A61 972410 6.19 58.17 medical or veterinary science; hygiene
C07 593444 5.12 53.73 organic chemistry
B81 5593 5.44 59.87 microstructural technology
C13 1402 4.39 30.79 sugar industry
C30 11506 4.91 40.13 crystal growth
A01 100309 5.85 63.02 agriculture; forestry; animal husbandry; hunting; trapping; fishing
G06 333626 6.53 65.50 computing; calculating; counting
G01 344963 5.58 51.31 measuring; testing
A23 61675 5.48 46.07 foods or foodstuffs; their treatment, not covered by other classes
H04 568514 6.83 75.78 electric communication technique
G10 23284 5.76 56.73 musical instruments; acoustics
C01 41421 4.29 47.89 inorganic chemistry
G11 83155 5.11 58.21 information storage
G02 102565 5.86 48.71 optics
H01 424044 6.05 54.34 basic electric elements
C05 5273 4.57 30.47 fertilisers; manufacture thereof
H03 69422 5.77 44.52 basic electronic circuitry
C04 35183 4.63 46.52 cements; concrete; artificial stone; ceramics; refractories
B01 190188 5.30 46.96 physical or chemical processes or apparatus in general
A21 8172 4.82 42.78 baking; equipment for making or processing doughs; doughs for baking
C23 41382 5.25 40.21
coating metallic material; coating material with metallic material; chemical surface
treatment; diffusion treatment of metallic material; coating by vacuum evaporation,
by sputtering, by ion implantation or by chemical vapour deposition, in general;
inhibiting corrosion of metallic material or incrustation in general
C22 34422 4.76 34.53 metallurgy; ferrous or non-ferrous alloys; treatment of alloys or non-ferrous metals
G09 43751 6.19 56.42 educating; cryptography; display; advertising; seals
C08 337146 4.60 38.52 organic macromolecular compounds; their preparation or chemical working-up;
compositions based thereon
E21 29316 5.24 47.15 earth or rock drilling; mining
C03 28360 4.56 35.42 glass; mineral or slag wool
C09 132566 5.30 42.99 dyes; paints; polishes; natural resins; adhesives; compositions not otherwise
provided for; applications of materials not otherwise provided for
G07 33692 5.32 48.83 checking-devices
C02 23804 5.02 38.74 treatment of water, waste water, sewage, or sludge
G03 75366 5.70 43.36 photography; cinematography; analogous techniques using waves other than
optical waves; electrography; holography
G05 42106 5.67 47.14 controlling; regulating
D01 17333 4.65 36.77 natural or man-made threads or fibres; spinning
G21 15183 5.76 36.89 nuclear physics; nuclear engineering
C11 39725 5.19 43.40 animal or vegetable oils, fats, fatty substances or waxes; fatty acids therefrom;
detergents; candles
B09 4642 3.61 37.39 disposal of solid waste; reclamation of contaminated soil
A62 10026 3.53 30.05 life-saving; fire-fighting
C25 19093 5.50 42.70 electrolytic or electrophoretic processes; apparatus therefor
47
B06 2548 6.36 42.86 generating or transmitting mechanical vibrations in general
A63 29203 5.37 51.46 sports; games; amusements
C10 60809 5.02 44.54 petroleum, gas or coke industries; technical gases containing carbon monoxide;
fuels; lubricants; peat
B03 5096 6.76 45.44 separation of solid materials using liquids or using pneumatic tables or jigs;
magnetic or electrostatic separation of solid materials from solid materials or
fluids; separation by high-voltage electric fields
B22 25745 4.66 40.40 casting; powder metallurgy
B32 54028 5.47 44.85 layered products
G08 27261 7.00 60.63 signalling
C06 2871 4.76 44.14 explosives; matches
H05 74628 6.24 50.92 electric techniques not otherwise provided for
A24 5848 6.08 21.92 tobacco; cigars; cigarettes; smokers' requisites
B28 10629 3.89 32.62 working cement, clay, or stone
D21 30511 4.28 39.52 paper-making; production of cellulose
C21 18066 5.00 44.47 metallurgy of iron
B05 43991 5.20 38.17 spraying or atomising in general; applying liquids or other fluent materials to
surfaces, in general
F21 43450 7.27 65.79 lighting
B41 71628 5.32 42.43 printing; lining machines; typewriters; stamps
D04 14911 5.36 35.52 braiding; lace-making; knitting; trimmings; non-woven fabrics
B29 118121 4.36 36.21 working of plastics; working of substances in a plastic state in general
H02 117217 5.71 42.32 generation, conversion, or distribution of electric power
B44 6913 5.72 33.29 decorative arts
D02 4015 3.63 32.48 yarns; mechanical finishing of yarns or ropes; warping or beaming
D06 40523 4.71 41.65 treatment of textiles or the like; laundering; flexible materials not otherwise
provided for
B64 21484 5.76 38.66 aircraft; aviation; cosmonautics
B24 20637 5.42 50.81 grinding; polishing
B42 9003 5.36 36.31 bookbinding; albums; files; special printed matter
B31 6146 4.19 27.36 making articles of paper, cardboard or material worked in a manner analogous to
paper; working paper, cardboard or material worked in a manner analogous to
paper
F04 45443 5.94 41.67 positive-displacement machines for liquids; pumps for liquids or elastic fluids
B08 8716 5.09 43.42 cleaning
F17 6312 4.95 30.87 storing or distributing gases or liquids
A41 8219 5.05 37.20 wearing apparel
F41 9474 5.16 42.72 weapons
B27 9313 4.72 35.58 working or preserving wood or similar material; nailing or stapling machines in
general
B25 26692 4.96 37.78 hand tools; portable power-driven tools; handles for hand implements; workshop
equipment; manipulators
B04 4760 4.38 34.24 centrifugal apparatus or machines for carrying-out physical or chemical processes
C14 1182 1.81 29.61 skins; hides; pelts; leather
B23 73056 5.77 49.40 machine tools; metal-working not otherwise provided for
A43 10643 5.37 41.87 footwear
F01 64495 6.18 51.59 machines or engines in general; engine plants in general; steam engines
F03 14724 5.80 47.15 machines or engines for liquids; wind, spring, or weight motors; producing
mechanical power or a reactive propulsive thrust, not otherwise provided for
G04 9336 6.27 35.19 horology
B21 33691 4.14 37.22 mechanical metal-working without essentially removing material; punching metal
48
A46 4839 3.86 31.61 brushware
B02 6196 3.75 33.50 crushing, pulverising, or disintegrating; preparatory treatment of grain for milling
F25 32049 6.53 43.86 refrigeration or cooling; combined heating and refrigeration systems; heat pump
systems; manufacture or storage of ice; liquefaction or solidification of gases
F15 11656 4.17 37.40 fluid-pressure actuators; hydraulics or pneumatics in general
F23 26346 4.58 40.73 combustion apparatus; combustion processes
B67 9452 5.21 32.54 opening or closing bottles, jars or similar containers; liquid handling
F42 6421 5.56 38.51 ammunition; blasting
F27 11853 5.33 41.85 furnaces; kilns; ovens; retorts
F02 113956 6.27 50.83 combustion engines; hot-gas or combustion-product engine plants
B43 3051 5.02 39.64 writing or drawing implements; bureau accessories
B07 5771 3.38 32.54 separating solids from solids; sorting
F26 7296 3.87 26.95 drying
A45 12126 5.66 30.46 hand or travelling articles
E01 13889 4.20 34.12 construction of roads, railways, or bridges
B63 14208 4.62 29.88 ships or other waterborne vessels; related equipment
F28 23852 5.01 31.40 heat exchange in general
F22 3790 8.22 25.98 steam generation
D07 1360 2.72 24.97 ropes; cables other than electric
B65 165572 4.76 37.60 conveying; packing; storing; handling thin or filamentary material
E02 16012 5.45 50.15 hydraulic engineering; foundations; soil-shifting
A44 5202 5.14 48.71 haberdashery; jewellery
D05 2044 2.61 39.50 sewing; embroidering; tufting
B60 176856 6.27 58.58 vehicles in general
A47 59501 5.46 37.05 furniture; domestic articles or appliances; coffee mills; spice mills; suction cleaners
in general
E04 47926 4.74 37.33 building
D03 7892 3.08 47.50 weaving
B62 48784 7.01 53.75 land vehicles for travelling otherwise than on rails
F24 37270 6.76 37.86 heating; ranges; ventilating
F16 191793 5.81 45.69 engineering elements or units; general measures for producing and maintaining
effective functioning of machines or installations; thermal insulation in general
A22 3717 1.81 28.32 butchering; meat treatment; processing poultry or fish
B30 6065 6.81 31.84 presses
B61 9889 4.79 26.03 railways
B66 18098 5.22 41.57 hoisting; lifting; hauling
B26 14447 3.50 30.29 hand cutting tools; cutting; severing
E03 9308 4.55 42.68 water supply; sewerage
E06 15805 3.43 33.77 doors, windows, shutters, or roller blinds, in general; ladders
E05 37663 6.04 43.50 locks; keys; window or door fittings; safes
49
Appendix D: Elasticity estimation relying on an alternative functional form (log-log)
(1) (2)
VARIABLES Log(Legitimation time) Log(Technological potential)
Technological component characteristics
Similarity 3 digits (dummy) -0.28*** -0.12***
log(Science-based content) 0.027** 0.12***
log(Familiarity) -0.021*** -0.032***
log(Appropriability) -0.089* -0.17***
Inventor’s characterisics
Inventors' experience (dummy) -0.047*** 0.00077
log(Inventors' team size) -0.024** 0.036***
Applicant characteristics
log(Applicants' experience) -0.0030 0.0040
University applicant (dummy) 0.042 0.071**
More than one applicant (dummy) -0.21*** 0.070***
log(Concentration) -0.010 0.088***
Dummy Technology class (3 digits) yes yes
Dummy Applicant’s country yes yes
Dummy technology entry year yes yes
Constant 2.71*** 5.08***
Observations 11,009 11,009
R-squared 0.183 0.211
50
Appendix E: OLS estimation using actual data
(1) (2)
VARIABLES Actual Legitimation time Actual Technological potential
Technological component characteristics
Similarity 3 digits (dummy) -0.91*** -12.0***
log(Science-based content) 0.32*** 16.1***
log(Familiarity) -0.22*** -5.26***
log(Appropriability) -1.09*** -14.4***
log(Concentration) 0.14*** 8.32***
Inventor’s characterisics
Inventors' experience (dummy) -0.21*** -0.93
log(Inventors' team size) -0.077* 4.49***
Applicant characteristics
log(Applicants' experience) -0.0085 0.65**
University applicant (dummy) 0.021 9.74***
More than one applicant (dummy) -0.98*** 5.07**
Dummy Technology class (3 digits) yes yes
Dummy Applicant’s country yes yes
Dummy technology entry year yes yes
Constant 12.3*** 172***
Observations 11,009 11,009
R-squared 0.232 0.140
51
Appendix F: Splitting the sample between technologies that reach their ceiling within the
first 20 years from their appearance and the others
In Figure F1, we plot the actual cumulated number of patents at the end of our observation period
(20 years) against the estimated technology potential (the ceiling of the s-curve). The plot should
be interpreted as follows:
- The closeness of the points to the diagonal indicate that the estimated ceiling is close to
the actual number of patents;
- Staying on the diagonal means that the technology reaches its ceiling within 20 years.
Figure F1: Actual cumulated number of patents at the end of the observation period
(when t=20 years) against the estimated technology potential (the ceiling of the s-curve)
-
52
Table F1 apply the regression model reported in Table 2 to two different subsamples of novel
technologies. Subsample A includes the technologies reaching their ceiling within the first 20 years
(+/-10%), while subsample B includes technologies not reaching their ceiling within the first 20
years.
Table F1: Legitimation and technological potential estimations on two subsamples
(1) (2) (3) (4)
Subsample A: Technologies reaching their
ceiling within the first 20 years
Subsample B: Technologies NOT reaching their
ceiling within the first 20 years
68% of the cases 32% of the cases
Legitimation (10%) Technology potential Legitimation (10%) Technology potential
Component characteristics
Similarity 3 digits (dummy) -1.23*** -14.1*** -1.07*** -11.5*
log(Science-based content) 0.22*** 16.8*** 0.29*** 17.3***
log(Familiarity) -0.11*** -4.56*** -0.24*** -10.2***
log(Appropriability) -0.70*** -12.6* -1.01*** -21.6*
log(Concentration) -0.12** 4.17** 0.58*** 29.5***
Inventor’s characterisics
Inventors' experience (dummy) -0.23*** -2.41 -0.05 3.88
log(Inventors' team size) -0.076 5.44*** -0.063 4.54
Applicant characteristics
log(Applicants' experience) -0.0065 0.74* 0.00016 0.69
University applicant (dummy) 0.2 8.67** 0.17 14.6*
More than one applicant (dummy) -0.83*** 4.34 -0.94*** 5.96
Dummy Technology class (3 digits) Yes Yes Yes Yes
Dummy Applicant’s country Yes Yes Yes Yes
Dummy technology entry year Yes Yes Yes Yes
Constant 9.69*** 149*** 16.1*** 330***
Observations 7,540 7,540 3,469 3,469
R-squared 0.26 0.133 0.267 0.257
53
Appendix G: OLS estimations restricting the sample to novel technologies having a diffusion
curve for which the large majority of the patented inventions (at least the 90%) is mono-
technology (3,221).
(1) (2)
Legitimation (10%) Technological potential
Component characteristics
Similarity 3 digits (dummy) -1.91*** -33.2***
log(Science-based content) -0.042 14.1***
log(Familiarity) -0.17*** -7.50***
log(Appropriability) -1.84*** -43.6***
log(Concentration) -0.35*** 9.12**
Inventor’s characterisics
Inventors' experience (dummy) -0.43*** -1.01
log(Inventors' team size) -0.11 8.98***
Applicant characteristics
log(Applicants' experience) -0.016 1.38*
University applicant (dummy) -0.011 7.33
More than one applicant (dummy) -0.65*** 10.3*
Dummy Technology class (3 digits) yes yes
Dummy Applicant’s country yes yes
Dummy technology entry year yes yes
Constant 16.1*** 368***
Observations 3,221 3,221
R-squared 0.444 0.196
54
Appendix H: OLS estimations based on a restricted sample of 8,557 novel technologies
having only one patented invention the first year when they appear.
(1) (2)
Legitimation (10%) Technology potential
Component characteristics
Similarity 3 digits (dummy) -1.31*** -13.6***
log(Science-based content) 0.18*** 16.6***
log(Familiarity) -0.14*** -5.68***
log(Appropriability) -1.02*** -24.8***
log(Concentration) 0.080 9.97***
Inventor’s characterisics
Inventors' experience (dummy) -0.16** -2.48
log(Inventors' team size) 0.078 2.71*
Applicant characteristics
log(Applicants' experience) -0.0018 0.48
University applicant (dummy) 0.072 5.30
More than one applicant (dummy) -0.18 -6.32
Dummy Technology class (3 digits) Yes Yes
Dummy Applicant’s country Yes Yes
Dummy technology entry year Yes Yes
Constant 9.69*** 149***
Observations 8,557 8,557
R-squared 0.227 0.139
55
Appendix I: 2-step Heckman selection model
In the main text we consider the diffusion of successful novel technologies and, as a consequence,
we limit our observations to all the novel combination that reach a minimum level of diffusion. If
we eliminate this restriction and we consider all the novel technologies, our study sample includes
191,627 observations. However, only 11,009 technologies diffuse. Not including in our analysis
of the diffusion determinants the 180,618 (191,627-11,009) novel technologies that do not diffuse,
may generate a selection bias problem. To correct for the selection bias associated with the loose
definition of novelty we adopt a 2-step Heckman estimation strategy. In the first step, we assume
that the determinants of diffusing or not depends only on the characteristics of the patented
inventions embedding the novel technology at the first year of its appearance. In table I1, we
estimate a Probit model where the dependent variable is a dummy that equals 1 if the novel
technology diffuses, 0 otherwise. The independent variables are the characteristics of the patented
inventions during the first year when the technology appears. In the second step, we estimated a
model explaining legitimation and technological potential including as explanatory variables the
characteristics of the combined technological components and add the Inverse Mill’s ratio to
correct for the selection bias. Table I2 reports the results of our estimations without and with the
correction for the selection bias, respectively columns 1-2 and 3-4.
56
Table I1: Heckman first step, probability of a novel technology diffusion
(1)
Probit
VARIABLES Diffusion
Inventors' experience (dummy) -0.021**
log(Inventors' team size) 0.090***
log(Applicants' experience) 0.013***
University applicant (dummy) 0.23***
More than one applicant (dummy) 0.14***
Constant -1.81***
Observations 191,627
Dummy application year Yes
Dummy technology class (3 digits) Yes
Dummy applicant's country Yes
Pseudo-R2 0.034
Table I2: Heckman second step, estimation of legitimation and technological potential
equations correcting for selection bias
(1) (2) (3) (4)
OLS OLS OLS OLS
VARIABLES
Legitimation
(10%)
Technological
potential
Legitimation
(10%)
Technological
potential
Similarity 3 digits (dummy) -1.28*** -14.2*** -1.24*** -15.2***
log(Science-based content) 0.22*** 19.8*** 0.27*** 18.3***
log(Familiarity) -0.18*** -6.78*** -0.16*** -6.95***
log(Appropriability) -1.06*** -25.3*** -0.99*** -26.0***
log(HH-index) 0.036 11.2*** 0.043 11.0***
Inverse Mill's ratio
(nonselection hazard) 1.94*** -53.0***
Constant 12.4*** 264*** 8.68*** 363***
Observations 11,009 11,009 11,009 11,009
R-squared 0.217 0.139 0.231 0.150
Dummy application year Yes Yes Yes Yes
Dummy technology class (3 digits) Yes Yes Yes Yes
We find a substantial coherence between the results reported in Table I2, columns 3 and 4 and the
results reported in the main text in Table 2, columns 5 and 6. This allows us to claim that analyzing
successful novelty and restricting our sample to 11,009 novel combinations does not introduce any
57
selection bias on the estimated impact of the characteristics of the combined components on
diffusion.
Documents De travail GreDeG parus en 2018GREDEG Working Papers Released in 2018
2018-01 Lionel Nesta, Elena Verdolini & Francesco Vona Threshold Policy Effects and Directed Technical Change in Energy Innovation 2018-02 Michela Chessa & Patrick Loiseau Incentivizing Efficiency in Local Public Good Games and Applications to the Quantification of Personal Data in Networks2018-03 Jean-Luc Gaffard Monnaie, crédit et inflation : l’analyse de Le Bourva revisitée2018-04 Nicolas Brisset & Raphaël Fèvre François Perroux, entre mystique et politique2018-05 Duc Thi Luu, Mauro Napoletano, Paolo Barucca & Stefano Battiston Collateral Unchained: Rehypothecation Networks, Concentration and Systemic Effects2018-06 Jean-Pierre Allégret, Mohamed Tahar Benkhodja & Tovonony Razafindrabe Monetary Policy, Oil Stabilization Fund and the Dutch Disease2018-07 Pierre-André Buigues & Frédéric Marty Politiques publiques et aides d’Etat aux entreprises : typologie des stratégies des Etats Membres de l’Union Européenne2018-08 Jean-Luc Gaffard Le débat de politique monétaire revisité2018-09 Benjamin Montmartin, Marcos Herrera & Nadine Massard The Impact of the French Policy Mix on Business R&D: How Geography Matters2018-10 Adrian Penalver, Nobuyuki Hanaki, Eizo Akiyama, Yukihiko Funaki & Ryuichiro Ishikawa A Quantitative Easing Experiment2018-11 Lionel Nesta & Stefano Schiavo International Competition and Rent Sharing in French Manufacturing2018-12 Melchisedek Joslem Ngambou Djatche Re-Exploring the Nexus between Monetary Policy and Banks’ Risk-Taking2018-13 Dongshuang Hou, Aymeric Lardon, Panfei Sun & Theo Driessen Compromise for the Per Capita Complaint: An Optimization Characterization of Two Equalitarian Values2018-14 Gérard Mondello & Evens Salies The Unilateral Accident Model under a Constrained Cournot-Nash Duopoly2018-15 Stéphane Gonzalez & Aymeric Lardon Axiomatic Foundations of a Unifying Concept of the Core of Games in Effectiveness Form2018-16 Claire Baldin & Ludovic Ragni François Perroux : Echange pur contre échange composite - Controverses et enjeux de justice2018-17 Guilhem Lecouteux What Does ‘We’ Want? Team Reasoning, Game Theory, and Unselfish Behaviours2018-18 Ludovic Dibiaggio, Benjamin Montmartin & Lionel Nesta Regional Alignment and Productivity Growth
2018-19 Nicolas Brisset L’organe comme marchandise fictive : une mise en perspective 2018-20 Dongshuang Hou, Aymeric Lardon, Panfei Sun & Hao Sun Procedural and Optimization Implementation of the Weighted ENSC Value2018-21 Frédéric Marty Intelligence Artificielle et Organisation Industrielle : quels enjeux pour l’économie numérique2018-22 Michele Pezzoni, Reinhilde Veugelers & Fabiana Visentin Is This Novel Technology Going to be a Hit? Antecedents Predicting Technological Novelty Diffusion