argue
h Ind
icate
place in that period. Furthermore, the distri
natur
tradit
the re
f mod
ritica
historians such asMacLeod (1988), O'Brien et al. (1995) andmore recently Nuvolari (2004) andMoser (2005) have suggested that
Explorations in Economic History 48 (2011) 97–115
Contents lists available at ScienceDirect
Explorations in Economic History
j ourna l homepage: www.e lsev ie r.com/ locate /eeha sizable share of inventive activities was undertaken outside the coverage of patent protection. Therefore, one should be
extremely cautious in gauging the dynamics of invention during the Industrial Revolution by looking at trends in patent counts.
Furthermore, patents differed greatly in their quality and, as a result, the use of simple patent counts for reconstructing the
patterns of technical change during this period may be unwarranted.
In this paper we examine the issue of the quality of patents during the Industrial Revolution. We provide a comprehensive
appraisal of the quality of all English patents granted in the period 1617–1841 using a historical source that so far has been
neglected. This source is the Reference Index of Patents of Invention, 1617–1852 edited by Bennet Woodcroft and published in 1855.initially restricted only to a handful o
Sullivan's findings are, however, c☆ We are particularly grateful to Christine MacLeod,
Allen, Patrick O'Brien, Giulio Bottazzi, Pietro Dindo, S
previous drafts. Davide Pirino has provided precious
seminars in Cambridge, London, Geneve, Strasbourg, a
Society Conference in Warwick for stimulating discus
⁎ Corresponding author.
E-mail addresses: alessandro.nuvolari@sssup.it (A.
0014-4983/$ – see front matter © 2010 Elsevier Inc.
doi:10.1016/j.eeh.2010.09.005ernized sectors (Crafts and Harley, 1992).
lly dependent on the reliability of patent counts as indicators of innovation. In fact,pointing to the relatively widespread
providing evidence in support of the “
technical change, while contradictingd that patent statistics can shed light on the ongoing debates on the timing and the
ustrial Revolution. The time series of English patents exhibits a significant structural
, at least according to Sullivan, a momentous acceleration of technical progress taking
bution of patents across industrial sectors displays a rather low level of concentration
e of inventive activities. Considered together these two findings may be regarded as
ional” interpretation of the Industrial Revolution as a phase of rapid and widespread
visionist view put forward by Crafts and Harley arguing for a more gradual dynamics,1. Introduction
Richard Sullivan (1989, 1990) has
nature of innovation during the Britis
break around 1760 and this would indBennet Woodcroft and the value of English patents, 1617–1841☆
Alessandro Nuvolari a,⁎, Valentina Tartari b
a Laboratory of Economics and Management (LEM), Sant'Anna School of Advanced Studies, Piazza Martiri della Liberta' 33, 56127, Pisa, Italy
b Innovation and Entrepreneurship Group, Imperial College Business School, Imperial College, South Kensington Campus, London, SW7 2AZ, United Kingdom
a r t i c l e i n f o a b s t r a c t
Article history:
Received 13 March 2009
Available online 9 November 2010
We examine the potentialities of a new indicator measuring the value of English patents in the
period 1617–1841. The indicator is based on the relative visibility of each individual patent in
the contemporary technical and legal literature as summarized in Bennet Woodcroft's
Reference Index of Patents of Invention. We conclude that the indicator provides a reasonable
proxy for the value of patents and that it can be usefully employed to shed light on the timing
and scope of innovation during the Industrial Revolution. In particular, our indicator offers a
suitable reconciliation between the patent records evidence and the Crafts–Harley view of the
Industrial Revolution.
© 2010 Elsevier Inc. All rights reserved.
Keywords:
Patents
Invention
Industrial Revolution
EnglandTim Leunig, Thomas Burr, Karl Gunnar Persson, Paola Criscuolo, Ken Lipartito, Jochen Streb, Jim Bessen, Bob
ean Bottomley, Marco Grazzi, Federico Tamagni and two anonymous referees for helpful suggestions on
advice in the implementation of some statistical exercises. We would like also to thank participants a
nd at sessions of the 2008 Social Science History Conference in Miami and of the 2009 Economic History
sions.
Nuvolari), v.tartari@imperial.ac.uk (V. Tartari).
All rights reserved.t
reconc
Indust
centur
scope
high q
produ
98 A. Nuvolari, V. Tartari / Explorations in Economic History 48 (2011) 97–115Industrial Revolution as a two-stage process, as suggested byMokyr (1999, pp. 20–23) andmore recently by Allen (2009, pp. 135–
155). The first stage, broadly coinciding with the “take-off” of the traditional chronology (1760–1800) is the phase when, in a
number of key-sectors, critical technical breakthroughs, or macroinventions in the sense of Mokyr (1990, pp. 13–14), such as
steam engines and textile machinery were invented. The second stage, corresponding to the period (1810–1840), may be seen as
the phase during which the potential of the macroinventions became fully realized by virtue of streams of microinventions that
greatly improved their performance. Obviously, it is in the second phase that we should expect to find a significant impact of
technical change on productivity growth.
2. Estimating patent quality
One of the well-known limitations of the use of simple patent counts as indicator of technical progress is the different quality of
the inventions covered by individual patents. This point is effectively made by O'Brien et al. (1996, p. 165):
In their quantitative work, cliometricians and economists are prone to aggregate recorded inventions into an index,
purporting to represent annual and cyclical variations in the volume of technological change within particular industries
or across national economies as a whole. Such an index would be extremely useful to historians, but, except for entirely
limited purposes, no such indicator can be constructed, since innovations recorded in patents and other documents are
unknown and potentially variable proportion of changes in the total flow of invention. Even recorded inventions cannot be
aggregated without some system of weighting to account for variations in their economic and technological significance.
So ideally, one would like to be able to assign each patent a weight reflecting its technological and economic significance.
Sullivan (1995), although acknowledging the issue of variations in patent quality, proposes that it may not be so severe in practice.
We should be aware that English patents until the reform of 1852 required the payment of a very expensive fee, well above the
average yearly household income. Thus, according to Sullivan, given the high costs of taking a patent, we could expect that
inventors carried out informed assessments of the economic value of the invention in question before patenting it. In other words,
we can imagine that the high patent fees acted as a filtering device, screening out inventions of particularly low quality.
In our view, Sullivan is too optimistic. We must remember that the English system was one of registration and not of
examination and this means that patents were not subjected to any check concerning their technological feasibility. Furthermore,
as noted by MacLeod (1988), in this period there were several heterodox uses of the patent system (e.g. using patents not for
protecting innovations, but as an advertisement or reputation device; this, for example, was a common practice in the medical
business). This clearly aggravates the problem of variations in patent quality. In this respect, detailed examinations of the contents
of patents for specific industries may provide us with important insights. An exercise along these lines has been recently carried
out by MacLeod et al. (2003). They examined a sample of 2010 British patents in steam engineering in the period 1800–1900 and
found that 365 of these patents (corresponding to a sizable 18.1%) were granted to “perpetual motion” machines or other
inventions which were not technically feasible. Notably, 217 of these impossible patents were granted in the period 1860–1900,
that is well after the formulation of the principles of classic thermodynamics by Clausius and Kelvin in the early 1850s, which
scientifically proved the impossibility of a perpetual motion engine.ntrated than the distribution of total patents. In this way, our proposed indicator of patent quality seems to offer a way of
iling the patent evidence with the “revisionist view” put forward by Crafts and Harley (1992). Concerning the timing of the
rial Revolution, our findings seem in line with the traditional chronology, confirming that the second half of the eighteenth
y (1762–1801) was characterized by a clustering of critical technical breakthroughs (high quality patents). In terms of the
of the change, our findings indicate that, although patents were relatively widespread across industries, patents of relatively
uality were localized in a more restricted number of sectors. Our findings can be reconciled with the dynamics of
ctivity growth posited in the Crafts–Harley view (Crafts and Harley, 1992) by interpreting the pattern of innovation of theFor each patent, the Reference Index volume provides a list of references (either to technical and engineering literature or to legal
proceedings and commentaries) where the patent specification is mentioned. Our basic assumption is that the relative “visibility”
of each patent inWoodcroft's Reference Index provides a reasonable proxy for its relative technical and economic significance (only
patents of non-trivial economic value are likely to be extensively discussed in the technical literature or at the centre of litigations).
This approach is analogous to the use of patent citations as measures of the value of patents in the contemporary literature (Jaffe
and Trajtenberg, 2002).
On the basis of Woodcroft's Reference Index, we assign a quality score to each patent in our period of interest. The first step of
our study is to assess the reliability of our new indicator of patent quality. We establish that patents classified as radical
innovations by Baker (1976) and patents taken by inventors listed in the 2004 edition of the Oxford Dictionary of National
Biography and in another selection of “great inventors” recently compiled by Allen (2009) are systematically characterized by high
quality scores (also after controlling for other patent characteristics). This result, in our interpretation provides significant
corroboration of the general reliability of our indicator of patent quality. Furthermore, the distribution of the quality scores both at
the aggregate level and at the level of individual industries is highly skewed and similar to those characteristic of modern patent
data (Jaffe and Trajtenberg, 2002).
The second step of our analysis is to examine the distribution of high quality patents both over time and across industrial
sectors. We find that, in comparison with the distribution of total patents over time, the distribution of high quality patents is
clustered on an earlier period. Additionally, the distribution of high quality patents across industries is significantly more
conce
However, recent developments in the economics of innovation suggest that the “nihilist” view of O'Brien, Griffiths and Hunt
may not be completely warranted and that the problem of variations in the value of patents can be tackled rather effectively by
constructing indicators of quality using patent characteristics that are likely to be positively correlated with the economic value of
patents.
The first example of this approach to the measurement of the value of patents is the use of renewal data pioneered by
Schankerman and Pakes (Schankerman and Pakes, 1986; see also Bessen, 2008 for a more recent contribution in this vein). Most
modern patent systems require patent owners to pay a renewal fee at fixed time intervals in order to keep the patent in force. In
fact, only few patents are kept in force for their full lifetime. Thus, it is possible to estimate the value of an individual patent by
considering the amount of renewal fees that the owner of the patent has paid to keep the patent “alive”. The underlying
assumption is that a patent owner will pay the renewal fee for a given period only if he expects that this payment will be lower
than the discounted streams of profits generated by the patent. Sullivan (1994) has applied this method to historical patent data
by providing estimates of the value of British patents in the period 1852–1876.1 The approach, however, cannot be used for the
English patent system in the period 1617–1852 because the system did not impose the payment of renewal fees after the granting
of the patent.
The second approach pioneered by Trajtenberg (1990) is the assessment of the economic value of patents using the number of
99A. Nuvolari, V. Tartari / Explorations in Economic History 48 (2011) 97–115citations received. The intuition is relatively straightforward: when a patent has received many citations, this means that it
contains knowledge that was used in a large number of subsequent technological developments. The actual existence of a positive
correlation between citations received and the economic value of patents was documented by Trajtenberg (1990) for the case of
US patents in computed tomography and it has been subsequently confirmed in a number of empirical studies both for US and
European patents (see Sampat and Ziedonis, 2004 and van Zeebroeck, in press for useful surveys). Another indicator that is gaining
popularity is the use of information on the filing of legal oppositions or on patent litigations (Harhoff et al., 2003; Allison et al.,
2004). The rationale underlying the use of this information is the idea that oppositions and legal disputes will tend to revolve
around patents of higher economic value.2
In this paper, we study the feasibility of employing an approach similar to the contemporary practice of constructing citation-
based indicators of patent quality to the case of the English patent system in the Industrial Revolution period. The English patent
system of the time did not prescribe the use of citations to previous patents for defining prior art. Therefore, we should look for an
alternative historical source suitable of being used for constructing a plausible proxy for the value of patents.
3. Bennet Woodcroft and the Reference Index
Before the reform of 1852, a patent application could be lodged in anyone of these three Public Offices in London: Rolls Chapel
Office, Petty Bag Office and Enrolment Office. In this way patent specifications were dispersed in three different locations.
Furthermore, the system also lacked an effective search catalogue providing easy access to the specifications of existing patents.
This was seen as an important problem: for an inventor was almost impossible to have a clear picture of the state of art covered by
existing patents (Gomme, 1946; Hewish, 2000).
From the early 1830s, several patent agents had begun to construct lists and indexes of existing patents. With the reform of the
patent system in 1852, the new Patent Office Commissioners decided to address this problem by funding a major publication of
indexes and abridgments of the patents granted from 1617 to 1852. The Commissioners entrusted this task to Bennet Woodcroft,
who had already been working on his own at the construction of patent indexes for specific industries such as steam navigation
and textile machinery.3 Woodcroft was probably the optimal choice for this task. He was energetic and his efforts were sustained
by a strong belief in the beneficial role of patents not only as a system of incentives for innovation, but also as a powerful
“information system” for engineers (and also for historians).4
Woodcroft and his team of clerks undertook the construction of the system of indexes following a straightforward approach.
Each patent was assigned a progressive number (on the basis of its date). The first volume published was a Chronological Index,
followed by an Alphabetical Index and this in turn was followed by a Subject Index. These three indexes provided an indispensable
1 The findings of MacLeod et al. (2003) suggest a cautionary attitude towards the use of renewal data. In their study they find that many potentially valuable
steam engineering patents were not renewed (this was most probably due to the limited financial resources of many patent holders). Vice versa, even some
technically impossible inventions were kept in force for the full patent duration.
2 A third approach in this stream of literature is the estimation of patent values by means of econometric models linking the market value of firms to their
patent portfolio (see Bessen, 2009 for a recent example). Also this approach is not feasible for the period of the first industrial devolution because the large
majority of example were owned by individuals.
3 Bennet Woodcroft (1803–1879) was himself a talented inventor, who took several patents (at least two of major technical importance). During his life, he
enjoyed friendships with some of the most important engineers of the time such as J. Whithworth, J. Nasmyth and R. Roberts. In 1843 he opened in London an
office as patent agent and consulting engineer. In 1847 he was appointed professor of machinery at University College. In 1852 with the passing of the Patent Law
Amendment Act, Woodcroft was appointed assistant to the commissioners. He was in charge of the publication of all the specifications of patents for the period
1617–1852 together with the relative series of indexes. On Woodcroft's life and achievements, see Hewish (1982) and Harrison (2006, pp. 55–66).
4 In 1851 Woodcroft in front of the Select Committee of the House of Lords on patent laws insisted on the advantages of implementing an effective index
system and of printing the full patent specifications: “Anyone who had the ambition to become the historian of inventions, could not do better than take such a
work on patents, because he would there not only find the true course of inventions, but he would also find every futile effort made in that direction…..It would
be the most valuable encyclopaedia of invention ever published” (House of Lords, 1851, p. 403). For an account of the publication of Woodcroft's indexes, against
the background of contemporary debates on the reform of the patent system, see MacLeod (2007, pp. 251-264).
Table 3 contains a list of all publications that were referenced more than 10 times in Woodcroft's Reference Index over the
period 1617–1841. In order to illustrate the changing coverage of different publications over time, the columns of the table show
the number of references for different subperiods, each of these covering a time interval during which 2000 patents
chronologically arranged, were granted. For example the first column contains the number of references for patents from number
1 to number 2000 (granted over the period 1617–1794). The last column gives the total number of reference throughout the entire
period 1617–1841. Overall, the publications used in the compilation ofWoodcroft's Reference Index can be classified in three broad
categories: i) publications reporting latest developments in science and technology (in particular those embodied in patents
recently granted), ii) engineering journals and books containing discussion of merits and limitations of specific technical solutions
iii) legal commentaries on patent laws and cases. We should note that this classification gives just a preliminary orientation to the
contents of Woodcroft's Reference Index and that in several cases a specific publication may be not straightforwardly classified in
one of the three categories. The first category contains specialized journals edited by patent agents that published regularly
selections of patent specifications.8 This specialized literature represented an important channel of information fuelling the
emergence of the market for patented inventions identified by Dutton (1984) in the first half of the nineteenth century. The first
important publication of this kind was the Repertory of Arts and Manufactures, first published in 1794, whose aim was to establish
“a vehicle, bymeans of which new discoveries and improvements in Arts andManufactures, may be transmitted to the public”. The
5 There first edition of the Reference Index was published in 1855. A second edition based on a slightly more extensive number of references was published in
1862. In this paper we use this second edition.
6 The publication of these indexes was followed by a further attempt to summarize and classify by subject all the existing patent specifications by publishing a
series of volumes Abridgments of Patent Specifications. Each of these volumes contained a succinct description of all the patent specifications pertaining to specific
technological subject.
7 On Symington's improved Newcomen engine design, see Harvey and Downs-Rose (1974).
8 See Harrison (2006, pp. 224–226) for an overview of the publishing activities of some early patent agents.
Table 1
Entry in Woodcroft's Reference Index for James Watt's patent of the separate condenser.
Patent number Reference
913 Repertory of Arts, vol I, p. 217
Mechanics Magazine, vol I, p. 4
Practical Mechanics' Journal, vol I, p. 285
Register of Arts and Sciences, vol IV, p. 4, etc.
Engineers' and Mechanics' Encyclopaedia, vol 2, p. 725
100 A. Nuvolari, V. Tartari / Explorations in Economic History 48 (2011) 97–115,
,orientation in the field for would-be patentees. The usefulness of these sources is also confirmed by the very intense use that
historians of technology have done of this material. The set of indexes was completed in 1855 by the publication of the Reference
Index (Hewish, 2000, p. 35–36). The index is structured in chronological/numerical order and for each patent it reports the office of
enrolment where the specification was filed. Additionally, for each patent, the index gives a list of references providing
information on the patent in question. These references comprisementions in technical journals and books, law commentaries and
reports, Record Office reports and other official publications such as Parliamentary Select Committees.5 Remarkably, this source so
far has received very little attention by historians (to the best of our knowledge the index so far has only been employed by Dutton
for examining the outcome of a number of legal disputes over patents, 1984, pp. 78–79).6
A typical entry of Woodcroft's Reference Index is represented in Table 1. The patent in question is the one granted in 1769 to
James Watt for the separate condenser. The entry gives references to the technical and legal literature where the patent is
mentioned, while the last line of the table indicates in which office the specification was lodged (in this case Rolls Chapel). Table 2
provides the example of another entry. This is for a patent covering an improvement in the Newcomen engine developed by
William Symington. This was surely a valuable invention, but whose economic and technological significance was relativelyminor
in comparison to Watt's separate condenser.7 For this patent, as one would have expected, the Reference Index contains a much
lower number of references.
Webster's Reports, vol I, p. 31, etc.
Webster's Patent Law, p. 46, etc.
Webster's Letter Patent, p. 6, etc.
Blackstone's Reports, vol II, 463
Carpmael's Report on Patent Cases, vol I, p. 117, etc.
Davies on Patents, p. 155, etc.
Collier's Law on Patents, p. 71, etc.
Parliamentary Report, 1829, p. 187, etc.
Vesey, junr.' S Reports, vol III, p. 140
Holroyd on Patents, p. 35, etc.
Durnford and East Term Reports, vol VIII, p. 95
Patentee's Manual, p.8
Billing on Patents, p. 20, etc.
Rolls Chapel Reports, 6th Report, p. 160
Extended by Act of Parliament for 25 years
Rolls Chapel
editors noted that a selection of “specifications of patents will form a considerable, and, it is presumed, interesting part of this
work.” (Anon., 1794, pp. i–ii). In the 1820s two noteworthy new journals that published regularly selections of paten
specifications were launched: the London Journal of Arts and Sciences edited by William Newton, whose declared goal was to
publish the “earliest information relative to every useful discovery and invention in practical mechanics, as well as such other
novel inventions as are applicable to the arts, manufactures and agriculture” (Newton, 1820, p. i) and the Register of Arts and
Sciences first issued in 1824, that had similar editorial scope. Finally, The Inventors' Advocate and Patentee Recorder first issued in
1839 aimed to become “an efficient medium of communication between inventors, patentees, capitalists and the public at large”
In terms of contents, the journal set itself the task to give its readers “earliest, and exclusive information with matters connected
with science and art, discoveries and inventions”. The journal would publish the full specification of the “most important” patents
(Anon., 1839, p. 1). The second category contains engineering journals and books that did not limit themselves to summarize the
contents of patents, but discussed in more depth the merits and limitations of the technical solutions contained in some patented
inventions. For example, Luke Hebert, a famous patent agent, with the The Engineer's and Mechanic's Encyclopedia intended to offer
a “judicious selection of all those machines, engines, manipulations, processes and discoveries, that now lie scattered throughou
several hundreds of volumes of the scientific journals or are inscribed in obsolete characters upon the rolls of the Court of Chancery
in the form of specifications of patent inventions” (Hebert, 1836, p. ii). Mechanics' Magazine and The Artizan can be regarded as
journals that falls in this category. The aim of Mechanics' Magazine was “to promote a better acquaintance with the history and
principles of the arts” together with “earlier information…of new discoveries, inventions and improvements” (Anon., 1823)
Technical treatises covering specific technology fields such as Ure's CottonManufacture and Stuart'sHistory of the Steam Engine also
belong to this second category. The third category comprises publications that were clearly more aimed to be digests of paten
cases, such as the famous patent treatises by Carpmael, Holroyd and Webster. However, even in these publications, lega
Table 2
Entry in Woodcroft's Reference Index for William Symington's patent of an improved
Newcomen engine design.
Patent number Reference
2544 Mechanics' Magazine, vol XVII, p. 385, etc.
Rolls Chapel Reports, 6th Report, p. 151
Rolls Chapel
Holroyd's on Patents (a Treatise) 12 15 4 0 0 31
Jurist (Reports) 0 1 4 15 11 31
Law Times (Reports) 0 2 3 12 12 29
Davies on Patents (Reports) 16 9 0 0 0 25
Practical Mechanics' Journal (Glasgow, 1848) 2 0 2 8 14 26
Stuart's History of the Steam Engine (London, 1825) 7 7 4 0 0 18
Moore's Privy Council Cases (Reports) 0 0 4 9 3 16
Transactions of the Society of Arts 0 0 0 5 10 15
Meeson and Welsby's (Reports) 0 1 5 5 2 13
Ure's Philosophy of Manufactures (London, 1835) 0 0 4 7 0 11
101A. Nuvolari, V. Tartari / Explorations in Economic History 48 (2011) 97–115Table 3
Publications with most references in Woodcroft's Reference Index.
Publications 1–2000
[1617–1794]
2001–4000
[1794–1816]
4001–6000
[1816–1830]
6001–8000
[1830–1839]
8001–9210
[1839–1841]
Total
London Journal of Arts and Sciences (Newton's; London, 1820) 0 2 1330 1290 463 3085
Repertory of Arts and Manufactures (5th series, London, 1794) 169 931 1124 833 335 3392
Rolls Chapel Reports, 6th, 7th and 8th 738 968 360 245 0 2311
Mechanic's Magazine (London, 1823) 32 22 194 249 641 1138
Inventors' Advocate and Patentees' Recorder (London, 1839) 0 0 0 38 901 939
Register of Arts and Sciences (2nd series, London, 1824) 23 35 602 212 1 873
Engineers' and Mechanics' Encyclopaedia (by Luke Hebert,
London, 1836)
30 44 315 102 0 491
Carpmael's Patent Cases (Reports) 25 36 44 25 1 131
Webster's (Reports) 22 20 44 41 5 132
Webster's Patent Law (a Treatise) 22 28 32 23 1 106
Billing on Patents (a Treatise) 13 18 36 28 5 100
Engineers' and Architects' Journal (London, 1837) 1 2 6 45 100 154
Law Journal (Reports) 0 2 13 27 18 60
Parliamentary, 1829 Patent Law (Reports) 15 26 17 0 0 58
Artizan. A Monthly Journal of Operative Arts (London, 1843) 2 4 7 17 17 47
Patentees Manual (by Henry Johnson, London, 1853) 6 10 12 7 2 37
Ure's Cotton Manufacture (London, 1836) 6 6 23 10 0 45
Websters Letters Patent (London, 1848) 6 9 14 5 0 34
Patent Journal and Inventor's Magazine (London, 1846) 0 0 5 24 12 41t
.
t
.
t
l
interpretation, whereas renewals provide a direct assessment of patent values in monetary terms.With respect to patent citations
the number of references has the disadvantage of being a composite indicator of the visibility of a patent in the contemporary
specialized literature (i.e., publications can mention the same patent for different reasons). Instead patent citations are the
outcome of a fairly “regulated” search process aimed at defining the prior art of the patented invention. However, in comparison to
modern patent citations, the number of references in the Reference Index has the advantage of being the product of a relatively
homogenous source (Woodcroft and his team of clerks), whereasmodern patent citations are generated by heterogeneous sources
(inventors, patent attorneys and patent examiners). Furthermore, it is also increasingly recognized that citing behavior in modern
patent systems is affected by strategic considerations, e.g. an inventor may be reluctant to cite a patent that may disrupt some
novelty claims. This problem, instead, is not present in the case of the Reference Index. In fact, it is worth noting that the nature of
the references of the Reference Index is more similar to that of citations in modern scientific literature than to contemporary patent
citations.11 In this sense, our approach is akin to bibliometric studies providing assessments of the importance of a scientific paper
on the basis of the citations received in the subsequent literature.
9 A similar exercise has been carried out by Sullivan (1989, pp. 431–433). His quality indicator is simply the number of different classes in which a patent is
listed in Woodcroft's Subject Index. This would be analogous to the count of patent classes in contemporary patents, so in our view, it should be properly
considered a measure of generality rather than of quality (Jaffe and Trajtenberg, 2002).
10 The Reference Index volume was prepared in the early 1850s. This means that Woodcroft and his team of clerks, due to lack of hindsight, may have faced more
difficulties in preparing accurate and complete list of references for the most recent patents. In order to minimize this problem, in this paper we restrict our
analysis to the period 1617–1841. This means that each patent in our sample can at least enjoy a period of ten years for becoming “fully visible” in the technica
and legal literature.
11 On strategic considerations affecting citing behavior in patents, see Lampe (2010) and for cautionary considerations on the use of patent citations as
indicators of patent value, see Bessen (2008)).
250
300
350
400
450
500
102 A. Nuvolari, V. Tartari / Explorations in Economic History 48 (2011) 97–115,
lconsiderations were often interwoven with technical discussions. Overall, Table 3 indicates that the bulk of the publications used
in the compilation of the Reference Index is largely of technical nature (either specialized journals publishing systematically
selections of patent specifications or more elaborated technical commentaries of specific patents). The extension of this publishing
activity suggests that, even if before 1852 patent specifications were of difficult access because dispersed in three different offices,
a considerable amount of the technical information embodied in patent specifications was actually placed in the public domain by
virtue of the growth of this specialized literature that reported and discussed the contents of patents (Mokyr, 2009, p. 409; Moser,
2010).
We suggest that the number of references listed inWoodcroft's Reference Index provides a good indication of the “visibility” of a
specific patent in the contemporary technical and legal literature. In this paper we study the feasibility of constructing an index of
the economic value of patents based on the number of references listed in Woodcroft's Reference Index.9 Our assumption is that
patents which are more significant from a technical point of view will tend to be cited more often in the technical literature.
Furthermore, we also assume that patents with high economic importance will be more likely to become the subject of legal
controversies. Thus, we propose that the number of references listed in Woodcroft's Reference Index can serve as a reasonable
proxy of the economic value or “quality” of the patent.10 This approach is obviously analogous to the use of patent citations as
proxy for the economic value of patents adopted in the modern literature on innovation. On reflection, there are two main
limitations of the Reference Index as indicator of patent value when compared to patent renewals and citations. With respect to
renewals, the number of references has the disadvantage of not being a metric susceptible of a straightforward economic
0
50
100
150
200
1617 1637 1657 1677 1697 1717 1737 1777 1797 1817 1837
Patents
1757
Fig. 1. Number of English patents granted per year.
4. The construction of the patent quality indicator
Our approach is to assign to each patent a quality score that is equal to the number of references listed inWoodcroft's Reference
Index. In our sample, this indicator has a lower bound of 0 (patents with no references and for which the index contains only
information concerning the public office in which the specification was lodged) . We will refer to this indicator as Woodcrof
Reference Index (WRI).12
Fig. 1 shows the annual number of granted patents over the period 1617–1841. The gap in the series corresponds to the period
of the civil war, Commonwealth and the Protectorate (1641–1660) when the patent system was practically dismantled and no
patents were granted (MacLeod, 1988, p. 16). Fig. 2 displays the yearly average number of references per patent in Woodcroft's
Reference Index as a thin line. Fig. 2 also shows as a thick line the average number of references for the following subperiods: 1617–
1701, 1702–1721, 1722–1741, 1742–1761, 1762–1781, 1782–1801, 1802–1811, 1812–1821, 1822–1831, 1832–1841 and as a
dotted thick line the yearly average number of references smoothed using the Hodrick–Prescott filter.13 It is quite clear that the
time series of the yearly average number of references shows a cyclical behavior around an upward trend, revealing a propensity o
more recent patents to be mentioned in a higher number of references. This increase in the number of references reflects both the
sustained expansion of the English literature on science and technology taking place during the eighteenth and early nineteenth
century (Mokyr, 2009, pp 46–48) and the growth of the specialized literature on patents, also related to the growing public
awareness of the working of the patent system (MacLeod, 1988, pp 146–147). Clearly, if one were to use simply the number o
references as indicator of patent quality when comparing patents granted in different years, he could obtain results that are
possibly biased by the variations over time in the number of references per patent.14 This type of problem is indeed present in
12 Thus, going back to the examples of Tables 1 and 2, patent 913 is assigned a WRI score of 20 and patent 2544 a WRI score of 2.
13 We have used a parameter of 6.25 for the Hodrick–Prescott filter as suggested by Ravn and Uhlig (2002).
14 The growth in the average number references per patent is mostly accounted for the increasing number of specialized periodicals reporting and commenting
the specifications of selections of contemporary patents.
3
3.5
4
Woodcroft
(raw)
Adjusted Woodcroft
(yearly mean)
Adjusted Woodcroft
(mean subperiods)
Adjusted Woodcroft
(HP filter)
Adjusted Woodcroft
(Goldsmith Kress Library)
Woodcroft (raw) 1
Adjusted Woodcroft
(yearly mean)
0.8256 1
Adjusted Woodcroft
(mean subperiods)
0.8748 0.9384 1
Adjusted Woodcroft (HP filter) 0.8722 0.9717 0.9782 1
Adjusted Woodcroft
(Goldsmith Kress Library)
0.8266 0.8285 0.8966 0.8803 1
All coefficients significant at 1%.
103A. Nuvolari, V. Tartari / Explorations in Economic History 48 (2011) 97–115t
f
f0
0.5
1
1.5
2
2.5
1617 1637 1657 1677 1697 1717 1737 1757 1777 1797 1817 1837
References per patent Mean (subperiods) HP filter
Fig. 2. Average number of references per patent (yearly and subperiods).
Table 4
Spearman rank correlation matrix of different fixed-effects adjustments.
Khan and Sokoloff (1993) havefirst used inclusion in biographical dictionaries as amethod for identifying “great inventors” (i.e.
those responsible for the most historically significant inventions) in the US case. More recently Khan and Sokoloff (2008) have
carried out a comparative study of American and British “great inventors” using the same method. Khan and Sokoloff (2008) have
constructed their British sample of “great inventors” using the 2004 edition of the Dictionary of National Biography (DNB). Here we
will follow this type of approach and consider patents awarded to inventors included in the Dictionary of National Biography as of
particular historical significance. In our period of interest, we have been able to retrieve 218 patenteeswhose biographical profile is
included in the DNB. These inventors were responsible for 723 patents. This is our first list of “important patents”.
15 We have retrieved the number of publications using the search catalogue of the digital edition of the Goldsmiths'-Kress Library published by Gale publishing
Table 5
Overlap between DNB, Allen and Baker patents.
104 A. Nuvolari, V. Tartari / Explorations in Economic History 48 (2011) 97–115,
.modern patent data. Also in this case the propensity to cite other patents is not constant over time. For example, the
computerization of patent databases during the 1980s enhanced the search of prior art for inventors and patent examiners leading
to an increase in the average number of citations per patent (Hall et al., 2002, p. 418–419). A common way to solve this issue is to
divide the citations received by a given patent by the mean of citations received by all patents belonging to the same time cohort.
This means using as a benchmark citation intensity for assessing the quality of an individual patent, the average number of
citations received by patents of the same time cohort. This procedure is usually referred to as “fixed effects” approach (Hall et al.,
2002, pp 437–441). Here we adopt the same type of adjustment by dividing the number of references of each individual patent for
the average number of references received by all patents in the time cohorts corresponding to the fixed subperiods used for the
drawing the thick line of Fig. 2. This is our adjusted indicator of the economic value of patents, which we shall call adjusted
Woodcroft Reference Index (WRI*).
As a first step, it is important to check how sensitive the quality indicator is to the choice of the time intervals employed for
benchmarking the reference intensity. Table 4 reports the Spearman rank correlationsmatrix between patentswhose quality has been
measuredusingdifferent typesoffixed-effects adjustments. Thefirst column(“Woodcroft raw”) refers simply topatentswhosequality
was measured by the “raw” number of references without any type of adjustment, the second column refers to patents whose quality
was measured using as benchmark the yearly average number of references, the third column refers to patents whose quality was
measured using as benchmark the average number of references for the fixed subperiods of Fig. 2, the fourth column refers to patents
whosequalitywasmeasuredusingasbenchmark the timeseries of theyearly average number of patents smoothedusing theHodrick–
Prescott filter, the fifth column refers to patents whose quality has been measured by dividing their number of references by the
number of publications retrieved in theGoldsmiths'-Kress Libraryof Economic Literature using thekeyword “patent” in thepublication
years corresponding to the fixed subperiods of Fig. 2.15 The Goldsmiths'-Kress Library is one of the largest collections of publications in
English covering the period 1450–1850. It is important to note that, notwithstanding its title, the collection is not restricted only to
economic subjects, but it contains a large amount of engineering publications and legal treatises (includingmost publications listed in
Table 3). Therefore, the aim of this last column is to examine the effects of a benchmark that does not reflect only the variations over
time of the references inWoodcroft's index, but instead captures themore general trends in the broad literature (books and journals)
related topatents. Table 4 shows that all indicators of patentquality calculatedusingmore sophisticated adjustmentprocedures appear
strongly correlated with each other. Furthermore, even the simple “raw” counting of the number of references is strongly correlated
with the indicators calculated using the more sophisticated adjustment procedures. The main reason for this result is that historically
significant patents (such as James Watt's patent for the separate condenser) have, in comparison to all other patents, a significantly
higher number of references. For this reason, even without fixed-effects adjustment, they will receive a high quality score even when
compared to patents belonging to later time cohorts characterized by higher average reference numbers. Overall, the results of Table 4
indicate that the use of different approaches for benchmarking the reference intensity is not likely to havemajor effects on the relative
assessment of patents (in the sense that the different approaches will tend to single out highly overlapping groups of patents as those
with ahighquality). In the rest of this paperwewill use the time cohorts corresponding to thefixed subperiodsof Fig. 2 to construct our
quality indicator adjustment using the fixed-effects procedure. Table 4 also suggests that theuse of a quality indicator based number of
references is better suited for the identification of groups of high quality patents, rather than as a precise index for measuring the
quality of individual patents. Accordingly, in this paper we will employ the quality indicator to study the characteristics of patents
belonging to different quality percentiles (limiting in this way the impact of measurement errors at the level of individual patents).
5. Assessing the reliability of the patent quality indicator
To substantiate the reliability of the indicator of patent quality based on Woodcroft's Reference Index we need to compare it
with some independent measure of patent quality. To perform this task, wewill construct four lists of “important patents” that can
be used to validate our patent quality indicator.
DNB Allen “Great Inventor” Allen “Macro Inventor” Baker
DNB 723 104 26 55
Allen “Great Inventor” 133 27 32
Allen “Macro Inventor” 27 12
Baker 137
recognized as a fundamental aspect for the legal validity of the patent (MacLeod, 1988, p. 49). The final row considers all patents
excluding those subjected to trials in court.17 Since patent lawsuits were very costly (MacLeod, 1988, pp. 58–74), it seems unlikely
that patents of minor economic importance were subjected to extensive litigation. Hence, references discussing legal issues may
be expected to be related to the economic value of patents. In this sense, it seems appropriate to have an indicator that combines
references to technical and legal literature (also because in several cases it is difficult to classify precisely the nature of the source
used in the index). On the other hand, patents subjected to court trials can attract references because the lawsuit in question
represented an important legal precedent, rather than by virtue of their technological significance, so by removing these patents
from the sample we may expect to have a more restricted set of patents in which the distribution of references is reflecting, to a
major degree, the genuine technological content of the patents, rather than legal issues. Therefore it is surely important to check
16 As a robustness check we have also carried out the Mann–Whitney–Wilcoxon test, which is the traditional test to assess the equality of medians between two
samples, obtaining results that are fully consistent with those of Table 6.
17 A list of patents subjected to court trials is provided in Woodcroft (1862, pp. 669–710).
Table 6
Fligner–Policello tests of stochastic equality.
DNB Baker Allen “Great Inventor” Allen “Macro Inventor”
105A. Nuvolari, V. Tartari / Explorations in Economic History 48 (2011) 97–115Allen (2009, pp. 242–271) has also recently constructed two samples of British “great inventors” with a specific focus on the
economic significance of the inventions they produced. Allen's first list of “great inventors” has been constructed by considering all
the inventors active in Britain between 1660 and 1800 mentioned in Singer et al. (1957, 1958). This sample has been integrated
also considering Mokyr (1990) and Mantoux (1928). This list contains 54 inventors that were responsible for 133 patents in total.
Allen's second list of inventors is more restricted and contains 10 “superstar” inventors that can be regarded as responsible for
genuine “macroinventions” in the sense of Mokyr (1990). These “macro inventors” were responsible for 27 patents. Allen's lists of
“great” and “macro” inventors are the sources of our second and third list of “important patents”.
Another interesting source for the identification of important patents is Baker (1976). Baker's list is meant to include the “most
important” patents granted in Britain over the period 1691–1971. An initial selection was originally compiled by the staff of the
enquiry desk of the British patent office in the early 1970s. This selection was extended by Baker through an extensive search in the
technical and historical literature (Baker, 1976, pp. 7–25). Baker's list of important patents has been employed by Kleinknecht (1987)
and Silverberg and Verspagen (2003) for testing the Schumpeterian hypothesis of the existence of temporal clusterings of radical
innovations. In our period of interest, we have 137 patents belonging to the Baker list. This is our fourth list of “important patents”.
Table 5 shows that there is an imperfect overlap between the lists of patents identified using theDNB, Allen's two lists of inventors
and the Baker list. In Table 5, the diagonal cells contain the total number of patents in each of these lists. Cells outside the diagonal
instead contain the number of patents that are included simultaneously in two lists. The most significant differences are clearly
between the Baker list and the other great inventor lists (both DNB and Allen's). In particular, respectively 60% (i.e., 82 patents) and
77% (i.e., 105 patents) of the patents in the Baker list are not included in theDNB list and the Allen's lists. In this perspective, these lists
of “important patents” seems to reflect a number of relatively independent assessments of the historical significance of inventions and,
for this reason, they can be an interesting yardstick for gauging the reliability of our indicator of patent quality.
Our first step is to check whether there are significant differences in the quality scores between patents included in our four
lists of “important patents” and the rest of the sample. This is done by performing the non-parametric test of stochastic equality
suggested by Fligner and Policello (1981), which is particularly suited for our case in which wemust compare samples of different
numerosity possibly characterized by non normal distributions of unknown shape. Given two random variables X and Y, the
Fligner–Policello statistic determines whether the Prob [XNY]N0.5. In other words, the Fligner–Policello statistic reveals whether
by randomly selecting two patents, one from the “important patents” list and one from the rest of the sample, the probability that
the patent from the “important patents” list has a higher quality score measured using WRI* is higher than 0.5.16
Table 6 contains the results of these Fligner–Policello tests of stochastic equality for the four different lists of “important
patents”. The first row of Table 6 reports the results for the period 1617–1841. We perform these tests of stochastic equality also
for subsets of our total patent sample. The second row of Table 6 considers only patents granted after 1741. In this way, wewant to
remove from the sample the erratic procedures of the early patent system. In particular, in the early period, patentees were asked
only to provide a cursory description of the invention. The filing of a complete written description of the invention (specification),
became established practice only in the 1730s (MacLeod, 1988, pp. 48–49). The third row considers only patents granted after
1781. With this third subset we would like to control for the effect of Liardet vs Johnson (1778) decision that established that the
specification should enable anyone skilled in the art to construct the invention. After this decision the specification was definitely
Entire sample 1617–1841
Fligner–Policello statistic 8.416*** 6.892*** 7.573*** 9.141***
Year N1741
Fligner–Policello statistic 7.787*** 6.877*** 10.542*** 12.469***
Year N1781
Fligner–Policello statistic 6.549*** 5.858*** 10.054*** 9.408***
(removing patent cases)
Fligner–Policello statistic 7.906*** 5.307*** 6.842*** 7.941***
*,**,*** indicate significance levels of 10%, 5%, 1%.
higher odds of being in the highest percentiles of the distribution of our quality indicator, WRI*. In particular, we consider three
cases: patents with values that are in the top 50%, 10% and 1% percentiles of the distribution ofWRI*.18 Also in this casewe perform
this exercise for different selections of the patent sample. The results are displayed in Table 7. The table shows in all cases odds-
ratios that are greater than 1 (an odds-ratio of 1 indicates that the odds of a patent of being in the high percentiles quality group is
the same for patents in the “important patents” lists and in the rest of the sample) and all significant at the 1% level. This means
that patents that are in the lists of important patents have consistently a much higher probability to appear in the three outcome
groups (top 50%, top 10% and top 1%) than patents that are not included in the lists. Note that the value of the odds-ratios are
higher when we consider as outcome the inclusion of the patent in the top 1% percentile of the quality distribution.
The positive relationship between “important patents” and the quality indicator is also obtained through multivariate analysis
that controls for the possible influence of other factors. In this case, our dependent variable is the number of references listed for
each patent in Woodcroft's Reference Index. As we have mentioned, WRI is an integer number that can take values between 0 and
the maximum number of references. Thus, the appropriate estimation technique is a negative binomial regression. Our covariates
are the following:
i) Dummy variables indicating whether the patentee is a DNB inventor, Allen “Great Inventor”, Allen “Macro Inventor” or the
patent is in the Baker list.
ii) Engineer: a dummy variable indicating whether the occupation of at least one of the patentees is related with engineering
type of trades.
iii) Number of inventors: a variable indicating the number of inventors.
iv) Patent experience: a dummy variable indicating whether at least one of the patentees had already been granted at least a
patent before the one in question.
v) Foreign communication: a dummy variable indicating whether the patent is the outcome of a communication from abroad
18 See Agresti (2002, pp. 36–104) for an introduction to the use of odds-ratios in case–control data.
Table 7
Odds-Ratios of “important patents” of being in high WRI* percentiles.
Treatment
DNB Baker Allen “Great Inventor” Allen “Macro Inventor”
106 A. Nuvolari, V. Tartari / Explorations in Economic History 48 (2011) 97–115.whether the relationship between the quality indicator and the “important patents” is confirmed also for this restricted set of
patents depurated by lawsuits.
The results presented in Table 6 provide a first validation for the construction of quality weights based on the Reference Index. In
all cases, the hypothesis of stochastic equality is rejected at a significance level of 1%, indicating that the patents belonging to the
list of important patents have a higher probability of assuming higher quality scores than patent in the rest of the sample.
The second approach we adopt in order to examine the relationship between “important patents” and the quality index is to
consider our lists of “important patents” as a “treatment” and to examine whether patents that receive this type treatment have a
Entire sample 1617–1841
Outcomes Odds-Ratios
WRI* (top 50%) 1.625*** 2.909*** 3.964*** 22.846***
WRI* (top 10%) 2.336*** 4.220*** 3.894*** 5.684***
WRI* (top 1%) 4.641*** 12.248*** 13.933** 24.326***
YearN1741
Outcomes Odds-Ratios
WRI* (top 50%) 1.525*** 2.774*** 6.334*** 22.729***
WRI* (top 10%) 2.240*** 3.986*** 4.164*** 11.163***
WRI* (top 1%) 5.076*** 12.098*** 13.420*** 33.647***
YearN1781
Outcomes Odds-Ratios
WRI* (top 50%) 1.543*** 2.809*** 9.253*** 28.547***
WRI* (top 10%) 1.979*** 3.831*** 4.283*** 9.172***
WRI*(top 1%) 3.697*** 13.919*** 9.269*** 27.181***
(removing patent cases)
Outcomes Odds-Ratios
WRI*(top 50%) 1.543*** 2.528*** 3.853*** 20.317***
WRI* (top 10%) 2.183*** 3.373*** 3.908*** 9.976***
WRI* (top 1%) 14.712*** 64.496*** 62.281*** 40.523***
*,**,*** indicate significance levels of 10%, 5%, 1% (chi-squared test for Odds-Ratios different from 1).
Information on town sizes was retrieved from Wrigley (1985) and Mitchell and Deane (1962).
vii) Insider: a dummy variable indicating whether the invention patented is related with the occupation of the patentee (e.g., a
medicine for physician or a plough for a farmer). Note that the variable has been constructed in such a way to consider only
the cases inwhich the inventor was clearly connectedwith the occupation of the patentee.When this dummy variable takes
a value of 0 this does not mean that the inventor in question is an outsider, but simply that it was not possible to establish
with full certainty whether he was an insider in relation to the subject matter of the patent in question. Given the degree o
uncertainty in the definition of the variable, we should obviously interpret the estimates of this coefficient with caution.
On the basis of the description of the invention contained in Woodcroft's Chronological Index (Woodcroft, 1854), we have also
classified patents in 21 industries (agriculture, carriages, chemicals, clothing, construction, engines, food, furniture, glass
hardware, instruments, leather, manufacturing, medicines, metallurgy, military, mining, paper, pottery, shipbuilding, textiles).20
This classification is very similar to the one adopted by Moser (2010). We include in the regressions, dummy variables for
controlling for industry effects (textiles is the base reference) andwe use time dummies for controlling for the rise over time in the
number of references per patent. For the time dummies we have adopted the same subperiods used for computing the adjusted
index (WRI*). In this case the period 1832–1841 is the base reference.
The results of the negative binomial regressions are reported in Table 8. The coefficients of the dummy variables related to
significant patents (DNB, Allen “Great Inventor”, Allen “Macro Inventor”, Baker) are all positive and significant. Furthermore, as one
wouldhaveexpected, the coefficient forAllen'smore restricted list of “Great Inventors” is higher than theone forDNB inventors and, in
turn, the one for the “Macro Inventors” is higher than the one for Allen's great inventor list. Concerning the other variables, in some
specifications thevariable “engineer” appears to bepositive and significant. Thisfinding is consistentwith the literature that pointed to
mechanical engineering as the critical innovative sector of the first Industrial Revolution (von Tunzelmann, 1995, pp. 104–122). In
some specifications, the variable “foreign communication” is significantwith a negative sign suggesting that patented inventions tha
were imported from abroad were not of particularly high quality. Finally, the coefficient of the “insider” variable is negative and
significantwhich seems to indicate that the inventionsproducedby “insiders”wereof somewhatminor quality relatively to the rest. In
19 We have also carried out estimations defining the variable “metropolitan” in terms of residence in towns with more than 100,000 inhabitants, obtaining
analogous results.
20 In our sample of 9210 patents covering the period 1617-1841, we have not been able to assign only 11 patents to a specific industry due to unclear o
insufficient description of the invention.
Table 8
Determinants of Woodcroft Reference Index (1617–1841).
(1) (2) (3) (4) (5) (6)
DNB inventor 0.352*** 0.356***
107A. Nuvolari, V. Tartari / Explorations in Economic History 48 (2011) 97–115f
,
t
rvi) Metropolitan: a dummy variable indicating whether the residence of at least one of the patentees is in a town with more
than 50,000 inhabitants.19 Information on patentees' occupations, number of inventors, previous patents, foreign
communication and patentees' addresses were all retrieved from Woodcroft's Chronological Index (Woodcroft, 1854).
(0.0460) (0.0461)
Number of inventors 0.00366 0.00525 0.0143 0.00589 0.00682 0.00275
(0.0290) (0.0291) (0.0289) (0.0287) (0.0291) (0.0291)
Previous patents −0.0243 −0.00775 0.00182 0.00867 0.0132 −0.0224
(0.0218) (0.0210) (0.0208) (0.0208) (0.0212) (0.0218)
Engineer 0.0184 0.0273 0.0456* 0.0381 0.0481* 0.0244
(0.0264) (0.0268) (0.0264) (0.0262) (0.0265) (0.0262)
Foreign Communication −0.0438 −0.0630* −0.0645* −0.0673* −0.0625* −0.0497
(0.0370) (0.0371) (0.0371) (0.0365) (0.0371) (0.0372)
Metropolitan −0.0201 −0.0152 −0.0132 −0.0222 −0.0194 −0.0216
(0.0208) (0.0207) (0.0208) (0.0206) (0.0209) (0.0211)
Allen “Great Inventor” 1.015***
(0.120)
Allen “Macro Inventor” 1.667***
(0.255)
Baker patent 0.894***
(0.107)
Insider −0.0391*
(0.0211)
Constant 0.492*** 0.499*** 0.469*** 0.488*** 0.491*** 0.617***
(0.0508) (0.0503) (0.0501) (0.0505) (0.0510) (0.0389)
Time dummies Yes Yes Yes Yes Yes Yes
Industry dummies Yes Yes Yes Yes Yes No
Observations 9199 9199 9199 9199 9199 9210
Log-likelihood −13852 −13819 −13839 −13816 −13910 −13900
Pseudo R2 0.0836 0.0858 0.0844 0.0860 0.0798 0.0814
Note: negative binomial regressions (dependent variable is WRI), robust standard errors in parenthesis; *,**,*** indicate significance levels of 10%, 5%, 1%.
WRI* scores across industries. This is particularly evident when looking at the median values and maximum values of the quality
distributions. Fig. 3 shows the distribution of WRI* across industries using histograms. All the distributions are right-skewed (this
also confirmed by the summary statistics reported in Table 9).21 This means that the majority of patents tend to be negligible (the
number of patents with WRI*=0 is indicated in the last column) or of very low value and that only few patents have high quality
scores. Further, all industrial sectors seems capable of producing at least some “high quality” patents, although “technologica
blockbusters” (patents with quality scores that are more than one order of magnitude higher than the time cohort average) are
experienced only in a few sectors. This is fully in line with the findings of the modern literature on the value of patents. All the
modern indicators of patent quality (number of citations received, renewal data and survey data based on inventors' self-
assessment) have sharply right-skewed distributions similar to those of Fig. 3 (see Silverberg and Verspagen, 2007 for a thorough
discussion). Notably, this holds also at the level of individual technology fields (Schankerman, 1998). Therefore, in our
interpretation, the shape of the distribution ofWRI* provides further corroboration of its plausibility as indicator of patent quality
6. The distribution of high quality patents over time and across industries
What are the implications of the indicator of patent quality based onWoodcroft's Reference Index for the debate concerning the
timing and scope of the Industrial Revolution? Fig. 4 charts the cumulative distribution of patents of different quality over time
The thin lines represent the cumulative distribution of patents that are in the top 0.5%, 1% and 5% percentiles in terms of their
quality scores measured using WRI*. The thick line represents the cumulative distribution of the total number of patents. Fig. 4
may be interpreted as comparing the evolution over time of the stock of knowledge embodied in patented macroinventions (the
top percentiles) and in patented microinventions (the total number of patents). The dotted line represents the cumulative
distribution of a set of “important patents” (this set contains all the patents that have been mentioned in at least two lists of
important patents used in the previous section: DNB, Allen “Great Inventor”, Allen “Macro Inventor” and Baker). We have plotted
Table 9
Descriptive statistics for WRI* (1617–1841).
21 The hypothesis of normality is rejected both for the entire sample and for all technology classes (Shapiro–Wilks test).
108 A. Nuvolari, V. Tartari / Explorations in Economic History 48 (2011) 97–115l
.
.themodern innovation literature it has been indeed pointed out that radical innovations are frequentlymade by outsiders, who enjoy
the advantage of an “uncommitted mind” and in this way have fresh insights on the possible solutions of specific technological
problems (Jewkes et al., 1969). In the period we are considering, consistently with our findings, O'Brien et al. (1996) have contended
that, in the textile industries, the most important inventions were made by outsiders who had a pre-professional interest (scientific
and technological curiosity, fascination for mechanical contrivances, etc.) in invention, whereas the inventive activities of insiders
were mostly of incremental nature (on the possible advantages of outsiders as inventors in this period, see also O'Brien, 1997). The
results of Table 8 are also confirmed when performing the regression analysis on subsets of patents restricted to the periods 1742–
1841 and 1782–1841 (in order to remove the early periods in which the legal status of the specificationwas not fully developed) and
on the subsets excluding the patents disputed in court cases (to limit the analysis only to the subsample of patents less likely to be
mentioned in the legal literature). The results of these regressions are reported in Appendix (Tables A.1, A.2, A.3).
Table 9 reports descriptive statistics forWRI*scores across industries. Table 9 suggests the existence of systematic differences in
Industry Number Mean Standard deviation Median Skewness Kurtosis Min Max Number of 0s
Agriculture 287 0.829 0.720 0.798 0.773 4.178 0 3.989 88
Carriages, vehicles, railways 513 1.064 1.424 1.091 11.291 192.174 0 26.214 97
Chemical and allied industries 753 1.113 1.256 1.091 5.286 46.051 0 15.701 148
Clothing 196 0.893 1.126 0.786 3.824 25.714 0 8.738 65
Construction 400 1.127 1.614 1.091 7.494 83.257 0 22.154 83
Engines (steam engines, water wheels) 1177 1.092 2.287 1.091 17.893 428.160 0 61.167 282
Food and drink 529 0.949 1.116 0.798 4.383 36.000 0 12.502 156
Furniture 473 0.848 0.911 0.786 4.103 35.272 0 10.467 120
Glass 89 1.157 2.206 0.798 5.498 37.584 0 17.476 27
Hardware (edge tools, locks, grates) 628 0.961 1.111 0.940 6.641 85.237 0 17.476 147
Instruments (scientific instruments, watches,
measuring devices)
410 0.889 0.779 1.082 2.016 14.020 0 6.547 106
Leather 158 1.050 0.809 1.137 1.169 6.697 0 5.063 33
Manufacturing machinery (other) 457 0.901 1.012 0.798 3.824 28.534 0 10.229 128
Medicines (drugs, surgical and dental instruments,
other medical devices)
244 0.804 0.927 0.786 4.035 29.221 0 7.851 73
Metal manufacturing 466 1.112 2.636 0.798 16.142 310.156 0 52.429 115
Military equipment and weapons 216 0.979 1.471 0.786 7.372 77.248 0 17.476 46
Mining 62 1.052 1.044 1.114 1.509 6.601 0 5.111 20
Paper, printing and publishing 337 1.107 1.407 1.091 6.510 59.536 0 14.775 52
Pottery, bricks, artificial stone 169 1.044 1.199 1.091 3.474 20.518 0 8.738 45
Shipbuilding 481 0.953 1.081 0.786 3.504 21.272 0 8.738 119
Textiles 1154 0.972 1.653 0.633 7.615 89.369 0 27.692 288
Total sample 9210 1 1.530 0.798 15.239 450.939 0 61.167 2245
the cumulative distribution of this set of patents in order to compare it with those of the patents in the top percentiles of WRI*.
Fig. 4 shows that the cumulative distribution of high quality patents tends to “anticipate” the cumulative distribution of the total
patents. In particular, the cumulative distribution of the top 0.5% percentile reaches a level of 50% in 1794 (in the same year the
cumulative distribution of “important patents” is also reaching a level of 50%), whereas the cumulative distribution of total patents
reaches a level of 50% only in 1823, almost thirty years later. It is instructive to compare this time profile with the estimates of
productivity growth produced by Crafts and Harley. According to Crafts' most recent estimates, total factor productivity growth
was negligible in the period 1760–1780. It increased to 0.3% per year in the period 1780–1831 and from there to 0.75% per year in
109A. Nuvolari, V. Tartari / Explorations in Economic History 48 (2011) 97–1151831–1873 (Crafts, 2004, p. 522). It is relatively straightforward to put forward an explanation that can accounts for the temporal
patterns of Fig. 4: the classical take-off period (1760–1801) should be regarded as the phase in which several macroinventions in
the sense of Mokyr (1990) emerged. In this interpretation, the time profile of the high quality patents in Fig. 4 (in particular the top
0.5%) is capturing the time dynamics of these macro inventions. However, the impact of these macroinventions on productivity
growth became fully manifest only after a stream ofmicroinventions (possibly represented in Fig. 4 by the cumulative distribution of
total patents) improved their technological performance and cost effectiveness. In this way, the distribution over time of high quality
patents may be reconciled with the dynamics of productivity growth posited in Crafts and Harley (1992) revisionist account.
Turning our attention to the scope of technical change, it is possible to use the quality indicator to carry out a simple accounting
exercise aimed at singling out the relative contribution of different industries to overall technical change during this period. Given the
imperfections of our indicator, we should consider this accounting exercise nothing more than a rough back of envelope type of
calculation. In particular, since we are going to use themean ofWRI* as a weight for the quality of patents, wewill obtain results that are
smoothing the impact of radical innovations over many patents, neglecting the distinction between macro and micro innovations. The
results of theexercise are reported inTable10. Thefirst twocolumnsof the table contain thenumberof apatents ineach industry and their
averagequality (WRI*).22 The third columnof the table contains thepatenting rates calculated byMoser (2010). These patenting rates are
computed as the share of British inventions at the Crystal Palace exhibition of 1851 that were patented. The purpose of this column is to
provide some indication of the different propensity to patent of different industries. The fourth column reports the share of each sector in
total inventive output computed using the number of patents. The fifth column reports the share of each sector in total inventive output
computed as number of patents weighted by their average quality. The sixth column reports the shares in inventive output weighted by
averagequalitywhenweuse thepatenting rates ofMoser (2010) for trying to adjust for thedifferentpatent propensity of the sectors. This
adjustment consists in dividing the number of patents in each sector for the patenting rate.23 The results of Table 10 seems to point to a
pattern of technical change that is widespread and not localized in few sectors (although, it is interesting to note that, for all our three
approaches tomeasurement, three industries such as textiles, engines and chemicals account formore than 30% of inventive output). The
adjustment for the average quality of patents does not seem to have amajor effect the industrial concentration of inventive output and as
such these results are broadly consistent with those obtained by Sullivan (1990) who has used simple patent counts for measuring
inventive output. Interestingly enough, the adjustment for the propensity to patent has instead a much stronger effect on the shares of
inventive output. In particular, when adjusted for patenting propensity, the share of the paper industry becomesmuch larger. In fact, the
paper industrywas an industrywitnessing important technical changes, but characterizedby avery lowpatent propensity (Moser, 2010).
However, even after adjusting for the propensity to patent, inventive output remains fairly widespread across industries.
However, the sharply skewed nature of the distribution of the quality of patents discussed in the previous section, suggests
that, in order to shed light on the nature of technical change, is probably more revealing to consider the distribution across
industries of the patents in the top percentiles of the quality scores. In this case, our underlying assumption is that is the restricted
group of macroinventions situated in the tails of the quality distributions that are critical for productivity growth. Table 11 shows
the concentration of patents of different quality across industries over the period 1702–1841measured using the Herfindahl index
and concentration ratios.24 Patents in the top percentiles of the quality scores (top 0.5%, top 1%) exhibit a remarkably higher
degree of concentration than patents of lower quality (top 50% and the total patent sample). In Table 11we have also displayed the
degree of industrial concentration for the set of important patents (defined as in Fig. 4 as the set of patents that have been
mentioned simultaneously in at least two lists of “important patents”). The level of concentration for this set of patents is similar to
that of the top 0.5% percentile. Overall, Table 11 suggests that, although total patents were relatively widespread across industries
as pointed out by Sullivan (1990),25 technological blockbusters (patents of very high quality) were remarkably more localized.
This result, in our view, provides an interesting hint for reconciling the Crafts–Harvey view with the patent evidence.
7. Concluding remarks
In this paperwehaveproposedanew indicator (WRI*)of thequality of Englishpatents in theperiod1617–1852basedonWoodcroft's
Reference Index. We have also explored the properties of this indicator and our preliminary results appear quite encouraging. We have
established that the quality indicator is positively correlated with four different lists of “important patents” (patentees included in the
22 Note that since we have used the average number of references to benchmark the quality of patents, the average patent quality at the level of the total
sample is equal to 1.
23 For the sectors for which patenting rates were not available we have used the aggregate patenting rate of 0.12.
24 The Herfindahl index was computed as H=∑si2 where si is the share of patents of given quality in industry i. The higher the value of H the higher the degree
of concentration: in this formulation, the index ranges from 1/21 (≅0.048) when patents are evenly distributed across industries, to 1 when all patents are
concentrated in one single industry. It is also instructive to consider the equivalent number (1/H) that indicates the number of industries with equal size
corresponding to the level H of concentration. The C3 concentration ratio is the sum of the shares of the three industries with the largest shares.
25 The Herfindahl index for columns 4, 5 and 6 in Table 10 are respectively equal to 0.07 (this, of course, is the same as column 6 in Table 11), 0.07 and to 0.09.
110 A. Nuvolari, V. Tartari / Explorations in Economic History 48 (2011) 97–115
111A. Nuvolari, V. Tartari / Explorations in Economic History 48 (2011) 97–115DNB, in the two lists of “great inventors” recently compiled by Allen (2009) and in the list of significant patents constructed by Baker
(1976)). Oneof themerits of theWRI* indicator in comparisonwith the “great inventors” approach is that it provides aproxy for quality at
the invention rather than at the inventor level. One further advantage (in comparison both to the great inventor lists and to Baker's list of
significant patents) is thatWRI* index canbe calculated for all patents. Insteadoneof the obvious limitations of the adjustedWRI* index in
comparison with “great inventors” approach is that the former is restricted to patented inventions, whereas the latter can include also
Fig. 3. Distribution of WRI* across industries.
inventions that were not patented. We have also established that the distribution of WRI*, both at aggregate level and at the level of
individual industry, is sharply right-skewed and similar to the empirical evidence found in contemporary studies of value of patents.
In a broader perspective, we think that the WRI* indicator and, more generally, Woodcroft's Reference Index have some very
interesting potential for helping us to shed further light on some of the ongoing debates on the timing and scope of innovation
Fig. 4. Cumulative distribution of patents of different quality, 1702–1841.
(WRI*) (Moser, 2010) output (patents) output (patents×WRI*) (patents/patenting rates)×WRI*
Agriculture 272 0.875 0.228 0.031 0.027 0.010
Carriages, vehicles, railways 500 1.039 0.12 0.057 0.059 0.043
Chemical and allied industries 723 1.111 0.063 0.082 0.091 0.125
Clothing 185 0.851 0.103 0.021 0.018 0.015
Construction 383 1.155 0.141 0.043 0.050 0.031
Engines (steam engines,
water wheels)
1133 1.050 0.261 0.128 0.135 0.045
Food and drink 492 0.985 0.054 0.056 0.055 0.088
Furniture 466 0.861 0.061 0.053 0.045 0.065
Glass 76 1.010 0.104 0.009 0.009 0.007
Hardware (edge tools, locks, grates) 610 0.961 0.151 0.069 0.066 0.038
Instruments (scientific instruments,
watches, measuring devices)
406 0.898 0.088 0.046 0.041 0.041
Leather 152 1.091 0.093 0.017 0.019 0.018
Manufacturing machinery (other) 445 0.925 0.296 0.050 0.047 0.014
Medicines (drugs, surgical and dental
instruments, other medical devices)
240 0.818 (–) 0.027 0.022 0.016
Metal manufacturing 430 1.063 (–) 0.049 0.052 0.037
Military equipment and weapons 208 0.932 0.135 0.024 0.022 0.014
Mining 54 1.207 0.038 0.006 0.007 0.017
Paper, printing and publishing 324 1.151 0.023 0.037 0.042 0.160
Pottery, bricks, artificial stone 159 1.055 (–) 0.018 0.019 0.014
Shipbuilding 444 1.013 (–) 0.050 0.051 0.037
Textiles 1132 0.968 0.065 0.128 0.124 0.166
Total sample 8834 1 0.12 1 1 1
Note: For calculating the shares of column (4), inventive output is computed as number of patents. For calculating the shares of column (5), inventive output is
computed as column (1)×column (2). For calculating the shares of column (6), inventive output is computed as [column(1)/column(3)]×column (2).
112 A. Nuvolari, V. Tartari / Explorations in Economic History 48 (2011) 97–115during the Industrial Revolution. TheWRI* indicator presents the advantage of being of relatively easy computation and it seems
capable to provide a reasonable proxy for the economic value of patents, which can fruitfully complement simple patent counts
as indicator of innovation in this historical period. In particular, it has been frequently pointed out that the patent evidence lends
support to a traditional view of the Industrial Revolution as a dramatic acceleration of technical change taking place in the second
Table 10
Inventive Output measured using patents and WRI*, 1702–1841.
Patents Mean Patenting rates Shares in inventive Shares in inventive Shares in inventive output
(0.0211)
Constant 0.475*** 0.482*** 0.453*** 0.472*** 0.473*** 0.606***
(0.0509) (0.0505) (0.0501) (0.0506) (0.0511) (0.0390)
Time dummies Yes Yes Yes Yes Yes Yes
Industry dummies Yes Yes Yes Yes Yes No
Observations 8623 8623 8623 8623 8623 8629
Log-likelihood −13549 −13520 −13538 −13519 −13599 −13598
Pseudo R2 0.0547 0.0567 0.0555 0.0568 0.0511 0.0519
Note: negative binomial regressions (dependent variable is WRI), robust standard errors in parenthesis; *,**,*** indicate significance levels of 10%, 5%, 1%.
Table 11
Industrial concentration of patents of different quality (Herfindahl indexes), 1702–1841.
113A. Nuvolari, V. Tartari / Explorations in Economic History 48 (2011) 97–115half of the eighteenth century and affecting simultaneously many sectors (Sullivan, 1989, 1990; Temin, 2000, p. 845). Our
proposed indicator of patent quality seems instead to offer a way of reconciling the patent evidence with the “revisionist view”
put forward by Crafts and Harley (1992). Concerning the timing of the Industrial Revolution, our findings seem in line with a
traditional chronology and confirm that the second half of the eighteenth century (1762–1801) was the critical historical phase
with a clustering of critical technical breakthroughs (top-quality patents). However, it is important to take into account that the
full impact of these macro “prototype” inventions on productivity growth became visible only after a phase of adaptation,
improvement and refinement bymeans of streams of microinventions. Thus the time profile of high quality patents that we have
reconstructed using the WRI* indicator appears to be consistent with the dynamics of productivity growth estimated by Crafts
and Harley (1992). In terms of the scope of the change, our findings indicate that, although total patents were relatively
widespread, top-quality patents (i.e., those covering technological blockbusters) were much more concentrated across
industries. If we regard productivity growth as an outcome of the sustained improvement and extension (also to other industrial
applications) of these macroinventions, our results indicate that the patent records evidence may indeed be consistent with a
view of the Industrial Revolution as a process driven by a few revolutionary industrial innovations localized in a relatively
circumscribed segment of the economy.
It is worth to conclude with an obvious word of caution. Our findings on the patterns of technical change during the Industrial
Revolution are exclusively based on patented inventions. We should not forget however, that in this historical period a very significant
amount of inventive activities was undertaken without the coverage of patent protection. Thus, the findings of this paper do not imply
that the search for indicators of technical change based on historical sources that are alternative to the patent records is going to become
less important. Real progress in our understanding of the historical process of technical change is likely to emerge only by tackling the
subject combining systematically different type of indicators and approaches to measurement both on patented and not patented
inventions.
Appendix
Table A.1
Determinants of Woodcroft Reference Index (1742–1841).
(1) (2) (3) (4) (5) (6)
DNB Inventor 0.329*** 0.333***
(0.0453) (0.0453)
Number of inventors 0.0144 0.0163 0.0244 0.0165 0.0176 0.0134
(0.0292) (0.0293) (0.0291) (0.0288) (0.0292) (0.0292)
Previous patents −0.0245 −0.00914 −0.00001 0.00653 0.0104 −0.0228
(0.0217) (0.0210) (0.0209) (0.0208) (0.0211) (0.0218)
Engineer 0.0221 0.0298 0.0472* 0.0402 0.0496* 0.0265
(0.0264) (0.0268) (0.0264) (0.0262) (0.0265) (0.0262)
Foreign communication −0.0444 −0.0621* −0.0636* −0.0663* −0.0618* −0.0502
(0.0370) (0.0371) (0.0371) (0.0365) (0.0371) (0.0372)
Metropolitan −0.0206 −0.0164 −0.0142 −0.0226 −0.0199 −0.0217
(0.0208) (0.0208) (0.0208) (0.0206) (0.0210) (0.0211)
Allen “Great Inventor” 0.969***
(0.120)
Allen “Macro Inventor” 1.590***
(0.255)
Baker 0.839***
(0.107)
Insider −0.0386*
Top 0.5% Top 1% Top 5% Top 10% Top 50% Total patents Important patents
Herfindahl (H) 0.12 0.10 0.07 0.08 0.07 0.07 0.12
Equivalent number (1/H) 8.07 9.86 13.40 13.27 14.42 14.37 8.68
C3 0.51 0.42 0.37 0.37 0.34 0.34 0.51
Table A.3
Determinants of Woodcroft Reference Index (removing patent cases).
(1) (2) (3) (4) (5) (6)
DNB Inventor 0.203*** 0.212***
(0.0308) (0.0311)
Number of inventors 0.00950 0.00813 0.0146 0.0115 0.0114 −0.000359
(0.0211) (0.0211) (0.0209) (0.0210) (0.0210) (0.0217)
Metropolitan −0.0269* −0.0261* −0.0262* −0.0296* −0.0278* −0.0268*
(0.0157) (0.0156) (0.0157) (0.0157) (0.0157) (0.0157)
Previous patents −0.0237 −0.0159 −0.00625 −0.00404 −0.00142 −0.0221
(0.0165) (0.0160) (0.0160) (0.0160) (0.0161) (0.0166)
Engineer 0.0624*** 0.0635*** 0.0758*** 0.0729*** 0.0785*** 0.0787***
(0.0210) (0.0210) (0.0209) (0.0210) (0.0211) (0.0214)
Allen “Great Inventor” 0.759***
(0.0843)
Allen “Macro Inventor” 1.180***
(0.181)
Baker patent 0.526***
(0.0839)
Insider −0.0382**
(0.0158)
Constant 0.275*** 0.284*** 0.265*** 0.271*** 0.272*** 0.495***
(0.0386) (0.0385) (0.0385) (0.0385) (0.0386) (0.0299)
Time dummies Yes Yes Yes Yes Yes Yes
Industry dummies Yes Yes Yes Yes Yes No
Observations 8937 8937 8937 8937 8937 8948
Log-likelihood −11947 −11919 −11942 −11942 −11966 −12019
Pseudo R2 0.107 0.109 0.108 0.108 0.106 0.103
Note: negative binomial regressions (dependent variable is WRI), robust standard errors in parenthesis; *,**,*** indicate significance levels of (10%,5%,1%).
Table A.2
Determinants of Woodcroft Reference Index (1782–1841).
(1) (2) (3) (4) (5) (6)
DNB Inventor 0.258*** 0.265***
(0.0417) (0.0420)
Number of inventors 0.0207 0.0238 0.0258 0.0217 0.0230 0.0164
(0.0281) (0.0282) (0.0282) (0.0277) (0.0283) (0.0283)
Previous patents −0.0209 −0.00645 0.000184 0.00298 0.00636 −0.0181
(0.0218) (0.0210) (0.0209) (0.0207) (0.0211) (0.0219)
Engineer 0.0322 0.0368 0.0522** 0.0449* 0.0529** 0.0303
(0.0262) (0.0267) (0.0265) (0.0262) (0.0265) (0.0262)
Foreign communication −0.0510 −0.0646* −0.0655* −0.0678* −0.0648* −0.0560
(0.0370) (0.0371) (0.0371) (0.0365) (0.0371) (0.0372)
Metropolitan −0.0127 −0.00954 −0.00974 −0.0164 −0.0115 −0.0118
(0.0211) (0.0211) (0.0212) (0.0208) (0.0212) (0.0212)
Allen “Great Inventor” 0.806***
(0.107)
Allen “Macro Inventor 1.228***
(0.224)
Baker 0.767***
(0.104)
Insider −0.0336
(0.0212)
Constant 0.442*** 0.447*** 0.431*** 0.443*** 0.440*** 0.598***
(0.0501) (0.0500) (0.0499) (0.0498) (0.0503) (0.0382)
Time dummies Yes Yes Yes Yes Yes Yes
Industry dummies Yes Yes Yes Yes Yes No
Observations 7896 7896 7896 7896 7896 7901
Log-likelihood −12658 −12647 −12665 −12628 −12688 −12711
Pseudo R2 0.0465 0.0472 0.0459 0.0487 0.0442 0.0430
Note: negative binomial regressions (dependent variable is WRI), robust standard errors in parenthesis; *,**,*** indicate significance levels of 10%, 5%, 1%.
114 A. Nuvolari, V. Tartari / Explorations in Economic History 48 (2011) 97–115
115A. Nuvolari, V. Tartari / Explorations in Economic History 48 (2011) 97–115References
Agresti, A., 2002. Categorical Data Analysis, 2nd ed. Wiley, New York.
Allen, R.C., 2009. The British Industrial Revolution in Global Perspective. Cambridge University Press, Cambridge.
Allison, J.R., Lemley, M.A., Moore, K.A., Trunkey, R.D., 2004. Valuable patents. Georgetown Law Journal 92, 435–479.
Anon [Newton, W.], 1820. Preface. London Journal of Arts and Sciences 1, i–v.
Anon., 1794. Advertisement. Repertory of Arts and Manufacures 1, i–iv.
Anon., 1823. Preface. Mechanic's Magazine 1, i–iv.
Anon., 1839. Introductory address. The Inventors' Advocate and Patentee Recorder 1, 1–2.
Baker, R., 1976. New and Improved…Inventors and Inventions that Have Changed the Modern World. British Library, London.
Bessen, J., 2008. The value of US patents by owner and patent characteristics. Research Policy 37, 932–945.
Bessen, J., 2009. Estimates of patent rents from firm market value. Research Policy 38, 1604–1616.
Crafts, N.F.R., 2004. Productivity growth in the industrial revolution: a new growth accounting perspective. Journal of Economic History 64, 521–535.
Crafts, N.F.R., Harley, C.K., 1992. Output growth and the British Industrial Revolution: a restatement of the Crafts−Harley view. Economic History Review 45,
703–730.
Dutton, H.I., 1984. The Patent System and Inventive Activity during the Industrial Revolution. Manchester University Press, Manchester.
Fligner, M.A., Policello II, G.E., 1981. Robust rank procedures for the Behrens–Fisher problem. Journal of the American Statistical Association 73, 162–168.
Gomme, A.A., 1946. Patents of Invention. Origins and Growth of the Patent System in Britain. Longmans, London.
Hall, B. H., Jaffe, A. B, Trajtenberg, M., 2002. “The NBER Patent-Citations Data File: Lessons, Insights, Methodological Tools” in Jaffe, A. B. and Trajtenberg, M. (2002),
Patents, Citations and Innovations, (Cambridge (MA): MIT Press).
Harhoff, D., Scherer, F.M., Vopel, K., 2003. Citations, family size, opposition and the value of patents. Research Policy 32, 1343–1363.
Harrison, J., 2006. Encouraging Innovation in the Eighteenth and Nineteenth Centuries, The Society of Arts and Patents, 1754–1904. High View, Gunnislake.
Harvey, W.S., Downs-Rose, G., 1974. Mr. Symington's improved atmospheric engine. Transactions of the Newcomen Society 48, 27–32 (1973-4).
Hebert, L., 1836. The Engineer's and Mechanic's Encyclopaedia. Kelly, London.
Hewish, J., 1982. The Indefatigable Mr. Woodcroft. The Legacy of Invention. British Library, London.
Hewish, J., 2000. Rooms near Chancery Lane. The Patent Office under the Commissioners 1852–1883. British Library, London.
House of Lords, 1851. [Select Committee of the House of Lords appointed to consider the Bills for the amendment of the Law touching Letters Patent for Invention],
‘Reports and Minutes of the Evidence’ (P.P. 1851, XVIII).
Jaffe, A., Trajtenberg, M., 2002. Patents, Citations and Innovation. A Window on the Knowledge Economy. MIT Press, Cambridge, MA.
Jewkes, J., Sawers, D., Stillerman, R., 1969. The Sources of Invention, 2nd ed. Macmillan, London.
Khan, B.Z., Sokoloff, K.L., 1993. ‘Schemes of Practical Utility’: entrepreneurship and innovation among ‘Great Inventors’ in the United States, 1790–1865. Journal of
Economic History 53, 289–307.
Khan, B.Z., Sokoloff, K.L., 2008. A tale of two countries: innovation and incentives among great inventors in Britain and the United States, 1750–1930. In: Farmer, R.
(Ed.), Macroeconomics in the Small and the Large. Edward Elgar, Cheltenham.
Kleinknecht, A., 1987. Innovation Patterns in Crisis and Prosperity: Schumpeter's Long Cycle Reconsidered. MacMillan, London.
Lampe, R., 2010. Strategic citation. Working Paper. Stanford University, available at http://ssrn.com/abstract=984123.
MacLeod, C., 1988. Inventing the Industrial Revolution. The English Patent System, 1660–1800. Cambridge University Press, Cambridge.
MacLeod, C., 2007. Heroes of Invention. Technology, Liberalism and British Identity, 1750–1914. Cambridge University Press, Cambridge.
MacLeod, C., Tann, J., Andrew, J., Stein, J., 2003. Evaluating inventive activity: the cost of nineteenth century UK patents and the fallibility of renewal data. Economic
History Review 56, 537–562.
Mantoux, P., 1928. The Industrial Revolution in the Eighteenth Century. MacMillan, London.
Mitchell, B.R., Deane, P., 1962. Abstracts of British Historical Statistics. Cambridge University Press, Cambridge.
Mokyr, J., 1990. The Lever of Riches. Oxford University Press, Oxford.
Mokyr, J., 1999. Introduction: the new economic history and the Industrial Revolution, In: Mokyr, J. (Ed.), The British Industrial Revolution. An Economic Perspective,
2nd ed. Westview, Boulder.
Mokyr, J., 2009. The Enlightened Economy. An Economic History of Britain 1700–1850. Yale University Press, New Haven.
Moser, P., 2005. How do patent laws influence innovation? Evidence from nineteenth century world fairs. American Economic Review 95, 1213–1236.
Moser, P., 2010. Do Patents Weaken the Localization of Innovations ? Evidence from World Fairs, 1851–1915. Working Paper. Stanford University, available at
http://ssrn.com/abstract=941571.
Nuvolari, A., 2004. Collective Invention during the British Industrial Revolution: the Case of the Cornish Pumping Engine. Cambridge Journal Economics 28,
347–363.
O'Brien, P.K., 1997. The Micro Foundations of Macro Invention: the Case of Reverend Edmund Cartwright. Textile History 28, 201–233.
O'Brien, P.K., Griffiths, T., Hunt, P., 1995. There is nothing outside the text and there is no safety in numbers: a reply to Sullivan. Journal of Economic History 55,
671–672.
O'Brien, P.K., Griffiths, T., Hunt, P., 1996. Technological change during the First Industrial Revolution: the paradigm case of textiles, 1688–1851. In: Fox, R. (Ed.),
Technological Change. Harwood, Amsterdam.
Ravn, M.O., Uhlig, H., 2002. On adjusting the Hodrick–Prescott filter for the frequency of observations. Review of Economics and Statistics 84, 371–380.
Sampat, B.N., Ziedonis, A.A., 2004. Patent citations and the economic value of patents. In: Moed, H.F., et al. (Ed.), Handbook of Quantitative Science and Technology
Research. Kluwer, Dordrecht.
Schankerman, M., 1998. How valuable is patent protection ? Estimates by technology field. Rand Journal of Economics 29, 77–107.
Schankerman, M., Pakes, A., 1986. Estimates of the value of patents in European countries during post-1950 period. Economic Journal 97, 1–25.
Silverberg, G., Verspagen, B., 2003. Breaking the waves: a Poisson regression approach to Schumpeterian clustering of basic innovations. Cambridge Journal of
Economics 27, 671–693.
Silverberg, G., Verspagen, B., 2007. The size distribution of inventions revisited: an application of extreme value statistics to citation and value measure of patent
significance. Journal of Econometrics 139, 318–339.
Singer, C., Holmyard, E.J., Hall, A.R., 1957. A History of Technology. Vol. III. From the Renaissance to the Industrial Revolution, 1500–1750. Clarendon, Oxford.
Singer, C., Holmyard, E.J., Hall, A.R., 1957. A History of Technology. Vol. IV. The Industrial Revolution 1750–1850. Clarendon, Oxford.
Sullivan, R.J., 1989. England's ‘Age of Invention’: the acceleration of patents and of patentable invention during the Industrial Revolution. Explorations in Economic
History 26, 424–452.
Sullivan, R.J., 1990. The revolution of ideas: widespread patenting and invention during the English Industrial Revolution. Journal of Economic History 50, 349–362.
Sullivan, R.J., 1994. Estimates of the value of patent rights in Great Britain and Ireland, 1852–1876. Economica 61, 37–58.
Sullivan, R.J., 1995. Patent counts and textile invention: a comment on Griffiths, Hunt and O'Brien. Journal of Economic History 55, 666–670.
Temin, P., 2000. A response to Harley and Crafts. Journal of Economic History 60, 842–846.
Trajtenberg, M., 1990. A penny for your quotes: patent citations and the value of innovations. Rand Journal of Economics 21, 172–187.
Van Zeebroeck, N., in press. The puzzle of patent value indicators. Economics of Innovation and New Technology. doi:10.1080/10438590903038256.
Von Tunzelmann, G.N., 1995. Technology and Industrial Progress. Edward Elgar, Aldershot.
Woodcroft, B., 1854. Titles of Patents of Invention Chronologically Arranged, 1617–1852. G.E. Eyre & W. Spottiswoode, London.
Woodcroft, B., 1862. Reference Index of English Patents of Invention, 1617–1852. G.E. Eyre & W. Spottiswoode, London.
Wrigley, E.A., 1985. Urban growth and agricultural change: England and the continent in the early modern period. Journal of Interdisciplinary History 15, 683–728.