U.S.

Patent and Trademark Office

OFFICE OF CHIEF ECONOMIST
Economic Working Paper Series

Patent Claims and Patent Scope

Alan C. Marco
U.S. Patent and Trademark Office

Joshua D. Sarnoff
DePaul University

Charles A. deGrazia
U.S. Patent and Trademark Office (ADDX Corp.) and Royal Holloway,
University of London

Original Version: January 2016

This Version: October 2016

USPTO Economic Working Paper No. 2016-04

Also available as Hoover IP2 Working Paper

The views expressed are those of the individual authors and do not necessarily reflect official positions
of the Office of Chief Economist or the U. S. Patent and Trademark Office. USPTO Economic Working
Papers are preliminary research being shared in a timely manner with the public in order to stimulate
discussion, scholarly debate, and critical comment.
For more information about the USPTO’s Office of Chief Economist, visit www.uspto.gov/economics.

The USPTO’s Patent Claims Research Dataset will be made available at www.uspto.gov/economics.

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=2844964
https://ssrn.com/abstract=2844964
 Patent Claims and Patent Scope
Alan C. Marco
U.S. Patent and Trademark Office

Joshua D. Sarnoff
DePaul University

Charles A. deGrazia
U.S. Patent and Trademark Office (ADDX Corporation) and Royal Holloway,
University of London

Original Version: January 2016

This Version: October 2016

Abstract
Patent scope is one of the important aspects in the debates over “patent quality.”
The purported decrease in patent quality over the past decade or two has
supposedly led to granting patents of increased breadth (or “overly broad”
patents), decreased clarity, and questionable validity. Such patents allegedly
diminish the incentives for innovation due to increased licensing and litigation
costs. However, these debates often occur without well-defined measurements of
patent scope. This paper explores two very simple metrics for measuring patent
scope based on claim language: independent claim length and independent claim
count. We validate these measures by showing that they have explanatory power
for several correlates of patent scope used in the literature: patent maintenance
payments, forward citations, the breadth of patent classes, and novelty. Using
these data, we provide the first large-scale analysis of patent scope changes during
the examination process. Our results show that narrower claims at publication are
associated with a higher probability of grant and a shorter examination process
than broader claims. Further, we find that the examination process tends to narrow
the scope of patent claims in terms of both claim length and claim count, and that
the changes are more significant when the duration of examination is longer.

Keywords: Patents, patent scope, patent claims, patent examination, patent

quality, USPTO
JEL Classification Numbers: O3, O31, O32, O34, O38

Acknowledgements: We would like to thank Robert Kimble, Jesse Frumkin,

Jamie Kucab, and Joseph Bailey, as well as numerous personnel at the US Patent
and Trademark Office. We would also like to thank seminar participants at
Harvard Business School and the Hoover Institution’s IP2 program.

The USPTO’s Patent Claims Research Dataset will soon be made available at
www.uspto.gov/economics.

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=2844964
https://ssrn.com/abstract=2844964
1 Introduction

For many years, debates over the effectiveness of the patent system have
focused on the central issue of “patent quality.” In 2002, then-former Assistant
Secretary of Commerce and Commissioner of the U.S. Patent and Trademark
Office (PTO) Gerald Mossinghoff noted a “real concern that with the dramatic
increase in the number of patent applications filed and patents granted - and with
the influx of new and unavoidably inexperienced examiners hired to handle the
workload - compromises to patent quality may be inevitable.”1 This increasing
number of patents of purportedly diminishing quality supposedly led to dramatic
increases in assertion of patents to extract rents through licensing and litigation,
particularly by non-practicing entities (NPEs).2 In turn, the purported decrease in
patent quality supposedly led to diminished innovation due to increased licensing
and litigation costs as well as to reduced sequential innovation in various
industries, particularly in regard to software patents.3

1 Gerald J. Mossinghoff & Vivian S. Kuo, Post-Grant Review of Patents: Enhancing the Quality of the Fuel
of Interest, 43 IDEA 83, 83 (2002).
2 See, e.g., Patent Quality Improvement: Hearing Before the Subcomm. on Courts, the Internet, and

Intellectual Property of the H. Comm. on the Judiciary, 109th Cong. 18 (2005) (statement of Richard J.
Lutton, Jr., Chief Patent Counsel, Apple) (“The current patent system has given rise to too many low quality
patents being issued, and a growing pattern of assertions of weak patents that threaten to damage productive
companies and stifle innovation.”); David L. Schwartz & Jay P. Kesan, Analyzing the Role of Non-Practicing
Entities in the Patent System, 99 CORNELL L. REV. 425, 429 (2014) (“Under this narrative, NPEs assert
marginal patents and read patent claims unreasonably expansively. Under any reasonable view, the patents
are likely invalid or not infringed by the NPEs’ targets. NPEs, who themselves do not innovate or introduce
any products into the marketplace, merely extract rents from the large, innovative companies that they sue.
They create fear of holdup by selecting venues where injunctive relief is available such as the International
Trade Commission. They seek and accept ‘nuisance’ settlement amounts, far below the cost of litigation, so
that the NPEs’ targets have no incentive to defend in costly litigation.”). But see John R. Allison & Ronald J.
Mann, The Disputed Quality of Software Patents, 85 WASH. U. L. REV. 297, 298 (2007) (“In general, we find
that patents the computer technology firms obtain on software inventions have more prior art references,
claims, and forward citations than the patents that the same firms obtain on nonsoftware inventions. We also
find that the patents that ‘pure’ software firms (those producing only software) obtain on software inventions
have more prior art references, claims, and forward citations than the software patents obtained by the firms
that derive revenues from other product lines.”).
3 See, e.g., Arti K. Rai, Improving (Software) Patent Quality Through The Administrative Process, 51 HOUS.

L. REV. 503, 505 (2013) (“Low-quality software patents … generate the usual negative static effects, in the
form of either unnecessary licensing fees or deadweight loss. They also generate deleterious dynamic effects,
as firms in the information and communications technology industries must accumulate large defensive
arsenals in order to avoid being sued.”). Cf. Jay Pil Choi & Heiko Gerlach, Patent Pools, Litigation, and
Innovation, 46 RAND. J. ECON. 499, 499 (2015) (“If patents are sufficiently weak, patent pools with
complementary patents reduce social welfare as they charge higher licensing fees and chill subsequent
innovation incentives.”).

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=2844964
https://ssrn.com/abstract=2844964
 Patent claim scope and claim clarity have been identified as significant
concerns for patent quality.4 Even the basic approach to determining claim
meaning has been called into question.5 Software patents in particular have been
criticized for having unduly broad and/or unclear claims employing functional
claiming language.6 Thus, in 2003, a Federal Trade Commission (FTC) Report
summarized hearings where “[m]any panelists and participants expressed the
view that software and Internet patents are impeding innovation.”7 And in 2004,
the authors of the most comprehensive litigation study of patents to that date
noted – after determining that litigated (and by hypothesis more valuable)
individual patents experienced significantly longer and more complex
prosecutions at the PTO – that this “could suggest that the much-maligned PTO is
doing a better job than expected in evaluating the patents that really matter, or it
could mean that patent examiners are buried in paper by those critical
applications.”8

4 See, e.g., Rai, supra note 3, at 512 (identifying as concerns raised by commentators “the grant of patents
with scope that exceeds their level of disclosure; and the grant of patents with unclear claim language that
fails to provide adequate notice”); Dargaye Churnet, Patent Claims Revisited, 11 NW. J. TECH. & INT. PROP.
501, 502 (2013) (“Reading an entire patent application and gaining a thorough understanding of the claims
may take weeks. Patent examiners, however, are expected to do so in less than 24 hours. It is no wonder,
then, that many have questioned the quality of patents the PTO has issued.”); Lee Petherbridge, On
Addressing Patent Quality, 158 U. PA. L. REV. PENNUMBRA 13 (2009) (Polk Wagner “thoroughly and
dispassionately identifies and examines the incentives that patent applicants and the Patent Office have to
draft and issue, respectively, large quantities of patents with opaque disclosures and indeterminate claims.”)
(citing R. Polk Wagner, Understanding Patent Quality Mechanism, 157 U. Pa. L. Rev. 2135 (2009)): cf. John
P. Zimmer, To Infinity and Beyond: The Problem of Open-Ended Claim Language in the Unpredictable Arts,
59 S. Car. L. Rev. 865, 867-68 (2008) (arguing for a more stringent enablement examination approach to
open-ended claims having uncertain scope at one end of a range of claimed values).
5 See, e.g., Dan L. Burk & Mark A. Lemley, Fence Posts or Sign Posts? Rethinking Patent Claim

Construction, 157 U. PA. L. REV. 1743, 1745 (2009) (“Claim construction is sufficiently uncertain that many
parties don’t settle a case until after the court has construed the claims, because there is no baseline for
agreement on what the patent might possibly cover. Even after claim construction, the meaning of the claims
remains uncertain, not only because of the very real prospect of reversal on appeal but also because lawyers
immediately begin fighting about the meaning of the words used to construe the words of the claims.”).
6 See, e.g., U.S. Government Accountability Office (GAO), Report to Congressional Committees, Intellectual

Property: Assessing Factors that Affect Patent Infringement Litigation Could Help Improve Patent Quality,
GAO-13-465, at 28-30 (Aug. 2013), available at
http://www.uspto.gov/sites/default/files/aia_implementation/GAO-12-
465_Final_Report_on_Patent_Litigation.pdf (last visited Nov. 3, 2015) (citing views of various
stakeholders). See generally Mark A. Lemley, Software Patents and the Return of Functional Claiming,
2013 WISC. L. REV. 905.
7 Federal Trade Commission, To Promote Innovation: The Proper Balance of Competition and Patent Law

and Policy ch. 3 at 56 (Oct. 2003), available at http://www.ftc.gov/os/2003/10/innovationrpt.pdf (last visited

Nov. 3, 2015).
8 John R. Allison, et al., Valuable Patents, 92 GEO. L.J. 435, 438-39 (2004).

Electronic copy available at: https://ssrn.com/abstract=2844964

 Of course, “patent quality” may have varying meanings, which depend on the
user and the context. There are at least five “dimensions” of patent quality on
which analysts of the patent system tend to focus, with the first three focused on
the patent instrument itself: “(1) a patent’s probable validity; (2) clarity of the
patent (to different audiences); (3) faithfulness of the patent to the scope of the
invention; (4) social utility of the patented invention; and (5) commercial success
of the patented invention.”9 It is commonly agreed that only valid patents can be
quality patents, but it is frequently disputed as to whether other measures of
quality should be considered.10 Further, some of the measures used in the past by
the PTO to assess the rate of granting patents have been criticized, given that the
various forms of continuing application practices11 – including requests for
continued examination (RCEs) in the same application12 – suggest lower grant
rates for applications that may ultimately issue, whether with identical or with
different claims.13 Such continued application practices increase the overall
demand for examination services, regardless of whether the overall supply of such
services is sufficient or reflects “rational ignorance” – i.e., reasonable limits on
examination time expenditures from what would otherwise result in improved
administrative validity decisions, given the substantial costs of expanding
examination resources to address many patents of low innovation utility or low
commercial value that will never be licensed or litigated.14 These (and related)

9 Christi J. Guerrini, Defining Patent Quality, 82 FORDHAM L. REV. 3091, 3096 (2014). See, e.g., James E.
Malackowski & Jonathan A. Barney, What is Patent Quality? A Merchant Banc’s Perspective, 43 LES
NOUVELLES 123, 124-28 (2008) (distinguishing low quality resulting from examination – validity – errors
from low quality resulting from low standards of patentability).
10 See Guerrini, supra note 9, at 3098 & n.27, 3099 & nn.28-30 (citing numerous sources).
11 See 35 U.S.C. §§ 120, 121 (2014); 37 C.F.R. § 1.78(a) (2014).
12 See American Inventors Protection Act of 1999, § 4403 (Title IV of the Intellectual Property and

Communications Omnibus Reform Act of 1999 (S. 1948)), as enacted by Pub. L. No. 106-113, § 1000(a)(9),
Division B, 113 Stat. 1501 (1999); 37 C.F.R. § 1.114 (2014).
13 See, e.g., Bruce A. Kaser, Patent Application Recycling: How Continuations Impact Patent Quality &

What The USPTO Is Doing About It, 88 J. PAT. & TRADEMARK OFF. SOC’Y 426, 427-35 (2006). See
generally Cecil D. Quillen Jr. & Ogden H. Webster, Continuing Patent Applications and Performance of the
U.S. Patent and Trademark Office, 11 FED. CIR. B.J. 1 (2001); Cecil D. Quillen Jr. & Ogden H. Webster,
Continuing Patent Applications and Performance of the U.S. Patent and Trademark Office - Extended, 12
FED. CIR. B.J. 35 (2002); Cecil D. Quillen Jr. & Ogden H. Webster, Continuing Patent Applications and
Performance of the U.S. Patent and Trademark Office - Updated, 15 FED. CIR. B.J. 635 (2005-2006); Cecil
D. Quillen Jr. & Ogden H. Webster, Continuing Patent Applications and Performance of the U.S. Patent and
Trademark Office – One More Time, 18 FED. CIR. B.J. 379, 387-94 (2008-2009); Christopher A. Cotropia,
Cecil D. Quillen Jr. & Ogden H. Webster, Patent Applications and the Performance of the U.S. Patent and
Trademark Office, 23 FED. CIR. B.J. 179 (2013).
14 See generally Mark A. Lemley, Rational Ignorance at the Patent Office, 95 NW. U. L. REV. 1495 (2001).

Electronic copy available at: https://ssrn.com/abstract=2844964

concerns led the PTO in 2007 to adopt rules to restrict continuation practice,15
which ultimately were withdrawn in 2009 following litigation.16

The PTO itself has expressed concerns about patent quality, and has adopted
many other initiatives over the last decade to address patent quality concerns.17
In early 2015, the PTO adopted an “Enhanced Patent Quality Initiative,”
supervised by a new Deputy Commissioner for Patent Quality, which focuses on
three “pillars” of “excellence”: (1) work quality; (2) measuring patent quality; and
(3) customer service.18 Some of the measures being considered by the PTO
include metrics of quality that go far beyond assessing validity of final
determinations, such as processing time, correctness of intermediate actions, and
(particularly) assuring clarity of claims and of other aspects of the prosecution
record.19 As noted by the PTO in 2011, its “previous focus on the correctness of
actions taken by an examiner in an individual application has been widened to
better encompass the entirety of the patent application and examination
process.”20 And as noted by the PTO in 2015:

15 See USPTO, Changes to Practice for Continuing Applications, Requests for Continued Examination
Practice, and Applications Containing Patentably Indistinct Claims, 71 Fed. Reg. 48, 58-61 (Jan. 3, 2006)
(proposing restrictions by amending Rules 78 and 114); USPTO, Changes To Practice for Continued
Examination Filings, Patent Applications Containing Patentably Indistinct Claims, and Examination of
Claims in Patent Applications, 72 Fed. Reg. 46716, 46837-41 (Aug. 21, 2007) (adopting amendments to
Rules 78 and 114).
16 See USPTO, Changes to Practice for Continued Examination Filings, Patent Applications Containing

Patentably Indistinct Claims, and Examination of Claims in Patent Applications, 74 Fed. Reg. 52686, 52689-
91 (Oct. 14, 2009) (withdrawing amendments to rules 78 and 114); Tafas v. Kappos, 586 F.3d 1369, 1370
(Fed. Cir. 2009) (en banc) (dismissing appeal without vacating District Court judgment); Tafas v. Doll, 328
Fed.Appx. 658 (Fed. Cir. 2009) (en banc) (granting en banc rehearing and vacating panel opinion); Tafas v.
Doll, 559 F.3d 1345, 1359-63 (Fed. Cir. 2009) (panel decision invalidating PTO rule restricting continuation
filings, but upholding PTO rule restricting RCEs); Tafas v. Doll, 541 F. Supp. 2d 805, 814-16 (E.D. Va.
2008) (District Court judgment invalidating amendments restricting both continuations and RCEs).
17 See, e.g., USPTO, 2010-2015 Strategic Plan (2010), available at http://

www.uspto.gov/about/stratplan/USPTO_2010-2015_Strategic_Plan.pdf (last visited Nov. 3, 2015)

(identifying quality improvement as a critical priority); USPTO, The 21st Century Strategic Plan 5 (2003)
[hereinafter “21st Century Strategic Plan”], available at http://
www.uspto.gov/web/offices/com/strat21/stratplan_03feb2003.pdf (last visited Nov. 3, 2015) (identifying
patent quality as the PTO’s “highest priority”).
18 See USPTO, Director's Forum: A Blog from USPTO's Leadership (Feb. 4, 2015), at

http://www.uspto.gov/blog/director/entry/uspto_launches_enhanced_patent_quality (last visited Nov. 3,

2015); USPTO, Enhanced Patent Quality Initiative, http://www.uspto.gov/patent/initiatives/enhanced-patent-
quality-initiative (last visited Nov. 3, 2015).
19 See, e.g., USPTO, Request for Comments on Enhancing Patent Quality, 80 Fed. Reg. 6475, 6476-80 (Feb.

5, 2015) (hereinafter “PTO Request for Comments 2015”). See also Guerrini, supra note 9, at 3099 & nn.31-
32 (citing various PTO documents).
20 USPTO, Adoption of Metrics for the Enhancement of Patent Quality Fiscal Year 2011, at

Electronic copy available at: https://ssrn.com/abstract=2844964

 [t]he USPTO recognizes that, in order for the patent system to fulfill
its critical role in promoting innovation, issued patents must not only
fully comply with all statutory requirements, but also contain an
Official record that is unambiguous and accurate. Such a complete
record provides patent boundaries that are clearly defined to the
benefit of the patent owner, the courts, third-parties, and the public
at large, giving inventors and investors the confidence to take the
necessary risks to launch products and start businesses, and the
public the benefit of knowing the precise boundaries of an
exclusionary right.21

Many of the recent debates over the effectiveness of the incentive

mechanisms created by patent rights have focused on the central issue of patent
quality, but have treated patent examination as a “black box” or have worked
backward from the characteristics of issued and litigated patents (or of patents that
underwent some form of post-grant administrative reevaluation).22 More recent
scholarship, however, seeks to address more directly patent examination
processes relating to patent quality, by looking at examination characteristics in
light of the greater availability of such data. For example, Frakes and Wasserman
(2015) have used information on application outcomes (including abandonments)
to test their hypothesis that under conditions of resource constraints the PTO is
more likely to grant applications in technology areas of higher continuation
application filings.23 Earlier, Frakes and Wasserman (2013) found – using PTO
data from before and after the PTO acquired fee-setting authority – that the PTO
was more likely to grant claims on “technologies with high renewal rates and
patents filed by large entities, as the PTO stands to earn the most revenue by
granting additional patents of these types.”24 Others, such as Régibeau and

http://www.uspto.gov/sites/default/files/patents/init_events/qual_comp_metric.pdf (last visited Nov. 3, 2015).

21 PTO Request for Comments 2015, supra note 19, at 6479.
22 See, e.g., Allison, et al., supra note 8; Shawn P. Miller, What’s the Connection Between Repeat Patent

Litigation and Patent Quality? A (Partial) Defense of the Most Litigated Patents, 16 STAN. TECH. L. REV. 313
(2013); John R. Allison, Mark A. Lemley & Joshua Walker, Patent Quality and Settlement Among Repeat
Patent Litigants, 99 GEO. L.J. 677 (2011); John R. Allison, Mark A. Lemley & Joshua Walker, Extreme Value
or Trolls on Top? The Characteristics of the Most-Litigated Patents, 158 U. PA. L. REV. 1 (2009); Stuart J.H.
Graham, et al., Post-Issue Patent “Quality Control”: A Comparative Study of US Patent Re-Examinations and
European Patent Oppositions, NBER Working Paper 8807, at 1-5 (2002), available at
http://www.nber.org/papers/w8807 (last visited Nov. 3, 2015).
23 See Michael D. Frakes & Melissa F. Wasserman, Does the U.S. Patent and Trademark Office Grant Too

Many Bad Patents?: Evidence From A Quasi-Experiment, 67 STAN. L. REV. 613 (2015).
24 Michael D. Frakes & Melissa F. Wasserman, Does Agency Funding Affect Decisionmaking?: An Empirical

Assessment of the PTO’s Granting Patterns, 66 VAND. L. REV. 67, 70 (2013).

Electronic copy available at: https://ssrn.com/abstract=2844964

Rockett (2010), have looked at the relationship between technology development
and invention importance compared to time in examination, using historical data
specific to patents on genetically modified crops.25 Yet others have focused on
theoretical modeling of the examination process or of application filing
behaviors,26 and seek to draw empirical support from cross-country comparisons27
or inferences for patent examination policy.28

As noted by Lemley and Sampat (2010), empirical analysis of actual

examination practices was first made feasible around 2001 “when the PTO began
publishing data on pending applications, and when the Patent Application
Information Retrieval (‘PAIR’) system allowed the public to track the fate of
those applications in real time.”29 The availability of patent examination data
permitted analysis of grant rates, continuation practices, appeals, and other
prosecution events.30 But the PAIR data do not identify (without further hand-
coding) the substantive grounds for the actions or the nature of the changes made
to any claims during prosecution.31

As noted above, applicant claiming and PTO examination practices have

been criticized, focusing on how purportedly “low quality” issued patents are
treated in litigation.32 However, there has been precious little empirical analysis of
initial application claiming practices and changes to claims during the
examination process. In contrast, numerous studies have looked at judicial

25 See Pierre Régibeau and Katharine Rockett, Innovation Cycles and Learning at the Patent Office: Does the
Early Patent Get the Delay?, 58 J. INDUS. ECON. 222, 222-24 (2010).
26 See, e.g., Stefano Comino & Clara Graziano, How Many Patents Does It Take to Signal Innovation

Quality?, 43 INT’L. J. INDUS. ORG. 66, 66-69 (2015) (positing that “true innovators” are forced to patent more
intensively in the presence of “bad patents”).
27 See, e.g., Florian Schuett, Patent Quality and Incentives at the Patent Office, 44 RAND J. Econ. 313

(2013).
28 See, e.g., Bernard Caillaud & Anne Duchêne, Patent Office in Innovation Policy: Nobody’s Perfect, 29

INT’L. J. INDUS. ORG. 242 (2011).

29 Mark A. Lemley & Bhaven Sampat, Examining Patent Examination, 2010 STAN. TECH. L. REV. 2, 2

(2010).
30 See, e.g., Mark A. Lemley & Bhaven Sampat, Is the Patent Office a Rubber Stamp? 58 EMORY L.J. 181,

182-83 (2008).
31 Note that the coding of actions in PAIR does not itself distinguish claim amendments from other

application amendments (such as changes in the specification), although it is possible to read the associated,
scanned documents to distinguish these types of amendments.
32 See, e.g., Petherbridge, supra note 4, at 16-18 (discussing various critiques focused on litigation); Alan

Marco et al., U.S. Patent and Trademark Office, Patent Litigation and USPTO Trials: Implications for Patent
Examination Quality 7-9 (January 2015) (discussing various studies of the relationship of patent quality to
litigation and post-grant administrative reviews).

Electronic copy available at: https://ssrn.com/abstract=2844964

construction of claims at the district court and appellate court level, suggesting
changes to how judges perform claim constructions and review thereof.33

This paper explores two claim-related metrics for patent scope.

Specifically, for each published application and patent in our dataset we calculate:

1. the number of words used in the shortest independent claim (which we

call independent claim length, or “ICL”);34 and,
2. the total number of independent claims (which we call “ICC”).

We are able to observe the claim language for applications at the date of
their pre-grant publication (“PGPub”) and for granted patents at the date of
issuance. We also calculate changes in ICL and ICC from publication to grant for
each published application resulting in a grant (“publication-grant pair”).
Moreover, we make the underlying claims data available for public use in order to
stimulate more research in the area.35 We validate ICL and ICC as measures of
scope by testing the explanatory power with respect to several patent scope
correlates from the previous literature: patent maintenance fee payments, forward
citations, the number of technology classes to which the patent was assigned, and
patent novelty as defined by Fleming (2001) and Strumsky et al (2012).

This paper presents the first large-scale analysis of patent application and
granted patent scope changes during the examination process. Our results reveal
several interesting features about the patent examination process. First, we find
that applications with narrower claims (in terms of ICL) are more likely to be

33 See, e.g., Shawn P. Miller, “Fuzzy” Software Patent Boundaries and High Claim Construction Reversal
Rates, 17 STAN. TECH. L. REV. 809 (2015); J. Jonas Anderson & Peter S. Menell, Informal Deference: A
Historical, Empirical, and Normative Analysis of Patent Claim Construction, 108 NW. U. L. REV. 1 (2014);
Thomas W. Krause & Heather F. Auyang, What Reversals and Close Cases Reveal About Claim
Construction: The Sequel, 13 J. MARSHALL REV. INTELL. PROP. L. 525 (2014); Thomas W. Krause & Heather
F. Auyang, What Close Cases and Reversals Reveal About Claim Construction, 12 J. MARSHALL REV.
INTELL. PROP. L. 583 (2013); R. Polk Wagner & Lee Petherbridge, Did Phillips Change Anything? Analysis
of the Federal Circuit’s Claim Construction Jurisprudence, in INTELLECTUAL PROPERTY AND THE COMMON
LAW (S. Balganesh ed. Cambridge U. Press 2011); David L. Schwartz, Practice Makes Perfect?: An
Empirical Study of Claim Construction Reversal Rates in Patent Cases, 107 MICH. L. REV. 223 (2008);
Kimberley A. Moore, Markman Eight Years Later: Is Claim Construction More Predictable?, 9 LEWIS &
CLARK L. REV. 231 (2005); Joseph Scott Miller & James A. Hilsenteger, The Proven Key: Roles and Rules
for Dictionaries and the Patent Office and the Courts, 54 AM. U. L. REV. 829 (2005).
34 We also considered alternative measures for ICL including the average independent claim length and the

length of the first independent claim. The results are largely insensitive to the definition of ICL.
35 The USPTO’s Patent Claims Research Dataset will soon be made available at www.uspto.gov/economics.

Electronic copy available at: https://ssrn.com/abstract=2844964

granted than those with broader claims. Second, the examination process itself
tends to narrow the scope of patents. Patent prosecution tends to add 45 words, on
average, to the shortest independent claim and tends to reduce the number of
independent claims by 0.4 claims. Third, we find that broader applications (in
terms of ICL and ICC) tend to have longer pendency times, both for abandoned
applications and granted patents. Further, longer pendency periods tend to
generate more significant narrowing of the patent between application and grant
(in terms of both ICL and ICC). We also find significant variation over time in the
breadth of patent applications and patent grants, contrary to conclusions drawn
from some earlier analyses that suggested a high level of stability in claim lengths
of issued patents over longer periods of time.36

This paper is organized as follows. In Section 2, we discuss the

relationship between patent scope and our measures of ICL and ICC as proxies
for the scope (breadth) of patent applications and granted patents. Section 3
describes the claims data and our resulting sample. Section 4 provides our
descriptive analysis. We examine differences in the statistical distributions of ICL
and ICC between abandoned applications and applications that are later granted,
as well as the evolution of claims during prosecution. We also look at time trends
for ICL and ICC, and cross-sectional differences for many types of application
characteristics. Lastly, we consider the relationship between patent breadth and
pendency. Section 5 provides several validations for using ICL and ICC as a
measure of patent scope. Section 6 briefly concludes. We include appendices that
provide a detailed description of the public use data sets, and the computer code
that generated them.

36See, e.g., Kristen J. Osenga, The Shape of Things to Come: What We Can Learn from Patent Claim Length,
28 SANTA CLARA COMPUTER & HIGH TECH. L.J. 617, 619-37 (2012). Cf. Johannes Koenen & Martin Peitz,
Firm Reputation and Incentives to “Milk” Pending Patents, 43 INT’L J. INDUS. ORG. 18, 18-23 (2015)
(discussing equilibrium effects of reputation to seek only meritorious grants and benefits from extending
beyond that and from examination errors); Stephen Yelderman, Improving Patent Quality with Applicant
Incentives, 28 HARV. J.L. & TECH. 77, 78-81 (2014) (arguing that various measures, such as fees, could be
used to affect applicant willingness to file overbroad claims).

Electronic copy available at: https://ssrn.com/abstract=2844964

2 Patent Claims and Patent Prosecution

Our analysis proceeds from the theoretical presumption that the word length
of an independent claim and the scope of the claim (equivalently, claim breadth)
tend to be negatively correlated: adding more words to a claim should generally
decrease its scope of potential application.37 A patent application contains two
distinct parts: the specification and the claims.38 The specification encompasses a
written description and background, along with drawings or figures. The claims
represent the legal metes and bounds of the invention. Importantly, the
specification may not be significantly altered after filing, whereas it is common to
amend claims during prosecution.

Typically, applicants have an incentive to file an application with the

broadest claims to which they think they are entitled. There is no incentive for the
applicant to excessively narrow the claims, ex ante, before the examiner has done
her search; that would be the legal equivalent of leaving money on the table.
Broader claims translate to a larger set of technologies that the owner can exclude
others from using, and making it more difficult for competitors to invent around.
During examination, a search may reveal prior art that renders the applicant’s
claim(s) unpatentable under novelty or obviousness standards. In that case, the
examiner rejects the application and the applicant typically amends the claim(s) or
abandons the application. In order to circumvent the prior art, claims must be
narrowed so that they are not so broad as to overlap with the prior art.
Consequently, amendments almost always involve narrowing. Further, this
process almost always involves adding words to the claim: modifiers, qualifiers,
or other details. The patent prosecution process itself provides it’s own support:
applicants have no incentive to narrow claims, except to respond to examiners’
rejections. Yet, as we show below, the vast majority of independent claims grow
longer during prosecution, in response to rejections. Thus, there is at least a

37 See, e.g., Benedikt Szmrecsányi, On Operationalizing Syntactic Complexity, in JOURNÉES INTERNATIONALS

D’ANALYSE STATISTIQUE DES DONNÉES TEXTUALLES 1037‐ 38 (2004) (“determining length in words — to
assess syntactic complexity is by all means one that is nearly as accurate as the more sophisticated and
cognitively, conceptually, or even psychologically ‘more real’ methods”). Cf. Nicholas van Zeebroeck,
Bruno van Pottelsberghe de la Potterie & Dominique Guellec, Claiming more: the Increased Voluminosity of
Patent Applications and its Determinants, Centre Emile Berneim working Paper No. 06/018, Université Libre
de Bruxelles – Solvay Business School, text at n. 21-22 (Mar. 2007) (“As technology becomes more
complex, more words may be required to describe and claim it.”). See generally Thomas Wasow, Remarks
on Grammatical Weight, in 9 LANGUAGE VARIATION AND CHANGE 81-105 (Cambridge U. Press 1997).
38 Technically the specification as defined by 35 USC 112 contains the written description and the claims.

However, it is common in the industry to refer to the “spec” as distinct from the claims.

Electronic copy available at: https://ssrn.com/abstract=2844964

correlation between the narrowness of claims and the length of claims, ceteris
paribus.

Patent prosecution generally involves several rounds of rejection and

amendment. A common practice is for applicants to include a very broad
independent claim, along with narrower dependent claims. The examiner may
reject the independent claim while indicating approval for a dependent claim,
which by law must have narrower scope. In that instance, the applicant may “roll
up” at least one dependent claim limitation into the original independent claim to
form a new, longer and narrower independent claim. For example, claim 1 of U.S.
patent application 10/495,059 was modified by the applicant to include most of
the language of claim 1 as originally filed, as well as the additional limitations of
dependent claims 5 and 6 as originally filed. 39 This additional language narrowed
the original independent claim 1 such that, as modified, issued independent claim
1 was allowable over the prior art of record. By legal construction, a dependent
claim incorporates the independent claim language and adds a limitation, which
requires adding words (e.g., “A device as described in claim 1, such that…”).
Thus, by definition a dependent claim roll up will be longer and narrower than the
original independent claim.

Where claim language is ambiguous or vague, the examiner may reject the
claim under section 112.40 Clarification by adding words normally narrows the
claim scope because it excludes a set of potential embodiments, whether by
restricting the meaning of the ambiguous or vague language or by specifying a
narrower conception of the things (or relevant properties of things) that the
meaning denotes. Note that the approach of treating additional words as
narrowing does not necessarily mean that comparing two different claims from
different patents on unrelated inventions will permit a general inference that the
longer claim implies the narrower scope. Rather, it only indicates that adding
words in a particular application tends (all else being equal) to add limitations that
reduce or otherwise restrict claim scope. However, comparing word lengths
within narrow technology groups may be appropriate. Further, comparing word
lengths across patents may enable us to observe general trends over time.

39 See Appendix A for the full text claim language of application 10/495,059 as reflected in U.S. pre-grant
publication 20050065799 (published March 24, 2005) and U.S. patent 7,769,690 (issued August 3, 2010).
40 35 USC 112.

10

Electronic copy available at: https://ssrn.com/abstract=2844964

 In particular instances it may be possible to add words to a claim without
narrowing its scope, even without regard to specialized claim formats. This is the
case when the claim contains a list of possible embodiments, each separated by
the word “or.” Adding another possible embodiment to the list would add words
and potentially add scope.41 We analyzed our presumptions for robustness against
two particular claim formats for which the addition of words may be more likely
to expand than to narrow claim scope: claims using the connecting word “or”; and
Markush claims that use the words “selected from.”

Our observations below focus on claim length and on changes to claim length
in particular patent applications during prosecution. Claims as published in the
PGPub are a good indication of the claims at filing, because only 8.1 percent of
patents have any claim amendments between the date of filing and the date of
publication. Further, the change in independent claim length acts as a good proxy
for assessing changes to patent scope during prosecution. In contrast, in an effort
to assess changes to claiming practices over decades, Osenga (2012) looked at
average independent and dependent claim length at grant alone, using small
samples of randomly selected patents. She found that claim length practices had
remained surprisingly stable over five decades, notwithstanding significant
doctrinal and technological changes.42 In contrast, we find significant variations
in claim length from 1976 to 2014 for granted patents and 2001 to 2014 for
published applications.

With respect to the number of independent claims we presume—based on

principles of patent prosecution—that more independent claims implies a broader
patent scope. That is, adding an independent claim should tend to increase a
particular patent’s scope,43 and should never decrease the patent’s scope. Claims
are subject to the interpretive principle of claim differentiation,44 and

41 Similarly, a Markush claim provides alternatives as being “selected from the group consisting of A, B, and
C” (MPEP 803.02). Adding more elements to the group would add words and increase the scope.
42 See Osenga, supra note 36, at 619-22, 632-37.
43 In comparison, many scholars have used a count of total claims. For example, Allison and Lemley (2000)

performed analyses on the total number of claims at grant, based on the assumption that comparative
increases across unrelated patents in the total number of claims should reflect either increased complexity or
increased value of the technology sought to be protected, given that additional claims will normally cost
patent applicants additional filing fees and drafting and prosecution costs. With respect to scope, however,
the number of independent claims is more accurate, because dependent claims may not be broader than their
independent claims.
44 See, e.g., World Class Tech. Corp. v. Ormco Corp., 769 F.3d 1120, 1125 (Fed. Cir. 2014) (“‘The doctrine

of claim differentiation creates a presumption that distinct claims, particularly an independent claim and its

11

Electronic copy available at: https://ssrn.com/abstract=2844964

consequently a new independent claim should not be entirely subsumed in the
scope of another independent claim. Thus, the number of independent claims
should be an indicator of the patent’s scope, and the change in the number of
independent claims during prosecution should reflect the narrowing or broadening
of scope due to claim amendments.

Accordingly, we provide our analyses under the presumptions that ceteris

paribus, a patent’s scope is correlated with (1) fewer words in its shortest
independent claim (broader claims), and (2) a greater number of independent
claims. Therefore, as the length of the shortest independent claim increases, and
as the number of independent claims decreases, the scope of patent should
narrow. We validate these presumptions in Section 5.

3 Data

We build our claims data sets45 from publicly available full-text

information on pre-grant publications and patent grants. Machine readable claims
information is readily available on published patent documents, including the
patent grant itself (since 1976), as well as the pre-grant publication (since 2001).
Unfortunately, the individual claim amendments during prosecution are only
available as image files in the electronic file record of the application (the “image
file wrapper”). Further, the bulk data files incorporate the entire text of the patent,
not just the claims, and the claims themselves are not individually parsed.

To develop the datasets, we first cleaned and identified the claims section
of each bulk file for published applications and patents. Second, we applied an
algorithm to the parsed files to identify individual claims as well as the
dependency relationships between claims. From the parsed claims text, we
measured the length of each claim based on word count. We created data sets at
the claim level and summary statistics at the document level.46

dependent claim, have different scopes.’”) (quoting Kraft Foods, Inc. v. Int'l Trading Co., 203 F.3d 1362,
1368 (Fed.Cir.2000)). See generally Joshua D. Sarnoff & Edward D. Manzo, An Introduction to, Premises of,
and Problems with Patent Claim Construction, in CLAIM CONSTRUCTION IN THE FEDERAL CIRCUIT § 0:4(2)
(2014 on-line ed. Thompson West). Again, this does not necessarily mean that comparing two unrelated
patents with different numbers of claims will indicate that the patent with the larger number of claims has the
greater scope.
45 The USPTO’s Patent Claims Research Dataset will soon be made available at www.uspto.gov/economics.
46 The data sets are provided at www.uspto.gov/economics. More information about the methodology and the

structure of the data sets can be found in the appendices.

12

Electronic copy available at: https://ssrn.com/abstract=2844964

 We provide claim-level data for each pre-grant publication of a U.S.
patent application (publication, or PGPub) filed after November 29, 2000 and
published before January 1, 2015. Similarly, we provide claim-level data for each
patent granted between January 1, 1976 and before January 1, 2015. We also
create publication-patent pairs for analysis. Unfortunately, the available sources
do not contain parsable claims at the time applications are abandoned, and thus
we cannot directly compare claims at the time of abandonment to claims at the
time of issue.

For the purposes of this paper we examine document-level (patent-level or

PGPub-level) claims statistics. We calculate two document-level statistics of
primary interest:
 ICL (independent claim length): the word count of the shortest
independent claim in the document. This is often, but not always,
the first claim.
 ICC (independent claim count): the number of independent claims
in the document.

From the full data set, we constructed publication-patent pairs, for those
applications for which we can identify both a PGPub and a granted patent. These
pairs enable the observation of changes in an application’s claims between
publication and grant. For these pairs we define ICL and ICC, which represent
the value of ICL or ICC at grant less the corresponding values at publication.
Note that the shortest independent claim at grant may be a different claim number
than the shortest independent claim at publication. First, claims may be
renumbered at various times during prosecution and the particular forms on which
claim amendments are made are not machine readable. Second, amendments may
cause the shortest independent claim on the PGPub to grow longer than another
independent claim.

Table 1 summarizes our final sample, which represents 3.9 million

PGPubs, 4.9 million granted patents, and 2.1 million publication-patent pairs. For
publications, the table shows that abandoned applications tend to have broader
claims relative to applications that are later granted; abandoned applications have
15 fewer words in their shortest independent claims, at the median. Further,
granted patents are narrower at grant than at publication; they tend to gain 45
words in their shortest independent claims, and lose 0.4 independent claim
between publication and issuance. We discuss these differences in greater detail in
Section 4, below.

13

Electronic copy available at: https://ssrn.com/abstract=2844964

4 Analysis

4.1 Comparing pre-grant publications and granted patents

We generated frequency distributions of ICL and ICC (Figures 1 and 2) for

all publications and grants, for application years 2001-2014. PGPubs are
separated based on whether they resulted in abandonments or grants (pending
applications are not shown). These distributions demonstrate how ICL and ICC
vary across all applications during prosecution and by disposal type.

From the ICL distributions, it is notable that: (1) applications with

narrower claims at the time of publication are more likely to be granted, and (2)
granted patents are narrower at the time of grant than at the time of publication.
These facts suggest that the prosecution process leads to narrower claims and
narrower patents. This is consistent with the practice to roll-up dependent claims
into independent claims; the practice would lead to longer independent claims.

We find that the different distributional characteristics between the ICL

for PGPubs and for patents indicate that examination not only increased ICL from
publication to grant, but also disproportionately decreased the concentration of
very short independent claims at grant; i.e., prosecution shifted the distribution to
the right (Figure 1). In other words, the prosecution and examination process on
average narrows the scope of applications by increasing ICL from 106 words at
publication to 156 words at grant for application years 2001 to 2014.
Unfortunately, we cannot observe the distribution of abandoned applications at
the time of disposal, which would provide insight into separating the relative
effects of examination from the initial filing choices by applicants. Nevertheless,
the overall distribution of ICL for applications that are later abandoned has the
same general shape as for those that are later granted, except that applications that
go abandoned have a larger mass of shorter claims for PGPubs. This confirms
that allowances are less frequent for applications that have claims of greater
scope.

On the other hand, abandoned applications tend to have fewer claims at

publication than those that are allowed. With regard to claim counts, the
distribution of ICC is right-skewed for PGPubs and for patents at grant. These
distributions differ significantly between PGPubs that are later granted (PGPub-
grants), PGPubs that are later abandoned (PGPub-abandonments), and granted
patents as shown in Figure 2. PGPub-grants have the highest concentration at
three independent claims, whereas PGPub-abandonments and patent grants have
the highest concentration at one independent claim. The mean of ICC is slightly

14

Electronic copy available at: https://ssrn.com/abstract=2844964

lower for PGPub-abandonments (2.99 claims) than for PGPub-grants (3.07
claims).

The different distributions for ICC suggest (contrary to what would be

expected) that abandoned applications have narrower scope compared to
applications that result in grants. However, two alternative potential explanations
are possible. First, applications with a single independent claim may be more
likely to go abandoned when that independent claim is rejected; applications that
have more than one independent claim are more likely to continue prosecution if
one independent claim is rejected. Second, it is possible that applications that
include very broad claims may be more likely to include fewer of such claims or
fewer categories of such claims (e.g., product and process claims). The mean
number of independent claims for PGPubs, 2.97, is consistent with the maximum
number of allowable independent claims per patent application that avoid
incurring excess claim fees.

4.2 Trends

Figures 3 to 10 show the trends over time in claims for PGPubs and for
grants, which provide some insights into applicant filing behavior as well as
potential changes in examination practice. For patents, we can observe claims
information from 1976 to 2014. For published applications, we can observe
claims data from 2001 to 2014. The figures graph annual arithmetic means for
three different cohort aggregations.

First, we define cohorts based on the year of their final disposition,

whether that was an abandonment or grant, and compare it to information on
granted patents using the year of their issue – which we refer to as “cohort
comparisons.” Second, we compare PGPubs and patent grants based on
publication date and the patent issue date, respectively. These “contemporaneous
comparisons” provide an indication of how both application and patent claims are
changing in a particular year, rather than by looking at the year in which a
published application was disposed. Third, we examine publication-patent pairs,
which we refer to as “paired comparisons.” This permits us to measure trends
over time in the change in claim language during prosecution, by computing

15

Electronic copy available at: https://ssrn.com/abstract=2844964

annual means of ΔICL and ΔICC within applications that are granted.47 Because
paired comparisons provide an indication only of the changes to ICL or ICC of a
given application, this measure is the least likely to suffer from problems of cross-
patent scope comparisons.

A few stylized facts emerge from examining the trends in Figures 3-10:

 There have been significant trends in ICL and ICC over time. Figures
3 and 4 show the trend in ICL and ICC, respectively for granted
patents. There is a notable shift towards broader patents from 1984-
2004, after which there is a shift towards narrower patents (2004-
2014). The trend holds for both ICL (Figure 3) and ICC (Figure 4).

 PGPub-grants tend to be narrower than PGPub-abandonments based

on ICL, and granted patents are narrower still (Figures 5 and 7). This
holds across the observable range (2001-2014) whether measured by
the cohort comparison (Figure 5) or the contemporaneous comparison
(Figure 7). It confirms what we observed for the full distribution in
Figure 1.

 PGPub-grants and PGPub-abandonments are virtually identical at the

means based on ICC, whether measured by the cohort comparison
(Figure 6) or the contemporaneous comparison (Figure 8).

 Granted patents are narrower than PGPubs as measured by ICC.

However, that difference has been getting smaller over the last decade
as measured by the cohort comparison (Figure 6), and virtually
disappearing at the mean as measured by the contemporaneous
comparison (Figure 8).

 There has been an upward trend in the number of words added to ICL
between publication and grant as shown in Figure 9. At the same time,
the number of independent claims removed from applications has gone
from -0.7 in 2001 to -0.2 in 2014 (Figure 10). These facts are
consistent with the cohort comparisons in Figures 5 and 6.

47 The paired comparisons are aggregated based on the date of disposal (issuance). As with cohort
comparisons, it is feasible to aggregate by date of publication, which may better highlight applicant filing
rather than examination behaviors.

16

Electronic copy available at: https://ssrn.com/abstract=2844964

 Overall, the trends reinforce the conclusion that the examination process
reduces the scope of patent applications. On average, examination adds words to
the shortest independent claim and reduces the number of independent claims.
Further, the trend over the last decade has been towards narrower patents. ICL for
patents has grown significantly since 2004 even while the claim length of
published applications has remained more flat (Figure 5). This indicates that the
examination process may have become more stringent. The contemporaneous
comparisons show that applicant’s may have responded to this in recent years by
filing narrower claims. However, one should note that the publication time series
are censored, because most applications published in 2014 were still pending by
the end of 2014. So, the values for recent years include only those applications
that received a fast allowance or that abandoned early. (We show below that
pendency is correlated with the scope of incoming applications.)
The change in trends beginning around 2004 may correspond to various
Patent Office examination quality initiatives adopted following the PTO’s 2003
21st Century Strategic Plan and July 2003 legislative hearings on patent quality,
including expanded reviews of primary examiners’ work, “second-pair-of-eyes”
reviews, and quality assurance reviews.48 USPTO quality initiatives adopted
around 2004 and later may have influenced examiner and subsequently applicant
behaviors, thus leading to narrower patents over the last decade.

4.3 Relationships between patent scope and examination pendency

The differences in scope between allowed and abandoned applications

suggests that there may be differences in patent prosecution based on the scope of
the incoming applications, aside from the difference in allowance itself. To
investigate this we focus on examination pendency, which is an issue central to
applicants, the PTO, and to Congress (Mitra-Kahn, et al, 2013).

We measure pendency by total pendency: the time from filing to final

disposal. We use this measure of pendency to determine how ICL or ICC at
publication are associated with the time in prosecution. Figures 11 and 12
represent scatter plots of ICL and ICC against total pendency. As one might

48See, e.g., 21st Century Strategic Plan, supra note 17, at 8-9 (discussing measures to improve examiner
competency and to enhance quality assurance techniques); National Academy of Public Administration,
Report for the U.S. Congress and the USPTO, U.S. Patent and Trademark Office: Transforming to Meet the
Challenges of the 21st Century 66-67 (Aug. 2005).

17

Electronic copy available at: https://ssrn.com/abstract=2844964

expect (or hope), we find a positive correlation between application scope and
total pendency. That is, broad applications tend to have higher pendency. This is
shown in Figure 11 by a negative correlation between ICL and total pendency
(i.e., fewer words indicates a longer pendency). Correspondingly, Figure 12
shows a positive correlation between the number of independent claims and
pendency.

These results are intuitive, particularly for ICC, as examiners should require
more time to evaluate each additional independent claim (which by hypothesis
should require independent evaluation). The results are identical if we restrict the
definition of pendency to examination pendency only (post-first-action
pendency).49

If broad patents have a longer pendency, a natural question is whether the

longer pendency has any impact on the resulting claims at the time of final
disposal. With our data we cannot observe claims at the time of abandonment.
However, we can investigate the relationship between pendency and the claims at
disposal for granted patents. More precisely, we are interested in the relationship
between pendency and the change in claims for granted patents.

For our publication-patent pairs, we calculate the change in ICL and the
change in ICC between publication and grant (as defined in Figures 9 and 10).
Our interest is in whether these differences are correlated with pendency. In both
cases, we find that greater pendency is associated with more narrowing of the
claims during prosecution. Figure 13 shows that there is a positive correlation
between pendency and ICL (more time is correlated with more words added to
the claim). Correspondingly, Figure 14 shows that greater pendency is associated
with more independent claims being removed during prosecution.

In short we find that broader applications are subject to longer pendency, and
longer pendency is associated with more significant narrowing of claims, both in
the length of claims, and the number of claims. This is confirmed in Figures 15
and 16, which show the change in ICL and ICC during prosecution, against the
values of ICL and ICC at publication, respectively. The scatter plot (Figure 15)
shows a negative correlation between ICL and ICL, indicating that broader

49Post-first-action pendency measures the time from the first action by an examiner to the time of final
disposal. This definition of pendency reflects the time under examination at the office, which is impacted by
both examiner and applicant behavior.

18

Electronic copy available at: https://ssrn.com/abstract=2844964

applications (a low value of ICL at publication) experience greater narrowing (a
larger value of ICL). Figure 16 also shows a negative correlation between ICC
and ICC, again indicating that broader applications (a high value of ICC at
publication) experience greater narrowing (a more negative value for ICC).
However, about 25% of applications do not have a change in ICL between
publication and grant, and over 50% do not have a change in the number of
independent claims.

4.4 Application characteristics

We find that different characteristics of applications can lead to statistically

significant differences in measures of scope. However, the general patterns about
scope discussed above hold for all groupings: narrower applications tend to be
granted, and the prosecution process tends to narrow applications.

Table 2 provides the ICL and ICC for PGPubs grouped by entity size,50
examination unit (technology center),51 technology category,52 and parent
application type.53 The technology center analysis was generally similar to that for
the NBER technology categories; thus, we restrict our discussion to the
technology centers. For each case, the ICL is higher for PGPub-grants relative to
PGPub-abandonments. The number of claims is not substantially different

50 Entity status is based on fee payments at the time of filing. Small and micro entities are combined as a
single category relative to large entities.
51 There are eight technology centers (TCs) used during our period of study, including Biotechnology and

Organic Chemistry (1600), Chemical and Materials Engineering (1700), Computer Architecture, Software,
and Information Security (2100), Computer Networks, Multiplex Communication, Video Distribution, and
the Security (2400), Communications (2600), Semiconductors, Electrical, and Optical Systems and
Components (2800), Transportation, Construction, Electronic Commerce, Agriculture, National Security and
License & Review (3600), and Mechanical Engineering, Manufacturing and Medical Devices/Processes
(3700).
52 NBER technology categories, as defined by Hall, Jaffe, and Trajtenberg (2001) and Marco, et al (2015)

are: Chemical (1), Computers and Communications (2), Drugs and Medical (3), Electrical and Electronics
(4), Mechanical (5), and Other (6).
53 Parent application type or application status relative to the parent. If there was no parent (a first time

filing), we identified the application as having “no parent” (not applicable, or USNA). For applications
having a parent application, we identified the type of such application. These were divided into applications
having a parent that was: a foreign application (Foreign, or FOR); a Patent Cooperation Treaty (PCT)
application (which was further subdivided by the designated office of the parent – either PCT-foreign or
PCT-US); a prior US non-provisional application (and if so, the relationship to that parent application as
discussed below), or a US provisional application (US-provisional, or US-PRO). If the application had a prior
US non-provisional application as its parent, we denoted the application’s relationship to the parent as a
continuation (CON), a divisional (DIV), or a continuation-in-part (CIP) application to a US application.

19

Electronic copy available at: https://ssrn.com/abstract=2844964

between PGPub-grants and PGPUB-abandonments across application
characteristics, which is consistent with the aggregate results in Figures 6 and 8.

There are some notable characteristics that differ from the means. With
regard to technology, applications in biotechnology (TC 1600) have the largest
difference in ICL between granted applications and abandoned applications:
approximately 28 words. This is driven by the very low values for PGPub-
abandonments, which are about 12-15 words below the PGPub-abandonment
mean of 94 (from Table 1). However, these applications tend to have the most
independent claims at filing. Further, biotech is the only examination unit for
which PGPub-abandonments have more claims, on average, than PGPub-grants.

TC 3600 (including transportation, construction, e-commerce, and

agriculture) tends to have the longest claims (125 words for PGPub-grants and
107 words for PGPub abandonments, relative to the means of 111 and 94,
respectively). These applications also tend to have the fewest independent claims.
Surprisingly, small and large entities look almost identical at the mean for ICL
and ICC. Applications with foreign parents tend to be narrower than average at
filing, having higher ICL and lower ICC. The broadest patents at filing tend to be
those with US provisional parents.

Table 3 provides the ICL and ICC for publication-patent pairs, by groups
based on application characteristics. By comparing claims at publication to claims
at grant, we can identify the average change in claims during patent prosecution.
The publication values in Table 3 match those found in Table 2 for granted
applications. There are several interesting facts that emerge from Table 3. Most
notably, for each group applications are narrowed between publication and grant,
in terms of ICL and ICC. We also see interesting differences between application
types.

Small and large entity applications tend to be similar at filing, but small
entities experience greater narrowing during prosecution, leading to 5 more words
and 0.25 fewer claims at issue relative to large entities. Biotech applications again
stand out relative to other examination groups: they are not significantly narrowed
with respect to ICL (only 11 words), but they lose an average of 1.5 independent
claims during prosecution. This is likely based on nature of the invention and the
terminological (nomenclature) conventions for how certain types of inventions
(particularly chemical products) are claimed. Computer-related patents—on the
other hand—are more subject to increases in ICL than to decreases in ICC.

20

Electronic copy available at: https://ssrn.com/abstract=2844964

 Parent types reveal some interesting facets about application sources. Both
foreign and PCT-foreign sourced applications are filed with the longest
independent claims (123 and 120 words, respectively), yet they are among the
highest with respect to changes in ICL during prosecution (44 and 48 words,
respectively). This means that the ICL of the resulting patents has an ICL of more
than 165 words—more than 15 words higher than the next highest parent type.
This is perhaps surprising because foreign applications may already have been
through an examination process in the home jurisdiction, and thus may be “pre-
narrowed.” Further, the other application types with significant narrowing during
prosecution are those with no parent and those with provisional parents (adding
51 and 48 words, respectively). Yet, those applications tend to be filed with the
broadest claims (99 words at the mean). One might expect that “new”
applications, with no previous non-provisional filings, would be filed with broad
claims. Thus, it is surprising that foreign applications and new applications are
narrowed by similar amounts.

Continuations and divisionals of regular US applications had the largest ICL at

publication and had the smallest ΔICL among U.S. applications (+29.6 and +31.5
words, respectively). It is likely that continuations tend to be narrower when filed
and require fewer changes from application to grant than other applications,
because continuations have already gone through at least one round of US
prosecution before the continuation was filed.

5 Validation

To validate our ICL and ICC measures of patent scope, we employ several
statistical tests to compare these measures with post-grant outcomes and other
variables traditionally correlated with patent scope, as shown in Tables 4a and 4b.
The tests extend the previous literature and examine the impact of patent scope—
based on ICL and ICC—on (1) forward citations, (2) the number of Cooperative
Patent Classification (CPCs)54 subclasses to which the patent was assigned, (3)
patent maintenance, and (4) a novelty measure based on whether the granted
patent was issued in a “new” US patent classification subclass. We use a variant
of the validation method from Lerner (1994), which analyzes the relationship
between a proxy for patent scope—the number of 4-digit International Patent

54The CPC classification system was jointly developed by the USPTO and European Patent Office (EPO)
and is a descendent of the IPC classification. For more information on t the CPC classification system, please
visit http://www.cooperativepatentclassification.org/.

21

Electronic copy available at: https://ssrn.com/abstract=2844964

Classifications (IPCs) a patent was assigned—to the number of forward citations
assigned to a given patent and to the incidence of litigation. We present evidence
that our measures of patent scope explain traditional scope proxies in a way
consistent with Lerner (1994). We also discuss how our measures relate to the
results from the USPTO’s Patent Litigation and USPTO Trials: Implications for
Patent Examination Quality, which examined the relationship between the
incidence of litigation and ICL and ICC at grant. Our results show that the
relationship between our measures of patent scope and the outcome variables
above is consistent with other validation tests of patent scope in the literature.

Lerner (1994) found that a proxy for patent scope, the number of 4-digit IPCs,
was positively and significantly related to the number of forward citations a patent
receives. An increase in the number of 4-digit IPCs assigned to a patent reflects
an increasing number of distinct technologies incorporated into the invention,
which can be interpreted as increasing broadness of a given patent. Lerner (1994)
used a simple Poisson regression to examine the relationship between the
dependent variable, a count of forward citations for a given patent, and the
independent variable, the number of IPCs. He also controlled for the time since
grant, to account for varying exposure time among patents in his sample of
biotechnology firms. The results show that as the number of IPCs increases, the
number of forward citations in a given patent increases as well.

We extend Lerner’s analysis to include maintenance rates and forward citations

(following van Zeebroeck, 2011) and a novelty indicator based on Fleming (2001)
and Strumsky et al (2012). Further, we include the number of CPCs as a
dependent variable. More precisely, our dependent variables include two count
variables and two binary indicators are defined as follows:

 Forward citations. A count of the number of citations received by the

patent within three years of the issue date.
 Number of subclasses. A count of the number of unique CPC
subclasses (4-digit) assigned to the patent.
 Fully maintained. A binary indicator of whether the patent was
maintained to its maximum statutory term (paying the requisite fees at
3.5, 7.5, and 11.5 years after grant).
 New subclass. A binary indicator of whether the patent was classified in
a “new” subclass, according the US Patent Classification system.
“New” is defined as being within 12 months of the first use of the
subclass (see Fleming, 2001 and Strumsky et al, 2012).

22

Electronic copy available at: https://ssrn.com/abstract=2844964

 We expect each indicator to be positively correlated with patent scope, along
the lines of Lerner’s argument and the findings in van Zeebroeck (2011). Forward
citations have long been used by economists as a correlate of patent value and
scope (van Zeebroeck, 2011). Patent maintenance is closely related to the
concepts of patent value and patent scope. According to Bessen (2008), “[t]he
implicit value of a patent is revealed when its owner pays a renewal fee, implying
that the patent is worth more than the fee required to keep it in force.” Broader
patents, ceteris paribus, have wider applicability than a narrower patent
representing similar underlying technologies, and should therefore be more
valuable.

First-movers in a technology space have the opportunity to patent fundamental

inventions. These seminal patents can be expected to have broader scope than the
incremental inventions that follow (Strumsky et al, 2012). Until the conversion to
CPC in 2015, the USPTO regularly re-evaluated the US Patent Classification
system. New classes or subclasses were created retrospectively based on whether
a significant volume of the “new” inventions had been filed, so that a new
subclass would make routing and search easier. For instance, class 977
(nanotechnology) was created in August, 2004.55 The classification effort had the
purpose to “[f]acilitate the searching of prior art related to Nanotechnology,” and
to “[f]unction as a collection of issued U.S. patents and published pre-grant patent
applications relating to Nanotechnology across the technology centers.” As such,
it added the cross-reference classification to already issued U.S. patents. The
earliest patent in class 977 is US patent 4,107,288, issued in 1978,56 a full 26
years before the creation of the class. These early patents represent the first
patents identified by the USPTO that are relevant for prior art search in the
technology.

We expect ICL and ICC to be negatively and positively correlated,

respectively, to our patent scope indicators. To confirm this hypothesis, we run
Poisson regressions with forward citations and the number of subclasses as
dependent variables, and a linear probability model (ordinary least squares) for
the fully maintained and new subclass indicators. We also include year fixed
effects and US Patent Classification fixed effects, to control for differences in

55See USPTO memo dated August 25, 2004 at

http://www.uspto.gov/web/offices/pac/dapp/opla/documents/nanotechdig.pdf (accessed August 9, 2016).
56

23

Electronic copy available at: https://ssrn.com/abstract=2844964

claim length and citation behavior by applicants between classes and across years.
Tables 4a and 4b present the results of the regressions.

For each of the four dependent variables we estimate three models based on the
explanatory variables: ICL, ICC, and ICL and ICC together. Each model includes
year fixed effects and US Patent Class fixed effects. Our expectation is that ICL
will have a negative coefficient and ICC will have a positive coefficient, both of
which correspond to a positive correlation between our scope measures and the
dependent variables of value and scope.

For ICC, all coefficients are positive and statistically significant at the 1
percent significance level for all specifications. For ICL all coefficients are
negative and statistically significant at the 0.1 percent significance level with
three exceptions. The coefficient is positive for forward citations when combined
with ICC (Model 6), and it is negative but not statistically significant for the new
subclass specifications. The robustness of the results across specifications implies
that ICL and ICC are useful measures of patent scope. Because the models that
include both measures tend to have the expected signs further imply that ICC and
ICL represent different aspects of patent scope.

As further evidence that ICL and ICC represent patent scope, we rely on results
from Marco et al (2015). There, the authors find that patent scope—as measured
by average independent claim length and independent claim count—is correlated
with the incidence of patent litigation. Lanjouw and Schankerman (2001) explain
why patent breadth should be positively correlated with litigation. Thus, the result
provides more evidence that ICL and ICC are indicators of patent scope.

6 Conclusion
This paper presents the first large-scale analysis of patent claim language as it
applies to patent scope. We define two document-level measurements of scope
that should be useful to researchers interested in patent value and patent quality:
independent claim length (ICL) and independent claim count (ICC). Our
hypotheses that ICL is negatively correlated with patent scope and ICC is
positively correlated with patent scope are born out in several ways. First, we find
that the narrowing process that occurs during examination tends to add words to
the shortest independent claim and tends to remove independent claims, leading to
greater ICL and lower ICC. Second, our formal validation exercise shows that
ICL and ICC independently explain patent maintenance, forward citations, the
breadth of patent classes, and—to a lesser extent—novelty.

24

Electronic copy available at: https://ssrn.com/abstract=2844964

 As shown above, using very simple claim length and claim count metrics to
model application and patent scope can provide useful information about patent
prosecution. For instance, we show that narrower applications tend to have shorter
examination times, and that longer examination times lead to more significant
narrowing of the original application claims. As a measure of scope, we expect
ICL and ICC to be the most meaningful for intra-application comparisons and
intra-technology comparisons. However, we believe that the results presented
here provide ample evidence that claim text can be usefully exploited by
researchers to measure patent scope.

Our continuing research agenda includes more in-depth analysis into the
examination process, as well as exploring how natural language processing
techniques can be applied to claim text. By making these data widely available we
hope to stimulate more research into the usefulness of analyzing claim text in
order to understand patent scope and its relationship to examination quality and
patent quality.

References

"Adoption of Metrics for the Enhancement of Patent Quality Fiscal Year 2011."
U.S. Patent and Trademark Office (blog). Accessed November 3, 2015.
http://www.uspto.gov/sites/default/files/patents/init_events/qual_comp_metric.pdf
.

Allison, John R., and Ronald J. Mann. "The Disputed Quality of Software
Patents." Washington University Law Review 85 (2007): 297.

Allison, John R., Mark A. Lemley, Kimberly A. Moore, and R. Derek Trunkey.
"Valuable Patents." Georgetown Law Review 92 (2004): 435.

Allison, John R., J. H. Walker, and Mark A. Lemley. "Patent Quality and
Settlement among Repeat Patent Litigants." Georgetown Law Review 99 (2011):
677.

Allison, John R., Mark A. Lemley, and Joshua Walker. "Extreme Value or Trolls
on Top? The Characteristics of the Most-Litigated Patents." University of
Pennsylvania Law Review 158 (2009): 1.

Allison, John R., and Mark A. Lemley. "Who's Patenting What? An Empirical
Exploration of Patent Prosecution." Vanderbilt Law Review 53 (2000): 2107.

25

Electronic copy available at: https://ssrn.com/abstract=2844964

Anderson, J. Jonas and Peter S. Menell, “Informal Deference: A Historical,
Empriical, and Normative Analysis of Patent Claim Construction,” Nw. U. L. Rev.
108 (2014): 1.

Burk, Dan L. and Mark A. Lemley, “Fence Posts or Sign Posts? Rethinking
Patent Claim Construction,” U. Pa. L. Rev. 157 (2009): 1743.

Caillaud, Bernard, and Anne Duchêne. "Patent Office in Innovation Policy:

Nobody's Perfect." International Journal of Industrial Organization 29, no. 2
(2011): 242.

Choi, Jay Pil. "Patent Pools And Cross-Licensing In The Shadow Of Patent
Litigation." International Economic Review 51, no. 2 (2010): 441.

Dargaye Churnet, “Patent Claims Revisited,” Nw. J. Tech. & Int. Prop. 11 (2013):
501.

Collins, Kevin E. "The Reach of Claim Scope Into After-Arising Technology: On

Thing Construction and the Meaning of Meaning." Connecticut Law Review 41
(2008): 493.

Comino, Stefano, and Clara Graziano. "How Many Patents Does It Take to Signal
Innovation Quality?" International Journal of Industrial Organization 43 (2015):
66.

Cotropia, Christopher A., Cecil D. Quillen, Jr., and Ogden H. Webster. "Patent
Applications and the Performance of the U.S. Patent and Trademark Office." The
Federal Circuit Bar Journal 23 (2013): 179.

Federal Trade Commission. To Promote Innovation: The Proper Balance of

Competition and Patent Law and Policy. October 2013. Accessed November 3,
2015. http://www.ftc.gov/os/2003/10/innovationrpt.pdf.

Fleming, Lee. “Recombinant Uncertainty in Technological Search.” Management

Science Volume 47, Issue 1, pp. 117-132 (2001).
http://dx.doi.org/10.1287/mnsc.47.1.117.10671.

Frakes, Michael D., and Melissa F. Wasserman. "Does the U.S. Patent and
Trademark Office Grant Too Many Bad Patents?: Evidence From A Quasi-
Experiment." Stanford Law Review 67 (2015): 613.

26

Electronic copy available at: https://ssrn.com/abstract=2844964

Frakes, Michael, and Melissa F. Wasserman. "Does Agency Funding Affect
Decisionmaking?: An Empirical Assessment of the PTO’s Granting Patterns."
Vanderbilt Law Review 66 (2013): 67.

Graham, Stuart J., Bronwyn Hall, Dietmar Harhoff, and David Mowery. "Post-
Issue Patent "Quality Control": A Comparative Study of US Patent Re-
examinations and European Patent Oppositions." NBER Working Paper 8807,
2002. Accessed November 3, 2015. http://www.nber.org/papers/w8807.

Guerrini, Christi J. "Defining Patent Quality." Fordham Law Review 82 (2014):

3091.

Hall, Bronwyn, Adam Jaffe, and Manuel Trajtenberg. "The NBER Patent Citation
Data File: Lessons, Insights and Methodological Tools." NBER Working Paper
8498, October 2001.

Kaser, Bruce A. "Patent Application Recycling: How Continuations Impact Patent

Quality & What The USPTO Is Doing About It." Journal of the Patent and
Trademark Office Society 88 (2006): 426.

Koenen, Johannes, and Martin Peitz. "Firm Reputation and Incentives to ‘Milk’
Pending Patents." International Journal of Industrial Organization 43 (2015): 18.

Krause, Thomas W. & Heather F. Auyang, “What Close Cases and Reversals
Reveal About Claim Construction,” J. Marshall Rev. Intell. Prop. L. 12 (2013):
583.

Krause, Thomas W. & Heather F. Auyang, “What Reversals and Close Cases
Reveal About Claim Construction: The Sequel,” J. Marshall Rev. Intell. Prop. L.
13 (2014): 525.

Lemley, Mark A. "Software Patents and the Return of Functional Claiming."

Wisconsin Law Review 2013 (2013): 905.

Lemley, Mark A. "Rational Ignorance at the Patent Office." Northwestern

University Law Review 95 (2001): 1495.

Lemley, Mark A., and Bhaven Sampat. "Examining Patent Examination."

Stanford Technology Law Review 2010 (2010): 2.

Lemley, Mark A., and Bhaven Sampat. "Is the Patent Office a Rubber Stamp?"

27

Electronic copy available at: https://ssrn.com/abstract=2844964

Emory Law Journal 58 (2008): 181.

Lerner, Joshua, “The Importance of Patent Scope: An Empirical Analysis,”

RAND J. Econ. 25 (1994): 319.

Malackowski, James E., and Jonathan A. Barney. "What Is Patent Quality? A

Merchant Banc’s Perspective." LES Nouvelles 43 (2008): 123.

Marco, Alan C., Michael Carley, Steven Jackson, and Amanda F. Myers. "The
USPTO Historical Patent Data Files: Two Centuries of Innovation." U.S. Patent
and Trademark Office Working Paper, 2015.

Marco, Alan C., Richard D. Miller, Kathleen Kahler Fonda, Pinchus M. Laufer,
Paul Dzierzynski, and Martin Rater, “U.S. Patent and Trademark Office, Patent
Litigation and USPTO Trials: Implications for Patent Examination Quality”
(January 2015). Accessed November 3, 2015.
http://www.uspto.gov/sites/default/files/documents/Patent%20litigation%20and%
20USPTO%20trials%2020150130.pdf

Miller, Shawn P. "What’s the Connection Between Repeat Patent Litigaiton and
Patent Quality? A (Partial) Defense of the Most Litigated Patents." Stanford
Technology Law Review 16 (2013): 313.

Mitra-Kahn, Benjamin, Alan C. Marco, Michael Carley, Paul D’Agostino, Peter

Evans, Carl Frey, and Nadiya Sultan, “Patent Backlogs, inventories, and
pendency: an International Framework” (June 2013). Accessed August 16, 2016.
https://www.gov.uk/government/uploads/system/uploads/attachment_data/file/31
1239/ipresearch-uspatlog-201306.pdf

Moore, Kimberley A., “Markman Eight Years Later: Is Claim Construction More
Predictable?,” Lewis & Clark L. Rev. 9 (2005): 231.

Mossinghoff, Gerald J., and Vivian S. Kuo. "Post-Grant Review of Patents:

Enhancing the Quality of the Fuel of Interest." IDEA: The Journal of Law and
Technology 43 (2002): 83.

National Academy of Public Administration, “Report for the U.S. Congress and
the USPTO, U.S. Patent and Trademark Office: Transforming to Meet the
Challenges of the 21st Century” (Aug. 2005).

Osenga, Kristen J. "The Shape of Things to Come: What We Can Learn from

28

Electronic copy available at: https://ssrn.com/abstract=2844964

Patent Claim Length." Santa Clara High Technology Law Journal 28 (2012): 617.

Petherbridge, Lee, “On Addressing Patent Quality,” U. Pa. L. Rev. PENNumbra

158 (2009): 13.

Quillen, Cecil D., Jr., and Ogden H. Webster. "Continuing Patent Applications
and Performance of the U.S. Patent and Trademark Office." The Federal Circuit
Bar Journal 11 (2001): 1.

Quillen, Cecil D., Jr., and Ogden H. Webster. "Continuing Patent Applications
and Performance of the U.S. Patent and Trademark Office - Extended." The
Federal Circuit Bar Journal 12 (2002): 35.

Quillen, Cecil D., Jr., and Ogden H. Webster. "Continuing Patent Applications
and Performance of the U.S. Patent and Trademark Office - Updated." The
Federal Circuit Bar Journal 15 (2006): 635.

Quillen, Cecil D., Jr., and Ogden H. Webster. "Continuing Patent Applications
and Performance of the U.S. Patent and Trademark Office – One More Time."
The Federal Circuit Bar Journal 18 (2009): 379.

Rai, Arti K. "Improving (Software) Patent Quality Through The Administrative

Process." Houston Law Review 51 (2013): 503.

Régibeau, Pierre, and Katharine Rockett. "Innovation Cycles And Learning At

The Patent Office: Does The Early Patent Get The Delay?" The Journal of
Industrial Economics 58 (2010): 222.

Sarnoff, Joshua D. and Edward D. Manzo, “An Introduction to, Premises of, and
Problems with Patent Claim Construction,” in Claim Construction in the Federal
Circuit § 0:4(2) (2014 on-line ed. Thompson West).

Schuett, Florian. "Patent Quality and Incentives at the Patent Office." The RAND
Journal of Economics 44, no. 2 (2013): 313-36.

Schwartz, David L., “Practice Makes Perfect?: An Empirical Study of Claim

Construction Reversal Rates in Patent Cases,” Mich. L. Rev. 107 (2008): 223

Schwartz, David L. and Jay P. Kesan, “Analyzing the Role of Non-Practicing

Entities in the Patent System,” Cornell L. Rev. 99 (2014): 425.

29

Electronic copy available at: https://ssrn.com/abstract=2844964

Strumsky, Deborah, José Lobo, and Sander Van der Leeuw, “Using patent
technology codes to study technological change.” Economics of Innovation and
New Technology 21 (3) pp. 267-286 (2012).
http://www.doi.org/10.1080/10438599.2011.578709.

Szmrecsányi, Benedikt. "On Operationalizing Syntactic Complexity."

JOURNÉES INTERNATIONALS D’ANALYSE STATISTIQUE DES DONNÉES
TEXTUALLES, 2004, 1037-038.

U.S. Government Accountability Office,. Report to Congressional Committees,

Intellectual Property: Assessing Factors That Affect Patent Infringement
Litigation Could Help Improve Patent Quality, GAO-13-465, August 2013.
Accessed November 3, 2015.
http://www.uspto.gov/sites/default/files/aia_implementation/GAO-12-
465_Final_Report_on_Patent_Litigation.pdf.

van Zeebroeck, Nicolas, Bruno van Pottelsberghe de la Potterie, and Dominique

Guellec. "Claiming More: The Increased Voluminosity of Patent Applications and
Its Determinants." Research Policy 38, no. 6 (2009): 1006-020.

van Zeebroeck, Nicolas. “The puzzle of patent value indicators,” Economics of

Innovation and New Technology, Volume 20, Issue 1 (2011).
http://dx.doi.org/10.1080/10438590903038256.

Wagner, R. Polk, “Understanding Patent Quality Mechanism,” U. Pa. L. Rev. 157

(2009): 2135.

Wagner, R. Polk & Lee Petherbridge, “Did Phillips Change Anything? Analysis
of the Federal Circuit’s Claim Construction Jurisprudence,” in Intellectual
Property and the Common Law (S. Balganesh ed. Cambridge U. Press 2011).

Wasow, Thomas. "Remarks on Grammatical Weight," in Language Variation and

Change (Cambridge U. Press 1997): 81.

Yelderman, Stephen. "Improving Patent Quality with Applicant Incentives."

Harvard Journal of Law & Technology 28 (2014): 77.

Zimmer, John P., “To Infinity and Beyond: The Problem of Open-Ended Claim
Language in the Unpredictable Arts,” S. Car. L. Rev. 59 (2008): 865.

30

Electronic copy available at: https://ssrn.com/abstract=2844964

Tables and Figures

Table 1. Distributional statistics for pre-grant publications (2001-2014) and patent grants
(1976-2014)
ICL ICC
Frequency Mean P25 P50 P75 Mean P25 P50 P75
Publications (2001-2014)
Later Abandoned 1089427 94.2 46 75 115 3.03 1 2 3
Later Granted 2113273 111.4 58 90 137 3.08 2 3 4
Pending* 790019 107.1 59 90 133 2.73 2 3 3
All 3992719 105.8 54 86 130 2.99 2 3 3

Grants
At Publication 2113273 111.4 58 90 137 3.08 2 3 4
At Grant (previously published) 2113273 155.9 93 136 195 2.70 1 2 3
At Grant (not previously published) 634235 141.0 82 121 176 3.12 2 3 4
At Grant (1976-2000) 2203409 155.6 92 137 198 2.43 1 2 3
* As of December 31, 2016

31

Electronic copy available at: https://ssrn.com/abstract=2844964

Table 2. Applications at publication by application characteristics (2001-2014)

IC Length IC Count
Later Issued Later Abandoned Difference Later Issued Later Abandoned Difference
Small entity status
Large 111.03 94.07 16.96 3.09 3.08 0.02
Small or Micro 112.24 94.03 18.21 3.03 2.94 0.08
Technology Center
1600 110.23 81.79 28.44 3.75 3.98 -0.23
Biotech, Organic Chem
1700 97.75 84.37 13.38 2.74 2.66 0.08
Chem & Mat Engineering
2100 107.80 95.68 12.11 3.60 3.50 0.10
Comp Architecture
2400 107.73 95.60 12.13 3.59 3.50 0.10
Comp Networks
2600 109.21 95.68 13.53 3.47 3.19 0.27
Communications
2800 110.99 95.65 15.35 2.89 2.62 0.27
Semiconductors, Electrical
3600 125.45 106.86 18.60 2.80 2.78 0.02
Trans, Constr, E-Comm, Ag
3700 117.04 99.93 17.11 2.84 2.67 0.17
Mech, Mfg, Products
NBER category
1 - Chemicals 102.07 95.20 6.87 2.91 2.84 0.07
2 - Comp & Comm 109.71 97.43 12.28 3.43 3.36 0.07
3 - Drugs & Medical 107.28 78.80 28.47 3.54 3.72 -0.18
4 - Electrical 110.82 95.41 15.41 2.85 2.63 0.23
5 - Mechanical 123.43 105.25 18.18 2.66 2.49 0.17
6 - Others 114.17 95.72 18.45 2.83 2.58 0.25
Parent application type
Foreign 122.95 101.84 21.10 2.69 2.66 0.03
PCT - foreign 119.92 97.11 22.81 2.66 2.81 -0.15
PCT - US 109.40 87.94 21.46 3.39 3.60 -0.21
CIP of US app 107.15 95.77 11.37 3.58 3.51 0.07
CON of US app 112.09 94.85 17.24 3.27 3.43 -0.17
DIV of US app 108.99 94.25 14.74 3.16 3.12 0.04
No parent 98.73 91.91 6.82 3.32 2.97 0.36
US provisional 98.83 83.19 15.64 3.67 3.44 0.23

IC Length is defined as the length of an application's shortest Independent Claim

32

Electronic copy available at: https://ssrn.com/abstract=2844964

Table 3. Publication-Patent Pairs (2001-2014)

IC Length IC Count
At publication At issuance Difference At publication At issuance Difference
Small entity status
Large 111.03 154.97 43.94 3.09 2.75 -0.34
Small or Micro 112.24 160.03 47.79 3.03 2.50 -0.53
Technology Center
1600 110.23 121.39 11.16 3.75 2.27 -1.48
Biotech, Organic Chem
1700 97.75 138.64 40.88 2.74 2.21 -0.54
Chem & Mat Engineering
2100 107.80 175.53 67.73 3.60 3.32 -0.28
Comp Architecture
2400 107.73 183.72 75.99 3.59 3.34 -0.25
Comp Networks
2600 109.21 159.73 50.53 3.47 3.26 -0.21
Communications
2800 110.99 145.36 34.36 2.89 2.66 -0.23
Semiconductors, Electrical
3600 125.45 179.36 53.90 2.80 2.58 -0.22
Trans, Constr, E-Comm, Ag
3700 117.04 168.23 51.18 2.84 2.53 -0.31
Mech, Mfg, Products
NBER category
1 - Chemicals 102.07 135.32 33.25 2.91 2.20 -0.71
2 - Comp & Comm 109.71 165.59 55.88 3.43 3.19 -0.24
3 - Drugs & Medical 107.28 138.32 31.04 3.54 2.47 -1.07
4 - Electrical 110.82 148.19 37.37 2.85 2.60 -0.25
5 - Mechanical 123.43 167.59 44.16 2.66 2.44 -0.22
6 - Others 114.17 165.71 51.54 2.83 2.53 -0.30
Parent application type
Foreign 122.95 166.83 43.88 2.69 2.49 -0.20
PCT - foreign 119.92 168.06 48.14 2.66 2.22 -0.44
PCT - US 109.40 150.78 41.38 3.39 2.58 -0.81
CIP of US app 107.15 149.30 42.15 3.58 3.02 -0.56
CON of US app 112.09 141.68 29.59 3.27 2.89 -0.38
DIV of US app 108.99 140.48 31.49 3.16 2.37 -0.79
No parent 98.73 150.02 51.30 3.32 3.03 -0.29
US provisional 98.83 146.68 47.85 3.67 3.04 -0.63

Note: 10,311 of 2,113,273 publication-patent pairs were lost due to data availability issues for application
characteristics. IC Length is defined as the length of an application's shortest Independent Claim

33

Electronic copy available at: https://ssrn.com/abstract=2844964

Table 4a. Validation Results
Fully Maintained New Subclass
VARIABLES (1) (2) (3) (4) (5) (6)

ICL at Grant -0.0229*** -0.0169*** -1.87e-05 -7.28e-06

(0.000435) (0.000444) (2.01e-05) (2.05e-05)
ICC at Grant 0.0127*** 0.0113*** 2.81e-05** 2.75e-05**
(0.000169) (0.000172) (9.61e-06) (9.80e-06)
Constant 0.518*** 0.449*** 0.479*** 0.0230*** 0.0229*** 0.0229***
(0.00167) (0.00159) (0.00177) (0.000170) (0.000169) (0.000173)

Observations 1,448,038 1,448,177 1,448,038 4,937,731 4,937,997 4,937,731

R-squared 0.017 0.019 0.020 0.006 0.006 0.006
Years 1994-2004 1994-2004 1994-2004 1976-2014 1976-2014 1976-2014
Standard errors in parentheses. The OLS models above include disposal year and USPC fixed effects.
*** p<0.001, ** p<0.01, * p<0.05

Table 4b. Validation Results

Forward Citations Number of 4-digit CPCs
VARIABLES (7) (8) (9) (10) (11) (12)

ICL at Grant -0.00297*** 0.0183*** -0.0219*** -0.0189***

(0.000667) (0.000608) (0.000372) (0.000377)
ICC at Grant 0.0372*** 0.0379*** 0.00831*** 0.00669***
(0.000157) (0.000157) (0.000158) (0.000163)

Observations 2,068,106 2,068,231 2,068,106 4,666,314 4,666,557 4,666,314

Years 2000-2011 2000-2011 2000-2011 1976-2014 1976-2014 1976-2014
Standard errors in parentheses. The poisson models above include disposal year and USPC fixed effects.
*** p<0.001, ** p<0.01, * p<0.05

34

Electronic copy available at: https://ssrn.com/abstract=2844964

Figure 1

Figure 2

35

Electronic copy available at: https://ssrn.com/abstract=2844964

Figure 3

Figure 4

36

Electronic copy available at: https://ssrn.com/abstract=2844964

Figure 5

Figure 6

37

Electronic copy available at: https://ssrn.com/abstract=2844964

Figure 7

Figure 8

38

Electronic copy available at: https://ssrn.com/abstract=2844964

Figure 9

Figure 10

39

Electronic copy available at: https://ssrn.com/abstract=2844964

Figure 11

Figure 12

40

Electronic copy available at: https://ssrn.com/abstract=2844964

Figure 13

Figure 14

41

Electronic copy available at: https://ssrn.com/abstract=2844964

Figure 15

Figure 16

42

Electronic copy available at: https://ssrn.com/abstract=2844964

Appendix A: Example of Dependent to Independent Claim “Roll Up”
In section 2, we provided an example of a published application (US
20050065799) in which dependent claims five and six at publication were “rolled
up” into the first independent claim at grant (U. S. Patent 7769690). The
application and patent text is provided below. Most, but not all, of the
application’s dependent claims five and six are incorporated into the granted
patent’s first independent claim. As you can see, the inclusion of the dependent
claims into the independent claim narrows the scope of the independent claim.

U.S. Patent Application (US 20050065799 – filed 10/21/2002)

Independent Claim

1. A method for supply of data relating to a described entity to a relying entity, the
method comprising:
generating a first digital certificate signed with an electronic signature by a first
signing entity and including:
one or more attributes of the described entity;
one or more attributes of the first digital certificate which include one or
more attributes identifying the first signing entity;
an indication of data relating to the described entity which is to be
supplied;
an indication of one or more sources for the data to be supplied; and
one or more attributes identifying one or more relying entities to which
the data is to be supplied;
the relying entity forwarding the first digital certificate for processing; and
a source supplying the data indicated in the first digital certificate.

Dependent Claims

5. The method of claim 1, wherein some or all of the data relating to the described entity
is supplied by a second digital certificate to the relying entity, the second digital
certificate signed with an electronic signature by a second signing entity and including:
one or more attributes of the described entity including the data which is to be
supplied;
one or more attributes of the second digital certificate which include one or more
attributes identifying the second signing entity; and
one or more attributes identifying one or more relying entities to which the data
is to be supplied.

6. The method of claim 5, wherein the first digital certificate authorises the relying entity
to use the first digital certificate to obtain a second digital certificate.

43

Electronic copy available at: https://ssrn.com/abstract=2844964

U.S. Patent (US 7769690 – Granted 8/3/2010)

Independent Claim

1. A method for supply of data relating to a described entity to a relying entity, the
method comprising:
generating, using a computer device, a first digital certificate signed with an
electronic signature by a first signing entity and including:
one or more attributes of the described entity;
one or more attributes identifying the first signing entity;
an indication of data relating to the described entity which is to be
supplied;
an indication of one or more sources for the data to be supplied; and
one or more attributes identifying one or more relying entities to which
the data is to be supplied;

the relying entity forwarding the first digital certificate for processing; and

after the processing, the one or more sources supplying the data indicated in the first
digital certificate to the relying entity,

wherein some or all of the data relating to the described entity is supplied by a second
digital certificate to the relying entity, the second digital certificate signed with an
electronic signature by a second signing entity and including:
one or more attributes of the described entity including the data which is to be
supplied;
one or more attributes of the second digital certificate which include one or more
attributes identifying the second signing entity; and
one or more attributes identifying one or more relying entities to which the data
is to be supplied, and

wherein the first digital certificate authorizes the relying entity to use the first digital
certificate to obtain the second digital certificate.

44

Electronic copy available at: https://ssrn.com/abstract=2844964

Appendix B: Methodology

We applied a natural-language Python57 algorithm to identify whether a claim is

independent or dependent, and to parse each individual claim for the full text of
each claim. To do so, we assumed that all dependent claims contain some
dependency language referring to (and thereby incorporating limitations from)
earlier claims, rather than actually reciting the language of limitations of the
claims from which they depend.

In the document_stats datasets, we aggregated the individual claims-level

data into patent/application-level summary statistics.58 Each observation contains,
for each application at publication and for each patent at grant, the number of
independent and dependent claims, the average number of words in all
independent claims, and a count of the number of words in the shortest
independent claim. Since this paper’s principal focus is the analysis of patent
application and granted patent claims and filing characteristics, the dissemination
and analysis of other patent-prosecution-related characteristics, such as data on
RCE filings, numbers and types of continuations generated, appeals, etc., will be
left to future dataset releases and analyses.

This Appendix details the data sources, methodology, descriptive statistics, and
some general trends that can be observed in the claims_stats,
claims_fulltext, and document_stats datasets. It is our hope that
researchers will be able to use this data to enhance understanding of the
examination process, including but not limited to assessing patent scope and how
it changes during examination.

Data Sources
Our primary data sources for the claims-level datasets include the Patent
Application Publication Full-Text and Patent Grant Full Text files provided by the
U.S. Patent and Trademark Office (USPTO).59 The Patent Application Publication
Full-Text data, provided in XML format and disseminated as separate files by

57 The Python code used to generate the USPTO’s Patent Claims Research Datasets will be made available
soon on GitHub.
58 The data were obtained from USPTO Electronic Bulk Data Products (http://www.uspto.gov/learning-and-

resources/electronic-bulk-data-products)
59 Full-text of patents and patent applications is available at http://patft.uspto.gov/. Bulk data is available at

http://www.uspto.gov/learning-and-resources/electronic-bulk-data-products.

45

Electronic copy available at: https://ssrn.com/abstract=2844964

years or ranges of years, contains the full-text of all patent applications published
from December 2000 to December 31, 2014. The Patent Grant Full-Text files,
provided in multiple file formats (XML, SGML, and APS), contain the full-text of
all patents issued from 1976 to December 31, 2014. These files were cleaned,
parsed, and appended to create the claims_fulltext datasets (one each for
PGPubs and patents), which includes the patent or application number, the full-
text of each claim, and an indicator variable to distinguish between independent
and dependent claims, in a STATA® data file format.60 In the claims_stats
datasets (again, one each for PGPubs and patents), we include claim-level
statistics (e.g. word count, number of “or”s, etc.) but not the full text of each
claim. This allows researchers to analyze claim-level data in a more manageable
dataset size.
For our analysis, but not included in our data release, we merged an in-house
USPTO patent application database with the document_stats datasets to link
certain filing and prosecution information at the application/patent-level for
publicly available (published and/or granted) applications with our measure of
patent scope. This information includes the nature of any parent application for
the subject application (e.g., having a parent that was a foreign or PCT
application) and the relationship to the parent of the subject application if the
parent is a regular utility application (e.g., the subject application is a divisional
application of that parent) and any filing priority information relating to the parent
application. The USPTO in-house database includes various post-filing
prosecution characteristics such as disposal type (disp_ty) and disposal date
(disp_dt), among others.61 We also used certain prosecution characteristics.
For example, we use the disposal date for an application (which includes the time
evaluating any requests for continued examination (RCEs) in the same
application) to determine total pendency from filing to abandonment or grant
(“disposal”62), and post-first-action pendency to measure the time from first-
action to disposal. While the dataset does not include claim counts or claim
lengths at the time of an abandonment, the our merged data on publications
included a variable to distinguish whether the application matured into a granted

60 Cancelled claims were identified in claims_fulltext but were not included in independent claim count and
length summary statistic calculations.
61 For a fuller description of all of the prosecution characteristic variables that were available for coding,

please see the variable descriptions in Appendix C.

62 There are two types of disposals: abandonment or grant. For more information on disposals and patent

prosecution, please see http://www.uspto.gov/patents-getting-started/general-information-concerning-

patents#heading-22. Please note that abandoned applications can sometimes be reinstated.

46

Electronic copy available at: https://ssrn.com/abstract=2844964

patent or ultimately went abandoned (disp_ty). Accordingly, many of our
analyses distinguish characteristics at publication of applications that result in
grants from applications that result in abandonments. Of course, abandonment (or
grant) of a particular application does not mean that prosecution ended on the
invention described in the abandoned (or granted) application, as various forms of
continuation applications may have been filed prior to final disposition of any
particular application.

Data Limitations
Relying on publicly available information on claims as captured from existing
databases limits our sample in several ways. First, we can observe the claim text
only at the time of publication and at the time of grant. This reliance also restricts
the time period, because pre-grant publication of patent applications has been
practiced by the USPTO only for applications filed after November 29, 2000.63
Since that time, and without a non-publication request (which requires foregoing
international protection on the patented innovation), publication has been required
by statute 18 months after the filing priority date requested in relation to the
earliest related parent application.64 Applications filed prior to November 29,
2000 are unpublished. Thus, although our source patent dataset (grants) extends
back to 1976, the bulk patent application data contains applications filed only
during and after 2000. We have calculated that since November 29, 2000,
approximately ten percent of filed applications have opted out of publication.
Further, in contrast to the captured data on claims from granted applications (at
publication and at issue), machine-readable claim text is not readily available for
abandoned applications (after publication). That is, we cannot observe the change
in claims between publication and abandonment. Consequently, we limit our
analysis of difference variables (dif_wrd_min, dif_wrd_avg, and
dif_clm_ct) to publication-patent pairs (i.e., to applications that resulted in
granted patents).
Although it is possible for claims in a particular application to change between
filing and publication, we believe this is a relatively infrequent event. Our
analysis shows that only 8.11 percent of total applications in the dataset have a
preliminary claims amendment filed after their actual (not priority) filing date but
before the publication date. Normal office practice is to incorporate preliminary

63 See 35 U.S.C. 122(b).

64 See 35 U.S.C. § 122(b)(2)(B); 37 C.F.R. § 1.213(a)(1)-(4).

47

Electronic copy available at: https://ssrn.com/abstract=2844964

amendments into the claims when they are published, and thus these claim
amendments (except for the possible few that are filed too close to publication to
be incorporated) are reflected in the publication data. Since the percentage of
applications with preliminary amendments submitted between filing and
publication is relatively small, we have treated for analysis the claims at
publication as a reasonable approximation of the claims at filing.65
As can be expected in any dataset of this size, the source data files (the Patent
Application Publication Full-Text and Patent Grant Full Text files) have some
errors. Specifically, some claim language was excluded from the text and word
length counts. The general (introductory) claiming language (e.g., “I claim” or
“What is claimed is”) has been excluded from the claims_fulltext
datasets.66 Similarly, we have not included the numeral associated with any claim
in the claim length counts; rather, we have included only the language following
the numeral for any particular claim (although the numeral is included in the
dataset).67 For example, U.S. patent 4,788,34968 was issued with three claims of
word lengths fourteen, two, and two, respectively. Excluding the general claiming
language – which is not included in the datasets and consists of the words, “I
claim:” – and the numeral assigned to the claims thus allows for one word claims
such as chemical compounds. The exclusion of the general claiming language and
numerals from the claim counts slightly biases the individual, average, and
minimum independent claim length downwards.

Claim Identification and Measurements

As stated above, we used full-text claims data for patents and applications
(claims_fulltext) to create patent-level summary statistics for both PGPubs
and patents. We computed the summary statistics by applying a Python-based
algorithm developed to distinguish independent claims from dependent claims
and to compute various measures of claim length and claim count, among other
variables.69 The algorithm identifies independent from dependent claims by

65 It should be noted that not all preliminary amendments are included in an application’s publication. See
MPEP 1121.
66 There are exceptions in the claims_fulltext data set: (1) the first claims of twenty-two utility patents begin

with the general (introductory) claiming language, “I claim”; and (2) claims in ten patents, such as patent
6,901,209, begin with the words, “I Claim.” For example, claim 5 states, “I claim the access system of claim
4 characterized by the addition of data manager means to allow a user to access the program.” This list is not
exhaustive.
67 The Claim number can be found as a separate field in the claims_fulltext data set (claim_no).
68 See https://www.google.com/patents/US4788349
69 See Python code in Appendix D

48

Electronic copy available at: https://ssrn.com/abstract=2844964

assuming that dependent claims will reference independent and other dependent
claims, but not vice-versa.70 Specifically, if claim language contains a direct
reference to another claim or a group of claims, we designated the referring claim
as a dependent claim (and coded it as such in the database). If the claim contains
no such language, we designated it as an independent claim. We repeated this
process for all applications at publication and for those that are granted at issue.
To measure the independent claim length (ICL), we used a simple count of the
number of words in each independent claim.71 To create a patent-level metric, we
measured ICL by using the minimum claim length among all independent claims
of an application or granted patent. Our metric for the number of independent
claims is a simple independent claim count (ICC). 72 We did not include in
document_stats the minimum claim length for dependent claims. 73

Following our assumption that patent scope depends on the length and number of
independent claims, it is important to provide the arithmetic difference in the
length and number of independent claims between publication and grant. These
differences from publication to grant provide an approximation of the changes in
breadth of the independent claims from filing to grant and thus of the change in
the scope of the applications during prosecution. For example, as a direct result of

70 It may be the case that a claim will contain referents to other claims that do not incorporate the other
claims’ limitations. However, we believe this to be a rare event.
71 Because the algorithm uses natural language processing, claims that separate portions of words with spaces

are automatically read as including separate words, which may thereby artificially increase the claim’s word
count. For example, chemical formula sometimes are written as a single word without spaces, but
occasionally may contain many spaces, which would artificially increase the word count by as many spaces
as are added. See US Patent 3,262,977, claim 4 (“N – [1’ -phenyl-propyl-(‘1)] – 1,1 diphenyl-propyl-(3)-
amine”).
72 Our algorithm also identifies specific words or phrases (e.g., “or” and “selected from”) that are more likely

to have the potential to broaden the scope of an independent claim by addition of other words, to permit
robustness checks.
73 To measure the dependent claim length (DCL), we would need to start with a simple count of the number

of words in each dependent claim, and then add the count of the limitations language of the claim(s) from
which the dependent claim depends and eliminate the count of the referential language in the dependent claim
(as such language would then become duplicative and unnecessary). Nevertheless, the data in
claims_fulltext are coded with the claim number(s) from which each dependent claim directly
depends. Accordingly, some automated counts to approximate the number of words of dependent claims are
possible to perform, e.g., by tracing the chains of dependency and adding the simple count of the words of
each dependent claim and of the claim(s) from which it depends. (Such simple counts would be slightly over-
weighted, by including counts of both the referential language and of the full text of the claim(s) to which
those dependent claims refer). Some dependent claims, moreover, reference multiple independent or
dependent claims that may have different lengths, which makes it more difficult to provide a count that is an
accurate length for any such dependent claim. (Of course, each such multiply dependent claim could be
decomposed into separate claims for further analysis.)

49

Electronic copy available at: https://ssrn.com/abstract=2844964

our assumption on patent scope, if the change in independent claim length (ICL)
from publication to grant is positive, then it follows that the patent scope at grant
should (generally) be narrower than at publication (and filing). If the change in
independent claim count (ICC) is positive, then the scope of the patent should
(generally) be broader at grant than at publication (and filing).

Appendix C: Variable Codebook

claims_fulltext (Patents and PGPubs – 2 separate datasets) Dataset

Variable Name Variable Description Notes Dataset

appl_id Application identification number U.S. patent application PGPub only

number issued to an
applicant at filing.
claim_no Claim number Both

claim_txt Claim text This variable includes the Both

full text of each claim,
dependent or indepdent
dependencies Referenced claims to which the claim is This variable includes the Both
dependent all claim references
within the text of the
observed claim
ind_flg Indicator of independent claim See identifying algorithm Both
in Appendix 12.1.3 for
more information
pat_no Patent number U.S. patent number issued Patents only
to an applicant at grant.

claims_stats (Patents and PGPubs – 2 separate datasets) Dataset

Variable Name Variable Description Notes Dataset

appl_id Application identification number U.S. patent application PGPub Only

number issued to an
applicant at filing.
char_ct Count of characters in each claim Both

claim_no Claim number Both

cns_ct Count of “consisting” in each claim Both

cnx_flg Indicator of a canceled claim PGPub Only

ind_flg Indicator of independent claim See identifying algorithm Both

in Appendix 12.1.3 for
more information
or_ct Count of “or”s in each claim Both

pat_no Patent number U.S. patent number issued Patents Only

to an applicant at grant.
pub_no Pre-grant publication number PGPub Only

sf_ct Count of “selected from” in each claim Both

word_ct Count of words in each claim Both

50

Electronic copy available at: https://ssrn.com/abstract=2844964

 document_stats (Patents and PGPubs – 2 separate datasets) Dataset

Variable Name Variable Description Notes Dataset

appl_id Application identification number U.S. patent application Both

number issued to an
applicant at filing.
pat_no Patent number U.S. patent number issued Patents Only
to an applicant at grant.
pat_clm_ct Number of independent claims at grant Patents Only

pat_dep_clm_ct Number of dependent claims at grant Patents Only

pat_wrd_avg Average count of words among independent Patents Only

claims at grant

pat_dep_wrd_avg** Average count of words among dependent Patents Only

claims at grant

pat_wrd_ct Number of words among independent claims Patents Only

at grant
pat_dep_wrd_ct** Number of words among dependent claims Patents Only
at grant
pat_wrd_min Minimum count of words among independent Patents Only
claims at grant
pat_dep_wrd_min** Minimum count of words among dependent Patents Only
claims at grant
pub_clm_ct Number of independent claims at PGPub Only
publication
pub_dep_clm_ct Number of dependent claims at PGPub Only
publication
pub_wrd_avg Average count of words among independent PGPub Only
claims at publication
pub_dep_wrd_avg** Average count of words among dependent PGPub Only
claims at publication
pub_wrd_ct Number of words among independent claims PGPub Only
at publication
pub_dep_wrd_ct** Number of words among dependent claims PGPub Only
at publication
pub_wrd_min Minimum count of words among independent PGPub Only
claims at publication
pub_dep_wrd_min** Minimum count of words among dependent PGPub Only
claims at publication
pub_no Pre-grant publication number PGPub Only

** While document-level dependent claim word count statistics are included in our datasets, these statistics
are not accurate measures of claim scope. The dependent claim word counts do not add the word counts of
referenced claims on which the given dependent claims are dependent. For more information, please see footnote
73.

51

Electronic copy available at: https://ssrn.com/abstract=2844964