Content uploaded by Reinhilde Veugelers
Author content
All content in this area was uploaded by Reinhilde Veugelers
Content may be subject to copyright.
In search of performance effects of (in) direct industry
science links
Bruno Cassiman, Reinhilde Veugelers and Pluvia Zuniga
DEPARTMENT OF MANAGERIAL ECONOMICS, STRATEGY AND INNOVATION (MSI)
Faculty of Economics and Applied Economics
MSI 0610
In Search of Performance Effects of (in) direct Industry
Science Links
Bruno Cassiman
1,3,4
, Reinhilde Veugelers
2,3,4
,
Pluvia Zuniga
4,5
1
IESE Business School, Universidad de Navarra, Barcelona (Spain)
2
European Commission (BEPA)
³ CEPR
4
Katholieke Universiteit Leuven
∗
5
OECD
Abstract
Using patent data from the European Patent Office combined with firm level data, we evaluate
the contribution of science linkages to the innovation performance of a firm at the patent level.
We examine the effect of i) firm level linkages to science (firms active in publication and co-
publication), and ii) invention-specific linkages (patents with citations to scientific
publications) on patent quality measures. Our results suggest that citations to scientific
publications are not significant in explaining forward citations but are positively related to the
scope of forward citations, both in terms of generality and geographical dispersion. Our main
finding is that it is the linkage to science at the firm level that matters more for forward
citations, except for patents in emerging technologies. In particular, non-science related patents
of firms with firm level scientific linkages are more frequently and more quickly cited than
comparable patents of firms without these science linkages.
Key words: Patent value, forward citation, science, industrial innovation
JEL : O32, O34, L13
∗
Faculty of Economics and Applied Economics. Steunpunt O&O Statistieken Dekenstraat 2, B-3000 Leuven,
Belgium. Tel: +32 (0) 16 32 57 93, fax: +32 (0) 16 32 57 99. Support from KU Leuven Research Fund
(OT/04/07) is gratefully acknowledged.
Acknowledgements: We are grateful to Rene Belderbos, Dirk Czarnitzki, Alfonso Gambardella, Bart Van Looy,
Yuichi Nagahara and Okazaki Teuro for remarks and suggestions, as well as participants to the EPFL Workshop
on "Technology Transfer from Universities" in Lausanne and our discussant F. Lissoni. We also thank Colin
Webb for useful comments and data provision (OECD-EPO Patent Citations).
1
1. Introduction
An important and recurrent concern in economics has been to understand to what extent
science explains technological progress. The answer to this question has profound implications
on public policy, notably on the decision to fund public research by public institutions. The
works by Jaffe (1989) and Adams (1990) have shown the importance of basic science (e.g.
public research expenditures or outputs, e.g. publications) for economic growth while research
by Acs, Audretsch and Feldman (1992) and others, have revealed the significant externalities
stemming from local academic research on private R&D and patenting.
1
More recent studies
suggest that the links to basic research by private firms have increased in the last decade and
they manifest themselves today in multiple ways: growing university-industry collaboration
(e.g. joint research, sharing of equipment and research tools) and contracting (Liebeskind et al,
1996; Darby and Zucker; 2001; Zucker et al, 2001; 2002), industry financing university
research (OECD, 2004), increasing university spin-offs and licensing (Jansen and Thursby,
2001; Thursby and Thursby, 2002; Agrawal, 2002), mobility of university researchers (Kim et
al, 2005; Geuna et al, 2005), and so forth. One of the most visible indications of growing
science linkages by industry is found in the citations to science in patents (Narin et al, 1997;
Hicks et al, 2001). For instance, Narin et al (1997) reported a threefold increase in the number
of citations to academic literature in industrial patents in the United States through the mid
1990s. Accordingly, 73% percent of the papers cited by industry patents were authored at
academic, governmental, and other public institutions.
2
These patterns evidence the increasing role played by in industry science links (ISL) in the
search for competitive advantages by private firms. In spite of these growing connections to
science our understanding of how these knowledge transmissions take place and how they
modify the innovation process by private firms still remains unclear. Due to the highly specific
nature of the know-how involved, only a select set of firms within specific industries are in fact
formally dealing with scientific know-how offered by universities or other science institutes.
For firms, science is more important as source of information for innovation in those
1
The importance of academic research for industrial innovation has also been corroborated in survey based
studies (Mansfield, 1991, 1995; Cohen, Nelson and Walsh, 2002).
2
Branstetter (2004) and Van Looy et al (2004) have also confirmed an increasing citation to academic
publications in patents.
2
technology fields where new breakthrough innovations can be achieved and transferred to new
products and processes.
Breakthrough innovations are not the only economic benefits of science for industrial
innovation. Research has shown that basic research influences in different and complex ways a
firm’s innovation process. By providing a map for research and codified forms of problem
solving (Fleming, 1997; Fleming and Sorenson, 2004), science helps firms to avoid wasteful
experimentation and focus on the most promising research paths, thereby increasing the
productivity of internal research (Evenson and Kislev, 1976; Gambardella, 1995).
4
It also
serves to expand firms’ absorptive capacity which allows firms to better screen and absorb
external information (Cockburn and Henderson, 1998). Empirical research on ISL has shown
that university-industry collaborations contribute to increase firms’ research productivity, but
that their contribution depends on firms’ research capabilities and abilities to absorb scientific
knowledge (Adams et al, 2001; Zucker et al, 2001; Zucker et al, 2002). One important
shortcoming in this line of research is the lack of evaluation of the contribution of science to
the quality of inventions and the specific mechanisms used by firms to link to scinece. This
paper fills this gap.
Using patent data from the European Patent Office combined with firm level data, we evaluate
the contribution of science linkages to the innovation performance of a firm at the patent level.
We examine the effect of i) firm level linkages to science and ii) invention-specific linkages
(citation in patents to science) on the quality of patents (as measured by forward patent
citation).
5
Our data consists of 1186 granted patents at the European Patent Office during 1995-
2001 for a sample of 79 Flemish firms, identified as innovation active through the Eurostat CIS
survey. Firms’ scientific linkages are defined through several connections of the patenting firm
with science, but are not necessarily directly related to the focal patent: i) the number of
assignees’ scientific publications and/or co-publications with universities; ii) cooperation links
with scientific institutes (as found in the Community Innovation Survey, 1998-2000); iii) use
of public sources of information (ibid). As a measure of invention-specific scientific
4
It contributes notably to overcome the difficulties attributed to the interdependency of technologies by reducing
the landscape for research and maximizing thus the probability of discovery (Fleming and Sorenson, 2005).
5
Following previous studies on patent quality (e.g. Harhoff et al, 1999; Reitzig, 2002), we use the number of
forward citations received by focal patent as an indication of technological impact of inventions.
3
connection, we use the scientific non-patent references cited by the (focal) patent (and found in
the ISI-Web of Knowledge database).
We contribute to the literature in several aspects. First, we bring the literature on industry
science links, which has focused mostly on determining who is engaged in ISL, to the level of
studying the impact of ISL on the production of patents, i.e. is there a citation premium related
to science linkage? Previous empirical research has shown that university patents, because they
rely on more fundamental knowledge, are broader in scope and cited more frequently than
private firm patents. By comparing private patents, we evaluate whether science linkages are
also valuable for firms’ by generating a citation premium.
Second, this paper attempts to expand the traditional analysis of determinants of quality of
patents. Past research has shown the highly skewed distribution of quality (e.g. forward citation,
renewal probability..) and economic value across patents within technologies (e.g. Scherer et al,
1999). As previously studied in the literature, the technological impact of patents is associated
to the nature of the invention (attributes of the patent) and explained by the individuals who
created that invention; e.g. experience or competences (Gay and Le Bas, 2003; Gambardella et
al, 2005). However, researchers have often ignored the characteristics of the firms’ or assignee
as determinants in the quality analysis of patents.
6
We claim that part of this skewed
distribution of value of patents can be explained by the heterogeneity across the patent owners,
in particular by the scientific capabilities of firms which allow them to decode advances in
fundamental knowledge, and transfer basic research into a sequence of technology applications.
These scientific capabilities of firms have been found to be equally skewed across firms, and
are therefore an interesting candidate for being matched with the skewedness in patent citations.
Both descriptive and econometric analysis based on patent forward citations lead us to the
following conclusions. First, our results suggest that citations to scientific publications
(invention linkage to science) are less relevant in explaining forward citation but do relate to
the scope of forward citation, both in terms of generality and geographical dispersion.
7
This
finding can be explained by the fact that patents citing science may contain more complex and
6
Researches have attempted to control for the nature of the organization owning the patent; e.g. public and private
organizations, universities versus corporations (Henderson et al, 1998; Scherer et al, 1999; Gambardella et al,
2005).
7
The irrelevance of citation to science in the patent could be related to the findings previously reported by
descriptive analyses that have cast doubt about the meaning of references to prior art in european patents (Meyer,
2000; Tijssen, 2001, 2002).
4
fundamental knowledge that not easily diffuses. Any potential application of this knowledge,
while indeed pioneering, is still far from the market. In contrast, a firm’s proximity to science
matters for patent quality: in particular, non-science related patents of firms with firm level
scientific linkages are more frequently and more quickly cited than comparable patents of firms
without these science linkages. This finding evidences the existence of internal spillovers
within scientific-oriented firms (knowledge transfer across inventors) allowing these firms to
write more valuable applied patents. It also suggests a process of innovation consisting in
achieving high impact inventions building on more fundamental innovations (patents with non
patent references), and transferring knowledge between inventors. In this way, this paper calls
for a broader perspective in the evaluation of determinants of patent quality and contribution of
science linkages at the firm level for innovation performance at the invention level. These
findings describe the importance of firm-level science linkages for patent quality, in general.
But when distinguishing according to the type of technology/industry, these firm-level science
linkages seem to become less important when the technologies concerned are highly evolving
technologies. In these cases, the invention-specific linkage to science seem to matter more.
The paper is organized as follows. Section 2 presents a summary of the literature and reviews
previous empirical work. Section 3 describes our data and presents the empirical model while
section 4 reports on the econometric results. The final section presents our conclusions and
identifies some policy implications.
2. Literature Background
2.1. The Growing Economic Importance of Science Links
Using a diversity of methodologies, economists have since long attempted to asses the
economic payoffs of basic research. Relying on the assumption of informational properties of
basic research (non-rival and non-excludable; Arrow, 1962; Dasgupta and David, 1994),
economists such as Griliches (1984) and Adams (1988; 1990) have shown the contribution of
basic research (e.g. public research expenditures and scientific publications) to economic
5
growth
8
This literature has demonstrated that knowledge flows from universities and public
research centres make a substantial contribution to industrial innovation and, consequently, to
public welfare. The rates of return to publicly funded research, for example, have been
estimated between 20% and 60% (Salter and Martin, 2001).
Complementary research based on survey studies, has provided alternative estimates of the
importance of basic research for industrial innovation and economic performance. For instance,
relying on a survey of 76 U.S. firms in seven industries, Mansfield (1991) found that 11% of
new product innovations and 9% of process innovations would not have been developed
(without substantial delay) in the absence of recent academic research; which represented
respectively 3% and 1% of sales. In addition, firms declared that 8% of their products were
developed with substantial input from recent academic research (6 % of process innovations).
9
Both the 1983 Yale Survey and the 1994 Carnegie Mellon Survey (CMS) of R&D have shown
the relevance of university research for innovation as conceived by managers. According to the
CMS, American firms consider publishing by universities and patenting amongst the most
important sources of knowledge to innovate Cohen et al, (2002).
10
A different perception is
found in European firms. The evidence from the Community Innovation Survey shows that
only a small fraction of innovative enterprises consider scientific information, i.e. from
universities and public research labs; as an important information source in their innovation
process. In the Eurostat-Community Innovation Survey CIS-III (1999-2000), of all reporting
innovative EU firms (excl UK) 4.5% rated universities as important sources of information,
while 68% indicated universities as not important at all. The CIS results also show the
importance of science as information source to be highly firm size and technology specific.
11
An additional indicator of the use of science constitutes the citation to scientific publications
(non patent references) in patents. Using this indicator, Narin et al (1997) revealed three
8
See Griliches (1995) for a review of the literature on estimates of private and social rates of return to private and
publicly funded R&D spending.
9
Using these figures, Mansfield estimated the rate of return from academic research to be 28%. In a follow-up
study in 1998, Mansfield found that the academic research was becoming increasingly more important for
industrial innovation: 15% of new products and 11% of new processes could not have been developed in the
absence of academic research, accounting in total for 5% of total sales. Mansfield second study also revealed that
the time delay from academic research to industrial practice has shortened from 7 years to 6.
10
The results indicate that the key channels through which university research impacts industrial R&D include
published papers and reports, public conferences and meetings, informal information exchange, and consulting,
amongst others.
11
Concentrating on surveying Europe’s largest industrial firms only, Arundel and Geuna (2004) find that public
science is amongst the most important sources of technical knowledge for the innovative activities
6
important patterns of science-linkages in the U.S. patents during the 1980s and 1990s: i) a
rapidly growing citation linkage to scientific research paper, ii) a strong national component in
citation linkage, with each country's inventors preferentially citing papers authored in their
own country, by a factor of between two and four; and iii) a highly skewed use of science
across technology fields (see also McMillan et al, 2000; Callaert et al, 2003; Van Looy et al,
2004).
12
In an attempt to disentangle the causes of these increasing linkages to science in the
U.S. patents, Branstetter (2004) and Branstetter and Ogura (2005) found that these trends are
best explained by a combination of the “changing composition” between scientific and
technology research fertility and the “changing methods of invention” associated to an
increased emphasis on the use of the knowledge generated by university-based scientists in
later years.
13
Nonetheless, their findings show that in spite of such dramatic rising, the new
technological opportunities generated by academic research are found to be overwhelmingly
concentrated in the “bionexus” area, that is, in the cross-field of biosciences and biotech-based
technologies (Branstteter, 2004; Branstteter and Ogura, 2005). This finding persistently comes
out in parallel studies (Callaert et al, 2003; Van Looy et al, 2004), suggesting in that way the
crucial role of science for new technologies and related industries experiencing technological
change (e.g. biotechnology and pharmaceuticals; information and telecommunication
technologies, etc.).
2.2. Uncovering the process through which science affects private
innovation
The available evidence indicates the wide firm heterogeneity in the importance of science links.
Several strategic advantages have been identified to explain the firm's choice on whether to
adopt or link to science. These include an increase of productivity and level of applied
research effort (Evenson and Kislev, 1976), substantial gains in overall R&D productivity
(Henderson and Cockburn, 1994; Gambardella, 1995), the development of absorptive capacity
(Cockburn and Henderson; 1998), labor cost reductions (Stern; 1999), amongst others. As they
12
For patents in biotechnology, McMillan et al (2000) also report a strong national bias in the citation patterns,
which can be explained by a more productive fertilization between national science and technology, an increase of
scientific publication activity and a renewal of research strategy by private firms, more intensively focused on the
opportunities given by open science.
13
Changes in the distribution of patenting across technologies and changes in the distribution of publications
across fields would appear to explain much more of the total variance in patent citations than does average
changes across fields in per-patent citation behavior over time (Branstetter, 2004; Branstetter and Oruga, 2005).
7
report successes and failures from basic research in a codified form of problem-solving,
science increases the efficiency of private research (Arrow, 1962; Nelson, 1982; Dasgupta and
David, 1994). The dissemination of scientific advances through open science reduces the
degree of redundant effort providing useful information about technological opportunities, new
industrial applications or re-combination of existing knowledge pieces (Sorenson and Fleming,
2004). As explained by Fleming and Sorenson (2004), science serves as a map for
technological landscapes guiding private research in the direction of most promising
technological venues avoiding thereby wasteful experimentation.
14
Further, the adoption of
open science -for instance through pro-publication incentives-(Cockburn & Henderson,1998),
helps firms to attract high quality academic researchers whose economic value might often be
higher than their actual remuneration. Stern (1999) has shown that researchers looking for
academic reputation, may want to pursue research projects leading to publications and are
therefore, prompt to accept lower salaries in exchange of permission to keep up with scientific
research. These researchers are twofold valued, they do not only imply important labor costs
reductions for the firm, but also they constitute the “bridge” (‘gatekeepers’ and “boundary
spanners”) with the scientific or academic world. In spite of such paybacks, the adoption of
science remains limited to a restricted set of firms, as the empirical evidence has detailed. Past
research has shown that adoption of science is not costless, it is highly conditional on human
capital and adoption of new organizational practices (Gambardella, 1995; Cockburn et al,
1999).
15
Research seems to confirm that, contingent upon internal absorption capacity, the difficulties
inherent to the transfer of tacit and complex knowledge often leads firms to look for
collaborative agreements with science.
16
Working jointly at the lab bench allows firms to
capture tacit upfront research, absorb spillovers and de-codification of scientific knowledge
generated by scientists at universities and public research centers, especially when tacit
14
According to Fleming and Sorenson (2004), scientific knowledge differs from that derived through ‘local’
search within the firm -which is closely related to firms’ prior research activities-, namely because the scientific
endeavour attempts to generate and test theories and fundamental ideas, whereas local search is focused on
finding new technological solutions within a predetermined pool of knowledge.
15
Cockburn, Henderson and Stern (1999) have shown the long lasting influence of “initial conditions’ (e.g. prior
involvement in scientific activities) to explain firms’ regular adoption of science in the pharmaceutical industry.
They show that firms that were engaged in science since the beginning of the sample period are the ones who are
also the most engaged today in adopting science for drug discovery.
16
As shown by Zucker et al (2001), the degree to which the scientific literature can produce such strong apparent
knowledge capture by firms depends on: a) the characteristics of tacit, complex knowledge that lead to natural
excludability (and market exclusion), and b) selection by firms of discoveries for which the degree of knowledge
capture is likely to offset sunk costs incurred in making the scientific discovery a commercial innovation.
8
knowledge is embodied in individual discovering scientists.
However, while the figures suggest
that university-industry collaboration are widely expanding, the rate of failure or dispute in this
kind of agreements remains important. Industry-university links are subject to tensions
regarding intellectual property, access and dissemination strategy of knowledge, and others,
inhibiting the chances to successfully translate scientific information into new products (Jansen
and Thursby, 2001; Thursby and Thursby, 2002
; Hall et al, 2001; Poyago-Theotoky et al,
2002
).
17
2.3. Empirical evidence on effects from science links
Mostly focused at the firm-level of analysis, the empirical literature has previously assessed the
role of scientific-connections, notably partnerships with university researchers, on firm
performance (e.g. Audretsch and Stephan, 1996; Zucker et al 1998; Cockburn and Henderson,
1998). Using university collaboration as a scientific-link, these papers coincide in the boosting
or complementary effect of cooperation on internal R&D (Adams et al, 2000), innovation
productivity and sales (Belderbos et al, 2005).
18
While they provide little explanation about the
process through which science affects private innovation, the studies relying on the patent
production function have found that science involvement and ties with academic star scientists-,
can lead to more technology (Henderson and Cockburn, 1996; Zucker et al, 2002; Cockburn
and Henderson, 1998); more “important” patents: i.e. international patents (Henderson and
Cockburn, 1994); and higher average of quality adjusted patenting (Zucker and Darby, 2001;
Zucker et al, 2002).
The work by Cockburn and Henderson (1998) has shown that not only absorption capacity
(Cohen and Levinthal, 1989; Kamien and Zang, 2000) in basic research matters but also direct
involvement into science. Using data on co-authorship of scientific papers for a sample of
pharmaceutical firms, they show that firms connected to science show a higher performance in
drug discovery and that this connectedness is closely related to the number of star scientists
17
For instance, in a survey based study on 38 Advanced Technology Projects (ATP), Hall et al (2001) found that
projects with university involvement tend to be in areas involving "new" science and therefore experience more
difficulty and delay but also are more likely not to be aborted prematurely. In a sample of 62 U.S, university
licensing officers, Jensen and Thursby (2001) find that over 75% of the inventions licensed by these universities
were in a very early, or embryonic stage. Further, 71% of the inventions licensed required cooperation between
the professor and the licensing firm in order to commercialize a product successfully (see Agrawal, 2001).
18
Lööf and Broström, (2004) have found complementarities between internal R&D and collaboration with
universities: the average R&D firm that cooperate on innovation with universities spend more money on R&D and
has a larger propensity to apply for patents compared to an almost identical R&D firm which has no such
collaboration.
9
employed by the firm.
19
Zucker et al (1998) and Darby and Zucker (2001; 2002) found that
location of top star scientists predicts firm entry into biotechnology (by new and existing firms)
both in the United States and Japan, while Darby and Zucker (2005) recently provided
evidence that firms enter nanotechnology where and when scientists are publishing
breakthrough academic articles.
20
For biotechnology in Japan, Darby and Zucker (2001) show
that collaborations between particular university star scientists and firms had a large positive
impact on firm research productivity, increasing the average firm's biotech patents by 34
percent, products in development by 27 percent, and products on the market by 8 percent as of
1989-1990.
Turning to the evaluation of scientific links at the level of inventions or patents, the little
research that exists on this, only offers a partial explanation. In a sample of 83 pharmaceutical
and biotechnology firms, Markiewicz (2004) shows that absorption capacity (R&D intensity
and publications) and co-publishing with universities alter the innovation process: both are
associated with more exploitation of published scientific research (citations to non patent
literature), shorter lag times between existing knowledge and new firm inventions. Mariani
(2003) shows that the R&D intensity and technological specialization of firms (jointly with
geographically localized spillovers), matter for the technological impact (forward citation) of
biotech and chemical patents.
21
Regarding the empirical evidence on the contribution of science-linkage (scientific non patent
reference) to patent quality, the literature is more inconclusive. One would expect that patents
relying on more fundamental knowledge would be more original and more likely to influence
different technologies. This argument has found some support in previous studies on university
patents (Henderson et al, 1999; Mowery and Ziedonis, 2002; 2002).
22
Nevertheless, the works
that have evaluated on firm patents the determinants of patent value provide more mixed
results. Research by Harhoff et al (2003), Fleming and Sorenson (2004), Sorenson and
Fleming (2004) or Markus and Reitzig (2003) show divergent results respect to the
19
Differences in the effectiveness with which a firm is accessing the upstream pool of knowledge correspond to
differences in the research productivity of firms of as much as 30%.
20
Furthermore, they report a similar pattern previously reported in biotech: breakthroughs in nanoscale science
and engineering appear frequently to be transferred to industrial application with the active participation of
discovering academic scientists.
21
Further, firms’ characteristics appear as the most significant drivers of patent quality for chemical patents.
22
Mowery and Ziedonis (2002) and Sampat, Mowery and Ziedonis (2003) have revised Hendersons’ et al (1998)
work and found university patents to consistently receive more citations than non-university patents; which
confirms the higher quality and broadness of academic inventions.
10
contribution of non patent references to patent quality. Harhoff et al (2003) found that NPRs
are informative about the technological (forward citation) and economic value (patent
opposition) of pharmaceutical and chemical patents, but not in other technical fields.
23
In a
study of US patents, Fleming and Sorenson (2004) show that having a “scientific” reference
matters for technological impact of patents but that the benefits of using science depend upon
the difficulty of the inventive problem being addressed: science only appears as beneficial
when researchers work with highly interdependent –or coupled- knowledge pieces –which
makes probability of discovery more uncertain.
3. Empirical Strategy and Variables
3.1. Variables for patent quality and science links
Our purpose in this paper is to evaluate the value of science linkages (invention and firm level)
to the quality of patents. In particular, we want to know first whether patents citing science
have a stronger technological impact, and are therefore more frequently cited. Second, we
want to asses to what extent the attributes of the firm, i.e. the scientific orientation of the firm,
contribute to explain the technological impact of inventions (patents). Following past studies
on patent quality (e.g. Henderson et al, 1998; Harhoff et al, 1999; Reitzig, 2002; 2003), we use
as our dependent variable the number of forward citations received for each patent since the
year of application (EPO-OECD, 2005). Our data on forward citation is available up to 2003
and it concerns all citations received from other European Patents and European patents with
PCT equivalent.
24
Previous research has shown that the number of citations a patent receives is
associated with its technological importance (Scherer et al, 1999) and social value
(Trajtenberg, 1990) and correlated to the renewal of patents, the estimated economic value of
23
In a study of European polymer patent opposition, Reitzig (2002) did not find significant differences between
patents reporting non patent references and those not reporting, to explain patent opposition likelihood; a similar
finding is reported by Harhoff and Reitzig (2002) for a sample of biotechnology and pharmaceutical patents.
However, an important shortcoming of these works is that they do not distinguish whether these non patent
references concern scientific publications or not.
24
Our forward citations measures are built based on the publication date of the patent application. This data comes
from the EPO-OECD Patent Citation Database (OECD, 2005). Because our last grant date is 2001, the maximum
number of years that our patents can be cited is 6 years (1998-2003) and the minimum is 3 (2001-2003). We take
into account citations received during the same year of publication. Our measure of forward citation is the total
count of forward citation made by other European patents plus citations made by European patents with WIPO
equivalents, for which the international search report is published by the World Patent Office. See Collin et al
(2005).
11
inventions and patent opposition (Lanjouw and Schankerman, 1999; Harhoff et al, 1999).
In addition to the count of forward citations, we also analyze the effects of science linkages on
nature of forward citations received. We calculate the generality index of patents across
different patent classes and the geographical scope of the citations received. The indicator for
generality is build as a Herfhindal index (Jaffe et al, 1997; Hall et al, 2001):
, where s
2
1
∑
−=
ni
i
ij
sGenerality
ij
denotes the percentage of citations received by patent i that
belong to patent class j, out of n
i
patent classes
25
. This measure has been used in previous
studies as indicative of the impact of a patent, a high generality score suggesting that the patent
presumably had a widespread impact, in that it influenced subsequent innovations in a variety
of fields (Hall et al, 2001). The index of geographical impact is built in a similar way (1-
Herfhindhal index of geographical concentration). Both measures are adjusted by the number
of groups reported (Hall et al, 2001). To measure the role of science on patent quality, we
consider the effect of the patent having a citation to a scientific publication (invention science
linkage) and the effect of the firm’s linkages to science, as measured by its participation in
open science and cooperation with universities (firm’s scientific linkage) .
• The invention science-linkage. A patent linkage to science is defined as a dummy variable
indicating whether the patent cites at least one NPR considered as a “scientific’ publication
(found in the ISI-Web of Science). The institutional features of the patent system validate
the appearance of citations to science and technology in patents. In order to be patentable, a
patent eligible invention must be shown to be both novel and non-obvious. To assess
whether an invention disclosed in a patent application satisfies these requirements, a patent
examiner reviews the prior art i.e. the search for state of- the-art technical and/or scientific
literature, embodied in references to other patents and printed publications. The cited
references both –patent and non patent–, define therefore the claims of the patent, its
specific uses and applications, methods and procedures, etc.
26
25
If a patent is cited by subsequent patents that belong to a wide range of fields the measure will be high (close to
one), whereas if most citations are concentrated in a few fields it will be low (close to zero).
26
Though examiners are officially responsible for constructing the list of prior art references against which
patentability is judged, they rely in part on applicant disclosure of the prior art submitted with the patent
application, on Information Disclosure Statements. The Non patent references (NPR) include scientific articles,
technical papers, conference proceedings, textbooks, disclosure bulletins, abstracting services.
12
While there is strong discussion about the validity of NPR as a ’causal’ and ‘direct’ link to
science in patented technology (Tijssen et al, 2000; Tijssen, 2002; Meyer, 2000), there is some
recognition of their use as indicators of interplay between science-technology (see Schmoch,
1997; Meyer, 2000). Past research has shown that the average level of non patent references is
an appropriate proxy for quantifying the relationship of a technology field to a scientific
domain (Van Looy et al, 2004; Callaert et al, 2004).
27
Researchers have argued however, that
NPR should be treated with caution (Tijssen, 2002; Meyer, 2000a, 2000b) as they hardly
represent a unidirectional direct link to science.
28
Others consider (e.g. Jaffe et al., 2000) patent
and non patent citations as a “noisy signal” of knowledge flows, with examiners adding much
of the noise.
29
As NPR and patent references are issued from the examiner revision of the prior
art, citations rarely reflect or coincide with the science used by inventors (Tijssen, 2002). This
is basically the case in the European Patent system where patent applicants are not required to
submit descriptions of the state of the art considered relevant to
the patentability of the
invention, contrary to the American counterpart (‘duty of candour’ doctrine).
30
• Firm’s scientific linkages (the indirect ISL). These are defined through various
connections of the patenting firm with science, but are not necessarily directly related to the
focal patent:
i) The cumulated number of scientific publications (found in the ISI Web of Knowledge)
by the firm (with publication dates 1990-1995). We also include a dummy indicating
whether the firm has been engaged into publication activity: takes the value of 1 if the
firm published at least one article up to the year in which the observed patent was filed
(application date).
ii) A dummy indicating whether the firm has been engaged into formal cooperation in
27
At the industry and country-level, the average intensity of science linkage in patents has been associated to
higher national technological productivity, notably in science-intensive fields (Van Looy et al, 2003).
28
Some NPR may reflect cases where technology is actually leading science, or refer to instances of reciprocal
relationships where both knowledge generation processes are intertwined.
29
Harhoff et al. (2003) advance the idea that inserted references by the examiner are frequently a response to a
broader patent scope envisaged by the applicant. Using survey based evidence from Dutch inventors, Tijssen
(2002) finds that the presence of non patent citations on patents is not related to the science dependence of an
invention itself as reported by the inventors themselves. In addition, no correlation is found between an indicator
of science-intensity of the firm and the science-linkage of patents.
30
Other limitations refer to the bias imposed by the examiners’ individual search strategy and cutting edge
technologies with no prior published scientific art (Sampat, 2005). The latter feature would imply that some
patents without –scientific- non patent references represent in fact very novel knowledge and cover new
technologies. Further, the intensity of citation to the scientific literature may be specific to examiners’
characteristics (e.g. teams of junior-senior examiners do appear to identify larger shares of prior art, ibid). Lastly,
the outcome of a patent examiners’ search is also influenced by external factors such as the availability and
treatment of the information (on line access and electronic bibliographic search tools, see Sampat, 2005).
13
R&D with universities and governmental research centers.
31
iii) A dummy indicating whether the firm considers public information a very important
source for innovation (firms scoring “3” of a scale of three for “using” scientific
information being very important).
32
Data on publications is collected from the ISI-Web of Knowledge database. Data on firms’
research strategies (e.g. engagement in research collaboration with universities and research
centers and importance of public information for innovation) come from the Belgian Third
Community Innovation Survey (1998-2000).
3.2 Hypotheses
Being aware of the noise around the use of citations to scientific references in patents, our first
research question examines whether patents citing scientific publications are different in nature
and scope of technological impact. As the invention they cover has a precedent or basis in the
prior scientific art, these patents –like academic patents-, are more likely to cover broader
inventions; and receiving therefore more citations in total but notably across different
technologies and perhaps from broader geographic areas.
33
Further, as patent citations have
been correlated to commercialization and acquisition of new technologies, a broader scope of
forward citation across technologies and countries can be interpreted as bigger opportunities
for market commercialization of patents (e.g. Sorenson and Fleming, 2004).
34
Our second research question stresses whether the technological impact of patents can be
explained by firm specific science linkages. Controlling for firm size, we argue that firms
engaged into science (as measured by their scientific publication profile) and being involved in
technology patenting, build up environments beneficial for developing inventions with high
31
Although this variable corresponds to activities reported during 1998-2001, we assume that firm’s strategies in
research, i.e. collaboration with universities and public research centers do not change radically in the previous 4
years at the last year surveyed: 1998 (1995-1997).
32
We have defined those firms “using” open science information as those declaring a three points score in the
survey (the rank is 0-3, 3 being very important).
33
Recall however that patents can be viewed as constituting an option for firms when the market scope for such
invention is not yet well established at the moment of discovery (or patent application). If the market potential is
not realized ex post, those basic patents will not be renewed later and no longer cited.
34
In their study of forward and backward patent citations for a sample of European patents belonging to CIS-
French firms, Duguet and McGarvie (2004) show that forward citation from countries other than France is
positively related to French firms’ dissemination of knowledge through contract R&D and equipment sales.
14
technological impact. Because they follow fundamental knowledge and are able to decode
scientific information, these firms are more likely to transfer advances in basic research into
potential high impact industrial applications. As shown by Cockburn and Henderson (1998)
among others, firms’ closeness to science matters to produce high quality patents. We expect
therefore a significant “firm individual” effect in the production of quality patents. We also
expect patents from scientific firms –independently from the invention-science linkage-, to be
broader in generality and geographical scope.
Our third hypothesis concerns the use of science and the nature of the firm. We argue that
“scientific firms” engaged into publications are more able and efficient in using open science;
and might therefore be more competent to develop higher quality inventions associated to the
cited scientific NPR. Therefore, once controlling for the invention-specific and firm specific
linkage, we still assume a positive impact of the “scientific firm” dummy; which would imply
that patents from scientific firms and having a linkage to science are superior in terms of
technological impact to similar patents by firms not involved in science.
Finally, we extend the analysis of patent quality determinants by including a second firm-level
linkage to “scientific” communities. We include a dummy variable indicating whether the firm
has been engaged into R&D cooperation with universities. Previous studies have shown that
having formal links to universities through cooperative agreements allow firms to better absorb
tacit and complex knowledge underlying scientific research thereby increasing the productivity
of internal R&D. We want to test whether cooperation status adds any explanatory power to
the quality of patent, once controlled for the scientific-involvement status of the firm through
its publications status. We also evaluate the interaction between cooperation status and the
scientific NPR reference.
3.3 Control Variables
Following the literature on patent quality using forward citations, we need to control for
intrinsic attributes of the patent (and technology) that may lead to a higher expected count of
forward citations. We include in our model the following variables:
15
¾ Patent citations made: Patent and non patent citations, determine the legal boundaries of
the property rights to an invention (Jaffe et al, 1993).
35
The number of backward citations
of a patent reflects the extent of technology dependence but also signals strategies to
avoid patent infringement. Some have argued that this is an indicator of cumulativeness
of the patent (Reitzig, 2002).
36
However, the empirical evidence tends to confirm a rather
positive impact of patent citations made on forward citation and patent opposition
likelihood (e.g. Harhorff et al, 2003; Reitzig, 2002).
¾ Backward citation lag: The citation lag (backward patent citations) is a measure of the
time necessary for a firm to assimilate prior technological information and to undertake
its inventions. The shorter is this time, the faster the focal patent has been built respect to
the cited patent (Fabrizio, 2004; Nagaoka, 2005). We measure this variable as the time
(years) between the publication of the cited patent application (in general, a patent cannot
be cited before it is published) and the publication date of the referencing search report
(OECD-EP Patent Citation Database, 2005).
¾ Patent scope (no. of classes and sub-classes): Following Lerner (1994), we measure
patent scope as a count of the number of international patent classes into which the EPO
assigns a patent. Research has shown that the scope of a patent may be an important
determinant of the efficacy of patent protection (extent of monopoly power) and
consequently, of economic value (see, e.g. Scotchmer, 1991). We employ two sets of
patent scope indices: one at the 4 IPC digit level and one at the 8 digit IPC level. As
explained by Lerner (1994), the more general the research content of the patent (the
broader the scope), the higher the probability to be cited by patents in different
technology classes.
37
¾ Family size: We use the number of countries where patent protection has been sought
within the EPO system for our focal patent. Putnam (1996) has shown that the number of
jurisdictions in which patent protection is sought for a particular invention is likely to be
35
Patent citations indicate the patent office assignment that a particular invention builds upon the cited knowledge
and that protection over this patent may not infringe the technological domains of the cited patents.
36
Fleming and Sorenson (2004) include the number of prior patent citations as a control for the degree of ‘local
search’. They argue that patent citations may also capture idiosyncratic differences in citation propensity that
technology class controls miss.
37
We also include a dummy for patents which only cover one IPC classification (8 digit level) to capture any
systematic difference between these inventions and those covering multiple subclasses (Fleming and Sorenson,
2004).
16
correlated with the value of the invention and thus with the value of any single national
patent right.
38
¾ Inventors’ Patenting Experience: this variable is measured as the number of patents filed
individually by each member of the team in the past (by the same or different assignee),
before the date of the focal patent application. It is assumed that inventors that have
previously and successfully searched for specific technological problems will more likely
to develop more important patents than inventors without any experience (Latham and Le
Bas, 2003; Arora et al, 2005).
39
¾ Past research has shown that the use of science and the science-linkage in patents differs
dramatically across technology classes.
40
We control for this pattern by including the
average intensity of scientific references (relative to the number of non patent references)
by technology class. Moreover, the propensity to cite other patents differs dramatically
across technology fields.
41
We control for permanent differences across industries in
expected forward citation by including dummies for 5 broad technology classes (IPC
Fraunhofer).
42
3.4. Methods and Data
We assume that the process of forward citations follows a Poisson distribution (Hausman et al,
1984), where citation is an event with relatively small probability of success:
!
)Pr(
i
y
i
i
y
e
yy
i
i
λ
λ
−
==
(1)
38
Scherer et al (1999) have shown that (international) family size and observed outcomes of opposition cases
contribute to an approximation of the patent right’s value both in terms of economic value (perceived selling price
of the patent) and forward citation.
39
Recent work by Latham et al (2003) shows that that there is a direct, positive relationship between the
involvement of a prolific (or foreign) inventor and the value of the new knowledge produced (forward citation).
40
The pharmaceutical and chemical, particularly the biotechnology classes, followed by information technologies
are amongst the principal classes reporting scientific publications within their non patent references.
41
The degree of dependence of past technology obligates the patentee to declare of the possible connections with
existing patents in order to avoid litigation; technologies such as semiconductors show consequently higher
backward citation intensity. The citation propensity over time also differs across technological areas. Citations in
Computers and Communications, come the fastest, followed by Electric and electronics (E&E), Drugs and
Medical technologies.
42
As illustrated by Hall et al (2001) in the USPTO patents, in general, the traditional technological fields cite
more and are cited less, whereas the emerging fields of Computers & Communication (C&C) technologies and
Drugs & Medical (D&M) technologies are cited much more but are in between in terms of citations made.
17
The mean value
i
λ
is parameterized by x
i
; the vector of patent and firm attributes ; and the
coefficient vector
β
: . Y
)'exp(
βλ
ii
x=
i
represents the number of forward citations received by
patent i (up to year 2003) after the publication of the patent application. There are two issues
that restrain the use of the Poisson model when dealing with skewed distributed count data.
First, the Poisson process assumes equi-dispersion, the mean and variance of parameters
i
λ
are
restricted to be equal, which is not frequently the case. Estimates of a Poisson model for
overdispersed data are unbiased, but inefficient with standard errors biased downward
(Cameron and Trivedi 1998). The Negative Binomial model (Hausman et al ; 1984 )
43
allows
for overdispersion and heterogeneity between subjects by including a unobserved (cross-
sectional) specific effect
into the
i
u
i
λ
so that :
)'exp(
iii
ux +=
λλ
(2)
It is assumed that is distributed gamma, then the integration leads to the negative binomial
model and the unconditional distribution for
can be expressed as:
i
u
i
y
1
1
1
)1(
)()1(
)(
)|Pr(
−
−
−
−
Γ+Γ
+Γ
==
α
α
α
i
y
i
i
y
i
ii
rr
y
y
xyy
i
i
(3)
Where
1−
+
=
αλ
λ
i
i
ri
. Equation (4) is the form of the negative binomial distribution with mean
i
λ
and variance )1(
ii
αλ
λ
+
for α>0. The likelihood ratio test for over-dispersion examines
the null hypothesis of α=0.
44
A second problem that arises is the over-dispersion of zeros in the data. The negative binomial
model assumes a same probability process for having a zero than any other discrete value.
Zero-inflated models (Zero Inflated Poisson ; ZIP ; and Zero Inflated Negative Binomial ;
ZINB) handle overdispersion by changing the mean structure to explicitly model the
production of zero counts (Long, 1997). The zero inflated models deal with two sources of
over-dispersion by portioning the observed variable into a qualitative and quantitative part. The
43
Both Long (1997) and Cameron and Trivedi (1998) note that the unobserved heterogeneity that can cause
overdispersion can also cause there to be “excess zeros”. In fact, Cameron and Trivedi (1998) review related work
by other authors that shows that for certain mixture models, the heterogeneity that gives rise to the overdispersion
will always raise the proportion of zeros.
44
The LR statistic follows the Chi-squared distribution with one degree of freedom. If the null hypothesis is
rejected, negative binomial is preferred to the Poisson regression.
18
ZINB model allows for "excess of zeros" in count models under the assumption that the
population is characterized by two regimes, one where members always have zero counts, and
one where members have zero or positive counts (Long 1997).
The likelihood of being in either regime is estimated using a logit specification, while the
counts in the second regime are estimated using a negative binomial specification.
45
The
probability of zero forward citation
)|0Pr(
ii
xz
=
given covariates x, can be modeled as a logit
distribution:
i
x
x
ii
e
e
XiFxz
'
'
1
)'()|0Pr(
γ
γ
γ
+
===
Then, we can write the unconditional probability of the count of forward citations as:
)1,|()|1()|0(),|Pr(
=
=
=
+
=
==
iiiiiiiiiii
zxyYPXzPXzPXxyy
or :
)|()|()'()'(),|Pr(
iiiiiiiiiii
xyYPxyYPXFXFXxyy =+=−==
γγ
This expression gives the ZINB that we will estimate. In our model the set of covariates X
i
(in
the logit model) and x
i
(in the count model) are the same; as there is no theoretical framework
defining a different set of determinants for the zero citation model and the expected count of
forward citations. The Poisson model and ZIP, and NBRM and ZINB cannot, however, be
tested by this likelihood ratio, since they are not nested respectively. The Voung’s statistic
compares these non-nested models.
46
3.5. Descriptive statistics
Table 1 reports the summary statistics of our sample. After restricting our sample of patents
with grant dates 1995-2001, in order to have at least three years for forward citation for the
patents in 2001, we ended with a total of 1186 patents for a sample of 79 firms. The average
forward citations for these Flemish patents is 1.08 citations coming from other European
patents (and EPO patents with WIPO equivalents); and 0.67 citations coming only from
45
See Long (1997) Chapter 8 and Greene (1997) for a discussion.
46
If the Vuong test is greater than 1.96, the ZIP or ZINB is favored. If the test is less than -1.96, the PRM or
NBRM is preferred (Long, 1997). The Vuong statistic is as calculated as :
⎥
⎦
⎤
⎢
⎣
⎡
=
)|(
)|(
2
1
ii
ii
i
Xyf
Xyf
LNm
where
and are the probability density functions of the ZINB and ZIP respectively. In our
case, the Vuong test yield a value of 5.59 with (Pr>Z=00) justifying our choice of the ZINB over ZIP.
)|(
1 ii
Xyf )|(
2 ii
Xyf
19
European counterparts. The unconditional standard deviation for the two measures of forward
citations is higher than the mean suggesting over-dispersion of data. Table 2 display the
distribution of patent science linkages across technology classes. The technology reporting the
highest percentage of citations to science is, not surprisingly, chemistry and pharmaceuticals
followed by electrical engineering. Process engineering and special equipment report however
the highest number of firms having scientific NPR in their patents. This technology also
reports a higher number of firms being involved in scientific publication. Table 3 displays the
number of patents by firm level science-linkage. Patents from scientific firms represent in total
64.50% of total patents. While patents from firms that cooperate with universities and public
research institutions is 82%. A similar high percentage of patents (62%) belongs to firms that
consider public sources of information (journals, proceedings; etc.) as a very important source
for innovation.
Table 4 displays t-student tests on the comparison of means for some variables of our interest.
We want to know whether the characteristics of patents (generality, number of IPC classes,
geographical dispersion, backward citation time) and their impact on subsequent technology
development (forward citations) differ between patents with and without scientific linkages
(both at the invention and at the firm level). In panel A and B, we compare patents that did not
have any citation to science to patents having at least one scientific NPR. There are no
significant differences between the two groups in terms of forward patent citations. However,
there are significant differences in terms of scope (number of sub-sub classes IPC-4) and
generality, with patents with scientific NPR displaying a larger scope and generality.
Panel B and C compare patents between firms involved in scientific publication and the rest of
firms (patents). Scientific firms have significantly more forward citations (both total and
relative (i.e. compared to the average in the same IPC class)), are broader in terms of no. of
IPC classes, and report a higher geographical impact. The panels D to G split firms’ patents
into patents -without and with- a linkage to science (scientific NPR) for each group of firms.
The panel D-E displays the patents for the group of firms with publications. It turns out that the
patents having a scientific NPR are only different in terms of generality (higher). Looking at
the panel F-G, in the group of patents from firms not engaged into science there are no
significant differences between NPR and non-NPR patents. The same holds for the panel G-E
which compares patents with NPR for the two types of firms.
20
However, the panel F-D comparing patents without NPR across the two types of firms, finds
significant differences. Non-NPR Patents from firms that are involved in science report higher
forward citation and intensity, they cover a large number of IPC classes, and have a higher
geographical impact but are not necessarily more general than non-NPR patents from firms’
not engaged into science. These results provide some first evidence, although partial, about the
importance of firm specific linkages rather than invention specific.
4. Econometric Results
4.1. The impact of science linkages on forward citation
Table 5 displays the step by step regressions using negative binomial and zero-inflated
negative binomial (ZINB) count regression. The goodness of the fit across the different
specifications is reported at the bottom in addition to the Vuong tests discriminating between
ZINB and negative binomial, and ZINB and zero inflated poisson. We include a set of 4
technology dummies (the reference being Mechanical and engineering technologies), and two
period dummies in all models. Robust standard errors are reported and adjusted for intra-group
correlation of errors (clustered by firm). The ZINB specifications split into the count model
(conditional on having a positive and superior to zero forward citation) and logit model
(probability of zero forward citation).
47
Before discussing the variables of interest, we first briefly discuss the fitness of the different
estimation methods. The Negative Binomial model is preferred over the Poisson model as
confirmed by the test of over-dispersion. These first estimations corroborate the previous
findings reported in the t-tests comparison of means. According to the estimates of the negative
binomial, patents having a citation to a scientific publication are not significantly different
from other patents in the expected technological impact; they do not receive more forward
citations. The convergence of the model improves slightly with the inclusion of the scientific-
firm dummy. Similar results are corroborated in the with the ZINB model. The Vuong test
corroborates the superiority of the Zero Inflated Negative Binomial over the standard negative
47
Logit coefficients can be interpreted like the normal logistic coefficients. In the logit part of the model,
coefficient with a negative sign mean that changes in the related variable are inversely related to the likelihood of
belonging to the zero citation group. In principle, the coefficients in the discrete part of the model (logit of zero
citation) are expected to be the opposite of the expressed in the count model.
21
model (5.99 at 1%).
48
Further, the likelihood ratio test comparing the ZINB to the ZIP
suggests the preference of the former. Thus, two sources of dispersions exist in our data; one
arising from the high number of zeros, and a second one coming from the between subject
heterogeneity.
49
Three main findings arise from the ZINB regressions. First, contrary to our expectations, but
confirming our findings on comparison of means, patents citing science seem not to have a
higher citation impact. The science linkage reflected in the citation to scientific publication
status is never significant across the different models. There is no relationship between citing a
scientific NPR and the expected count of forward citations even when correcting for over-
dispersion and heterogeneity. Second, the size of the firm (no. of employees) appears not to be
relevant to explain patent quality (columns 5 and 6) in the count part of the model but it is
significant in the logit model
50
: firm size is negatively and significantly associated to the
likelihood of not receiving a forward citation at all.
51
Recall that some theory (Schumpeter;
1945); argues that big firms might be more capable of producing high quality technology,
because of scale economies; advantages in accessing upfront knowledge in the market. And
third, the dummy identifying scientific firms (i.e. those engaged in scientific publication)
appears as positive (column 7), which confirms the hypothesis of a firm’ specific effect for
patent quality. Conditional on receiving a citation, the estimate of column 7 suggests that being
a firm engaged in open science increases the expected count of forward citation by 30%.
Expressed in factor changes, being a scientific firm increases the expected rate of forward
citation by a factor of 1.3.
52
The scientific firm dummy variable is however not significant in
explaining the probability of receiving forward citations.
48
The null hypothesis of the dispersion parameter (alfa) equal to zero is rejected at 1% (chi2(1)=99.72); which
confirms the overdispersion of our dependent variable; and justifies the adoption of negative binomial model over
a poisson specification.
49
As there is not underlying theoretical model distinguishing this probability from the expected count of citations
we cannot assume an equivalent direct economic interpretation as in the count model. The use of the ZINB model
attempts to correct for part of the “noise” characterizing the skewed distribution of citations associated to the
over-dispersion of data and the intrinsic heterogeneity; as well as the excess of zeros.
50
We also included information on firm R&D expenditures from the CIS-survey, but the data on this variable are
very noisy and do not generate robust results.
51
This result differs from the one reported by Gambardella et al (2005) on a sample (survey based) study on EPO
patents for France, UK, Germany, Italy, Netherlands and Spain. They found that small and medium firms report
higher economic value as conceived by the first or second inventor.
52
A log-likelihood ratio comparing the unrestricted versus restricted model gives a chi2(1)=6.48 with
Prob>chi2=0.032, while a Wald test on the significance of the dummy gives chi2(1)=5.86 with Prob>chi2=0.052
22
With respect to the other control variables, our results show that the higher the number of
backward patent citations, the larger the probability of not receiving a citation. Thus patents
that rely more intensively in previous technologies are of a more cumulative or incremental
nature. Further, patents that report a higher backward citation lag between the date of
publication and the backward cited patents, are less likely to be found in the zero-citation
group. Furthermore, patent quality is negatively associated to patents having a single IPC
technology class (4 digit IPC classes); that is, very narrow patents shows a lower expected
count of forward citations; although the effect is not significant. The results on the number of
jurisdictions covered in the European community (EPO) seem to indicate that inventions that
are filed internationally (although at the regional level) are more likely to cover quality patents.
In addition, patent quality appears positively associated to the amount of past patenting by
inventors. Experience enables inventors to develop more complex technologies, learn from
past research; and improve quality of their research.
We want to elucidate what is behind the higher forward citation reported in patents without
NPR in the previous descriptive analysis. The possible explanation for such patterns would be
the existence of internal spillovers -knowledge transfers across inventors - that could be the
link between patents with and without basic knowledge. Patents without NPR could be
inventions having high qualified inventors that have come up with successful technological
applications of existing fundamental knowledge. The very first patents with scientific NPR
might be too complex or representing technologies at very embryonic stage to already receive a
higher citation frequency. We investigate this issue in the Table 6. The columns 1 and 2 report
a ZINB model only in the sample of patents without NPR. The scientific-firm dummy appears
significant, although with a similar impact as reported in the total sample. This finding
suggests that even in the absence of invention-specific science linkages, firms that are formally
involved into scientific activities (open science) are capable of producing higher impact patents.
It also suggests that the achievement of higher quality inventions by scientific firms follows a
process of breaking up technology building on more fundamental innovations (patents with
non patent references). The rest of the model display coefficients close to those from the total
sample model.
Columns 3 and columns 4 report the ZINB model for the sample of patents in electrical
engineering technologies, instruments, chemicals and pharmaceuticals, i.e. those technologies
that typically are considered as science-intensive. Contrary to the regressions on the total
23
sample, the quality of inventions measured by the count of forward citations in these science
intensive industries appears to be largely influenced by individual science linkage: being a
firm engaged in open science increases the expected count of forward citation by more of 45%.
Expressed in factor changes, being a scientific firm increases the expected rate of forward
citation by a factor of 1.46. This finding is in line with the descriptive studies highlighting the
preponderant importance (frequency) of science linkages in emerging technology fields (Hicks
et al, 2000; Branstteter, 2004; Callaert et al, 2003). The firm-level science linkage however
appears as an irrelevant factor of patents’ technological impact. We repeat the exercise of
restricting the analysis to the sample of patents without NPR now for the science intensive
technologies (columns 5 and 6). Again, it turns out that the scientific firm dummy is not longer
important to explain technological impact of inventions. Hence, these findings suggest that the
firm-level linkages to science matter for patent quality in general, but that they are however
less important when the technologies concerned are highly evolving technologies. Likewise,
the invention-specific linkage to science seem to matter for more science based technologies
where breakthroughs are more frequently of taking place out of firms’ laboratories --more
likely to emerge at universities and public research centers-. We should take these results
however with caution as our sample represents just a small part of the picture of total patenting
in these technologies and our universe of firms is rather small.
A last exercise in this section, assessing to evaluate the contribution of ISL in patent quality,
consists in testing the other firm-level science linkages reported in the CIS. For purposes of
saving space, table 7 displays only the count part of the ZINB models and presents sequentially
the different ISL linkages. Column 2 reports the baseline model with the dummy on firms
considering public information sources as a very important channel for innovation. The
coefficient and its significance are not so much different from the scientific-firm dummy and
the tests on the goodness of fit do not change dramatically. This finding validates therefore the
use of the CIS variable (use of public information) as a good predictor of involvement in
scientific activities by firms. Given our small sample of firms and multicolinearity between
these two variables, we are not however successful to isolate the effects of these two variables
(column 4) on patent quality. We include the dummy on collaborating firms (with universities
and public institutions) in the column 3. Contrarily to previous findings, patents from firms that
cooperate in R&D with universities, do not generate any significant effect neither on the
likelihood nor on the count of forward citations. The model in column 5 includes the different
24
firm-level ISL. Again, given the limitation of our data the regressions are unsatisfactory to
properly assess the individual effects of the different firm-level linkage measures to science.
Lastly, the right part of the table 7 reports interactive terms in order to identify citation
premiums related to comparative advantages in the use of science by scientific-oriented firms.
None of the interactions appeared as significant drivers of technological impact of patents.
53
4.2. The impact of science linkages on the scope of forward citation
The last part of our empirical analysis evaluates our research questions with respect to the
scope of forward citation for patents and the impact of science linkages. Table 8 displays
regressions on the generality and geographical index as well as on the speed of forward
citations. The Herfindahl indices for generality across technology classes and countries are
estimated using Tobit regression (1-2 and 5-6). We have also estimated the probability of
receiving a forward citation in a different IPC 4 digit class and the probability of receiving a
citation by a foreign country using Probit models (models 3-4 and 7-8 respectively).
Regressions on the median forward lag and shortest citation lag are made with Ordinary Least
Squares. Marginal effects are reported for the Probit models.
Estimates in the Tobit model (1 and 2) show that patents having at least one citation to
scientific publications are indeed broader in scope: patents having a scientific reference have a
28% larger generality index. Furthermore, NPR patents have a 16% higher probability of
receiving a forward citation in a different IPC-4 technology class (model 3), holding all the
other variables constant at their sample means. The dummy variable for scientific firm is not
significant to explain neither the generality scope of patents being cited (columns 1 and 2) nor
the geographical impact of patents (columns 5 and 6). No additional effect is detected on the
interactive terms neither in the generality and geographical dispersion models nor in the Probit
models for different IPC class citations. Nevertheless, the estimation of the Probit model
(model 8) on the probability of receiving a citation from a foreign country appears to confirm
the wider international diffusion of patents when the two indicators of scientific linkages are
53
This result might seem to be in contradiction with most of the past literature at the firm level that has positively
evaluated cooperative links with academia on firms’ innovation performance (e.g. Darby and Zucker, 2001;
Zucker et al, 2002). Recall that the effects we are analyzing here are at the level of the patented invention and
therefore not directly comparable with firm level performance evaluations.
25
included jointly with their interaction term; the effects are however small. Lastly, the ordinary
least squares regression on the median citation time and the shortest citation time, reveals that
patents from firms with scientific publications report shorter forward citation lag (both in the
median and shortest time models); This result confirms the speed of diffusion related to
inventions coming from scientific firms.
VI. Conclusions
This research has attempted to bring new evidence on the contribution of science linkages. In
particular, this research has analyzed for a sample of Flemish patents the role of invention-
specific science linkages and firms-closeness to science on the quality of patents, as measured
by the number and scope of forward citations. .
Our results are summarized as follows. First, the descriptive and econometric investigations
show that references to scientific publications are not relevant to explain forward citation. This
finding can be explained by the observation that patents citing science may be uncovering
extremely complex and fundamental knowledge that is not easy to diffuse or may be yet far
from the market application. However, the citation to science is positively related to the scope
of technological impact; both in terms of generality and probability of being cited by a foreign
country. Second, we find the invention-specific linkage to science, measured by references in
the patent to a scientific publication, to appear only as a significant driver of patent quality in
the science based technologies but this result needs still to be verified in a larger sample. And
third, firm’s closeness to science matters for patent quality: in particular, non-science related
patents of firms with firm level scientific linkages are more frequently and more quickly cited
than comparable patents of firms without these science linkages. This finding supports the
existence of internal spillover within scientific-oriented firms (inventors’ collaboration) and
suggests an innovation process consisting in achieving high impact technology building on
more fundamental innovations (patents with non patent references). In this way, this paper calls
for a broader perspective in the evaluation of determinants of patent quality and the study of
science linkages and innovation performance.
Before drawing any firm conclusions, the results presented here are suggestive of further
research. The Flemish data set is limited in terms of the number of firms active in innovation,
26
patents and particularly science links as measured through publications. But the methodology
used in this paper, relying on a combination of internationally standardized datasets, can easily
be replicated for other countries to check robustness of results on a larger sample of active
firms. The small number of firm cases of science linkages in the Flemish data provides
nevertheless an opportunity to examine in more detail, the mechanisms in place in these firms
to realize intra-firm spillovers. A fruitful avenue for further work is to examine the patterns of
self-citations across scientific and non-scientific patents of these firms. But also the pattern of
core inventors & teams across firm patents needs to be brought in the analysis as a likely
mechanism for intra-firm spillovers. In any case, the level of inventor-specific science
linkages should be integrated with the invention- and firm-specific level to complete the
analysis.
References
Acs, Z. J., Audretsch, D. B. and Feldman, M.P. (1992) Real Effects of Academic Research.
American Economic Review, 82, 363–367.
Adams, J. D. (1990). Fundamental stocks of knowledge and productivity growth, Journal of
Political Economy, 98, 673-702.
Ahuja G., Katila R, (2004). Where do Resources Come From? The Role of Idiosyncratic
Situations. Strategic Management Journal, 25: 887–907.
Arora, A., and Gambardella, A. (1990). Complementarity and External linkages: the strategies
of the large firms in Biotechnology, Journal of Industrial Economics, 38, 361-379.
Arrow, K. J. (1962). Economic welfare and the allocation of resources for invention. Richard R.
Nelson, ed. The rate and direction of inventive activity, Princeton University Press, Princeton.
Arundel, A. and Geuna, A. (2004).Proximity and the Use of Public Science by Innovative
European Firms',
Economics of Innovation and New Technology, Vol. 13, pp559-580, 2004.
Audretsch, D.B. and Stephan, RE. (1996) Company-scientist locational links: the case of
biotechnology, American Economic Review, 86: 641-652.
27
Beath, John and
Donald S. Siegel (2002) Universities and Fundamental Research: Reflections
on the Growth of University-Industry Partnerships. Oxford Review of Economic Policy, 2002,
vol. 18, issue 1, pages 10-21
Branstetter, L. (2004), Exploring the Link Between Academic Science and Industrial
innovation, unpublished working paper.
Branstetter, L. and Yoshiaki, Ogura (2005). Is academic science driving a surge in industrial
innovation? Evidence from patent citations. NBER working paper 11561, August.
Callaert, Julie, Van Looy, Bart , Verbeek Arnold, Debackere Koenraad, Thijs, B. ( 2004)
Traces of prior art: An analysis of non patent references found in documents. KU Leuven,
mimeo.
Cameron C, Trivedi P. 1998. The Analysis of Count Data. Cambridge University Press:New
York.
Cohen W. M, Levinthal DA. (1990). Absorptive capacity, a new perspective of learning and
innovation. Administrative Science Quarterly 35: 128-152
Cohen W. M, Levinthal, DA. (1989). Innovation and learning: the two faces of R&D. The
Economic Journal 99: 569-596.
Cassiman, B. & R. Veugelers (2002), R&D Cooperation and Spillovers: some empirical
evidence from Belgium, American Economic Review, 92, 4, 1169-1184.
Cockburn, I., Henderson, R. and Stern, S. (1999).
The Diffusion of Science-Driven Drug
Discovery: Organizational Change in Pharmaceutical Research
, NBER document de travail no.
7359, National Bureau of Economic Research, Inc.
Cockburn, Iain and Rebecca Henderson, (1998). The Organization of Research in Drug 38,
Discovery, Journal of Industrial Economics, Vol XLVI, No. 2.
Cockburn, Iain, Rebecca Henderson, and Scott Stern (1999), The Diffusion of Science Driven
Drug Discovery: Organizational Change in Pharmaceutical Research, NBER Working Paper
No. 7359.
Cohen, Wesley M.; Nelson, Richard; Walsh, R. John P. (2002). Links and Impacts: The
Influence of Public Research on Industrial R&D, Management Science, Vol. 48, No. 1, January
2002 pp. 1–23
28
Darby, M. and Lynne, G. Zucker (2005) Grilichesian breakthrougus: Inventions of methods of
invention and firm entry in nanotechnology, Annales d’Economie et Statistique, forthcoming.
Darby, M. R. and Lynne, G. Zucker (2002), Going public when you can in biotechnology.
NBER working paper, no. 9854.
Darby, M. R., L. G. Zucker (2001). Change or die: The adoption of biotechnology in the
Japanese and U.S. pharmaceutical industries. Res. Tech. Innovation, Management, Policy 7:
85–125.
Duguet Emmanuel & Megan MacGarvie (2005) How well do patent citations measure flows of
technology? Evidence from French innovation surveys, Economics of Innovation and New
Technology, Taylor and Francis Journals, vol. 14(5), pages 375-393, July.
Evenson, Robert E., and Yoav Kislev, A Stochastic Model of Applied Research, Journal of
Political Economy 84 (April 1976): 265-281.
Fleming, L., and Sorenson, O. (2004) Science as a map in technological search. Strategic
Management Journal, 25, pp. 909-9280.
Gambardella, A. (1994). The changing technology of technical change: General and abstract
knowledge and the division of innovative labor. Research Policy, 23, 523-532.
Gambardella, A., Harhorff, D., Verspagen, B. (2005). The value of patents. Bocconi University,
mimeo.
Greene W H. 2000. Econometric Analysis. Prentice Hall: Upper Saddle River
Hall B. H, Jaffe AD, Trajtenberg M. (2001). Market Value and Patent Citations: A First Look,
Economics Department Working Paper E00-277, University of California
Hausman J, Hall BH, Griliches Z. (1984). Econometric models for count data with an
application to the patents-R&D relationship. Econometrica 52(4): 909-938
Guellec, D. and Van Pottelsbergue de la Potterie, B. (2000). Applications, grants and the value
of patents, Economic Letters 69 (1).
Guellec, D. and Van Pottelsbergue de la Potterie, B. (2001) The internationalization of
technology analysed with patent data, Research Policy, 30 (8), 1256-1266.
29
Hall, B. H., Jaffe, A., Trajtenberg, M., (2000). Market Value and Patent Citations: A First
Look. NBER, Cambridge, MA, mimeo.
Harhoff, D., Reitzig, M., (2004). Determinants of Opposition Against EPO Patent Grants: The
Case of Pharmaceuticals and Biotechnology, International Journal of Industrial Organization,
22/4, 443- 480.
Harhoff, Dietmar & Scherer, Frederic M. & Vopel, Katrin, (2003).
Citations, family size,
opposition and the value of patent rights
, Research Policy, Elsevier, vol. 32(8), pages 1343-
1363.
Harhoff, Dietmar and Narin, Francis, F. M. Scherer and Katrin Vopel (1999).
Citation
Frequency And The Value of Patented Inventions
, The Review of Economics and Statistics,
MIT Press, vol. 81(3), pages 511-515
Harhoff, Dietmar and Reitzig, Markus, (2002).
Determinants of Opposition Against EPO
Patent Grants - The Case of Biotechnology and Pharmaceuticals
, CEPR Discussion Papers
3645, C.E.P.R. Discussion Papers
Henderson, R., A. Jaffe, M. Trajtenberg (1998), Universities as a source of commercial
technology: A detailed analysis of University patenting, 1965-1988. Review of Economics and
Statistics, 65, 119127.
Hicks, D., Breitzman, T., Olivastro, D., Hamilton, K. (2001). The changing composition of
innovative activity in the US a portrait based on patent analysis. Research Policy, 30, 2001,
681–703.
Jaffe, A. B., Trajtenberg, M. Henderson, R. (1993). Greographic Localization of knowledge
spillovers as evidenced by patent citations. Quaterly Journal of Economics 108, 577-598.
Jaffe, A., (1989), The Real Effects of Academic Research, American Economic Review, 79 (5),
pp. 957-70.
Jensen, Richard, and Marie Thursby (2001). « Proofs and Prototypes for Sale : The Licensing
of University Inventions ». American Economic Review, 91(1) : 240-59.
Kim, Jinyoung; Lee, Sangjoon John and Marschke, Gerald (2005). The influence of university
research on industrial innovation NBER working Paper 11447, june 2005.
Lanjouw, J. O., Schankerman, M. (1999). The Quality of Ideas: Measuring Innovation with
Multiple Indicators. NBER, Boston, MA.
30
Latham, W., Gay, L. and C. Le Bas (2003).
Collective Knowledge, Prolific Inventors and the
Value of Inventions: An Empirical Study of French, German and British Owned U.S. Patents,
1975-1998
. University of Delaware, department of Economics. Mimeo.
Lerner, J., (1994). The importance of patent scope: an empirical analysis. RAND Journal of
Economics 25 (2), 319–333.
Liebeskind JP, Oliver AL, Zucker L, Brewer M. 1996. Social networks, learning, and
flexibility: sourcing scientific knowledge in new biotechnology firms. Organization Science
7(4): 428-442
Mansfield, Edwin (1991). Academic Research and Industrial Innovation. Research Policy, vol.
20(1), pages 1-12Mansfield 1992
Mansfield, E., (1995), Academic Research Underlying Industrial Innovations: Sources,
Characteristics, and Financing,” The Review of Economics and Statistics 77: 55-65.
Markiewicz, Kira, (2004), Absorptive Capacity and Innovation: Evidence from Pharmaceutical
and Biotechnology Firms, working paper, UC-Berkeley.
Meyer, M. (2000). Does science push technology? Patents citing scientific literature. Research
Policy 29, 409–434.
Mowery, D. C., B. N. Sampat and Ziedonis, A. A. (2001), Learning to patent: institutional
experience, learning, and the characteristics of US university patents after the Bayh-Dole Act,
1981-1992’, Management Science 48(1): 73-89.
Nagaoka, Sadao (2005), Patent quality, cumulative innovation and market value: Evidence
from Japanese firm level panel data. Hitotsubashi University, mimeo.
Narin, F. and R. P. Rozek (1988) ‘Bibliometric Analysis of US Pharmaceutical Industry
Research Performance’, Research Policy, 17: 139-15.
Narin, F., Hamilton, K., Olivastro, D., (1997). The increasing linkage between US technology
and public science. Research Policy 26, pp. 317–330.
Nelson, R. R. (1982). The role of knowledge spillovers in R&D efficiency. Quaterly Journal
of Economics 97, 297-306.
Nesta, L. and Vincent, M. (2005). The dynamics of innovation networks. SPRU Electronic
Working Paper Series 114, University of Sussex, SPRU.
31
Poyago-Theotoky, J. Beath and D.S. Siegel J. (2002). Universities and fundamental research:
reflections on the growth of university-industry partnerships, Oxford Review of Economic
Policy, 18 (1), pp. 10–21.
Reitzig, M. (2003), What do patent indicators really measure? A structural test of novelty and
inventive step as determinants of patent profitability, LEFIC WP 2003-1.
Reitzig, Markus (2002). Improving Patent Valuation Methods for Management, Validating
New Indicators by Understanding Patenting Strategies Lefic Working Paper 2002-09.
Rosenberg, N. (1990). Why do firms do basic research (with their own money)? Research
Policy, 19, 165-174.
Sorenson, Olav and Fleming, Lee (2004). Science and the diffusion of knowledge, Research
Policy, 33, pp. 1615-1634.
Stern, Scott (1999). Do Scientists Pay to Be Scientists?, NBER Working Papers 7410, National
Bureau of Economic Research, Inc.
Thursby J. G. and M.C. Thursby, (2002) Who is selling the Ivory Tower? Sources of growth in
university licensing. Management Science, 48, 90-104.
Tijssen, R. (2001). Global and domestic utilization of industrial relevant science: patent
citation analysis of science-technology interactions and knowledge flows. Research Policy 30:
35-54.
Tijssen, R. J. W. (2002). Science dependence of technologies: evidence from inventions and
their inventors. Research Policy 31 (2002), 509-526. 6.
Trajtenberg, M. (1990), A penny for your quotes: patent citations and the value of innovation
RAND Journal of Economics, 21 (1), 172-187.
Van Looy, Bart; Magerman, Tom; Debackere, Koenraad (2004). Developing technology in the
vicinity of science: An examination of the relationship between science intensity and
technological productivity within the field of biotechnology. KU Leuven, mimeo.
Veugelers, R. and Cassiman, B. (2005). R&D Cooperation between firms and universities:
Some empirical evidence from Belgian manufacturing. Forthcoming International Journal of
Industrial Organization.
32
Webb, C., Dernis, H. Harhoff, D. Hois, K. (2005).
Analysing European and International
Patent Citations: A Set of EPO Patent Database Building Blocks
, OECD Science, Technology
and Industry Working Papers
2005/9, OECD.
Zucker,. Lynne G.; Darby Michael R & Armstrong, Jeff S. (2002).
Commercializing
Knowledge: University Science, Knowledge Capture, and Firm Performance in Biotechnology,
Management Science, 48(1), 2002
Zucker, Lynne G and Darby, Michael R, (2001). Capturing Technological Opportunity via
Japan's Star Scientists: Evidence from Japanese Firms' Biotech Patents and Products
, The
Journal of Technology Transfer
, Springer, vol. 26(1-2), 37-58.
Zucker, Lynne, Michael Darby, and Michael Brewer (1998), Intellectual Capital and the Birth
of U.S. Biotechnology Enterprises, American Economic Review, 88, 290-306.
Zucker, Lynne, M. Torero. (2000). Determinants of embodied technology transfer from stars to
firms. Working paper, UCLA Anderson School, Los Angeles, CA.
33
Variable Description Obs Mean Std Dv Min Max
Forward citation (EP-WIPO) Total count of citations received by other
European patents and WIPO patents with
European equivalent (up to 2003)
1186 1.084317 1.957689 0 19
Generality Herfhindhal index (IPC classes) on forward
patents (Jaffe et al, 1998; Hall et al, 2000)
387 .0961174 .1933514 0 .72
Forward citation lag Shortest forward patent citation time (in years) 518 3.067568 1.88583 0 10
Median forward citation lag Median forward patent citation time (in years) 518 3.488417 1.949027 0 13
Geographical impact
Degree of international dispersion of forward
patent citations (across foreign countries)
493 .1842779 .2517489 0 .8703704
Backward citation time Time difference in years between the publication
date of the focal patent and the cited patent
reference
(OECD, 2005)
1114 9.720676 7.753028 -2 59
Age difference Time difference in years respect to the grant date
of the focal patent and the cited patent reference
1186 .96543 1.805192 -5 9
Unique IPC class Dummy referring to patents with single IPC 4
digit class.
1186 .3583474 .4797172 0 1
Family Size (Europe) Number of designated states within the
European Community
1186 8.123946 4.880581 2 19
No. of Sub classes (IPC 4 digit) No. of fields (4 digit IPC) covered by the patent
1186 1.230185 1.682413 0 21
Patent Citations made Backward patent citations made
(total found in EPO and WIPO search reports)
1186 3.953626 2.222336 0 15
Firm size Number of Employees
1186 7.75601 1.349705 2.484907 8.711114
No. of inventors Proxy for the cost of the invention: team size
1185 2.556962 1.352677 1 10
Past patenting by inventors
No. of previous patents (granted)
up to 1995 by inventors
1186 12.49916 11.77226 1 52
Prolific Inventor Dummy on inventors having more than 5 patents
granted in the past
1186 .6365936 .4811833 0 1
SEC1 Electrical Engineering
1186 .118887 .3237921 0 1
SEC2 Instruments
1186 .3684654 .482592 0 1
SEC3 Chemistry and pharmaceuticals
1186 .118887 .3237921 0 1
SEC4 Process engineering, special equipment
1186 .3229342 .4677951 0 1
SEC5 Mechanical engineering, machinery
1186 .0708263 .2566427 0 1
y1996_y1998 Dummy for patent cohort with grant dates 1996-
1998
1186 .4232715 .4942861 0 1
y1999_y2001 Dummy for patent cohort with grant dates 1999-
2001
1186 .5767285 .4942861 0 1
Table 1 Descriptive Statistics
34
35
Table 2: Distribution of Patents and Science-linkages (Invention specific)
Technology Class Patents
Patents with Scientific
Non Patent References
% Patents with
scientific NPR
No. of Firms
No. Firms with
scientific NPR
Electrical Engineering 141 16 11,35% 15 4
Instruments 437 47 10,76% 12 7
Chemistry and pharmaceuticals 141 32 22,70% 9 3
Process engineering, special equipment 383 31 8,09% 28 9
Mechanical engineering, machinery 84 4 4,76% 28 4
Total 1186 130 10,96% 92 27
Table 3: Distribution of Firm-Level Science Linkages
Technology Class
Scientific
Firms
(publications)
Patents of
scientific
firms
%
Cooperating
with
universities
Patents of
cooperating
firms %
Firms that use
Public
Information
Patents of
firms that use
Public
Information %
Electrical Engineering 3 96 68,09% 7 127 90,07% 2 98 69,50%
Instruments 4 291 66,59% 7 395 90,39% 4 407 93,14%
Chemistry and pharmaceuticals 4 121 85,82% 4 122 86,52% 3 121 85,82%
Process engineering, special equipment 8 221 57,70% 12 296 77,28% 6 211 55,09%
Mechanical engineering, machinery 5 36 42,86% 10 32 38,10% 15 15 17,86%
Total 24 765 64,50% 40 972 81,96% 30 852 71,84%
Note: Firms that use public information (intensively) are those declaring a three points score in the CIS survey (the rank is 0-3, 3 being very important).
36
Table 4: Comparison of Means (T-tests)
Variable Obs Mean Obs Mean t Pr(T < t)
A. Patents without scientific NPR B. Patents with Scientific NPR
Forward Citation 1056 1,08 130 1,10 -0,078 0,461
Intensity citation 1056 0,61 130 0,57 0,337 0,633
generality 355 0,09 32 0,18 -2,14 0,019
No. IPC clases 1056 1,20 130 1,46 -1,585 0,051
Geographic scope 447 0,18 46 0,23 -1,139 0,129
B. Patents from Non scientific firms C. Patents from Scientific firms
Forward Citation 321 0,79 865 1,19 -3,94 0,000
Intensity citation 321 0,48 865 0,66 -3,004 0,001
generality 86 0,09 301 0,10 -0,066 0,473
No. IPC clases 321 0,88 865 1,36 -5,37 0,000
Geographic scope 114 0,15 379 0,19 -1,657 0,049
D. No NPR patents from Scientific firms E. NPR-Patents from Scientific firms
Forward Citation 762 1,20 103 1,18 0,04 0,516
Intensity citation 762 0,67 103 0,61 0,379 0,647
generality 275 0,09 26 0,16 -1,613 0,059
No. IPC clases 762 1,33 103 1,56 -1,238 0,108
Geographic scope 340 0,19 39 0,21 -0,415 0,339
F. No NPR patents from Non Scientific firms G. NPR-Patents from Non Scientific firms
Forward Citation 294 0,79 27 0,78 0,039 0,518
Intensity citation 294 0,48 27 0,44 0,281 0,609
generality 80 0,08 6 0,24 -1,426 *
No. IPC clases 294 0,86 27 1,07 -0,608 0,276
Geographic scope 107 0,14 7 0,33 -1,531 *
G. NPR-Patents from Non Scientific firms E. NPR-Patents from Scientific firms
Forward Citation 27 0,78 103 1,18 -1,066 0,144
Intensity citation 27 0,44 103 0,61 -0,803 0,212
generality 6 0,24 26 0,16 0,696 *
No. IPC clases (8) 27 1,07 103 1,56 -1,253 0,108
Geographic scope 7 0,33 39 0,21 0,899 *
F. No NPR patents from Non scientific firms D. No NPR patents from Scientific firms
Forward Citation 294 0,79 762 1,20 -3,831 0,001
Intensity citation 294 0,48 762 0,67 -2,95 0,001
generality 80 0,08 275 0,09 -0,274 0,391
No. IPC clases 294 0,86 762 1,33 -5,152 0,001
Geographic scope 107 0,14 340 0,19 -2,008 0,023
37
Table 5: Correlation Matrix
Forward citation (EP WIPO) 1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Generality 2 0.3063* 1
Age difference 3 0.4656* 0.0238 1
Family size (no. of states) 4 0.0994* 0.0729 0.0024 1
No. Sub fields (4 digit IPC) 5 0.1263* 0.0267 0.0730* -0.1433* 1
Unique sub class IPC (4 digit) 6 -0.1028* -0.1089* -0.0620* -0.0690* -0.0592* 1
Patent citations made 7 -0.0345 0.0128 -0.0255 -0.0516* -0.0618* -0.1105* 1
Past patenting inventors 8 0.1485* 0.0335 0.1200* -0.0821* 0.3591* -0.0691* -0.0319
Scientific Non patent reference 9 0.0312 0.0626 -0.0534* 0.0746* -0.0245 -0.0638* -0.0750* 0.0183 1
Firm size (employees) 10 0.0393 0.0518 -0.0317 -0.0296 0.1721* -0.0182 -0.0952* 0.0039 0.1054* 1
Count of publications 11 0.2851* 0.0625 0.0248 0.3210* 0.1727* -0.1923* -0.1283* 0.0431 0.1735* 0.1204* 1
Count of co-publications 12 0.2859* 0.0656 0.0223 0.3023* 0.2110* -0.1963* -0.1384* 0.0400 0.1716* 0.1411* 0.9959* 1
Scientific firm dummy 13 0.0727* 0.0889* -0.0281 0.0983* 0.0762* -0.0989* -0.0271 -0.0113 0.0926* 0.3563* 0.1523* 0.1744* 1
Co-publication intensity 14 -0.1364* -0.0110 -0.1588* -0.1310* 0.0608* -0.0154 -0.0169 -0.0028 0.0762* 0.4481* -0.0288 -0.0065 0.5865*
Cooperating firms (with universities/PRO) dummy 15 0.0992* 0.0517 0.0340 -0.1065* 0.4231* -0.0055 -0.1708* 0.0395 0.0377 0.3089* 0.2288* 0.2727* 0.0449 1
Use of public information sources (dummy) 16 0.1018* 0.0831* -0.0505* 0.0498* 0.3838* -0.1217* -0.1139* 0.0105 0.0379 0.2433* 0.3592* 0.4265* 0.2868* 0.4913* 1
Cooperation and Use of public sources 17 0.1232* 0.0909* -0.0224 0.0510* 0.4211* -0.1340* -0.1365* 0.0093 0.0288 0.2346* 0.3813* 0.4529* 0.2855* 0.5995* 0.9439* 1
Note: * Significance of correlation at 5% and better
Table 5
Explained Variable: Forward patent citation counts (EPO and EPO-WIPO patents)
Negative Binomial Zero Inflated Negative Binomial
1 2 3 4 5 6 7 8
Count Count Count Logit Count Logit Count Logit
Patent Characteristics
Backward citation lag 0.359 0.106 0.106 -1.859 0.106 -1.891 0.109 -1.885
(0.038)** (0.014)** (0.014)** (0.217)** (0.014)** (0.220)** (0.013)** (0.220)**
No. of Sub-subclasses 0.074 0.082 0.082 0.022 0.083 0.031 0.065 0.032
(0.039) (0.027)** (0.027)** (0.071) (0.028)** (0.071) (0.017)** (0.078)
Single IPC (4 digit) -0.073 -0.129 -0.129 -0.314 -0.127 -0.316 -0.141 -0.350
(0.147) (0.079) (0.079) (0.277) (0.079) (0.289) (0.079) (0.285)
Family size (EP) 0.112 0.270 0.270 0.267 0.272 0.282 0.269 0.299
(0.168) (0.067)** (0.067)** (0.289) (0.067)** (0.281) (0.069)** (0.281)
No. of inventors 0.042 0.202 0.202 0.227 0.198 0.198 0.185 0.205
(0.200) (0.091)* (0.091)* (0.275) (0.091)* (0.277) (0.098) (0.262)
Past patenting (inventors) 0.116 0.074 0.074 -0.072 0.075 -0.064 0.058 -0.058
(0.033)** (0.027)** (0.027)** (0.083) (0.028)** (0.084) (0.029)* (0.071)
Patent citations made -0.120 0.099 0.099 0.366 0.100 0.358 0.129 0.347
(0.160) (0.069) (0.069) (0.203) (0.071) (0.214) (0.063)* (0.230)
Scientific NPR 0.258 0.258 0.172 0.262 0.254 0.251 0.250
(0.181) (0.181) (0.203) (0.182) (0.191) (0.177) (0.187)
Firm Characteristics
Firm size -0.002 -0.163 -0.007 -0.167
(0.023) (0.065)* (0.021) (0.066)*
Scientific firm 0.247 -0.060
(0.123)* (0.436)
Electrical
Engineering
0.478 0.300 0.300 -0.304 0.303 -0.346 0.270 -0.354
(0.311)** (0.154) (0.154)** (0.614) (0.152)** (0.624) (0.111)** (0.619)
Instruments
0.286 0.077 0.077 -0.112 0.079 -0.108 0.022 -0.114
(0.227) (0.189) (0.189) (0.354) (0.188) (0.360) (0.144) (0.367)
Chemistry &
pharmaceuticals 0.447 0.503 0.503 0.402 0.505 0.379 0.438 0.362
(0.331) (0.372) (0.372) (0.428) (0.376) (0.443) (0.367) (0.469)
Process engineering
0.308 0.085 0.085 -0.206 0.085 -0.231 0.071 -0.262
(0.218) (0.148) (0.148) (0.489) (0.150) (0.503) (0.114) (0.509)
y1998_y2001 -0.126 -0.195 -0.195 -0.181 -0.197 -0.198 -0.202 -0.193
(0.162) (0.072)** (0.072)** (0.379) (0.074)** (0.393) (0.077)** (0.391)
y1994_y1997 -0.179 -0.034 -0.034 0.064 -0.028 0.034 -0.057 0.025
(0.442) (0.158) (0.158) (0.361) (0.155) (0.378) (0.156) (0.373)
Constant -1.294 -0.864 -0.864 -0.314 -0.859 0.950 -0.898 1.002
(0.868) (0.245)** (0.245)** (1.395) (0.321)** (1.789) (0.331)** (1.888)
Observations 1185 1185 1185 1185 1185 1185 1185 1185
Dispersion Parameter
(ln alpha) 1.05 1.05 -1.28** -1.28** -1.29**
Overdispersion Test (alpha=0) 451.48*** 451.52*** 99.72***
Wald Test (Joint Sig.) 5829.50 5480.36 1278.81***
1340.26*** 1431.88***
Log Pseudo Likelihood -1529.13 -1527.35 -1412.3 -1409.686 -1406.872
Vuong Test (zinb vs negative
binomial) 5.99***
Note: Robust standard errors in parentheses. Standard errors clustered by firm. The log-likelihood ratio test discriminating betzeen zero inflated poisson and zero
inflated negative binomial : 99.75 (significance<1%). Assumption in the LHR Test zip versus zinb: zip nested in zinb.
*** p<0.01, ** p<0.05, * p<0.1
38
39
Table 6
Explained Variable: Forward patent citation counts (EPO and EPO-WIPO patents)
Non NPR Patent Science Intensive Non NPR Patents
Whole Sample Technologies
Science Intensive
Tech.
1 2 3 4 5 6
count logit count logit count logit
Patent Characteristics
Backward citation lag 0.119 -1.957 0.109 -1.741 0.088 -1.740
(0.017)** (0.250)** (0.030)** (0.370)** (0.031)** (0.375)**
No. of Sub-subclasses 0.083 0.055 0.075 -0.017 0.067 -0.038
(0.038)* (0.080) (0.063) (0.108) (0.072) (0.108)
Single IPC(4) patents -0.151 -0.468 -0.098 -0.300 -0.112 -0.289
(0.077)* (0.345) (0.137) (0.277) (0.147) (0.283)
Patent citations made 0.143 0.414 0.146 0.007 0.110 -0.033
(0.063)* (0.257) (0.138) (0.323) (0.140) (0.320)
Family size (EP) 0.248 0.181 0.322 0.611 0.328 0.573
(0.070)** (0.279) (0.134)* (0.270)* (0.139)* (0.268)*
No. of inventors 0.257 0.180 0.163 -0.039 0.162 -0.095
(0.067)** (0.292) (0.165) (0.396) (0.171) (0.396)
Past patents (inventors) 0.064 -0.076 0.046 0.099 0.045 0.071
(0.035) (0.076) (0.070) (0.142) (0.072) (0.143)
Scientific NPR 0.391 0.099
(0.195)* (0.351)
Firm Characteristics
Firm size -0.011 -0.169 -0.004 -0.279 0.005 -0.267
(0.019) (0.061)** (0.061) (0.097)** (0.062) (0.097)**
Scientific firm 0.245 0.077 0.519 0.385 0.550 0.563
(0.122)* (0.435) (0.272) (0.512) (0.319) (0.559)
Electrical Engineering
0.225 -0.319
(0.111)* (0.595)
Instruments
-0.032 -0.136 -0.284 0.242 -0.237 0.269
(0.195) (0.403) (0.167) (0.381) (0.176) (0.401)
Chemistry/pharmaceuticals
0.106 0.476 0.092 0.570 0.191 0.616
(0.324) (0.575) (0.228) (0.416) (0.256) (0.429)
Process engineering/equipment
0.041 -0.282
(0.117) (0.523)
y1998_y2001 -0.238 -0.469 -0.246 0.078 -0.238 0.017
(0.071)** (0.375) (0.128) (0.283) (0.133) (0.287)
y1994_y1997 -0.052 -0.242 -0.181 0.435 -0.244 0.379
(0.176) (0.269) (0.201) (0.385) (0.213) (0.385)
Constant -0.963 1.241 -0.992 0.959 -0.967 1.089
(0.312)** (1.865) (0.578) (1.339) (0.604) (1.325)
Observations 1055 1055 719 719 703 703
Dispersion Parameter (ln alpha) -1.28 -1.16 -1.12
Overdispersion Test 69.51*** 77.60*** 78.57***
Wald Test (Joint Sig.) 412.28*** 56.86*** 49.40***
Log Pseudo Likelihood -1248.32 -856.05 -832.5921
Vuong Test (zinb vs negative
binomial)
5.12*** 5.90*** 5.64***
Note: Robust standard errors clustered by firm in model displayed in columns 1 and 2. The goodness’ of the fit is reporting in
the LHR ratio test instead of the Wald test for the rest of the models (science intensive technologies) since they are not robust
to clustering by firms (26 firms only). For the sub-sample of science intensive technologies, the Vuong Test discriminating
between zinb versus negative binomial= 5,67 with Pr>z=0,0000. The Science intensive technologies are: electrical engineering,
instruments and chemical and pharmaceuticals (Franhuffer’ technology classes).
*** p<0.01, ** p<0.05, * p<0.1
Table 7
Patent Characteristics 123456789
Backward citation lag 0.110 0.110 0.108 0.110 0.110 0.110 0.110 0.110 0.111
(0.011)** (0.011)** (0.011)** (0.011)** (0.011)** (0.011)** (0.011)** (0.011)** (0.012)**
No. of Sub-subclasses 0.057 0.063 0.066 0.057 0.055 0.057 0.055 0.057 0.053
(0.019)** (0.022)** (0.022)** (0.019)** (0.018)** (0.018)** (0.018)** (0.019)** (0.018)**
Single IPC(4) patents -0.148 -0.147 -0.143 -0.149 -0.151 -0.148 -0.150 -0.146 -0.151
(0.073)* (0.071)* (0.073) (0.072)* (0.072)* (0.073)* (0.073)* (0.073)* (0.075)*
Patent citations made 0.115 0.111 0.100 0.116 0.118 0.115 0.119 0.113 0.119
(0.056)* (0.057) (0.060) (0.055)* (0.056)* (0.056)* (0.054)* (0.056)* (0.054)*
Family size (EP) 0.289 0.297 0.298 0.290 0.291 0.289 0.290 0.285 0.286
(0.075)** (0.072)** (0.071)** (0.076)** (0.077)** (0.076)** (0.075)** (0.077)** (0.077)**
Past patents (inventors) 0.085 0.086 0.098 0.084 0.084 0.085 0.085 0.086 0.087
(0.029)** (0.031)** (0.031)** (0.030)** (0.030)** (0.029)** (0.032)** (0.031)** (0.031)**
Scientific NPR 0.244 0.251 0.240 0.245 0.242 0.223 0.065 0.214 0.027
(0.190) (0.194) (0.187) (0.191) (0.189) (0.229) (0.332) (0.227) (0.343)
Firm Characteristics
Firm size -0.009 -0.015 -0.013 -0.010 -0.012 -0.009 -0.012 -0.006 -0.009
(0.022) (0.024) (0.024) (0.023) (0.021) (0.022) (0.020) (0.021) (0.020)
Scientific firm 0.271 0.251 0.250 0.269 0.247 0.351 0.388
(0.117)* (0.222) (0.216) (0.117)* (0.087)** (0.208) (0.235)
Use of Public sources (CIS) 0.239 0.025 0.006 -0.100 -0.201
(0.111)* (0.217) (0.226) (0.205) (0.249)
Cooperating firms (CIS) 0.170 0.049 0.056 0.125
(0.170) (0.167) (0.175) (0.221)
Scientific NPR*scientific firm 0.025 0.106
(0.342) (0.323)
Scientific NPR*Cooperating firm 0.179 0.121
(0.477) (0.431)
Scientific NPR*Public Sources 0.027
(0.338)
Electrical Engineering 0.295 0.216 0.250 0.285 0.270 0.297 0.268 0.340 0.321
(0.117)* (0.171)* (0.213)* (0.164)* (0.176)* (0.116)* (0.160) (0.160)* (0.169)
Instruments 0.060 -0.016 0.056 0.049 0.038 0.062 0.041 0.104 0.095
(0.162) (0.214) (0.234) (0.192) (0.197) (0.156) (0.190) (0.199) (0.196)
Chemistry/pharmaceuticals 0.470 0.390 0.472 0.458 0.447 0.471 0.452 0.519 0.511
(0.340) (0.359) (0.382) (0.365) (0.370) (0.349) (0.371) (0.373) (0.385)
Process engineering/equipment
0.102 0.030 0.057 0.093 0.079 0.103 0.078 0.140 0.118
(0.115) (0.171) (0.204) (0.155) (0.169) (0.114) (0.159) (0.155) (0.165)
y1998_y2001 -0.249 0.250 -0.187 0.149 -0.188 0.149 -0.166 0.146 -0.151
(0.041)** (0.042)** (0.046)** (0.041)** (0.042)** (0.043)** (0.045)** (0.043)** (0.047)**
Constant -0.910 -0.800 -0.840 -0.900 -0.900 -0.911 -0.900 -0.953 -0.973
(0.386)* (0.386)* (0.402)* (0.405)* (0.413)* (0.386)* (0.383)* (0.406)* (0.431)*
Observations 1186 1186 1186 1186 1186 1186 1186 1186 1186
Dispersion Parameter (ln alpha) -1.28 -1.28 -1.28 -1.28 -1.28 -1.28 -1.28 -1.28 -1.28
Wald Test (Joint Sig.) 429.28 511.54 639.12 429.28 1654.21 471.25 3602.43 1689.25 3199.67
Log Pseudo Likelihood -1409.547 -1411.272 -1412.08 -1409.547 -1409.489 -1409.545 -1409.365 -1409.149 -1408.32
Robust standard errors in parentheses clustered by firm.
* significant at 5%; ** significant at 1%
Explained Variable: Forward patent citation counts (EPO and EPO-WIPO patents)
Firm Level Science Linkages and Interactions (count regressions of the Zero Inflated Negative Binomial)
InteractionsOther ISL at the firm level
40
Tobit Tobit probit probit Tobit Tobit probit probit ols ols ols ols
-1 -2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12
Patent Radicalness 0.096*** 0.096*** 0.053*** 0.053*** 0.066*** 0.066*** 0.066*** 0.066*** -0.095*** -0.096*** 0.007 0.007
(0.017) (0.017) (0.013) (0.013) (0.013) (0.013) (0.014) (0.014) (0.018) (0.018) (0.040) (0.040)
Backward citation lag 0.022 0.021 0.009 0.009 0.025* 0.025* 0.017* 0.017* 0.024 0.024 0.037 0.036
(0.025) (0.025) (0.009) (0.009) (0.015) (0.015) (0.009) (0.009) (0.051) (0.051) (0.055) (0.056)
No. of Sub-subclasses 0.011 0.009 0.004 0.003 0.028 0.028 0.024 0.025 -0.105* -0.111* -0.085 -0.085
(0.040) (0.040) (0.014) (0.014) (0.026) (0.026) (0.019) (0.019) (0.056) (0.057) (0.052) (0.052)
Single IPC(4) patents -0.208** -0.214** -0.090*** -0.092*** 0.043 0.045 0.035 0.035 -0.132 -0.134 -0.165 -0.165
(0.104) (0.105) (0.024) (0.024) (0.065) (0.065) (0.044) (0.044) (0.229) (0.228) (0.209) (0.208)
Family size (EP) -0.071 -0.069 -0.040 -0.039 0.024* 0.024* 0.020* 0.020* 0.154 0.158 0.203 0.203
(0.097) (0.097) (0.029) (0.028) (0.062) (0.062) (0.038) (0.038) (0.132) (0.131) (0.161) (0.162)
No. of inventors 0.262* 0.265* 0.115** 0.117** -0.091 -0.092 -0.091** -0.092** 0.297* 0.300* 0.252 0.252
(0.136) (0.136) (0.055) (0.052) (0.090) (0.090) (0.042) (0.043) (0.169) (0.173) (0.171) (0.172)
Past patents (inventors) -0.019 -0.023 -0.009 -0.011 0.052 0.053 0.046** 0.046** -0.143*** -0.143*** -0.214*** -0.214***
(0.052) (0.052) (0.014) (0.014) (0.034) (0.034) (0.021) (0.021) (0.049) (0.049) (0.056) (0.056)
Scientific NPR 0.287** 0.281* 0.161* 0.157 0.044 0.044 0.031 0.031 0.283 0.282 0.176 0.176
(0.144) (0.144) (0.095) (0.096) (0.100) (0.100) (0.053) (0.052) (0.326) (0.331) (0.305) (0.306)
Scientific Firm (publications) -0.052 -0.041 -0.041 -0.036 0.055 0.052 0.039** 0.037** -0.629*** -0.606*** -0.662** -0.659**
(0.126) (0.127) (0.051) (0.053) (0.083) (0.084) (0.048) (0.049) (0.200) (0.202) (0.257) (0.262)
Firm size (employees) 0.026 0.027 0.009 0.010 0.015 0.014 0.010 0.010 -0.057 -0.056 -0.058 -0.057
(0.034) (0.034) (0.014) (0.014) (0.023) (0.023) (0.013) (0.013) (0.057) (0.057) (0.046) (0.046)
Scientific Firm*Scientific NPR -0.261 -0.112 0.055 0.038 -0.391** -0.054
(0.379) (0.077) (0.210) (0.056) (0.167) (0.352)
Constant -1.009** -1.010** -0.550* -0.551* 4.211*** 4.202*** 4.693*** 4.692***
(0.460) (0.460) (0.288) (0.288) (0.889) (0.892) (0.697) (0.697)
Sigma 0.590*** 0.590*** 0.538*** 0.538***
(0.057) (0.057) (0.034) (0.034)
Observations 386 386 386 386 492 492 492 492 517 517 517 517
Wald Test (Fisher) 65,91*** 66,43*** 3864,18*** 3662,61*** 53,57*** 53,64*** 990,18*** 1886,93*** 115,07*** 276,15*** 85,45*** 88,39***
Log LH/Log PLH -170,68 -170,42 -159,34 -158,97 -315,68 -315,65 -292,71 -292,68
Pseudo R-2/ R-2 0,16 0,16 0,19 0,19 0,07 0,08 0,094 0,094 0.135 0.136 0.160 0.160
Observed P. 0,204 0,204 0,37 0,37
Predicted P. 0,184 0,184 0,36 0,36
Censoring % 79,53% 79,53% 79,53% 79,53% 71,56% 71,56% 71,56% 71,56%
Other indicators of patent quality and diffusion
Table 7
Note: all regressions include the technology dummies and time cohorts (not reported to save space). Robust Standard errors in parentheses for the models probit and ordinary least squares (ols).
Generality Geographical Dispersion Forward Citation Lag
Generality Different IPC-4 digit Geographical Dispersion Abroad citation Median Time Shortest time
*** p<0.01, ** p<0.05, * p<0.1
41
