Research Policy 31 (2002) 13491367

R&D spillovers, patents and the incentives to innovate

in Japan and the United States
Wesley M. Cohen
a,
, Akira Goto
b
, Akiya Nagata
c
,
Richard R. Nelson
d
, John P. Walsh
e,b
a
Department of Social and Decision Sciences, Carnegie Mellon University, Pittsburgh, PA 15238, USA
b
University of Tokyo, Tokyo, Japan
c
Japan Advanced Institute of Science and Technology, Ishikawa, Japan
d
Columbia University, New York, NY, USA
e
University of Illinois at Chicago, Chicago, IL, USA
Received 17 September 2001; received in revised form 19 October 2001; accepted 10 January 2002
Abstract
National surveys of R&Dlabs across the manufacturing sectors in the USand Japan showthat intraindustry R&Dknowledge
ows and spillovers are greater in Japan than in the US and the appropriability of rents due to innovation less. Patents in
particular are observed to play a more central role in diffusing information across rivals in Japan, and appear to be a key reason
for greater intraindustry R&D spillovers there, suggesting that patent policy can importantly affect information ows. Uses of
patents differ betweenthe twonations, withstrategic uses of patents, particularlyfor negotiations, beingmore commoninJapan.
2002 Elsevier Science B.V. All rights reserved.
Keywords: Patents; R&D; Spillovers; Innovation; Appropriability
1. Introduction
The ability of rms to appropriate at least some of
the value created by their innovations is essential if
there is to be incentive to innovate. On the other hand,
the economy-wide benets from an innovation depend
to a considerable degree on the extent to which the
new knowledge associated with it becomes available
to others to use and build on. This article reports the
results of a survey research study that compares the
ability of the US and Japanese rms to appropriate
the returns to their innovations, how these rms pro-
tect their innovations (by using, e.g. secrecy, patents,

Corresponding author.
E-mail address: wc02@andrew.cmu.edu (W.M. Cohen).
lead time advantages) and the magnitude and channels
of intraindustry R&D information ows in the two
nations. We focus particularly on the role of patent-
ing since it is both a key policy instrument affecting
appropriability and an object of growing managerial
attention.
Our comparison of appropriability conditions and
spillovers between Japan and the US builds on prior
research that shows that, in most industries, rms
rely predominantly on mechanisms other than patents
to protect their innovations, including secrecy, rst
mover advantages and the exploitation of comple-
mentary capabilities (Scherer et al., 1959; Manseld,
1986; Levin et al., 1987; Arundel, 2001; Taylor and
Silberston, 1973). In some industries, such as drugs,
patents are, however, quite important. A comparison
0048-7333/02/$ see front matter 2002 Elsevier Science B.V. All rights reserved.
PII: S0048- 7333( 02) 00068- 9
1350 W.M. Cohen et al. / Research Policy 31 (2002) 13491367
between Japan and the US should add to these in-
sights by showing whether appropriability conditions
depend on cross-national differences in policy and
institutional environments, or factors, such as tech-
nology, that should vary little across nations.
The role and effectiveness of patents is of central
concern to this study. We will explore whether differ-
ences between the US and Japan in patent law, and ju-
dicial and administrative practice affect the degree to
which patents protect innovation. We will also exam-
ine whether the two nations different patent policies
affect the diffusion of technical knowledge, as sug-
gested by Ordover (1991) and Granstrand (2000).
In brief, although R&D spending as a percent of
GDP is higher in Japan than the US, intraindustry
R&D information spillovers appear to be greater in
Japan for almost all industries, and the appropriability
of rents due to innovation appears to be less. In contrast
to the US, Japanese respondents report secrecy as a
minor appropriability mechanism and patents as com-
parably effective to other major mechanisms. A key
reason for less appropriability and greater intraindus-
try R&D spillover in Japan appears to be that patents
diffuse information across rivals more readily than in
the US. We suggest that this cross-national difference
is not only due to the policy differences highlighted
by Ordover (1991), but also due to the different ways
in which patents are used in the two nations.
In Section 2, we review our sampling and survey
methods. Section 3 compares intraindustry R&D in-
formation ows and spillovers. Section 4 compares the
appropriability of rents due to innovation. Section 5
considers how the two patent systems inuence in-
traindustry R&D spillovers. Section 6 concludes the
paper.
2. Method, data, and samples
The data come from a survey of managers of R&D
units of manufacturing rms in the US and Japan. Re-
sponses were matched to each R&D units focus in-
dustry that represented the bulk of its R&D effort.
1
1
By obtaining answers for a rms activities and experience in
a given product market rather than the rm as a whole, we reduce
the measurement error associated with differences in conditions
across the several industries in which a rm might be active.
Where possible, our surveys employed objective re-
sponse scales, which tend to be both more interpretable
and more readily compared across respondents, which
is especially useful for comparisons across nations
where language and culture differ. First drafted in En-
glish, the questionnaire was translated into Japanese
and then independently back-translated into English to
assure accuracy. While most questions were identical
between the two surveys, a small number of questions
differed. The surveys were mailed during the sum-
mer of 1994. Both the American and Japanese teams
conducted follow-up mailings and phone calls to in-
crease response rates (cf. Dillman, 1978). To help in-
terpret our ndings, we also conducted 18 interviews
with R&D managers and intellectual property ofcers
at nine US rms, and 12 interviews at nine Japanese
rms, as well as with representatives of the Japanese
Patent Ofce (JPO).
The population sampled in the US included all R&D
units belonging to R&D performing manufacturing
rms located in the US. The US sample was drawn
from a frame consisting of eligible labs in the Direc-
tory of American Research and Technology (Bowker,
1994) or, if not included there, labs attached to rms
listed in Standard and Poors COMPUSTAT tapes. We
surveyed 3240 labs located in the US and received
1478 responses, for a raw response rate of 46% and an
adjusted response rate of 54% of the eligible sample.
2
The US survey data are supplemented with published
data on rm sales and employees from COMPUSTAT,
Dun & Bradstreet and similar sources.
The population sampled in Japan included R&D
performing, manufacturing rms with capitalization
over 1 billion yen (). The sample was drawn from
a list of 1219 rms amassed by Japans Science and
Technology Agency, or 71% of the 1722 rms es-
timated to meet the population denition (NISTEP,
1997). We received responses from 643 rms, yield-
ing a response rate of 52%.
In order to make the two national samples as compa-
rable as possible, we restricted our samples to respon-
dents from rms with annual sales of US$ 50 million
2
A nonrespondent survey found that 28% of the nonrespondents
in the US were not in the target population (for example, they did
no manufacturing). After correcting the sample size accordingly
for ineligible cases, the US response rate was adjusted upward to
54%.
W.M. Cohen et al. / Research Policy 31 (2002) 13491367 1351
Table 1
Firm and business unit size in the US and Japanese samples
Sales (US$
(in million))
Employees
US Japan US Japan
Firm size
Median 1245 558 8600 1476
Mean (unweighted) 6067 1956 30009 3720
Mean (weighted) 6020 1793 29272 3462
Business unit size
Median 300 361 1000 850
Mean (unweighted) 2403 1264 6282 2229
Mean (weighted) 2813 1122 6791 1945
Notes: 1 US$ = 102.8 , March 1994 exchange rate.
or more.
3
Foreign-owned labs were also dropped in
order not to confuse our cross-national comparisons.
In this truncated sample, there are 826 observations
from the US and 593 from Japan. Because the two
national samples can contain very different numbers
of observations for a given industry, we constructed
weights that correct for the differences in industry mix
in each country, and use these weights when compar-
ing national aggregate results in order to rule out the
possibility that country differences are due to varia-
tions in industry mix. When we make comparisons be-
tween these country averages, we use either weighted
least squares or weighted logistic regressions as ap-
propriate (see later). For comparisons across survey
item averages within a country, we use paired t-tests
or non-parametric tests (Agresti and Agresti, 1979).
Presenting summary statistics on rm and business
unit sizes for the two samples, Table 1 shows that the
US sample contains substantially larger rms, though
the business units (reecting rms activities in a given
market) are more comparable in size. Consistent with
public sources, our data indicate that Japanese rms
spend more on R&D relative to sales than compara-
ble US rms. For our sample, the comparison means
for own-nanced business unit R&D expenditures di-
vided by worldwide sales are 2.26% for the US and
3.70% for Japan (t = 6.89, P < 0.0001).
4
Our survey
3
We reran the analyses for the rms in the US sample with
sales revenue of less than US$ 50 million. The results are not
qualitatively different from the results reported in this paper.
4
Our sample data excludes outliers, dened as business units
with R&D intensities >100% or more than three interquartile
ranges from the median.
data suggest that the overall Japanese sample R&D
intensity exceeds that of the US largely because the
Japanese R&D intensities in the least R&D intensive
(i.e., <2%in the US) industries tend to be considerably
higher. In contrast, in the most R&D intensive indus-
tries such as drugs or semiconductors, the average US
and Japanese R&D intensities tend to be comparable.
5
3. Intraindustry R&D information ows and
spillovers
In this section, we compare the extent of R&D-
related information ows across rivals in the US and
Japanese manufacturing sectors. We begin by compar-
ing responses to a question in the survey about the
importance of information originating from a broad
range of information sources, of which rival rms
are one. Respondents report whether, over the prior 3
years, information from competitors either suggested
new R&D projects or contributed to the completion
of existing projects. Fig. 1 shows clear differences be-
tween the US and Japan. In Japan, information from
rivals suggests new R&D projects for 48% of the re-
spondents, exceeding the 41% gure for the US (
2
[1 d.f.] = 6.9, P < 0.01). This difference, though a
relative one of almost 20% and statistically signi-
cant, is dwarfed by the difference in the frequencies
with which respondents in Japan and the US claim
that information from competitors contributes to the
execution of existing R&D projects. In Japan, 51%
of the respondents claim that information from rivals
contributes to project execution versus only 13% in
the US (
2
[1 d.f.] = 213.9, P < 0.0001).
6
These
results suggest not only that intraindustry ows are
5
Public data show that over the last 20 years, Japan has consis-
tently spent more on non-defense R&D as a percent of GDP than
has the US. Over the last 5 years, the Japanese average has been
between 2.6 and 3.0%. During this same period, the US average
has been between 1.9 and 2.2% (National Science Foundation,
2000). We believe our aggregate results exceed the public data
largely because we sampled only R&D-performing rms.
6
The industry-level results show little deviation from the aggre-
gate patterns. For 25 of the 26 industries for which we have more
than ve observations for both samples, the percent of respondents
indicating that information from rivals contributes to R&D project
completion is greater in Japan than in the US. For 20 of 26 indus-
tries, the same comparison holds for the percent of respondents
indicating that information from rivals suggests new projects.
1352 W.M. Cohen et al. / Research Policy 31 (2002) 13491367
Fig. 1. Competitors as information sources for suggesting new projects or contributing to completion of existing projects, for Japan and
the US.
more extensive in Japan than in the US, but that they
play a different role in the two nations. In the US,
these ows predominantly guide R&D project selec-
tion, whereas, they affect project selection and project
completion comparably in Japan.
In another comparison of intraindustry R&D
knowledge ows, respondents reported when they
rst became aware of the R&D project leading to a
rivals major innovation. Respondents were permitted
to choose from one of the following four stages in the
innovation process: project initiation, research, devel-
opment, or product or process introduction. In the US,
only 15.6% of the respondents indicated they knew
about the rivals project prior to the development
stage, compared to 43.9% of the Japanese respon-
dents (
2
[1 d.f.] = 125.2, P < 0.0001), conrming
that Japanese rms tend to be better informed of rival
R&D activities than their American counterparts.
Our survey data thus suggest that R&D-related in-
formation ows among rivals are greater in Japan than
in the US. Does this difference correspond, however,
to one in intraindustry (nonpecuniary) R&Dspillovers,
which Griliches (1992) and others identify as a major
source of productivity growth? The key is whether the
observed information ows are largely the outcome of
market-mediated transactions or not. Our analysis of
the different channels through which rms might learn
about the R&D activities of rivals (discussed later)
show that of the ve channels that are the most im-
portant in both countries, fournamely publications,
public meetings or conferences, informal information
exchange and competitors products (via, for example,
reverse engineering)are not market-mediated, and
the fth, namely patents, is only partially so. We con-
clude, therefore, that not only are intraindustry R&D
information ows greater in Japan, but intraindustry
R&D spillovers as well.
4. Appropriability
While factors other than intraindustry R&D
spillovers might condition the appropriability of prof-
its due to innovation, such spillovers are a key deter-
minant. Thus, if the extent of intraindustry spillovers
are indeed greater in Japan than in the US, we should
expect, ceteris paribus, that Japanese rms appropriate
less of the returns to their R&D activities than Amer-
ican rms. Our survey provides two proxy measures
for the appropriation of prots due to innovation.
Building on Levin et al. (1987), our rst measure is
premised on the notion that the longer it typically takes
W.M. Cohen et al. / Research Policy 31 (2002) 13491367 1353
Fig. 2. Mean imitation lags for product and process innovations.
rivals within an industry to imitate a rms innovation
or to introduce a competing innovation, the greater will
be the innovating rms prots from their innovations.
We asked each respondent, separately for patented and
unpatented product and process innovations, how long
it took for a competing innovation to appear after they
had introduced a signicant innovation.
7
Consistent
with our nding of greater intraindustry spillovers in
Japan, we nd that the imitation lags for all four
types of innovation are much more compressed in
Japan than in the US. Presenting the results for the
two national samples, Fig. 2 shows much longer lags
in the US, ranging from 40% longer for unpatented
product innovations to about 80% longer for patented
process innovations (all country differences signi-
cant, P < 0.0001). These qualitative patterns apply to
all the industries except computers, where the imita-
7
The response categories for this question were less than 6
months, 0.51.5 years, 1.53 years, 35 years and >5 years. To
compute aggregate and industry averages, midpoints were em-
ployed, except for the last, open-ended category to which we as-
signed a value of 8 years.
tion lags associated with patented processes and un-
patented products and processes are longer in Japan.
To construct a second measure of appropriability,
we asked each respondent to estimate the percent
of their product and process innovations for which
patents, secrecy, lead time, complementary sales and
service, and complementary manufacturing facilities
and know-how were effective in protecting the
competitive advantage from those innovations.
8
Fig. 3 shows the average percent of product and pro-
cess innovations (based on category midpoints) for
which each mechanism is judged to be effective.
9
8
Note that secrecy is not limited to formal trade secrecy
protection, and that secrecy and patents are not exclusive. Firms,
for example, may try to maintain secrecy about a project or a
result until the patent is published. Also, rms may patent some
aspects of the innovation while trying to keep other parts secret.
More generally, Cohen et al.s (2000) analysis of the US data
show that the different appropriability mechanisms considered are
often used together, sometimes in quite systematic ways.
9
The response categories were 010%, 1140%, 4160%,
6190%, and 91100%.
1354 W.M. Cohen et al. / Research Policy 31 (2002) 13491367
Fig. 3. Effectiveness of appropriability mechanisms for product and process innovations.
W.M. Cohen et al. / Research Policy 31 (2002) 13491367 1355
With the exception of patents, which appear to be
comparably effective in Japan and the US in absolute
terms, US respondents report all mechanisms to be
more effective, scoring every mechanism, on average,
at least 20% higher (all country differences, except
patents, signicant, P < 0.01). Thus, consistent with
the nding of more compressed imitation lags and
greater intraindustry spillovers in Japan, the compar-
ison of absolute scores suggest that the effectiveness
of all appropriability mechanisms except patents is
greater in the US.
In addition to permitting an evaluation of overall
appropriability, the data presented in Fig. 3 allow a
comparison of the degree to which the different appro-
priability mechanisms are featured in the appropriabil-
ity strategies of rms in the two nations. Fig. 3 shows
clear differences in the reported absolute and relative
effectiveness of secrecy. For protecting product inno-
vations in the US, secrecy is pervasively ranked as one
of the most effective mechanisms across industries,
and reported to be the most effective mechanism along
with lead time in the aggregate sample. In contrast, of
all the major appropriability mechanisms, secrecy is
judged to be the least effective mechanism for protect-
ing product innovations in Japan. For process innova-
tions, US rms report secrecy as the most important
mechanism, over lead time and complementary man-
ufacturing capabilities (both signicant, P < 0.0001).
For Japan, in contrast, secrecy protects process inno-
vations only about as well as lead time (t = 1.06,
P > 0.25) and a little better than patents (t = 3.15,
P < 0.01), with complementary manufacturing as
the dominant mechanism (signicantly above secrecy,
t = 6.35, P < 0.0001).
10
In absolute terms, the US
aggregate scores for product and process secrecy are
approximately double those for Japan (P < 0.0001).
Thus, we conclude that secrecy is more central to the
appropriability strategies of the US rms.
Contrary to our initial conjecture that patents would
be similarly evaluated in Japan and the US, we ob-
serve a difference in the relative standing of patents
against other appropriability mechanisms. Consistent
10
Based on the Community Innovation Survey, Arundel (2001)
reports secrecy to be relatively more important than patents in
Europe, but not nearly as important as lead time, suggesting that
it gures more prominently in the appropriability strategies of
European than Japanese rms, but less than in those of American
rms.
with the ndings of Levin et al. (1987), Scherer et al.
(1959) and Manseld (1986) for the US, American
respondents report patents to be less important than
other major mechanisms in protecting product inno-
vations (signicantly below sales/service, t = 4.35,
P < 0.0001). In contrast, Japanese respondents report
patents to be about as effective as any other mechanism
for protecting product innovations, ranking slightly be-
low lead time (t = 1.27, P > 0.20) and a bit above
complementary manufacturing capabilities (t = 3.75,
P < 0.001).
11
The almost identical absolute scores
on patent effectiveness between the two nations sug-
gest caution, however, before concluding that patents
are stronger in Japan. Patents may not be more effec-
tive in Japan than in the US in any absolute sense.
Rather, other mechanisms may be weaker. Nonethe-
less, patents appear to be more heavily featured in the
appropriability strategies of Japanese rms.
5. Patents and intraindustry R&D spillovers
Ordover (1991, p. 48) argued that The Japanese
patent system is a complex web of policy choices more
or less consciously structured to affect R&D diffusion
while maintaining overall incentives for R&D invest-
ment. The Japanese patent system subordinates the
short term interests of the innovator in the creation of
exclusionary rights to the broader policy goals of dif-
fusion of technology. In the prior sections, we see
that intraindustry R&D spillovers are greater and ap-
propriability is less in Japan than in the US. In this sec-
tion, we build on Ordover (1991) to consider whether
the two nations distinct patent systems could affect
the observed cross-national difference in intraindustry
R&D spillovers.
5.1. Contrasting features of the Japanese and US
patent systems
Originally established in the 19th century not
only to encourage domestic inventors, but to facili-
tate technology transfer from the West, the Japanese
patent system emphasized the disclosure function of
11
Similar to our US respondents, European respondents report
patents to be the least effective appropriability mechanism for
protecting product and process innovations (Arundel, 2001).
1356 W.M. Cohen et al. / Research Policy 31 (2002) 13491367
patents.
12
This emphasis on diffusion is partly re-
ected by the Japanese requirement that all patent
applications be disclosed 18 months after the ap-
plications are led.
13
Since only 1730% (Westney,
1992; Okada, 1997; JPO, personal communication)
of the patent applications in Japan are approved, the
preponderance of patent applications in Japan are
disclosed with no ultimate benet of intellectual prop-
erty protection. In the US, in contrast, until this past
year, patent applications were only disclosed upon
issuance, with somewhere between 10 and 30% never
issuing (Westney, 1994, p. 162; Quillen and Webster,
2001).
14
Moreover, since the average pendency pe-
riod in the US was 20 months (US G.A.O., 1997), a
substantial fraction of the published applications were
published later than 18 months.
Ordover (1991) also argues that patents tend to be
applied for earlier in the innovation process in Japan
due to a rst-to-le rule of priority as opposed to the
rst-to-invent rule of priority that applies to the US
(cf. Scotchmer and Green, 1990). A procedural differ-
ence may also affect the disclosure function of patents.
At the time of our survey, Japanese patents were still
subject to pre-grant opposition. For 3 months after a
patent examiner gave notice that he intended to grant a
patent, competitors or anyone else could challenge the
validity of the prospective patent, and the examiners
often relied heavily on the evidence provided during
this period of pre-grant opposition in making their -
nal determination.
15
No analogous opposition process
12
Until recently, the Japanese patent policy has followed what
the Japanese refer to as the conventional catch-up style patent
system. Since the late 1990s, policymakers see Japan as beginning
to adopt a pro-patent policy that realizes speedy, strong and
broad protection of rights. (JPO Annual Report, 1999, p.2).
13
Reecting interviewed Japanese patent ofcials view that even
18 months is too long to effectively achieve the disclosure function
in fast-paced technologies, the JPO recently implemented a sys-
tem of pre-18 month disclosure where rms can opt to have their
application published once it passes the formality check (usually 5
months after submission), or any time after that up to 18 months
later. Current (February 2001) estimates are that only 2030 ap-
plications per month are published early.
14
With the American Inventors Protection Act of 1999, the US
will publish applications led after November 2000, 18 months
after ling. Strictly domestic patents are exempt.
15
In 1996, pre-grant opposition was replaced in Japan with a
system of post-grant opposition. Third parties now have 6 months
from the JPOs announcement of their intent to grant to challenge
the validity of the patent.
applies to the US. Rather, the validity of a US patent is
more typically challenged in the courts after a patent is
issued. Thus, the Japanese patent system places more
information in the public domain sooner, induces the
ling of a patent application sooner in the innovation
process, and the opportunity for pre-grant opposition
strengthened incentives to monitor competitors patent
lings early on.
In addition to different rules of priority and disclo-
sure and the presence of a pre-grant opposition system,
the analysis of Cohen et al. (2000) suggests yet an-
other dimension of policy and patenting practice that
may make patents in Japan more effective vehicles of
information disclosure than in the US. Specically,
they argue that the number of patentable elements in
a commercializable new product importantly affects
the way patents are used and, possibly, in turn, the
degree to which patents contribute to information dif-
fusion. They distinguish between complex versus
discrete product industries on the basis whether a
new, commercializable product or process is com-
prised of numerous separately patentable elements
versus relatively few.
16
In the US, for example, new
drugs or chemicals typically are comprised of a rel-
atively discrete number of patentable elements. In
contrast, electronic products in the US tend to be
comprised of a larger numberoften hundredsof
patentable elements and, hence, may be characterized
as complex. Consequently, in complex product indus-
tries, rms rarely have proprietary control over all the
essential complementary components of the technolo-
gies they are developing. Firms hold rights over tech-
nologies that others need, and vice versa, creating a
condition of mutual dependence that fosters extensive
cross-licensing, related negotiations and information
sharing. As compared to discrete product industries,
patents are used less often to enforce exclusivity and
more to secure market access (by conferring access
to key technology owned by others) and freedom of
operation (by conferring an ability to countersue if
sued, thereby discouraging suits).
While Cohen et al. (2000) distinguish complex
from discrete product industries largely on the basis of
technology and the physical character of the products,
16
Kash and Kingston (2000) and Kusunoki et al. (1998) make
a similar distinction; the latter use the terms systems and
materials industries.
W.M. Cohen et al. / Research Policy 31 (2002) 13491367 1357
policy and administrative practice can also play a
critical role. For example, the recent ruling in the US
that gene fragments are separately patentable suggests
that the commercialization of a single biotechnology
drug product may now require rights over numerous
patents (Heller and Eisenberg, 1998). In Japan, it is
conceivable that almost all products are complex
per our denition because, in Japan, patents encom-
pass fewer claims, and the claims themselves tend to
be interpreted more narrowly, generating more patents
per product and, in turn, widespread technological in-
terdependence (cf. Granstrand, 2000). For example, in
Japan in the mid 1990s, the average number of claims
per patent was less than ve (JPO, 1998), whereas in
the US, it was approximately 15 (Allison and Tiller,
2001).
17
Regarding the scope of claims, one former
Japanese patent examiner said he felt compelled to
narrow the claims to those spanned by the enable-
ments, rather than broadly interpret the claims, as is
US practice (see also Helfgott, 1990).
18
Our survey
nding that the average number of patent applications
per million US$ of R&D expenditure is 0.6 in the US
versus a much higher 2.8 in Japan (P < 0.0001) is also
consistent with the notion that there are more patents
per commercializable innovation in Japan than in the
US.
19
Thus, with a greater number of patents per prod-
uct and consequently greater interdependence across
17
Sakakibara and Branstetter (2001) explore the effect on R&D
and patenting of the 1988 reform where Japan moved away from
a one claim per patent requirement.
18
This was particularly the case in pharmaceuticals (and perhaps
chemicals more generally), where Japanese rms had lobbied for
stricter interpretations of claims in order to create space for do-
mestic patenting around European and American pharmaceutical
company patents. Recently, however, there have been attempts to
broaden claims in Japan by requiring examiners to justify in writ-
ing their narrowing of claims. In addition, a 1998 Supreme Court
ruling introduced the doctrine of equivalents, further broadening
patent scope.
19
A small portion of this difference may be due to the higher
propensity to patent observed in Japan. Our two surveys collected
data on rms propensities to patent, dened as the reported per-
centage of product or process innovations for which a respondent
applied for patents. The overall propensities to apply for prod-
uct and process patents in Japan are, respectively, 62 and 42%,
and in the US, are 54 and 32%, implying that the different na-
tional propensities to patent account for at most a tenth of the ob-
served differences in patents per R&D US$. While the difference
in patents per R&D US$ may also partly reect different R&D
productivities between the two nations, there is little evidence of
such enormous differences in R&D productivity.
rms technology holdings in Japan, we would expect
greater use of patents (as in US complex product in-
dustries) to protect market access via cross-licensing
and less use to enforce exclusivity. Reinforcing the
impact of greater interdependence, there is also less
incentive to enforce exclusivity in Japan due to a lower
expected payoff to ling infringement suits there.
20
A central question for our purpose is whether poli-
cies that increase the number of patents per commer-
cializable innovation stimulate uses of patents such
as cross-licensing that lead to greater information
sharing, or do such uses largely reect cross-industry
(but not cross-national) differences in product tech-
nology and characteristics. To compare the effects
of national policy versus technology on the uses of
patents, we divide our sample into discrete and com-
plex product industries. Lacking direct measures of
the number of patents per product or process, and fol-
lowing Kusunoki et al. (1998) and our interviews, we
operationalize the distinction between complex and
discrete product technologies by assigning industries
with ISIC codes <2900 (e.g. food, chemicals, drugs,
metals and metal products) to the discrete product
technology category and those with ISIC codes of
2900 or above (e.g. machinery, computers, electrical
equipment, instruments, and transportation equip-
ment) to the complex category, and exclude ISIC
3600 (other manufacturing). We conduct a series of
comparisons of the uses of patents between the two
industry types, across the two countries. While uses
of patents within each of the two groups of industries
are surely heterogeneous, such heterogeneity provides
for a conservative test of group differences by biasing
the differences toward zero.
21
Our expectation is that
20
Plaintiff success rates in infringement suits are much less in
Japan (Hylton, 1993) and, when the plaintiff is successful, payouts
tend to be less. Until 1998, for example, the highest payout in a
Japanese infringement suit was approximately US$ 3.2 million as
compared to over US$ 900 million in the 1990 Polaroid versus
Kodak decision in the US.
21
To compare the uses of patents between the two industry types
and two countries, we use a weighted logistic regression where:
the weights are each respondents country-standardized number of
patent applications, the dependent variable is a specic reason for
patenting, and the independent variables are two binary variables
(COMPLEX and JAPAN) and their interaction. We then report
the means for each cell of the two by two table, and the analysis
of variance estimates for the effects of each variable and the
interaction term. Analogous to a t-test for differences in group
1358 W.M. Cohen et al. / Research Policy 31 (2002) 13491367
the uses of patents that reect mutual dependence
will be more common in complex than in discrete
product industries. Also, we expect such uses to be
more common in Japan than in the US. Finally, and
this is the critical test of the effect of difference in
patent policy, the gap between the uses of patents in
discrete versus complex product industries should be
smaller in Japan than in the US.
To develop data on the uses of patents in the US and
Japan, we asked respondents to report the reasons mo-
tivating their decision to le their most recent patent
application and to report which of these was the most
important. The reasons considered include: to measure
the performance of their rms researchers, to obtain
licensing revenue, to improve their position in nego-
tiations (e.g. cross-licensing negotiations), to prevent
patent infringement suits, to prevent copying, to pre-
vent other rms frompatenting a related invention, and
to enhance the reputation of the rm. The aggregate
(patent application-weighted) results are presented in
Fig. 4.
Aside from the cross-industry differences, we nd
sharp cross-national differences. Generally, we nd
what might be called strategic uses of patents (e.g.
to block others patents, use in negotiations, to prevent
infringement suits) to be common in both nations but
more prevalent in Japan. For example, in the US, pre-
venting rivals from patenting related inventionswhat
we call patent-blockingwas almost as pervasive as
the prevention of copying as a motive for patenting.
22
means, this method permits us both to account for the dichotomous
(yes/no) character of the reasons to patent and, at the same time,
weight those reasons by the number of country-standardized patent
applications to better reect how the average patent is used.
Since some rms account for many more of the patent applica-
tions in an industry than others, using unweighted responses would
give small rms patents relatively greater inuence on the to-
tals. In addition, because of the very different average number of
patent applications between Japan and the US, the Japanese scores
overwhelm the US scores in the cross-industry comparisons. To
reduce this gap, we created a country-standardized patent appli-
cation measure to weight our regressions. We begin with each
countrys unadjusted mean and S.D. (US mean = 29, S.D. = 84;
Japan mean = 438, S.D. = 1220), and then transform the scores
by country for each respondent to create a country standardized
variable with a mean = 100 and S.D. = 250.
22
Since publication of a patent application after 18 months places
the technology in the public domain, the published application
alone, prior to any grant, can also prevent rivals from patenting
identical or closely related inventions. Thus, it is not surprising
For product innovations, 80% of US respondents re-
port blocking as a motive, second only to the preven-
tion of copying which was reported by 99% of the US
respondents (and signicantly above preventing suits,
t = 3.72, P < 0.01). We see a similar result for pro-
cess innovations (also signicantly above preventing
suits, t = 3.97, P < 0.0001). And yet, while this
use of patents is common in the US, Japanese rms
rated patent blocking even higher (country difference
is signicant,
2
[1 d.f.] = 3581, P < 0.0001), with
93% of respondents reporting blocking as a reason for
applying for a patent on a product innovation.
23
For
process innovations, patent blocking in Japan is more
important still (country difference signicant,
2
[1
d.f.] = 5628, P < 0.0001). Defensive patenting also
looms larger in Japan. The prevention of suits, listed
by 90 and 87% of Japanese respondents for product
and process inventions, respectively, is comparable in
Japan to patent blocking as a reason for patenting
(neither the product nor process scores differ signi-
cantly, P > 0.05). The Japanese gures also far ex-
ceed the comparable US numbers of 72 and 61%.
24
Use of patents in negotiations is substantial in Japan
at 86 and 80% for product and process inventions,
respectively, greatly exceeding the US gures of 55
and 46% (both country differences signicant, P <
0.0001).
25
that applicants do not request examination for a sizable share of
applications. For example, of the 341,095 applications submitted
in 1987, requests for examination were submitted by 1994 for
only 47.6% of them (JPO, personal communication).
23
The industry-level results also show the motive of patent block-
ing to be at least as pervasive in Japan as in the US in 18 of 26
industries.
24
The country differences are signicant (for product,
2
[1
d.f.] = 107.2, P < 0.0001; for process,
2
[1 d.f.] = 98.6, P <
0.0001).
25
There is one other contrast between the two national samples
with regard to reasons for patenting. Our data indicate that US
rms rarely (i.e. <10% of respondents) claim use of patents as
an internal performance measure as a reason to le a patent
application. In contrast, over half of the Japanese respondents
report using patents in this way (country differences signicant for
both product and process, P < 0.0001), although they almost never
cite it as the most important reason. Westney (1994) claims that it
is this measurement of internal performance that mainly explains
why rms patent so prolically in Japan. Our investigation suggests
that while performance measurement motivates Japanese rms to
patent, the other strategic rationales for patenting highlighted
earlier are more central to the way that patents are regarded in
both countries.
W.M. Cohen et al. / Research Policy 31 (2002) 13491367 1359
Fig. 4. Reasons to patent product and process innovations.
1360 W.M. Cohen et al. / Research Policy 31 (2002) 13491367
Fig. 5. Reasons to patent product innovations, by country and by discrete and complex product industries.
We now consider our conjecture that greater mu-
tual dependence in complex product industries and in
Japan leads to greater use of patents for negotiations.
Fig. 5 shows for each country the weighted averages
for discrete and complex product industries for various
reasons to patent product innovations. As noted earlier,
the use of patents to improve your position in negoti-
ations with other rms, for example, in cross-licensing
agreements (labeled Negotiate) is more common in
Japan than in the US. Use of patents in negotiations is
also more common in complex product industries than
in discrete product industries. The overall averages
across both countries (not shown in Fig. 5) are 82%
yes for complex product industries, but only 49%
yes for discrete product industries (P < 0.0001).
26
This conrms our expectations about the role of indus-
try type on uses of patents. However, Fig. 5 shows that
this gap is primarily among US rms because Japanese
26
The differences for process innovations are similar (73% for
complex versus 44% for discrete, P < 0.0001). In general, we
are less certain about our predictions for process patents, because
it is more difcult to know about the nature of the manufacturing
process, and, thus, whether commercializable process innovations
tend to be discrete or complex in a given industry.
rms are equally likely to patent for use in negotiations
in complex and discrete product industries. For the US
respondents, 78% of those in complex product indus-
tries report using patents for negotiations, while only
32% do so in discrete product industries. In contrast,
in Japan, the percent reporting use in negotiations as a
reason to patent is not only higher overall than in the
US, but also nearly the same across the two groups of
industries, about 85%. Thus, across the two groups of
industries, Japanese rms are more similar than Amer-
ican rms in their use of patents for negotiations.
As noted previously, patent blocking is another
common motive for patenting in both nations, though
more prevalent in Japan. Cohen et al. (2000) identify
two forms of patent blocking, though only one of
these tend to foster information sharing. In the rst
form of patent blockingwhich does not promote
information sharingrms patent numerous substi-
tutes for some core invention, adopting what we call a
fence strategy.
27
Alternatively, in complex product
27
For example, in the 1940s, du Pont patented over 200 substi-
tutes for Nylon to protect its core invention (Hounshell and Smith,
1988).
W.M. Cohen et al. / Research Policy 31 (2002) 13491367 1361
industries where a given innovation is comprised of
numerous complementary elements, rms hold rights
over technologies that others need to compete, and
vice versa, creating a condition of mutual depen-
dence. Exploiting this mutual dependence, blocking
in this case takes the form of patenting complements
(in contrast to substitutes in the fence strategy) to
preclude rivals commercial use of their patents. By
preventing other rms from controlling all the rights
necessary to commercialize products, a rm will as-
sure itself of player status in an industry in the
sense of assuring itself access to the technology of
rivals via cross-licensingor, at least conferring the
freedom of operation that comes from the ability to
countersue if sued.
28
To identify the use of patents for such a player
strategy, we create a variable, player, which is yes
if the rm uses the same patent both to prevent oth-
ers from patenting related inventions and for negoti-
ating cross-licensing agreements, and no otherwise.
In other words, we are identifying respondents who
use patents to block in order to compel other rms to
cross-license with them and to strengthen their power
in such negotiations. We expect the player strategy to
be more common in complex product than in discrete
product industries and more common in Japan than
in the US. The row labeled Player in Fig. 5 shows
28
In addition to the work of Hall and Ziedonis (2001) for the
US semiconductor industry and Granstrand (2000) for selected
Japanese rms, our interviews conrm the importance of this
player strategy. One American R&D manager from the com-
munications equipment industry (a complex product industry)
stated: Mostly, your patents are used in horse trading. You come
together and say, Heres our portfolio. In our industry, things all
build on each other. We all overlap on each others patents. Even-
tually we come to some agreement: You can use ours and we
can use yours. Japanese rms echoed this sentiment, noting that
products tended to span patent portfolios and that patents were
often used in cross-licensing agreements to secure freedom of
operation, the ability to make a product without interference from
rival patent holders. These rms further conrmed that patents are
traded largely to gain access to information and freedom of oper-
ation. One Japanese ofce equipment manager states, Patents are
not a good tool to get money. We always consider shaking hands
(royalty-free cross-license). This cross-license gives us more free-
dom of design. Money is not so important as getting the others
technology when we license to other companies. Similarly, an
IBM publication states: The IBM patent portfolio gains us the
freedom to do what we need to do through cross-licensingit
gives us access to the inventions of others that are key to rapid
innovation. (Think 1990, no. 5, pp. 10,11).
by country and industry type the percent of respon-
dents who report using a patent for blocking and for
negotiations (but not for licensing revenue). The over-
all averages across both nations (not shown in Fig. 5)
show that the use of the player strategy is indeed sig-
nicantly higher in complex than in discrete product
industries (73% versus 46% for product innovations,
60% versus 42% for process innovations, with both
differences P < 0.0001). Fig. 5 shows, however, that
this result is driven by the difference across indus-
try type only in the US. In Japan, the player strategy
is used equally across complex and discrete product
industries. Moreover, this strategy is employed more
intensively in Japan than in the US, with discrete
and complex product industries in Japan both scoring
higher than the US average for even complex product
industries (the interaction effect is signicant, P <
0.0001).
29
As described earlier, a fence strategy entails a sec-
ond form of patent blocking. In this case, rms patent
substitutes for core inventions in order to maintain ex-
clusivity over the technology. To the degree, however,
that protecting some core invention patent requires it
to be surrounded by numerous other patents, the trans-
action costs of licensing increases and the likelihood of
licensing declines.
30
Also, since the goal is exclusivity,
these patents tend not to be used for cross-licensing ne-
gotiations. To estimate how commonly the fence strat-
egy is employed, we create a variable, Fences, that
is yes when rms report using a patent for patent
blocking (i.e. to patent to prevent others frompatenting
related inventions), but not for licensing nor for use in
negotiations, and no otherwise. We expect the fences
strategy to be most common in discrete product indus-
tries, where the relatively low number of patentable
elements per innovation makes fence building feasi-
ble. We also expect this strategy to be more common
in the US, with a patent system that, as noted previ-
ously, provides greater incentive to enforce exclusivity
given higher expected payouts in infringement cases.
The row labeled Fence in Fig. 5 shows pat-
ent application-weighted percentages of respondents
29
We nd similar results for process patents.
30
Note that to the degree that patents are extremely effective in
preventing copying, as is especially true in pharmaceuticals, then
fence-building is not necessary since one or two patents can confer
control of the market.
1362 W.M. Cohen et al. / Research Policy 31 (2002) 13491367
reporting blocking as a motive for patenting, but not
use in negotiations or licensing. The overall averages
for product innovations (not shown in the gure) for
discrete product industries is 34%, but for complex
is only 10% (P < 0.0001). Fig. 5 shows, as conjec-
tured, that the use of patents to build such fences
to be much less common in Japan no matter the in-
dustry (both the country effect and the interaction
effect are signicant, P < 0.0001).
31
Thus, while in
discrete product industries in the US, patent blocking
is a mechanism to generate appropriability through
building patent fences, such fence building in Japan
is rare across either kind of industry.
Our results indicate that Japanese, compared to the
US, rms, are much less likely to use their patents
to achieve exclusivity (via, for example, a fence strat-
egy), and much more likely to use them as a means
of gaining market access, information from rivals and
freedom of operation and design (the player strategy).
By fostering technological interdependence across ri-
vals and diminishing the returns to the enforcement
of exclusivity, the Japanese patent system spawns a
patent strategy across its entire manufacturing sector
that is largely conned to the complex product tech-
nologies in the US. Pervasive interdependence across
Japanese rms patent holdings also encourages more
patenting because successful execution of the player
strategy requires the bargaining power that comes, in
part, with having many patents.
We conjecture, therefore, that by encouraging
cross-licensing and aggressive patenting, the Japanese
patent system generates more disclosure and exchange
of information. We suggest that these impacts on in-
formation ows of the greater number of patents per
product reinforce the effects of the Japanese systems
policies patent priority, bearing on disclosure and
patent opposition discussed earlier.
5.2. Patents as a channel of information ow
In this section, we consider whether patents are in-
deed a more important channel of intraindustry R&D
knowledge owin Japan than in the US as conjectured.
In our survey, we asked respondents to report the
most important reason for not applying for a patent on
the most recent invention which they decided not to
31
We nd a similar result for process patents.
patent. The reasons include: difculty in demonstrat-
ing the novelty of the invention, the ease of legally in-
venting around, the information disclosed in a patent
application, and the costs of applying for or defend-
ing a patent. The ndings conrm that patents more
effectively serve the information disclosure function
in Japan than in the US. Specically, Japanese rms
are much more likely to report concerns over disclo-
sure as the most important reason not to apply for a
patent (46% versus 26% in the US,
2
[1 d.f.] = 54.0,
P < 0.0001).
In our survey, we also examine the importance of
different channels of R&D information ow across ri-
vals. We asked both Japanese and the US respondents
to score on a four point Likert scale the importance to
a recently completed, major R&D project of each
of the following sources of information on rival R&D:
patents, informal information exchange, products (via,
for example, reverse engineering), publications, public
meetings and conferences, recent hires, licenses, joint
or cooperative ventures, contracts with other rms and
trade associations. Presenting the percent of respon-
dents reporting that a given information source was
important or very important (i.e. scoring at least a
three on the four-point scale), Fig. 6 provides com-
parisons for the aggregate samples.
First, we nd the same ve sources of informa-
tion on rival R&Dnamely publications, public
meetings or conferences, patents, informal informa-
tion exchange and competitors products (via, for
example, reverse engineering)to be the ve most
important sources in both nations.
32
A comparison
of these ve information sources shows the major
difference between the two countries is in the im-
portance of patents. The absolute score for patents
in Japan exceeds that of the US by 70% (
2
[1 d.f.]
= 175.7, P < 0.0001). In addition, patents are the
most important channel in Japan, scoring over 30%
32
Fig. 6 also supports the conclusion of Section 3 that overall
intraindustry R&D information ows are greater in Japan by show-
ing that eight of the ten sources are rated more highly in Japan
than in the US. (The country differences are signicant (P < 0.05)
for all but publications.) Moreover, the only two sources that are
rated in absolute terms more highly in the US, namely recent hires
and trade associations, are relatively unimportant in both nations
as sources of information on rival R&D. Recent hires are easily
the least important channel in Japan, which is consistent with the
convention of lifetime employment in Japan.
1364 W.M. Cohen et al. / Research Policy 31 (2002) 13491367
higher than the next most highly ranked channel in
Japan, publications (t = 9.25, P < 0.0001). In the
US, patents are ranked third behind publications and
informal information exchange, and quite close to
the sources of public meetings and conferences, and
competitors products (none of these are signicantly
below patents).
33
While patents play a role in diffus-
ing information across rivals in the US, they appear
to be the dominant channel in Japan.
34
Moreover,
while some of the information conveyed by patents
is conveyed via licensing, Fig. 6 shows that licenses
are a much less important source of information
33
Two other sources score signicantly higher in Japan, licenses
and contracts. However, these two sources are relatively unimpor-
tant sources of information in both nations. They are the two least
important information sources in the US (signicantly below the
next lowest, hires, t = 4.30, P < 0.0001) and are ranked only
seventh and eighth in Japan (a little below joint ventures, t = 1.8,
P < 0.10). The higher ranking of licenses and contracts in Japan,
and particularly their much higher absolute scores, is consistent
with an environment where patents play a more important role in
information diffusion since licensing typically requires a technol-
ogy having been patented, and contracts often involve patented
technologies.
34
Some interview respondents suggested that rms may be will-
ing to disclose information through publications or meetings only
after they apply for a patent to protect that information, imply-
ing that patents play a more important role in promoting infor-
mation ows than our results suggest. While we could not test
this conjecture directly, we did examine the industry-level cor-
relations between the mean importance of publications or meet-
ings/conferences as an information source and the amount of
patenting in an industry (measured as either the percent of re-
spondents who had patented in the prior 3 years, the percent of
product innovations patented, the percent of process innovations
patented or the mean number of the patent applications applied
for among our respondents). These eight correlations range from
0.14 to +0.30. The only two signicant (P < 0.10) correlations
are between publications as an information source and either the
propensity to patent process innovations (r = 0.24) or the mean
number of patent applications (r = 0.30). If we control for the
generic quality of information in an industry (which might in-
crease both patenting and publications) by controlling for the im-
portance of university research (measured as either the percent of
projects that build on research ndings, the percent of respondents
for whom university research suggested new projects, or the per-
cent for whom university research contributed to the completion of
existing projects), these relationships weaken and become insignif-
icant. Thus, while patenting may be a prerequisite for publishing
or disclosing information about an invention in public meetings
and that may well be the norm in selected cases such as the drug
industrywe nd only limited industry-level evidence suggesting
a link between patenting and the disclosure of research via publi-
cations or presentations for the manufacturing sector as a whole.
than patents generally, suggesting that a considerable
portion of the information contained in patents is
conveyed via public disclosure.
35
These ndings suggest that differences in the two
nations patent systems may provide an important
explanation for the differences in intraindustry R&D
information ows and spillovers. There are, however,
other possible explanations. These might include, for
example, cross-national differences in the signicance
of innovations (major versus incremental), manage-
rial practices regarding the monitoring of extramural
information, the role of R&D consortia, geographic
proximity, and market structure. Although our ability
to empirically probe these alternative explanations is
limited, we do so in a supplemental analysis (Cohen
et al., 2001). In this exercise, we nd little evidence
for the claims that the difference in R&D spillovers is
due to more incremental innovation in Japan, Japanese
rms expenditure of more effort on monitoring ri-
vals or greater geographic proximity of rivals. We
nd some evidence for the impact of Japanese R&D
consortia on information ows, and some evidence in
behalf of a role for market structure. Nonetheless, our
ndings suggest that patent policy plays an important
role in generating intraindustry spillovers, and, one
might conjecture, the efciency of innovative effort
as well.
6. Conclusion
Our comparison of appropriability conditions and
R&D-related information ows in the US and Japan
suggest some similarities, but mostly differences be-
tween the two nations. In both countries, the use of
lead time advantages and complementary manufactur-
ing and sales capabilities gure prominently in the way
rms protect their inventions. Differences are, how-
ever, numerous. Secrecy, for example, is prominent in
the appropriability strategies of US but not Japanese
rms. Though comparably rated in absolute terms,
patents are reported to be the least effective among the
major appropriability mechanisms in the preponder-
ance of US industries, but among the most effective
35
The difference between the scores for patents and licenses are
signicant in each country (the paired t-test results for the US are
t = 14.87, P < 0.0001; for Japan, t = 18.90, P < 0.0001).
W.M. Cohen et al. / Research Policy 31 (2002) 13491367 1365
in Japan. Notwithstanding the relative effectiveness of
particular mechanisms, our dataespecially our mea-
sures of imitation lagsalso suggest that it is more
difcult to protect rents due to invention in Japan than
in the US in almost all manufacturing industries (ex-
cept computers). Consistent with this observation, we
observe that R&D-related information ows across ri-
vals and, we infer, R&D spillovers as well, are signi-
cantly greater in Japan.
Our cross-national, within-industry comparisons
suggest that not only industry characteristics differen-
tiate appropriability conditions and R&D spillovers,
but also factors that cut across industries, such as
the institutions, policies, cultures or norms that are
distinctive to nations. Building upon the work of
Ordover (1991), Granstrand (2000) and others, we
proposed that one difference between the two nations,
namely their respective patent systems, may at least
partly account for the differences we observe, espe-
cially in intraindustry R&D information ows. These
scholars argue that differences between the two na-
tions in rules of patent priority, disclosure and patent
opposition strengthen the contribution of patents to
information ows across rms in Japan. Our data
showing that patents are a more important channel
of R&D information ow across rivals in Japan sup-
port this claim. In fact, we nd that patents are the
most important channel for such information ow
in Japan.
Building on our analysis of appropriability condi-
tions and the role of patents in the US (Cohen et al.,
2000), we suggest that yet another difference between
the US and Japanese patent systems might further
explain this nding. We argued that the lower num-
ber and narrower construction of claims per patent
that yield a greater number of patents per product
in Japan should spawn greater mutual dependence
across the technological holdings of rivals, and, in
turn, a greater tendency to use patents for mutual
blocking, cross-licensing and negotiationsuses that
tend to promote greater information sharing across
rivals. We nd that this player strategy, which in
the US is concentrated in complex product industries,
applies equally across discrete and complex product
industries in Japan, suggesting a policy impact on a
behavior that Cohen et al. (2000) conjectured to be
driven largely by physical differences between prod-
ucts. This pattern of use of patents also implies that
Japanese rms tend to use patents more to assure
their inclusion and freedom of operation even in dis-
crete product markets where American rms tend to
use patents more for exclusion. The use of patents in
Japan for maintaining market access and freedom of
operation rather than for exclusion is reinforced by
the comparatively low payoffs to infringement suits
in Japan.
In addition to strengthening R&D spillovers,
the technological interdependence induced by the
Japanese patent system also likely stimulates patent-
ing activity. A rms successful execution of the
player strategy depends in part on the bargaining
power that comes with possession of a larger patent
portfolio. Patents become weapons in mutually rein-
forcing, noncooperative strategic interactions where
rms feel increasingly compelled to patent either be-
cause they need to protect themselves from suits or
from being blocked, or they want to block rivals or use
patents as bargaining chips in negotiations (cf. Cohen
et al., 2000; Hall and Ziedonis, 2001; Granstrand,
2000). Consequently, rms in complex product indus-
tries in the US and across the entire manufacturing
sector in Japan may be accumulating more patents,
and incurring more related litigation and other costs
than what is collectively optimal. Our results suggest
that if patenting is excessive, however, the associated
social costs may be offset more in Japan to the degree
that patents, ceteris paribus, diffuse information more
effectively and the incentives to litigate are less.
An important limitation to our study is that we
are observing similarities and differences in R&D
spillovers and appropriability conditions between the
US and Japan at a point in time. Patent law and
administrative and judicial practice have been chang-
ing in both nations, however, even since our survey
was administered in 1994. Since differences in the
patent systems impact the way patents are used and
the degree of protection they afford, the patterns
we observe today may not endure. Policy changes
also raise the question of the degree to which differ-
ences in institutional environments across nations are
exogenous, reecting nations cultures and unique
histories, or are endogenous to global economic com-
petition. In the case of patent policies, there appears
to be an important endogenous component, stimulated
by high prole infringement suits between US and
Japanese rms, trade disputes and, more generally,
1366 W.M. Cohen et al. / Research Policy 31 (2002) 13491367
competition between the two nations industries and
rms.
36
Mazzoleni and Nelson (1998) point out that the
academic literature and much of the policy debate
on patents have focused on the impact of patents
on appropriability (an exception is Aoki and Spiegel
(1998)). By showing that patents can have important
and pervasive effects on information diffusion and
R&D spillovers, and perhaps, in turn, on the efciency
of innovative effort, our results imply that the infor-
mation diffusion effects of patents deserve at least
equal time from policymakers and academics. The
US has recently adopted a policy of automatic dis-
closure of (most) patent applications after 18 months.
One may question whether automatic disclosure will
signicantly affect R&D spillovers in the absence of
other judicial and administrative practices and poli-
cies that condition Japanese rms incentives of when
to patent, what to patent and how to use patents
all in ways that foster technology diffusion.
37
The
rst-to-le priority rule induces rms to patent ear-
lier in the innovation process. The opposition pro-
cess strengthened rms incentives to monitor rivals
patents. Narrow construction of claims and fewer
claims per patent (and, hence, more patents per prod-
uct) foster interdependence and cross-licensing, while
low infringement awards further diminish the returns
to a strategy of exclusion.
Although we would like to see policymakers
and academics pay more attention to the impact of
patents on technology diffusion, we are not argu-
ing that a patent system that promotes diffusion to
the extent observed in Japan is, on balance, socially
36
Ironically, some of the differences between the two nations
patent systems may reect different responses to the same inter-
actions. For example, Japanese rms have aggressively embraced
patents partly in reaction to prominent infringement suits led by
American against Japanese rms over the prior 1015 years. At
the same time, American rms may have drawn a different lesson
from the same events, namely that patents provide only partial
protection and must be supplemented when possible by other ways
of protecting their inventions, such as secrecy.
37
In a theoretical treatment, Aoki and Spiegel (1998) consider
the impacts on R&D spending and patenting of adoption of an
18 month automatic disclosure rule alone. Their model predicts
that such a move would discourage patenting in industries where
patents are only moderately effective, and that R&D would decline.
As we note earlier, however, Japanese R&D intensities and patent
propensities are on average higher than those of the US.
desirable.
38
A key question is how more disclosure
affects R&D investment incentives. To the degree
that disclosure does not dampen incentives to conduct
R&D itself, a patent regime that strongly supports
disclosure yields an unambiguous social welfare ben-
et, and, indeed, Japanese rms R&D intensities and
patent propensities exceed US rms overall. Before
we can conclude, however, that a patent system that
aggressively promotes diffusion yields a net social
benet, we require a more complete empirical anal-
ysis that considers the offsetting incentive and com-
plementarity (or efciency) effects of intraindustry
information ows on R&D,
39
the mutual determi-
nation of R&D and patenting, as well as the other
factors conditioning R&D.
Acknowledgements
For comments, we thank Anthony Arundel, Setsuko
Asami, Ashish Arora, Marco Ceccagnoli, Jesper
Christensen, Rebecca Henderson, David Hounshell,
Steven Klepper, Robert Kneller, Mariana Mazzucato,
Ichiro Nakayama, Manny Schechter, members of
the Institute of Intellectual Property (Tokyo), partic-
ipants in the DRUID Nelson and Winter Conference
of June 2001, the White House Working Group on
Intellectual Property, the NBER Workshop on R&D
and Productivity, the Innovation Survey Data Confer-
ence (Ministry of Industry, Paris), the University of
Chicagos Law and Economics Seminar, the Wharton
and UCLA strategy seminars and the Purdue Univer-
sity economics seminar. Principal research support
was provided by the Alfred P. Sloan Foundation, the
Center for Global Partnership, the National Institute
of Science and Technology Policy of Japan, and Grant
No. 99-76384 of the US National Science Foundation.
Supplemental support was provided by the Carnegie
Bosch Institute of Carnegie Mellon University and
Hitotsubashi University Institute of Innovation Re-
search. Authors are listed alphabetically.
38
One possibility is that different patent systems may be better
suited to some industries versus others.
39
Studies that have considered the offsetting effects of intrain-
dustry information ows on R&D productivity and incentives in-
clude Spence (1984), Levin and Reiss (1984), Cohen and Levinthal
(1989) and Cohen and Walsh (2000).
W.M. Cohen et al. / Research Policy 31 (2002) 13491367 1367
References
Agresti, A., Agresti, B., 1979. Statistical Methods for the Social
Sciences. Dellen, San Francisco, CA.
Allison, J.R., Tiller, E.H., 2001. Statistical Analysis of Internet
Business Method Patents, Draft report to the National
Academies STEP Board, October 7.
Aoki, R., Spiegel, Y., 1998. Public Disclosure of Patent
Applications, R&D, and Welfare, Working Paper No. 30-98.
Tel Aviv University.
Arundel, A., 2001. The relative effectiveness of patents and secrecy
for appropriation. Research Policy 30, 611624.
Bowker Press, 1994. Directory of American Research and
Technology. Bowker Press, New York.
Cohen, W.M., Levinthal, D.A., 1989. Innovation and learning:
the two faces of R&Dimplications for the analysis of R&D
investment. Economic Journal 99, 569596.
Cohen, W.M., Nelson, R.R., Walsh, J.P., 2000. Protecting their
Intellectual Assets: Appropriability Conditions and Why US
Manufacturing Firms Patent (or Not), NBER Working Paper
no. 7552.
Cohen, W.M., Nelson, R.R., Walsh, J.P., 2001. R&D Spillovers,
Patents and the Incentives to Innovate in Japan and the United
States. Carnegie Mellon University, Mimeo.
Cohen, W.M., Walsh, J.P., 2000. R&D Spillovers, Appropriability
and R&D Intensity: A Survey-Based Approach. Report to
the Economic Assessment Ofce, NIST Advanced Technology
Program.
Dillman, D.A., 1978. Mail and Telephone Surveys: The Total
Design Method. Wiley, New York.
Granstrand, O., 2000. The Economics and Management of
Intellectual Property. Edward Elgar, London.
Griliches, Z., 1992. The search for R&D spillovers. Scandinavian
Journal of Economics 94 (Suppl. 3), 529547.
Hall, B., Ziedonis, R.M., 2001. The patent paradox revisited: an
empirical study of patenting in the US semiconductor industry,
19791995. Rand Journal of Economics 32, 101128.
Helfgott, S., 1990. Cultural differences between the US and
Japanese patent systems. Journal of the Patent and Trademark
Ofce Society 72, 231238.
Heller, M.A., Eisenberg, R.S., 1998. Can patents deter innovation?
Science 280 (May 1), 671698.
Hounshell, D.A., Smith, J.K., 1988. Science and Corporate
Strategy. Cambridge University Press, Cambridge.
Hylton, K., 1993. Asymmetric information and the selection of
disputes for litigation. Journal of Legal Studies 22, 187
210.
Kash, D.E., Kingston, W., 2000. Patents in a World of Complex
Technologies. George Mason University, Mimeo.
Kusunoki, K., Nonaka, I., Nagata, A., 1998. Organizational
capabilities in product development of Japanese rms. Organi-
zation Science 9, 699718.
Levin, R.C., Klevorick, A.K., Nelson R.R., Winter, S.G., 1987.
Appropriating the Returns from Industrial R&D, Brookings
Papers on Economic Activity. pp. 783820.
Levin, R.C., Reiss, P.C., 1984. Tests of a schumpeterian model
of R&D and market structure. In: Griliches, Z. (Ed.), R&D
Patents and Productivity. University of Chicago Press for the
NBER, Chicago.
Manseld, E., 1986. Patents and innovation: an empirical study.
Management Science 32, 173181.
Mazzoleni, R., Nelson, R.R., 1998. The benets and costs of strong
patent protection. Research Policy 27, 273284.
National Science Foundation, 2000. National Patterns of R&D
Resources: 2000. Data Update, //www.nsf.gov/sbe/srs/nsf01309.
NISTEP (National Institute of Science and Technology Policy),
1997. Technological opportunities and appropriating the returns
from innovationcomparison of survey results from Japan and
the US, Report no. 48. Tokyo.
Okada, Y., 1997. Patent system and R&D in Japan. In: Goto,
A., Odagiri, H. (Eds.), Innovation in Japan. Clarendon Press,
Oxford.
Ordover, J., 1991. A patent system for both diffusion and exclusion.
Journal of Economic Perspectives 5 (1), 4360.
Quillen, C.D., Webster, O.H., 2001. Continuing patent applications
and performance of the US Patent Ofce. Federal Circuit Bar
Journal 11 (1), 121.
Sakakibara, M., Branstetter, L., 2001. Do stronger patents induce
more innovation? Evidence from the 1988 Japanese Patent Law
Reforms. Rand Journal of Economics 32, 77100.
Scherer, F.M., et al., 1959. Patents and the Corporation, 2nd
Edition. Boston, privately published.
Scotchmer, S., Green, J., 1990. Novelty and disclosure in patent
law. Rand Journal of Economics 21, 131146.
Spence, A.M., 1984. Cost reduction, competition, and industry
performance. Econometrica 52, 101121.
Taylor, C.T., Silberston, Z.A., 1973. The Economic Impact of the
Patent System: A study of the British Experience. Cambridge
University Press, Cambridge.
US G.A.O., 1997. Intellectual Property: Comparison of Patent
Examination Statistics for Fiscal Years 1994 and 1995, GAO/
RCED-97-58. General Accounting Ofce, Washington, DC.
Westney, E., 1992. Country patterns in R&D organization: the
United States and Japan. In: Kogut, B. (Ed.), Country Compe-
titiveness. Oxford University Press, New York, pp. 3653.
Westney, E., 1994. The evolution of Japans industrial research
and development. In: Aoki, M., Dore, R. (Eds.), The Japanese
Firm. Oxford University Press, New York.