Content uploaded by Balazs Kovacs
Author content
All content in this area was uploaded by Balazs Kovacs on Aug 16, 2018
Content may be subject to copyright.
NATURE BIOTECHNOLOGY VOLUME 36 NUMBER 4 APRIL 2018 307
Kyle Jensen, Balázs Kovács and Olav Sorenson
are at the Yale School of Management, Yale
University, New Haven, Connecticut, USA.
e-mail: kyle.jensen@yale.edu, balazs.kovacs@
yale.edu or olav.sorenson@yale.edu
ber of inventors listed on the patent, and the
size of the organizational assignee. Detailed
descriptions of the models, as well as their
robustness, are provided in the Supplementary
Data. For example, we demonstrated the
robustness of the results to using subsamples of
patent applications—such as applications with
only US inventors and only applications from
large organizations—and to using alternative,
nonlinear specifications of the inventor-team
composition.
Results and discussion
In Figure 1, the dark-blue bars depict ‘raw’ gen-
der differences, and the light-blue bars depict
gender differences after introduction of a fixed
effect for each application’s primary technology
class in the United States Patent Classification
(USPC) system (https://www.uspto.gov/
web/patents/classification/). (In the USPC,
each application submitted to the USPTO
is assigned to one or more of >400 different
USPC classes, which reflect the subject mat-
ter of the application, in categories as diverse
as ‘apparel’, ‘music’, ‘surgery’, and ‘molecular
biology and microbiology’. These classifica-
tions are used to assign the patent application
to particular groups of patent examiners.) As
the figure illustrates, men and women dif-
fered less in their outcomes after adjustment
for the technology class. For example, the two
topmost blue bars indicate that women inven-
tors were 21% less likely than men inventors to
have their application accepted, but that dif-
ference declined to 7% after technology-class
fixed effects were included. This effect could be
viewed as an example of ‘Simpson’s paradox’6;
that is, two-thirds of the diminished prob-
ability of women’s applications being accepted
stemmed from women applying at higher
rates than men to technology classes with
lower acceptance rates. In those classes, it is
Gender differences in obtaining and maintaining
patent rights
Kyle Jensen, Balázs Kovács & Olav Sorenson
An examination of the prosecution and maintenance histories of approximately 2.7 million US patent applications
indicates that women have less favorable outcomes than men.
Although women make up half of the popu-
lation, they represent just 10% of US pat-
ent inventors and only 15% of inventors in
the life sciences1–4. By tracking patent appli-
cations through the prosecution process, we
found disparities between men and women
inventors in the processes of obtaining and
maintaining patent rights. Patent applications
by women inventors were found to be more
likely to be rejected than those of men, and
those rejections were less likely to be appealed
by the applicant team (inventor, assignee, and
prosecuting attorney). Conditional on being
granted, patent applications by women inven-
tors had a smaller fraction of their claims
allowed, on average, than did applications by
men. Further, those claims allowed had more
words added during prosecution, thus reduc-
ing their scope and value. The granted patents
of women inventors also received fewer cita-
tions than those of men and were less likely to
be maintained by their assignees. Surprisingly,
many of these effects were larger in the life sci-
ences than in other technology areas.
Methodology
Our study examined the individual prosecu-
tion histories of approximately 2.7 million US
utility patent applications from the years span-
ning 2001 to 2014 (ref. 5). The US Patent and
Trademark Office (USPTO) recently released
these data in aggregate. In the past, researchers
could access these data only one application at a
time, through the USPTO’s Patent Application
Information Retrieval system (https://portal.
uspto.gov/pair/PublicPair/), thus hindering
the large-scale empirical study of patent-pros-
ecution outcomes. We joined these prosecu-
tion histories with the maintenance-fee and
full-text patent databases available from the
USPTO. The joined data allowed us to inspect
the communication between applicants and
examiners, the manner in which application
claims changed during prosecution, the dates
of various communications, the payment of
maintenance fees, the influx of forward cita-
tions, and other phenomena.
We determined the probable gender of each
inventor by using forename gender distribu-
tions available from the US Social Security
Administration (https://www.ssa.gov/oact/
babynames/limits.html) and from two com-
mercial databases (see Supplementary Data).
In the covered population, 94.1% of forenames
were associated at least 95% of the time with
only one gender. If an inventor had a highly
gendered forename, we accordingly classified
that inventor as either a man or a woman. This
approach allowed us to classify the probable
genders of 89% of the inventors listed on the
applications (detailed discussion and analysis
of the classification process, including possi-
ble selection issues, in Supplementary Data).
Because most applications listed multiple
inventors, we calculated a ‘proportion women’
variable: the number of women inventors
divided by the total number of inventors on
each application. When we refer to effect sizes,
the disparities between men and women repre-
sented a shift in this variable from 0% to 100%,
from all men to all women inventors.
We used a series of linear regressions and
Poisson count models to estimate the asso-
ciations between gender and various patent-
prosecution outcomes (Fig. 1). These models
included controls for a variety of patent attri-
butes, such as the number of claims, the num-
PATENTS
© 2018 Nature America, Inc., part of Springer Nature. All rights reserved.
308 VOLUME 36 NUMBER 4 APRIL 2018 NATURE BIOTECHNOLOGY
Patents by women inventors in the life sciences
also received 28% fewer forward citations from
other inventors.
The data available did not allow us to isolate
the mechanisms responsible for the gender dif-
ferences in Figure 1—we were able to assess
only the direction and magnitude of these
differences. However, the natural variation in
forename frequencies allowed us to gain some
insight into the degree to which these differ-
ences arose from the applicant side—the inven-
tor, assignee, and attorney—as compared with
other parties. The inventors themselves are
obviously aware of their own gender. Similarly,
their employers and the attorneys represent-
ing them probably have firsthand knowledge
of the inventors’ genders. In contrast, the pat-
ent examiners and others must generally infer,
either consciously or subconsciously, the gen-
der of the inventors according to the forenames
more challenging for anybody to get a patent
approved, regardless of gender.
Even after adjustment for the differences
across patent technology classes, however,
women inventors still had less favorable expe-
riences in nearly all outcomes. All else equal,
relative to a team of all-men inventors, pat-
ent applications by teams of all women were
2.5% less likely to be appealed if rejected.
Conditional on being granted, these applica-
tions, on average, had the number of indepen-
dent claims reduced by one-fifth of a claim; had
the number of words in their claims increased
by 2.5%, thus narrowing the scope of these
claims7; were 4.3% less likely to be maintained
by their assignee; received 11% fewer citations
from other patent applicants; and received
3.5% fewer citations from patent examiners.
Forward citations trace the acknowledged
contributions of prior art and are often used as
measures of a patent’s importance, scope, and
value8,9. (These statistics all refer to the light-
blue series in Fig. 1, which includes technol-
ogy-class fixed effects, and appear in tabular
form in the Supplementary Data.)
Although women might be expected to fare
better in the life sciences, given their relatively
higher representation in those fields, the data
show no such pattern. The pink bars in Figure
1 depict the gender differences within the
subset of patents bearing life science classifi-
cations (description of how these are identified
in Supplementary Data). For all outcomes that
differed for the life sciences subset compared
with the population of patents as a whole, the
disparities in the life sciences appeared more
disadvantageous to women. For example, in the
life sciences, a team of all-women inventors was
found to be 11% less likely than a team of all
men to have its patent application accepted.
Figure 1 Estimated differences for teams of all women inventors relative to teams of all men, in the processes of obtaining and maintaining patent rights.
Wide bars, point estimates; narrow bars, 95% confidence intervals (full model specifications in Supplementary Data). Teams with higher proportions of
women had more negative outcomes during patent prosecution. For example, the topmost dark-blue bar indicates that patent applications by teams of all
women inventors were 21% less likely to be granted than similar applications by teams of all men. The light-blue bar accounts for technology-class fixed
effects (women are overrepresented in technology areas with lower acceptance rates); the topmost bar for example, indicates that even after accounting for
technology-class fixed effects, all-women teams had a 7% lower probability of acceptance. The pink bars indicate the differences for patents in technology
classes related to the life sciences. The final two bars (light and dark green) depict the estimated effects within two subsets of single-inventor patent
applications. By examining the effects for inventors with common versus rare names, they provide an indication of the degree to which the gender differences
stem from the applicant side—inventor, assignee, and attorney—versus the examiner side. The first two green bars, for example, suggest that approximately
two-thirds of the lower probability of acceptance for applications with women inventors comes from the examiner side.
1.0 0.8 0.6 0.4 0.2 0.0
Effect size: inventorship by women relative to men
Proportional difference in
forward citations from examiners
Proportional difference in forward
citations from other applicants
Difference in probability of
assignee paying maintenance fee
Proportional difference in time
between application and issuance
Proportional difference in
independent-claim word count
Difference in number of
independent-claims accepted
Difference in probability of
appealing a final rejection
Difference in probability of
patent being granted
Without patent-class FE
With technology-class FE
Life sciences patents (with FE)
Most common names (with FE, single inventor)
Most rare names (with FE, single inventor)
PATENTS
© 2018 Nature America, Inc., part of Springer Nature. All rights reserved.
NATURE BIOTECHNOLOGY VOLUME 36 NUMBER 4 APRIL 2018 309
the lower probability of acceptance for applica-
tions with women inventors stemmed from the
examiner side. Second, future patent applicants
cited the patents of women with common names
30% less frequently than those of men with com-
mon names. The patents authored by women
with rare forenames, and who were therefore not
easily identified as women, were cited approxi-
mately 20% more often than the average patent
by a male inventor with a rare forename, all else
equal. To the extent that citations reflect patent
quality, this result suggests that women inven-
tors must clear a higher hurdle than men and
therefore that the average patent granted to a
woman inventor is of higher quality than the
average patent granted to a man.
Conclusions
These results should interest inventors, pat-
ent holders, and policymakers. In advanced
economies, technical progress appears to be
the primary driver of economic growth10. The
patent system, moreover, is one of the principal
public-policy mechanisms for promoting this
progress: governments grant patent holders a
limited monopoly in exchange for a thorough
disclosure of their inventions, so that oth-
ers may build upon those inventions11. That
women inventors are underrepresented in
this system and appear disadvantaged in the
process of obtaining and maintaining patents
suggests that changes to the patent system and
its prosecution process would increase fairness
and might even stimulate economic growth.
A thorough discussion of possible adjust-
ments to the patent system is beyond the scope
of this paper, but we can imagine many pos-
sibilities worth consideration. It may help, for
example, to make the patent-prosecution pro-
cess more ‘blind’ to the identity of participants.
Patents and patent applications could list only
the initials of the forenames of the inventors
on patent applications and could require com-
munication between examiners and applicants
to occur on a platform that would maintain
the anonymity of the applicants. Such blind
processes have eliminated gender inequality
in other settings: For instance, when orches-
tras introduced opaque screens to conceal the
identities of those auditioning, they hired more
women and placed more women in leadership
positions12. The introduction of such practices
at the patent office could help to ameliorate the
gender differences in patenting both in the life
sciences and in other technological areas.
Note: Any Supplementary Information and Source Data
files are available in the online version of the paper.
ACKNOWLEDGMENTS
The authors are grateful to IPwe for access to patent
data via Zuse Analytics and to T. Botelho and I.
Fernandez-Mateo for comments on earlier versions
of this paper.
AUTHOR CONTRIBUTIONS
The authors each contributed substantially to the
work in this manuscript.
COMPETING INTERESTS
The authors declare no competing interests.
1. Ding, W.W., Murray, F. & Stuart, T.E. Science 313,
665–667 (2006).
2. Wittington, K.B. & Smith-Doerr, L. Gend. Soc. 22,
194–218 (2008).
3. Hunt, J. et al. Res. Policy 42, 831–843 (2013).
4. Sugimoto, C.R. et al. PLoS One 10, e0128000 (2015).
5. Marco, A.C. et al. The USPTO Historical Patent Data
Files: Two Centuries of Invention. USPTO Economic
Working Paper no. 2015-1 www.uspto.gov/sites/default/
files/documents/USPTO_economic_WP_2015-01_
v2.pdf (2015).
6. Albers, C.J. Proc. Natl. Acad. Sci. USA 112, E6828–
E6829 (2015).
7. Marco, A.C. et al. Patent Claims and Patent Scope.
USPTO Economic Working Paper no. 2016–04http://
doi.org/10.2139/ssrn.2844964 (2016).
8. Trajtenberg, M. Rand J. Econ. 21, 172–187 (1990).
9. Reitzig, M. Res. Policy 33, 939–957 (2004).
10. Fagerberg, J. J. Econ. Lit. 32, 1147–1175 (1994).
11. Mazzoleni, R. & Nelson, R.R. Res. Policy 27, 273–284
(1998).
12. Goldin, C. & Rouse, C. Am. Econ. Rev. 90, 715–741
(2000).
listed on patents and patent applications.
(Examiners do sometimes meet with inventors
in person and by telephone; robustness checks
related to these scenarios are shown in the
Supplementary Data.) For common names,
such as ‘Mary’ and ‘Robert,’ outsiders can infer
gender with a high level of confidence, but for
thousands of rare names each held by only a
few individuals, they cannot make such infer-
ences. ‘Jameire’ and ‘Kunnath’, for example, are
also strongly associated with gender, with the
first being male and the second being female,
but because they are rare names, few people
would be aware of these associations. The
gender differences associated with common
names therefore should capture both differ-
ences in behavior on the applicant side as well
as differences in treatment of those inventors
by others. Any gender differences associated
with rare names, in contrast, should stem only
from the behavior of the applicant side.
The two series of green bars in Figure 1
show how the frequency of an inventor’s fore-
name moderates the effects of gender on various
outcomes. Because these models also include a
control for forename frequency, they account
for any association between the rarity of a
name and the underlying quality of the patent,
for example, because those patents might dis-
proportionately represent foreign applicants.
To avoid complications in aggregating across
names of varying frequency, these models
include only single-inventor patents. Among
those, two outcomes had large and statistically
significant differences between inventors with
common forenames and those with rare fore-
names. First, among inventors with common
names, women had an 8.2% lower probability
of having their application accepted than did
men. In contrast, among inventors with rare
names, women had only a 2.8% lower probabil-
ity of acceptance than did men. This combina-
tion suggests that approximately two-thirds of
PATENTS
© 2018 Nature America, Inc., part of Springer Nature. All rights reserved.
