See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/360579167

The Mechanics of the Industrial Revolution

Article in Journal of Political Economy · May 2022

DOI: 10.1086/720890

CITATIONS READS
18 1,008

3 authors:

Morgan Kelly Joel Mokyr

University College Dublin Northwestern University
52 PUBLICATIONS 1,813 CITATIONS 162 PUBLICATIONS 9,326 CITATIONS

SEE PROFILE SEE PROFILE

Cormac Ó Gráda
University College Dublin
274 PUBLICATIONS 3,129 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

The Industrial Revolution. View project

The Myth of Europe's Little Ice Age View project

All content following this page was uploaded by Morgan Kelly on 16 May 2022.

The user has requested enhancement of the downloaded file.

 The Mechanics of the Industrial Revolution
Morgan Kelly, Joel Mokyr, and Cormac Ó Gráda∗

Abstract
Although there are many competing explanations for the Industrial Revolution,
there has been no effort to evaluate them econometrically. This paper analyses how
the very different patterns of growth across the 41 counties of England between the
1760s and 1830s can be explained by a wide range of potential variables. We show that
although spatial standard error corrections are of limited practical utility, semipara-
metric regressions that include a spatial smoothing spline return reliable results. For
textiles we find that industrialization occurred in areas that began with low wages but
high mechanical skills, whereas other variables such as literacy, banks and proxim-
ity to coal have little explanatory power. Against the view that living standards were
stagnant during the Industrial Revolution, we find that real wages rose sharply in the
industrializing north and declined in the previously prosperous south.

1 Introduction.
The question of British leadership in the Industrial Revolution remains one of the central
topics in economic history as well as in the larger literature on the origins and nature of
modern economic growth. That literature is enormous, with attributions ranging from
geography to institutions, and from colonial exploitation to culture, or possibly even luck.1
One influential interpretation holds that the Industrial Revolution can be explained by
induced technological innovation, with cheap coal power being substituted for expensive
∗
University College Dublin; Northwestern University; University College Dublin. The paper has benefited
from the comments of seminar participants in Belfast, Ben Gurion University, Buenos Aires, Haifa, Lund,
Princeton, Stanford and Warwick; and those of Nick Crafts and Peter Solar. We would especially like to thank
the editor (Harald Uhlig) and three referees for their detailed criticisms of the submitted manuscript. Any
errors are ours.
1
Geographical explanations focus on natural resources, especially coal and iron, and location; institutional
explanations highlight the role of the Whig Revolution and, more generally, a pro-business framework of laws
and social institutions; cultural explanations refer to respect for science, innovation, and middle-class values;
and explanations invoking empire invoke privileged access to raw materials and the slave trade. On luck
see Crafts (1977). Surveys include Clark (2014), Crafts (2021), Jones (2010) McCloskey (2016), and Mokyr
(2009). Our emphasis on skills is perhaps closest to Harris (1976) and Berg (1994, 7) who traced Britain’s
success to “the extraordinary industry and inventiveness of her manufacturing people.”

1
labour (Allen, 2009a). For many contemporaries, by contrast, Britain’s success was rooted
in its uniquely deep and diverse pool of artisans, in metalworking especially, whose skills
could be readily adapted to developing the new machinery and manufacturing processes
that began to appear in the mid-eighteenth century.
Despite the importance of the Industrial Revolution, there have been no previous at-
tempts to evaluate these competing explanations econometrically. The purpose of this
paper is to test how the very different patterns of industrial growth across the 41 coun-
ties of England can be explained by factors including high wages, energy sources, literacy,
banks, and, especially, mechanical skills. On the way we will challenge claims that the In-
dustrial Revolution was induced by a desire to substitute cheap steam powered machinery
for expensive workers, and that living standards were static at this time.
The technology of the late eighteenth century is often dismissed as having been fairly
rudimentary (which raises the question of why it was not invented a good deal earlier). In
fact, the two iconic machines of early industrialization—Arkwright’s spinning frame with
its intricately meshing train of gears, spindles and rollers, and Watt’s steam engine with its
precisely bored cylinder and complicated valves—were unusually complex technologies
by the standards of the time, and each relied on co-opting local artisanal skill for its success.
Our approach is to focus on a simple process where the accumulation of artisan skill
drives technological progress, in a way that mirrors the historical pattern of early indus-
trialization. Specifically, as transport networks began to improve and English markets in-
tegrated from the late seventeenth century onward, regions specialized according to their
comparative advantage. This meant that areas of poor agricultural potential (reflected in
their low wages) increasingly specialized in manufacturing activities. Naturally, many of
these proto-industrial activities, such as making nails or low quality textiles, required only
rudimentary skills and offered no possibilities of technological advances.
However, a few forms of manufacturing—especially in exacting forms of metal work
such as watch making, iron founding, and tool making—created pools of skilled and ver-
satile workers, artisans whose skills could be readily be adapted and transferred to the
increasingly sophisticated machinery and manufacturing processes of the early Industrial
Revolution. This simple framework leads to the specific empirical prediction that suc-
cessful industrialization relied on existing concentrations of suitable skills, and that these
concentrations were to be found primarily in low wage areas already specializing in tech-
nologically demanding production, in metalworking especially.
The prediction that low wage areas should be the ones that industrialized runs directly
counter to the influential argument of Allen (2009a) that the British Industrial Revolution
involved a substitution of steam power for labour.2 This claim is historically question-
able on several counts. Certainly, unskilled English labourers enjoyed considerably higher
wages than their French counterparts but, because they were stronger and better fed as a
2
Criticisms of these data and methodology include Kelly, Mokyr and Ó Gráda (2014), Mokyr (2009, 267–
272), Stephenson (2018), Humphries and Schneider (2019), and Crafts (2021).

2
result, they were proportionately more productive: high wages need not mean expensive
labour (Kelly, Mokyr and Ó Gráda, 2014). Moreover, when it comes to the skilled artisans
who were the real drivers of the Industrial Revolution, it was France that was the high
wage economy, as shown by the thousands of British artisans who migrated there after
1815 to set up railways, cotton mills and ironworks (Bensimon, 2011).
Turning to England, we will show that it was the low wage north that mechanized,
whereas the high wage regions in the south that had dominated the textile industry for
centuries lacked the technical skills to adopt the new machinery and slid into terminal
decline. As a result, living standards rose substantially in the north, overtaking the pre-
viously prosperous south where real wages fell markedly. The widespread notion that
average national living standards were static during the Industrial Revolution (Feinstein,
1998) is simply a statistical artefact of aggregating together two regions that were moving
in sharply opposite directions.
Moving on from expensive labour, the claim that the mechanization of the textile indus-
try depended on steam power is also inaccurate. Early steam engines were complicated,
expensive and unreliable contraptions which meant that textile machinery was powered
wherever possible by water. As late as the 1820s, when most weaving was still done by
hand, von Tunzelmann (1978, 295) estimates that the cotton industry derived about as
much energy from human muscle as from coal.
Strikingly, the fact that the early textile industry made little use of coal made it highly
unusual among British industries of the time. By 1750 the British economy was already
distinguished by the way that almost every industrial process involving the transformation
of materials used coal as a source of heat (as opposed to motive power) in contrast to
the wood and charcoal employed elsewhere (Harris, 1976). For centuries coal had been
used to work iron in forges, but a decisive breakthrough came with the appearance of
heat reflecting (“reverberatory”) furnaces in the late sixteenth century. These provided a
general purpose technology to produce clean heat from contaminating coal that came to
be applied to an increasing range of industrial processes, culminating in wrought iron in
the 1780s.
In summary then, the analysis suggests that mechanized textile production in the mid-
nineteenth century would be located in regions that combined low wages and high mech-
anical skills in the mid-eighteenth century; and that widespread iron working would be
found in regions that had been accumulating skill in coal based metallurgy over centuries.
To test these predictions, we analyze the patterns of male employment across the 41 coun-
ties of England (although the Industrial Revolution was very much a British phenomenon,
data for Scotland and Wales are sparse, for early wages in particular).
Our regressions use spatial data, and such regressions often generate spuriously high
t-statistics, both because the included variables display strong directional trends and be-
cause nearby points closely resemble each other leading to effective sample sizes that are
smaller than they appear. The latter issue has received considerable attention in economet-
rics, starting with the kernel adjustment of Conley (1999). Later corrections include large

3
cluster estimators (Bester et al. 2016, Canay, Romano and Shaikh 2017, and Ibragimov and
Müller 2010), and Müller and Watson (2021) principal components (BCH, CRS, IM and
MW). Unfortunately, different adjustments commonly return significance levels that can
vary over an order of magnitude. A CRS correction with, say, eight clusters will often not
only differ markedly from an MW one that imposes an average correlation of 0.01, but each
in turn will differ entirely from CRS with 12 clusters or MW with correlation of 0.05.
In this paper we introduce a simple way to circumvent these difficulties by employing
a semiparametric regression that incorporates a spatial smoothing term. Semiparametric
regressions go back to Engle, Granger, Rice and Weiss (1986) who added to linear regres-
sions a smoothing spline that optimally adapted itself to fit time time trends of unknown
functional form. Despite their elegance, simplicity and power, semiparametric regressions
never took off in economics. However, they have continued to be actively developed in
statistics under the name Generalized Additive Models (Wood, 2017), and have become
popular in machine learning as estimators that are often almost as powerful as black box
techniques but whose results are immediately interpretable (James et al., 2021, 289–310).
Where Engle et al. added a one dimensional spline in time, we add a two dimensional
one in longitude and latitude. This allows us to separate out the spatial structure of the
regression as a nuisance variable and then carry out standard inference on the parameters
that interest us. Using Monte Carlo simulations we find that OLS standard errors are not
only too small but, in the presence of spatial trends in the data, its coefficient estimates
can be severely biased, and this bias can worsen as arbitrary polynomials in longitude and
latitude are added.
For textiles our dependent variable is the share of male employment in textile produc-
tion in 1831. To measure the supply of pre-existing skill we use data from the 1851 Census
on the occupations of workers aged 60 and above born in each county: men who would
mostly have been apprenticed in the late 1790s to masters trained by a previous genera-
tion. We find that the percentage of men working as mechanics and toolmakers, alongside
low wages in the 1760s, explain 70 per cent of the variation in textile employment in 1831.
The effect sizes are substantial: the supply of skilled workers has an elasticity of two, and
wages have an elasticity of minus five. Given the limited use of steam in textiles, proximity
to coal has as little explanatory power as we would expect.
Some explanations of the Industrial Revolution have emphasized schooling, property
rights, or financial markets. To assess the importance of these factors we add estimates of
literacy and number of booksellers for human capital; and number of attorneys for security
of property rights, as well as the density of local banks to assess the importance of access
to external finance. The contribution of these variables is negligible.
The natural reservation about these regressions is that the supply of mechanical skill in
the 1790s might have been endogenous: new textile industries could have attracted skilled
workers rather than vice versa. To control for this, we instrument the employment of
skilled workers with the cost of becoming a watchmaking apprentice between 1750 and
1779. The identification strategy is that there appears to be no direct path connecting the

4
cost of becoming a skilled watchmaker to the number of factory hands labouring in tex-
tile mills several generations later. The apprentice fee instrument, largely supply-driven,
is strongly negatively correlated with mechanical skill, and the estimated elasticities are
little changed from OLS.
Turning to iron working, we analyse employment in 1851. Although a nearby sup-
ply of coal for smelting and forging was necessary for an industry to emerge, it was not
sufficient. Instead, heavy metallurgy concentrated in areas around Birmingham that had
been accumulating expertise in the iron trades since the seventeenth century (Berg, 1994).
Explanatory power is again high.
By way of a control, we analyse the employment patterns of large traditional sectors
where technology was fairly static at this time: food processing, woodworking, garments
and shoes (as Clapham 1939, 169 observed, England in 1831 had many more cobblers than
coal miners). In these sectors the variables that drive the regressions for the progressive
textile and iron manufactures have little explanatory power.
The rest of the paper is as follows. In Section 2 we show the role of artisanal skill in
developing some of the best known technologies of the Industrial Revolution; the role of
coal as a source of heat for metalworking; and the historical supply of skilled workers.
A simple specific factors framework for understanding British industrialization is given in
Section 3 and 4 examines the patterns of growth across regions. Sections 5 to 7 describe our
regression results for textiles, metals, and traditional industries and Section 8 concludes.

2 Skills and the Industrial Revolution

For contemporary observers, the successful development of Britain’s factory system stemmed
from an abundance of long existing artisan skills combined to new purposes. This com-
bination can be seen in late eighteenth century insurance contracts where the value of a
factory is divided into its “millwright’s work” or power, and its “clockmaker’s work” or
machinery.3 Here we briefly discuss the contribution of Lancashire watchmakers to the
development of early textile machinery, and the role of the Birmingham metal trades in
developing Watt’s steam engine.4

2.1 Textile Machinery

Despite the widespread view that the technology of textile production during the Indus-
trial Revolution required little advanced skill or formal knowledge, much of it was any-
thing but simple. Hargreaves’ spinning jenny, the first major advance in cotton spinning,
was indeed a rudimentary, hand-powered machine that could be built by any competent
3
For instance, in 1799 the textile mill of Bissett and Co. was insured for £2,950, made up of £350 for mill-
wright’s work, £950 for clockmaker’s work, with the remainder for buildings and stock (Tann, 1970, 33).
4
Another important group, millwrights, is examined by Mokyr, Sarid and van der Beek (2021)

5
local carpenter.5 However one of the leading Manchester cotton spinner John Kennedy
recalled in 1815, with the appearance of Arkwright’s water frame in 1789 and its intri-
cately meshing metal rollers, spindles, and gearing, “a higher class of mechanics such as
watch and clock-makers, white-smiths, and mathematical instrument makers began to be
wanted; and in a short time a wide field was opened for the application of their more ac-
curate and scientific mechanism” (Kennedy, 1819, 124).6 A particular advantage enjoyed
by the early Lancashire cotton industry was that its easy access to the skills of a large ag-
glomeration of watchmakers and watch-toolmakers whose highly developed division of
labour and specialized tools made them Adam Smith’s chosen example of technological
progress (Allen 2009a, 205–206, Kelly and Ó Gráda, 2016). 7
The soft brass gears of early textile machines were soon replaced by durable cast iron
ones. This meant that their construction was first taken over by iron founders and makers
of large clocks whose facility with heavy lathes and gear cutters readily scaled from brass
to iron (Kelly and Ó Gráda, 2021). Rapidly, however, the large scale of the cotton industry
led to the emergence of firms of specialized machine builders, but the gearing of a textile
machine was still invariably known as its clockwork.8

2.2 Steam Engines

Just as Lancashire’s agglomeration of watch makers was vital to the successful develop-
ment of powered textile machinery, so was the concentration of metal trades in Birming-
ham for that of the steam engine. Although Newcomen had come up with the idea of an
atmospheric engine before 1710 in Cornwall, he could not get it to work for its intended
purpose of pumping mines without the help of “ingenious workmen” from Birmingham
(Desaguliers, 1744, 533). Birmingham metalworking was equally instrumental to the suc-
cess of Watt’s engine. Although Watt had built his first experimental engine in Scotland
in 1768, his progress was frustrated by its flimsy, poorly fitting cylinder until, in 1774, he
moved to Boulton’s large metal works in Birmingham.
Although Britain started with a uniquely large and flexible supply of artisans, the rapid
rise in demand for their services in the late eighteenth century created notable skill short-
ages. Predictably, as the producers of the most complicated large machine of its time, some
5
Allen (2009b) views the jenny, which he deems “not rocket science”, as encapsulating the “Industrial
Revolution in Miniature.” McCloskey (2010, 355–365) too, subscribes to the “practical tinkerers” view. While
this was an apt description of some famous inventors, many others such as William Strutt, John Smeaton,
William Murdoch were connected to practical science and advanced engineering.
6
This view is supported by the large number of newspaper advertisements from the 1770s onward, looking
for smiths, watchmakers, and toolmakers (Musson and Robinson, 1969, 427–458).
7
When Arkwright, assisted by a local watchmaker John Kay, began to build the first spinning frame, a local
engineer “agreed to lend Kay a smith and watch-tool maker, to make the heavier parts of the engine, and Kay
undertook to make the clock-maker’s part of it. . . ” (Baines 1835, 151).
8
The skills that British artisans brought to the cotton industry were more than just mechanical aptitude:
skilled chemists were vital in the development of bleaching powder and soda making (Christie, 2018).

6
of the severest shortages of skills were suffered by Boulton and Watt, resulting in unreli-
able engines, long delivery lags, and a complete absence of after-sales service (Tann, 1970,
83). To compensate, Boulton and Watt pioneered a form of industrial organization of un-
usual sophistication and complexity where a standardized range of engines sizes (allow-
ing a stock of spare parts to be kept on hand for customers experiencing breakdowns) was
made on a systematic production line with a detailed and explicit division of labour (Roll,
1930, 179–184).

2.3 Coal and Heat Skills: Metallurgy

Against the widespread belief that Britain’s large coal resources gave it a cheap source of
motive power for early textile machinery, in reality coal’s early contribution was minor.
Kanefsky (1979, 338) estimated that water power accounted for 70 per cent of industrial
power in 1800 compared with 20 per cent for coal, and was only overtaken by coal around
1830.9 Cotton was woven almost entirely by hand until the 1820s and von Tunzelmann
(1978, 179) estimates that as late as 1800 a quarter at most of Lancashire cotton was spun
using steam.
The reasons are clear: besides paying for fuel (on top of heavy annual royalties to Boul-
ton and Watt), water cost only about one quarter as much per horsepower as steam to in-
stall around 1790 (Chapman, 1970, 241); the efficiency of water wheels improved dramat-
ically through the experimental work of Smeaton and others; and the lifetime of new iron
water wheels was measured in decades compared with years for steam. Steam was only
adopted in the early nineteenth century when the reliability and efficiency of the engines
was improved, machinery was increasingly made of iron rather than wood, and suitable
sites for water power were becoming scarce.
However, the fact that spinning and weaving relied so little on coal before the early
nineteenth century made them highly unusual among British industries at the time. As
Harris (1976) noted, the distinguishing feature of Britain’s economy in the early eighteenth
century was its almost exclusive reliance on coal as a source of industrial heat, rather than
the wood or charcoal employed elsewhere.
Coal had been used since the middle ages for boiling water for brewing beer, extracting
salt, cleaning raw wool and heating dye for textiles. For metal working, the first impor-
tant use of coal from medieval times was for working iron in forges. However, for other
applications, although coal offered a cheap source of intense heat it was very polluting,
contaminating any materials it came in contact with.10
9
It is notable that Wrigley (1988)—who popularized the idea of the British Industrial Revolution as a tran-
sition from an inherently limited “advanced organic economy” of wood and animal power to one powered
by inorganic energy in the form of coal—mentions water power only twice in passing (pages 27, 75), and it is
absent from the index.
10
Abraham Darby came up with the partial solution of using degassed coal or coke that could be used to
make cast iron, but demand for this was limited at first (Hyde, 1977).

7
 The vital generic technology for using coal as a clean source of heat came in the form
of the reverberatory furnace where heat from coal gases was reflected downwards from a
domed roof. As experience and competence in the use of these furnaces grew, coal spread
from smelting copper in the early seventeenth century, to glass making and pottery, and
then in the 1740s to making steel in clay crucibles. Cort’s 1783 puddling process of smelt-
ing wrought iron using a reverberatory furnace was less the revolutionary substitution of
mineral for organic energy that it is often portrayed to be, but “one more conversion” of
fuel from wood to coal. It was for this lack of originality that Cort’s patent was revoked
(Harris, 1988, 26).
The growth of technological competence, then, was a centuries-old evolution where
the skills developed in mastering one process were applied to other, more demanding pro-
cesses. Expertise grew in designing and building furnaces, making refractory bricks and
crucibles, and maintaining exact temperatures, portfolios of interlocking artisanal skills
that could not be readily exported piecemeal.
Many of the tools, instruments, and machines that typified the Industrial Revolution
required high quality iron and steel. Birmingham was known for its extensive variety
of metal trades ranging from forging and casting in the neighbouring Black Country, to
more intricate work like guns, clocks, locks, and mass-produced “toys” (decorative metal
goods). Sheffield, already producing the best files in Europe by 1700 (Harris, 1998, 95),
was the birthplace of crucible steel which gave Britain a unique advantage in the high
quality metal-working tools that were increasingly in demand for shaping machine parts.

2.4 The Market for Artisan Skill

Naturally, England owed its large supply of proficient metalworking craftsmen to many
more things than a fortuitous abundance of coal. On the supply side, the relative limited
power of guilds meant that rapidly growing sectors could swiftly attract extra appren-
tices Ben Zeev, Mokyr and van der Beek (2017).11 On the demand side, from the mid-
seventeenth century the English were becoming an increasingly “polite and commercial
people” with rising prosperity caused by and causing urban expansion, growing over-
seas trade, intensified agriculture, and the improved transportation networks that made
regional specialization possible.
Britain’s relatively large “middling class” meant that the demand for better quality up-
market goods, many involving complicated manufacturing, was on average higher than
elsewhere. The relative affluence of the working class, shown in such things as the quality
of their everyday clothes, was often noted (Styles, 2007, 13), and the fairly decent quality
of their diets resulted in considerably taller, stronger and more productive workers than
elsewhere. This meant that although English workers did earn wages that were substan-
11
See also Leunig, Minns and Wallis (2012) and Ogilvie (2019, chapter 9). The impact of apprenticeship
institutions on the transmission of knowledge is demonstrated by de la Croix, Doepke and Mokyr (2018).

8
tially higher than their French counterparts, they were also more productive: high wages
did not necessarily mean expensive labour (Kelly, Mokyr and Ó Gráda, 2014).12
The need for a region to be close to large concentrations of diverse mechanical skill in
order to industrialize successfully suggests a possible answer to a central question in the
economic history of the Industrial Revolution: Why did some areas of large scale cottage
industry (“proto-industry”) like northern and midland England go on to industrialize
successfully, when others, such as the west of England, southern Ireland, and northern
France, failed (Coleman, 1983)? The decisive characteristic that distinguished winners
from losers, we argue, was a supply of mechanical expertise; a claim that we will test
directly in Section 4 below.

3 Regional Integration and the Accumulation of Skill.

From the early eighteenth century, market integration in England was driven by the rapid
growth of transportation networks, with rivers made navigable and roads, many previ-
ously only usable by packhorses, being turnpiked. Just as important for a large island, but
often overlooked, was the rapid expansion of coastal shipping (Armstrong, 1987).
These large improvements in transport infrastructure allowed regions to specialize ac-
cording to their comparative advantage. Specifically, areas with poor soil began to spe-
cialize in manufacturing activities and to import food. This is most easily understood in
a simple specific factors model where falling food prices reallocate unskilled labour to in-
dustry, and increase the incomes of skilled factors specific to that sector: see Figure 1.
In thinking about the British economy on the eve of the Industrial Revolution, it is use-
ful to divide it into two regions, North and South. Each region has two sectors, agriculture
and manufacturing, and each sector has a specific factor: fertile land for agriculture, and
skilled artisans for manufacturing. There is a common pool of unskilled workers that can
work in either sector. We suppose that the North has less fertile land and a larger supply
of skilled artisans than the South. There is assumed to be no mobility of workers between
regions which again is historically fairly accurate because most internal migration in an
era before railways occurred over short distances (Redford, 1964).
Growing integration leads low-wage areas with large endowments of skilled workers to
specialize in manufacturing while more prosperous areas with high agricultural potential
de-industrialize.13 We begin with a textbook diagram in Figure 1 (depicting the North)
where the supply of unskilled labour gives the length of the x-axis and the labour demand
of each sector is drawn on opposite sides so that equilibrium wage and the employment
levels of each sector are given by the intersection of the curves.
12
At the French firms of Charenton and Le Creuzot, English workers were paid about 50 percent more than
their French counterparts (Belhoste and Woronoff, 2005, 90–91).
13
The classic historical account of this is by Jones (1968). See also Jones (2010).

9
 MPLM pMPLA

W Aut
W Trade

Aut Trade
OM NM NM OA
Mfg. −→ ←− Ag.
Workers Workers

Figure 1: A fall in the relative price of agricultural goods caused by market integration
leads areas with poor soil to specialize in various manufacturing activities. Some of these
activities have the potential to generate new skills of the sort that underlay British indus-
trialization.

As transportation costs fall and trade between regions increases, the relative price of
agricultural goods in the North falls, causing the labour demand curve for agriculture
to fall. This causes the output of manufactures to rise, along with the income of skilled
artisans, given by the triangle between the equilibrium wage and the labour demand curve.
At the same time, the South de-industrializes and increasingly specializes in agriculture.14
The income of skilled artisans will rise further if there is an influx of unskilled work-
ers, who are a complementary input to skilled workers in industrial production (e.g., the
immigration of the Irish settling in the new industrial cities after 1800: Clapham, 1939, 59–
62). This raised their numbers as well, as parents had more of an incentive to apprentice
their sons into skilled occupations.
The result is that a region with poor agricultural productivity will accumulate increas-
ing numbers of artisans with specific manufacturing skills, (besides offering cheap labour
for new manufacturing enterprises) and some of these skills may be conducive to the sub-
14
This simple discussion of historical specialization mirrors the modern literature on structural transforma-
tion (for a review see Herrendorf, Rogerson and Valentinyi 2014), such as Moscona’s (2019) finding that the
Green Revolution reduced urbanization in areas where agricultural productivity rose sharply.

10
 1760s 1830s

Q1
Q2
Q3
Q4

Figure 2: Cartograms of England in the 1760s and 1830s. The area of each county is scaled
in proportion to its aggregate labour income (wage times population), and shaded accord-
ing to its wage rate.

sequent development of new industrial technology of the sort that occurred historically in
parts of England.15 We test these predictions below.

4 The Great Reversal and the Living Standards Puzzle: England

1760–1830
Pre-industrial England fell into three regions. The first was the prosperous, high wage
agricultural region of the south and east which for centuries had been the heartland of
England’s main industry, the manufacture of woollen cloth. Next was the urban giant of
London which, as well as being a port, was a major industrial centre. By 1750, it contained
over 10 per cent of England’s population and was the largest city in Europe (Jones 1968;
Wrigley 2010, 61).
15
In the working paper version we analyze a process where market integration leads to a mutually reinforc-
ing growth of skill and production technology through parental investment and learning by doing, and show
how this process can lead to a sudden transition from a low technology-skills steady state to a high one.

11
 The final region was the upland North and West. Despite low wages reflecting the re-
gion’s poor agricultural potential, its population had been growing rapidly since the seven-
teenth century in response to the widespread non-agricultural employment opportunities
offered by outwork and small-scale cottage industry. As well as fast flowing streams to
provide water power this region had ample supplies of “cheap and amenable female and
child labour” (Humphries, 2013) that eventually became a vital input into the early factor-
ies of the Industrial Revolution, combined with a fairly well nourished population for un-
dertaking heavy physical labour (Horrell and Oxley, 2012). Above all, this region also pos-
sessed a flexible supply of highly competent workers with useful mechanical skills—clockmakers,
mechanics, toolmakers—who would play a key role during the Industrial Revolution, and
several of whom would become inventors and factory owners in their own right (Cookson,
2018).

4.1 The Great Reversal

Figure 2 shows a map of England where the counties are re-scaled in proportion to their
aggregate labour income: the wage of agricultural labourers times population. Counties
are shaded according to the wage rates of agricultural labourers in each period. Given the
absence of any restrictions on occupational mobility these are likely to have been close to
the wages earned by unskilled labourers in other sectors.
In the 1760s, Figure 2 shows that the English economy was still dominated by London
and its environs, and that southern wages were higher than northern ones, reflecting their
higher agricultural productivity. By the 1830s we observe what can only be called a great
reversal: Northern counties that were in the bottom quartile of wages are now in the top;
and the aggregate income of the textile areas of Lancashire and West Yorkshire has be-
come as large as London’s. At the same time manufacturing in the south, London and its
environs excepted, sharply declined, a phenomenon described by Jones (2010, pp. 47-50)
as “the anomaly of the South”. While these changes took two full generations to be com-
pleted, they underline that the Industrial Revolution, while perhaps not as dramatic as the
word “revolution” conjures up, brought about radical structural changes in the economic
geography of Britain.

4.2 The Standard of Living Puzzle.

This analysis also sheds light on one of the more durable puzzles of the British Industrial
Revolution, the supposed failure of real wages to increase appreciably despite rapid tech-
nological progress and industrialization (Mokyr 1999, 113–116; Feinstein 1998). Figure
2 shows that the puzzle is in large measure a statistical artefact caused by looking at na-
tional wages rather than regional ones. National real wages were indeed static between
1760 and 1830: weighted by population, the average national money wage rose by 50 per
cent, as did the national CPI estimated by Clark (2011). However, this disguises the 80–90

12
per cent rise in nominal wages in industrializing counties, compared with only 15 to 25
per cent in agricultural ones, so that northern wages not only caught up on southern ones
but overtook them.
Wage dispersion remained constant with a coefficient of variation of 13 per cent in
both the 1760s and 1830s: the poorest counties in the 1760s had wages that were 70 per
cent of the highest ones (for comparison, average French wages at the time were 80 per
cent of English ones). One possible caveat here is that differences in the cost-of-living
might account for some of the regional variation in wages. Yet Crafts (1986, 68) and Hunt
(1986) have shown that regional cost-of-living differences in the 1840s were minor. What
of earlier? Frederic Eden’s The State of the Poor (1797) is a comprehensive source on regional
price variations: it indicates little difference between the cost of provisions in northern and
southern counties in the mid-1790s; if anything prices seem to have been slightly higher in
the north.
The differential labour demand that drove these wage rises led to very different pat-
terns of population growth. Between 1761 and 1831, the population of the depressed agri-
cultural counties in the south and east grew only 25–33 per cent, whereas that of the in-
dustrial counties and those around London more than doubled, with that of Lancashire
more than quadrupling.

5 Regression Results: Textiles.

Although the regional specialization model outlined above is extremely simple, it makes
the very specific and testable prediction that areas that industrialize successfully will have
had poor agricultural potential combined with high endowments of skills that could be
applied to new textile manufacturing activities.

5.1 Data
The specific factors model predicts that before the growth of industry, low agricutural
potential translated into low wages. This turns out to be the case for England. For agri-
cultural wages in the 1760s the correlation with the median county level of suitability for
wheat, estimated by the FAO, is 0.5; with the average land tax per acre in 1707 (leaving
out London) is 0.6; and their correlation with the age of the dominant rock type in each
county—hard, ancient rock weathers into less fertile soil—is 0.7. We report results below
using agricultural wages as the explanatory variable, but using any of these three measures
of agricultural potential gives similar results.
Although it is straightforward to measure wages and agricultural potential, the chal-
lenge comes in measuring the availability of skill at the beginnings of industrialization. We
do have detailed data on the supply of one type of skilled artisan in the mid-eighteenth cen-
tury: watch and clock-makers, where the records of the London Watchmaker’s Company
(guild) detail every one of its apprentices during the eighteenth century (Moore, 2003).

13
 Textile Employment 1831 Agricultural Wage 1767 (Inverse)

Q1
Q2
Q3
Q4
Q5

Mechanics and Toolmakers 1790s Watchmaking Apprentice Fee 1750−79 (Inverse)

Figure 3: County levels, shaded by quintile, of textile employment in 1831, the inverse
wage of agricultural labourers, the share of mechanics and toolmakers in the labour force
in the 1790s, and the inverse cost of becoming a watch-making apprentice. The paper pre-
dicts that successful textile areas in the nineteenth century will have had low wages and
high levels of mechanical skill in the eighteenth century. The cost of becoming a watch-
making apprentice is used as an instrument to deal with potential issues of endogeneity
and measurement error in the mechanics and toolmakers variable.

14
However, as Cummins and Ó Gráda (2022) demonstrate, roughly half of English watch-
makers never apprenticed to the guild (and about 80 per cent in the main watchmaking
region Lancashire), making this a potentially unreliable measure.
To test our hypotheses, we instead take advantage of the fact that the 1851 census details
the numbers of workers in each occupation broken down by age. By examining elderly
men (aged sixty and over, most of whom would have been apprenticed around age 14
in the late 1790s) we can get an idea of the geographical availability of skill at an earlier
stage of the Industrial Revolution.16 For nearly every county and every skill, the number of
these men with a particular skill residing in a given area closely matches the cohort size of
men with the skill born in that county, suggesting that most of these skilled workers were
apprenticed locally and that inter-county migration does not confound the analysis.17
We focus on the share of men over sixty born in each county who had potentially useful
skills. Specifically we look at blacksmiths, millwrights (both traditional skills), watch- and
instrument makers, gunsmiths and locksmiths, toolmakers, sheet-metal workers, and me-
chanics. These last workers made, assembled and maintained the machinery we associate
with the Industrial Revolution.
Given their historical importance for early industrialization it might seem surprising
at first that the number of watchmakers and lock- and gun-smiths has little explanatory
power. It is important to remember, however, that specialized industrial skills were trans-
formed and adapted rapidly: many of the men in our sample may have been trained by
masters who started out as watch tool makers or millwrights but by 1851 were making a
living as industrial tool makers or machine builders (Mokyr, Sarid and van der Beek, 2021).
Terms such as millwright and blacksmith, moreover, by 1851 meant different things in dif-
ferent places. In agricultural areas they were largely engaged in traditional practices of
maintaining water mills, and making farm implements and shoeing horses respectively;
whereas in industrial areas millwrights were increasingly morphing into engineers, and
blacksmiths forged machinery parts.
We have data for textile employment from two censuses, 1831 and 1851. What we ob-
serve in Table 1 is that the distribution of textile employment is largely unchanged during
this time: the regression results are very similar except that the coefficients in 1851 are
around one standard deviation smaller.
The broad patterns of the data are shown in Figure 4 which plots textile employment
share in 1851 against the supply of mechanical skill in the 1790s, and the agricultural wage
in the 1760s. It is immediately evident that successful regions were those that had com-
bined low wages with high mechanical skills allowing them to adopt new machinery.
The points for Gloucestershire (GLC) and West Yorkshire (YWR) are particularly re-
vealing. In the mid-eighteenth century Gloucestershire and the west of England domi-
nated the English woolen textile industry (other centres like Norfolk and Suffolk had al-
16
Using men aged fifty and over gives practically identical results.
17
Historical research finds that most factory workers were recruited from the surrounding rural hinterland
surrounding the new industrial centers (Redford 1964, Anderson 1971).

15
 NOT YWR
YWR
20.0 LCR NOT LCR
DRB
CHE DRB CHE
10.0 CUM CUM
Textile Employment Share 1851.

WML
WRC WRC
WML
NRF
5.0 WTS GLC WRW WRW SMS GLC WTS
NRF
SMS DRH DRH
YNR MSX YNR
NHB STF STF BER
OXD DRS DRS OXD
SUR BER NHB SUR
DVN NHP NHP
1.0 SFF
ESE DVN
SFF
HTF HTF ESE
0.5
YER YER

HRF RTL
LNC HRF RTL LNC

CMB CNW BUC HNT CNW CMB HMP KNT

HMP SUS BED KNT SHP SHP BED HNT BUC SUS
20 40 60 80 80 90 100 110
Mechanical Skills 1790s. Wage 1760s.

Figure 4: Supply of mechanical skill in the 1790s, and agricultural wages in the 1760s
versus the percentage of males employed in textiles in 1851. Logarithmic axes. Note how
the decline of Gloucestershire (GLC) (the dominant woollen textile centre in the mid-
eighteenth century) and the rise of West Yorkshire (YWR) can be predicted by their re-
spective supplies of mechanical skills in the 1790s.

ready started to decline) but it failed to mechanize and by 1851 had become a comparative
backwater, with the industry dominated by the factories of West Yorkshire. Although the
West Country around Gloucestershire was actually a large producer of charcoal iron, in the
absence of coal this iron had to be sent to be worked in Birmingham so that the region failed
to develop the skilled labour that its textile industry needed to mechanize successfully.18
Similarly, Jones (2010, 86) observes that the industrial decline of the west of England owed
little to coal prices. Coal prices were indeed one third higher in the west of England com-
pared with Yorkshire but coal accounted for only a small fraction of production cost.19

5.2 OLS Results

Table 1 gives the results of regressing textile employment in 1831 and 1851 on a variety of
possible explanatory variables. First is mechanical skill in the 1790s which we will treat
as exogenous for now. Next we have agricultural wages to test if high wages in the 1760s
drove regions to substitute machinery for expensive labour, or whether regions of poor ag-
ricultural potential and low wages came to specialize in manufacturing as national markets
integrated.
18 16coalfield nearby, no heavy iron or engineering indus-
In the words of its leading historian “With no large
tries, and no other local industry requiring precision engineering, the area lacked a pool of skilled labour to
draw upon...” (Tann, 1974).
19
Even in 1870, the average cotton mill was driven by a 120 horsepower engine (less than most small cars):
Samuel (1977, Table 1).
 Dependent variable: Share of men employed in textiles.
1831 1851

Skills 1790s 2.022 2.466 2.020 1.151 1.079 1.007

(0.503) (0.668) (0.511) (0.398) (0.532) (0.377)
Wage 1760s -4.776 -5.402 -4.783 -3.035 -3.004 -1.623
(2.742) (2.817) (2.937) (2.109) (2.138) (2.056)
Mkt Potential 1750 0.712 0.592 0.712 0.503 0.514 0.403
(0.301) (0.320) (0.306) (0.217) (0.227) (0.209)
Distance to Coal 0.720 -0.127
(0.685) (0.650)
Water Power 0.001 -0.664
(0.425) (0.348)
Adj. R2 0.684 0.695 0.676 0.813 0.810 0.865
AIC 131.1 130.6 132.9 88.7 89.5 74.3
S(lon, lat) 0.641 0.534 0.768 0.070 0.076 0.013

Dependent variable is the employment share of textiles among all males over 20 in 1831,
and among males aged 20–29 in 1851. Semiparametric regressions with a thin plate spline
in longitude and latitude: S(lon, lat) gives the approximate significance level for this
term. Standard errors in parentheses. Regressions include a dummy in 1831 for Shrop-
shire, and in 1851 for Staffordshire. All other variables are in logs. 41 observations.
Table 1: Textile employment in 1831 and 1851, semiparametric spatial regressions.

Given the importance of market integration to our story, we include the size of the po-
tential market (measured as the product of 1760s wages and 1750 population, with weights
declining with the square of distance) emphasized by Crafts and Wolf (2014). In addition
we include proximity to coalfields, and water flow. Other factors are considered below.
In any regression using spatial data there is a real possibility that the results are spuri-
ous artefacts of underestimating standard errors or fitting directional trends. There are
many of standard error corrections that attempt to control for the fact that the effective
sample size can be smaller than it appears because many observations closely resemble
their neighbours. These include the Conley (1999) kernel adjustment; the large cluster
procedures of Bester, Conley and Hansen (2011), Canay, Romano and Shaikh (2017), and
Ibragimov and Müller (2010); and the principal components approach of Müller and Wat-
son (2021). However, as we demonstrate in Appendix A, these corrections commonly
return significance levels that not only vary widely between estimators (sometimes by an
order of magnitude or more), but are often highly sensitive to the assumed value of their

17
 Dependent variable: Share of men employed in textiles.
1831 1851

Literacy c1800 -1.671 (1.937) -0.140 (1.580)

County Banks 1796 -0.064 (0.431) -0.342 (0.368)
Lawyers 1730 0.016 (0.444) -0.378 (0.386)
Booksellers 1761 0.023 (0.358) 0.155 (0.294)
Adj. R2 0.649 0.652
AIC 137.6 114.1
S(lon, lat) 0.682 0.709

Semiparametric spatial regressions. Dependent variable is the em-

ployment share of textiles among all males over 20 in 1831, and
among males aged 20–29 in 1851. Standard errors in parentheses.
The approximate significance of the thin plate spline is given by
S(lon, lat). Regressions include skill, wages, market potential, and
county dummies from first columns of Table 1. Literacy is the per-
centage of convicts that were literate. Density is the density of 1700
population relative to county farmland. Booksellers is the number
of booksellers, lawyers is the number of attorneys, and banks is the
number of country banks, all per capita.
Table 2: Other explanatory variables for textile employment, semiparametric spatial re-
gressions.

tuning parameters. For these adjustments to be useful they require moreover that low
frequency spatial trends have already been properly removed from the the regression to
ensure unbiased coefficient estimates.
Given the infeasibility of current approaches to spatial regressions we adopt instead a
simple semiparametric approach in the spirit of Engle et al. (1986). Alongside the explan-
atory variables of interest we add a thin plate regression spline in longitude and latitude
that optimally absorbs the spatial structure—both low frequency directional trends and
local correlation—and allows us to proceed with standard inference techniques. We out-
line the approach in Appendix A and provide Monte Carlos to demonstrate its accuracy.
In particular we find that OLS standard errors are not only too small but, in the presence
of spatial trends in the data, its coefficient estimates can be severely biased.
What Table 1 indicates is that the growth of textile production occurred in areas with
vigorous supplies of skilled metalworkers, and low wages, alongside access to substantial
markets. The size of the coefficients is notable: the elasticity of textile employment in 1831
with respect to skill supply is in the region of 2, while wages have an elasticity of around

18
−5 although the associated standard error is large. Market potential has a considerably
smaller elasticity of 0.6. Given the limited early use of steam, the unimportance of coal,
once we control for skills, is not surprising. For 1851 the elasticity of employment with
respect to skill has fallen to around 1, and the wage elasticity has declined somewhat to
−3.
The S (lat, lon) term gives the approximate significance level of the penalized spline in
longitude and latitude. For the 1831 regressions this has little explanatory power, whereas
for 1851 there is a somewhat stronger spatial pattern to the observations. If we compare the
semiparametric estimates in Table 1 with the OLS ones reported in Table 3 below, there is
little material difference between them. One potential complication with semiparametric
estimation is that any spatial component in an explanatory will get confounded with the
spatial component in the regression equation but this can be handled by a standard two
step procedure outlined in Appendix A. The two step estimates do not differ noticeably
from the unadjusted semiparametric ones and are not reported here.
Table 2 considers a variety of other factors often mentioned in discussions of the causes
of the Industrial Revolution, adding extra variables to the base regressions in the first
columns of Table 1. These are literacy around 1800 and booksellers per capita in 1761,
as measures of human capital. The number of lawyers per capita in 1730, which largely
reflects the intensity of local land enclosures (Aylett, 1987), is added as a measure of the
security and easy transfer of property rights.
Population density is the ratio of population in 1700 to agricultural land, reflecting the
hypothesis that areas of cottage industry that were able to support large non-agricultural
populations developed the skills and attitudes that subsequently drove industrialization
(Coleman, 1983). If our argument is correct, proto-industrial areas that lacked mechanical
skills failed to industrialize so this variable should have no impact.
Finally, we include the number of country banks in the 1790s. Quite apart from the
need for trade credit, if industrialization was driven by a desire to replace expensive work-
ers with machinery, this investment would have been facilitated in areas with extensive
banking networks (Pressnell 1956; Neal 1994). The reported coefficients are for a regres-
sion that also includes wages, mechanical skill, and market access. It can be seen that none
of these variables adds any noticeable explanatory power.

5.3 Instrumental Variable Results

The natural concern with these OLS results showing how skill supply is strongly correl-
ated with textile employment is that the supply of skills was endogenous: new industries,
even as early as the 1790s, may have encouraged inward migration of skilled workers, or
caused men in traditional industries, like millwrights and blacksmiths, to become special-
ized machine builders. To address this concern, as well as the possibility that our skills
variable is mismeasured, we employ an IV strategy where we instrument the measure of
skills derived from old skilled workers in the 1851 census.

19
 80
NOT
YWR LCR

STF DRB
60 LCS
WRW
Mechanical Skill 1790s.
40
SHP WRC
DRH
GLC CHE
MSX BER
SUR NHB
KNT
CUM HNT
WML BED YER RTL
YNR DRS SMS
20
NHP
BUC WTS
NRF
DVN SFF CNW
HTF HRF
LNC ESE SUS
CMB
OXD HMP

10 20
Apprenticeship Cost 1750−1780.

Figure 5: Apprenticeship fees for watchmaking 1750–1780, and the supply of mechanical
skill in the 1790s. Logarithmic axes.

Our instrument for skill supply is the median fee charged to become an apprentice
watchmaker between 1750 and 1779 taken from the records of the Watchmakers’ Company,
assembled by Moore (2003). The justification for this instrument is that there would ap-
pear to be no direct path linking the cost of becoming a watchmaker in the mid-eighteenth
century with the number of mostly unskilled factory hands in textile mills over half a cen-
tury later.
At the same time we would expect these fees to be a strong proxy for the supply of
mechanical skill on the eve of the Industrial Revolution. As Ben Zeev, Mokyr and van der
Beek (2017) show, similar trades demanded similar fees, so we can be quite confident that
counties that had a larger supply of masters making things like locks, guns, and instru-
ments would charge lower fees in the market for apprenticeship. The close relationship
between mid-eighteenth century apprenticeship fees and the number of skilled workers
trained in each county around 1800 is apparent in Figure 5.

20
 Dependent variable: Share of mechanics in 1851.

Apprentice Fee -0.570 -0.491 -0.531

(0.239) (0.164) (0.133)
Apprentice Fee/Wage -0.475
(0.138)
Land Tax 1707 -0.085 -0.080 -0.083
(0.141) (0.115) (0.120)
Pop Density 1700 0.818 0.552 0.533
(0.144) (0.133) (0.139)
Distance to Coal -0.596 -0.603
(0.140) (0.147)
Adj. R2 0.533 0.841 0.898 0.888
AIC 30.0 -13.0 -31.8 -27.7
S(lon, lat) 0.105 0.004 0.003 0.005

Dependent variable is the share of males aged 60-69 listed as mechan-

ics in 1851, by county of birth. Semiparametric spatial regressions:
S(lon, lat) gives approximate significance level of a thin place spline
in longitude and latitude. Standard errors are parentheses. Regres-
sions including population density include a dummy for London. All
other variables are in logs.
Table 3: Factors affecting the supply of mechanical skill, semiparametric spatial regres-
sions.

The determinants of skill supply are analysed in Table 3. It indicates that the share of
mechanics in 1851 was strongly predicted by apprenticeship fees and proximity to coal,
and population density relative to farmland in 1700, an indicator of the presence of proto-
industry. Land quality measured by average land tax per acre in 1700 has a smaller impact.
It might be argued that low apprenticeships fees are simply a proxy for low wages but, as
the final column shows, the results are little changed if the ratio of apprenticeship fees to
wages is used.
Results for the IV regressions are given in Table 3 for textile employment in 1831 and
1851, alongside OLS and semiparametric estimates. The IV standard errors are somewhat
larger but, as the Wu-Hausman test indicates, the coefficient estimates change little after
instrumenting, suggesting that the tighter OLS confidence intervals would seem to be ap-
propriate.

21
 Share of men employed in textiles.
1831 1851

SP OLS IV SP OLS IV
Skills 1790s 2.022 2.125 3.178 1.151 1.703 2.466
(0.503) (0.459) (1.422) (0.398) (0.332) (1.119)
Wage 1760s -4.776 -6.960 -5.169 -3.035 -4.908 -3.686
(2.742) (1.599) (2.634) (2.109) (1.136) (2.026)
Mkt Potential 1750 0.712 0.713 0.484 0.503 0.499 0.340
(0.301) (0.228) (0.481) (0.217) (0.188) (0.392)

Adj. R2 0.684 0.655 0.606 0.813 0.613 0.567

Wu-Hausman . . 0.286 . . 0.388
Weak Instruments . . 0.007 . . 0.015
S(lon, lat) 0.641 . . 0.070 . .

Dependent variable is the employment share of textiles among all males

over 20 in 1831, and among males aged 20–29 in 1851. Apprentice fees
are used as the instrument for mechanical skills. SP denotes spatial semi-
parametric regressions. Heteroskedasticity robust standard errors are in
parentheses. Regressions include longitude and latitude and a dummy in
1831 for Shropshire, and in 1851 for Staffordshire. All other variables are
in logs. Weak Instruments, and Wu-Hausman are the significance levels
of tests for weak instruments, and differences in OLS and IV coefficients.
Table 3: Textile Employment in 1831 and 1851: Semiparametric, OLS and IV.

6 Regression Results: Metallurgy

The indispensable role of coal in early British industrialization beyond being a source of
heat, we have argued, is in the array of metalworking skills—ranging from blacksmith
work and iron smelting, to casting watch springs and pinions—that it allowed to accumu-
late. Here we examine three heavy metalworking classifications: metal manufacturing,
metal products, and sheet metal. Whereas the 1831 categories for textile employment are
systematic, those for metalworking activities are chaotic reflecting the novelty of many of
these activities, so we restrict ourselves to 1851 data.
The relationship between aggregate employment in metal industries and coal distance
is shown in Figure 6. Naturally, the relationship with coal proximity is strong but, what is
equally important and again somewhat disguised by the logarithmic axes, is the concentra-
tion of the industry in three neighbouring counties in the West Midlands—Staffordshire,
Warwickshire (which includes Birmingham), and Worcestershire— and to a lesser extent

22
 WRC
STF
15 WRW

Employment in metallurgy 1851. 10 YWR

DRH
NHB
DRB
SHP

LCS

5
CHE CNW
MSX
YER
CUM GLC SFF
DVN
NOT LNC HMP
HRF SMS DRS
YNR WTS NRF SUS
WML BER
LCR ESE
RTL BED HTF KNT
NHP OXD SUR
CMB
HNT BUC
20 40 60 80 160
Distance to coal.

Figure 6: Distance to coal and employment in metallurgy 1851. Logarithmic axes. It can be
seen that proximity to coal was a necessary but not sufficient condition for a large industry:
the three west midland counties around Birmingham, and West Yorkshire around Sheffield
had been accumulating skill in metal working since at least the sixteenth century.

in West Yorkshire, around Sheffield, where suitable coal had made them centres of metal-
working skill since at least the seventeenth century.
Table 4 shows regression results for these three activities. It can be seen that for all
three activities, market potential, paired to being in the traditional centres of the West
Midlands are strong predictors, especially for metal manufacturing. After controlling for
these, proximity to coal (with or without a quadratic term) has no explanatory power. For
randomized standard errors, the significance of location for the first three activities was
so strong that they were unmatched in any simulation, so the Heteroskedasticity robust
values are reported instead. The mechanical skill variable adds no explanatory power ex-
cept for metal products. Despite their importance to the modern sector, the metallurgical
industries even as late as 1851 still depended heavily on traditional skills in forging and
smelting, and the mechanical aptitude that was required in the construction and mainten-

23
 Dependent Variable: Employment in Metallurgy, 1851.
Total Metal Mfg Metal Prods Sheet Metal

Traditional Metal Area 1.197 1.966 1.114 0.864

(0.159) (0.411) (0.145) (0.283)
Distance to Coal -0.145 -0.185 0.039 -0.244
(0.166) (0.293) (0.121) (0.299)
Mkt Potential 1750 0.139 0.558 0.097 0.383
(0.065) (0.188) (0.059) (0.114)
Skills 1790s 0.314 0.444 0.323 0.138
(0.144) (0.359) (0.128) (0.257)

Adj. R2 0.874 0.756 0.819 0.670

AIC -4.8 78.5 -6.4 40.8
S(lon, lat) 0.071 0.056 0.077 0.218

Dependent variable is the share of males aged 20-29 employed in me-

tallurgy. Standard errors in parentheses. Semiparametric regressions:
S(lon, lat) is the approximate significance of a thin place spline term in
longitude and latitude. Traditional Metal Area is a dummy for three West
Midland counties: Staffordshire, Warwickshire, and Worcesteshire. Metal
Manufacturing contains a dummy for Rutland. All other variables are in
logs.
Table 4: Determinants of employment in metallurgy in 1851.

ance of textile machinery were less important for the location of this industry: different
skills mattered. 20
The importance of metalworking skills was early on recognized early on by James Watt.
In a letter to a fellow inventor in 1785 he wrote that “you ask what is the principal hin-
drance to erecting engines? It is always smith-work.” Watt found the skills he needed in
Birmingham, where business partner Matthew Boulton had promised him artisans who
could work “with as great a difference of accuracy as there is between the blacksmith and
the mathematical instrument maker.” In the words of the statesman Richard Cobden, “Our
strength, wealth, and commerce grow out of the skilled labour of the men working in met-
als. They are at the foundation of our manufacturing greatness.” 21
20
The other new industry reliant on coal for heat in the eighteen century was pottery but it was overwhelm-
ingly concentrated in Staffordshire, which was a major centre for centuries before Josiah Wedgwood pioneered
the mass production of quality tableware.
21
The quotations are from Smiles (1865, 179,203) and Smiles (1863, 331).

24
 Dependent Variable: Share of men employed in traditional industries, 1851.
Food Garments Shoes Wood

Skills 1790s -0.042 0.154 0.025 0.028

(0.054) (0.112) (0.084) (0.152)
Wage 1760s -0.600 -0.652 -0.523 0.290
(0.314) (0.654) (0.498) (0.885)
Mkt Potential 1750 -0.010 -0.021 0.028 -0.116
(0.034) (0.070) (0.051) (0.095)

Adj. R2 0.476 0.372 0.728 0.017

AIC -48.1 11.9 -23.0 36.4
S(lon, lat) 0.000 0.312 0.692 0.780

Dependent variable is the employment share of males aged 20–29 in 1851.

Semiparametric spatial regressions: S(lon, lat) gives the approximate sig-
nificance level of a thin plate spline in longitude and latitude. Standard er-
rors in parentheses. Regressions include a dummy for London, and Shoes
includes one for Northhamptonshire. All other variables are in logs.
Table 5: Employment shares of traditional industries in 1851, semiparametric spatial re-
gressions.

7 Regression Results: Traditional Industry

Having discussed textiles and metallurgy, the two sectors almost synonymous with the
Industrial Revolution, as a falsification test we consider how well the factors we have em-
phasized explain the location of traditional manufacturing activities: in other words, can
they appear to explain employment in sectors that they should not be able to explain? The
variables we stress here (mechanical skills and low wages for textiles, and a tradition of
using coal as a source of thermal energy for metallurgy) explain modern industry but, as
Table 5 shows, they do not correlate systematically with traditional manufacturing sectors
whose technology was relatively static during our period. Table 5 shows that for the major
traditional sectors of food, garments, shoes, and woodworking, wages are the only variable
important variable for food and garments, and skills have no explanatory power.

8 Conclusions.
Mechanical and related skills, then, were crucial to the success of the Industrial Revolu-
tion. Britain’s advantage on other European nations in this respect explains why it was

25
the technological leader of Europe for over a century. The question why this was so has
a number of components. First and foremost, as already noted, regional specialization
due to market integration stimulated the demand for artisan skills in the regions special-
izing in those products. Second, the apprenticeship system in Britain was more flexible
and market-oriented than elsewhere, and the institutional structure that enforced the in-
stitution was more effective (Mokyr 2019, 2021). Third, the lower inequality of income
distribution in Britain in the first half of the eighteenth century meant that there was con-
siderable demand for middle-class luxuries that demanded high levels of skills. Finally,
as argued in Kelly, Mokyr and Ó Gráda (2014), the average quality of the entire British
labor force was higher than on the Continent.
If we assume, with some simplification, that the quality of mechanical competence can
be summarized in a single variable that is distributed symmetrically over the labor force,
it is a characteristic of symmetrical distributions that fairly small differences in the means
of two populations are amplified to much larger differences in the density of the respec-
tive upper tails. Hence the very top artisans of Britain, what has been termed upper tail
human capital, had considerably more mass than their continental neighbors, providing
Britain with a serious advantage in an age when such artisans and their tacit knowledge
of mechanical skills may have mattered more than either before or after.
For a generation now the debate on the origins of the Industrial Revolution itself has
been overshadowed by the realisation that its macroeconomic impact was at first mod-
est, reflecting the fact that the sectors that grew fastest had started out small (Crafts and
Harley, 1992). From this spread a widespread belief that the Industrial Revolution was
less of an epochal change in human history than a narrow event confined to a few sectors
like cotton, iron, and steam in an economy that was otherwise fairly static.22
Although our empirical analysis here has centred on textiles and iron, this narrow view
of the Industrial Revolution has increasingly been recognized as untenable. First, Broad-
berry et al. (2015) have demonstrated that slow but persistent output growth across a
broad range of industrial sectors was already under way by the late seventeenth century.
Second, technological change is increasingly seen as sustained improvements, most of
them anonymous and incremental, across many important activities—as varied as watch-
making, shipping, ceramics, glass-making, brewing, road transport, paper making, candle-
making, gas lighting, water power and machine tools—in many cases starting in the early
eighteenth century.

Appendix A Spatial Standard Errors and Semiparametric Regres-

sions
In this Appendix we show first how current spatial standard error corrections tend to re-
turn such widely differing estimates as to be of limited utility in practice, and then intro-
22
For early dissenting views, see Berg and Hudson (1992) and Temin (1997).

26
duce a simple way to estimate accurate results by using a semiparametric regression that
includes a spatial spline.

A.1 Standard Error Corrections

Regressions with spatial data frequently generate spuriously large t-statistics for two rea-
sons. Firstly, geographical variables often show strong directional trends (for example,
per capita income rises as one moves away from the equator). Secondly, because observa-
tions tend to resemble their neighbours, effective sample sizes are smaller than they appear
leading to underestimated standard errors.
Starting with the kernel adjustment of Conley (1999), several standard error correc-
tions have been proposed including the large cluster methods of Bester et al. (2016), Ibrag-
imov and Müller (2010) and Canay, Romano and Shaikh (2017); and the principal compo-
nents approach of Müller and Watson (2021). However, all of these adjustments require
the user to set, more or less arbitrarily, the value of a tuning parameter value (the correla-
tion range, the number of clusters, or the average correlation between points respectively)
and as Table A1 shows, the estimates can be extremely sensitive to small changes in these
assumed parameters.
The Table gives the significance level of the main explanatory variable in three well
known historical persistence studies for Africa, Germany, and countries around the world.
Successive rows contain increasing polynomials in longitude and absolute latitude varying
in degree from zero to quadratic. It is immediately evident that almost any significance
level from 0.01 and higher may be obtained by an appropriate choice of estimator, assumed
parameter, and polynomial in longitude and latitude.

A.2 Semiparametric Spatial Regressions

The approach that we propose here is to view spatial structure, both trends and local cor-
relation, as an issue not of standard errors but of specification. The standard approach
with spatial data is to run a regression of the form

yi = Xi β + ηi (1)

for observations at geographical sites i = 1, . . . , n where η ∼ N (0, Σ), and then to attempt
to adjust standard errors for spatial correlation. However, leaving standard errors aside,
the coefficient estimates β̂ will only be unbiased if (1) is correctly specified.

27
 Table A.1: Significance levels of the main explanatory variable in three regressions using different standard error
corrections. In each case significance levels from 0.01 upwards may be obtained by an appropriate choice of correction
and assumed tuning parameter, along with the degree of the included polynomial in longitude and latitude.

Conley BCH CRS IM MW

Degree Robust Clustered 50 100 4 6 8 12 8 12 0.01 0.05
Africa. N = 379
0 0.000 0.001 0.000 0.000 0.084 0.032 0.031 0.020 0.174 0.300 0.005 0.046
1 0.000 0.002 0.000 0.000 0.084 0.032 0.125 0.964 0.103 0.415 0.004 0.043
2 0.000 0.006 0.001 0.001 0.168 0.058 0.352 0.963 0.256 0.465 0.013 0.103
Germany. N = 324
0 0.008 0.054 0.014 0.034 0.185 0.095 0.508 0.613 0.486 0.620 0.103 0.109
1 0.014 0.047 0.018 0.050 0.304 0.108 0.805 0.127 0.808 0.178 0.106 0.167
28

2 0.009 0.041 0.015 0.055 0.287 0.179 0.961 0.340 0.943 0.275 0.128 0.228
Global. N = 114
0 0.000 0.032 0.000 0.001 0.319 0.190 0.016 0.999 0.273 0.717 0.059 0.288
1 0.005 0.037 0.004 0.006 0.259 0.139 0.211 0.692 0.300 0.735 0.022 0.123
2 0.007 0.001 0.006 0.010 0.122 0.068 0.805 0.834 0.526 0.952 0.042 0.065

Significance levels for different standard error corrections. Degree gives the degree of polynomial in longitude and absolute latitude added
to the regression, from none to quadratic. Robust denotes heteroskedasticity robust standard errors. Cluster gives clustered standard
errors when the observations were divided into twenty groups of nearest neighbours by k-means clustering. Conley denotes Conley
(1999) standard errors with a Bartlett kernel and cutoff distances of 50 and 100 km for the German data and 500 and 1000 otherwise.
Bester, Conley and Hansen (2011), Canay, Romano and Shaikh (2017) and Ibragimov and Müller (2010) (BCH, CRS, and IM) are based
on large clusters, again chosen by k-means clustering, with the assumed number of clusters given below each. MW gives Müller and
Watson (2021) values assuming average correlation values between residuals of 0.01 and 0.05.
 In general however, any spatial regression will omit many factors that each has a direc-
tional trend and local correlation, so a less restrictive specification is

yi = g (si ) + Xi β + i (2)

where  ∼ N 0, σ 2 I and g is some unknown function of si = (s1 , s2 ), the longitude and

latitude of the observations, that reflects the impact of the omitted, spatially correlated
explanatory variables. This semiparametric approach allows the spatial structure of the
regression to be separated out as a nuisance variable when estimating the coefficients of
interest β.
Suppose initially that yi = g (si ) + i . Thin-plate spline smoothing estimates g by find-
ing the function fˆ that minimizes ky − f (s)k2 + λJ (f ) where J (f ) is a function that pe-
nalizes overfitting, measured by changes in slope, and λ is a smoothing parameter that
controls the tradeoff between data fitting and the smoothness of f . Specifically, the thin
plate spline fˆ is the solution to the constrained minimization problem
ˆ
 !2 !2 !2 
X ∂ 2 f (s) ∂ 2 f (s) ∂ 2 f (s) 
min (yi − f (si ))2 + λ +2 + ds (3)
f 
i
∂s21 ∂s1 ∂s2 ∂s22 

The equation of fˆ is given in Wood (2017, 216). The optimal value of the roughtness
penalty λ is chosen by a generalized cross-validation.23
To illustrate the reliability of this semiparametric procedure compared with standard
error corrections and polynomials in latitude and longitude we run Monte Carlo simula-
tions on three sets of geographical coordinates: the first is the 41 English counties, and the
second is a random sample of 150 of the locations of African ethnic groups; and the third
is a sample of 100 German counties. The last two were taken from well known historical
studies. Coordinates were rescaled so that points lie on a unit square.
We simulate normal variables with mean zero and covariance between sites si , sj at
distance h equal to Σ (si , sj ) = ρ exp (−h/θ) + σ 2 1ij where 1ij = 1 when i = j and zero
otherwise. The parameter ρ gives systematic correlation while σ 2 represents idiosyncratic
noise The range parameter θ controls how fast correlation decays with distance: correlation
falls to about 0.1 at distance 2θ (Gneiting and Gutthorp, 2010). In the simulations here
we suppose a fairly strong and empirically realistic spatial structure where θ = 0.1, ρ =
0.9, σ 2 = 0.1. In simulations where a spatial trend was added to each variable, it took the
form of two peaks on a northwest-southeast diagonal used by Wood (2003).
Each entry of Table 2 gives the fraction of simulations where a 95 per cent confidence in-
terval contained the true coefficient value of zero. Robust least squares and Conley (1999)
23
The procedure extends to having multiple splines for explanatory variables, and to a wide family of
non-gaussian regressions including binomial, multinomial and count data. These regressions are straight-
forwardly estimated using Wood’s R package mgcv.

29
standard errors with a cutoff of 0.1 were added for comparison. Successive rows give re-
sults when no longitude and latitude terms were added to the regression, and then when
they were included linearly and quadratically.
Starting with the set of 41 counties both OLS and Conley perform well in terms of cov-
erage in the absence of spatial trends, but their coverage falls to around 0.7 when a trend is
added to the variables. For African and German coordinates without trends, they give cov-
erage of 70–80 per cent and this falls to 40–60 per cent when trends are added, even though
directional polynomials have been added to reduce spatial structure. However, coefficient
estimates are severely biased when there are spatial trends, and this bias sometimes rises
as polynomial terms are added.
Turning to semiparametric estimates, both uncorrected and two step estimates perform
almost perfectly in the absence of trends. When a trend is added, both continue to perform
well except in the case of England where the uncorrected regressions have a coverage of
only 85 per cent but two step regressions again have close to nominal coverage.

Appendix B Transportation
By 1760, England already had 1,400 miles of navigable rivers connecting places like Manch-
ester and Sheffield to the sea, and by 1830 it had added a further 2,600 miles of canals
(Satchell, 2017). Between 1750 and 1770, 10,000 miles of road, were turnpiked, increasing
to 20,000 miles by 1830; and Bogart (2005) estimates that between 1750 and 1820, road
freight charges fell by around 40 per cent. Furthermore, Britain had no internal tariffs,
unlike fragmented Italy and Germany, and even seemingly unified ancien regime France
and Spain.
Between 1760 and 1783, the tonnage of ships moving bulk goods around the coasts
rose from 155,000 to 270,000 tons, and by 1824, now including Scotland and Ireland, this
had risen to 833,000 tons. In the first half of the eighteenth century the average annual
shipment of coal from the north-east to London has been estimated at half a million tons a
year and had risen to 1.5 million tons by 1780, reaching 5.7 million by 1829; and on the eve
of the Industrial Revolution, there were no fewer than 580 locations in England and Wales
that were accessible to coal shipments by water. (Armstrong and Bagwell 1983, Tables 15,
19–22; Hausman 1987, Table 2; Szostak 1991).

Appendix C Data Sources and Construction

Measures of skill of men aged 60 and over born in each county or resident in each county
are taken from the 1851 Census in the UK Data Archive http://icem-nesstar.data-archive.
ac.uk/webview/. Workers are assigned by HISCO Code as follows: Blacksmiths 83120–83150.
Toolmakers 83210–83400. Gunsmiths and Locksmiths 83210–83700. Mechanics 84110,

30
 Bias 95% Coverage SE

Degree SP OLS SP OLS Conley SP OLS Conley

England, No Trend
0 -0.02 -0.01 0.90 0.78 0.79 0.53 0.27 0.29
1 . -0.00 . 0.88 0.86 . 0.36 0.28
2 . 0.00 . 0.88 0.82 . 0.37 0.27
England, Trend
0 -0.05 0.43 0.92 0.67 0.69 0.60 0.29 0.30
1 . 0.19 . 0.92 0.89 . 0.46 0.36
2 . 0.21 . 0.90 0.86 . 0.45 0.36
Africa, No Trend
0 0.01 -0.02 0.92 0.37 0.68 0.38 0.17 0.39
1 . 0.00 . 0.55 0.75 . 0.49 0.94
2 . 0.01 . 0.56 0.69 . 0.37 0.58
Africa, Trend
0 -0.01 0.01 0.93 0.42 0.75 0.39 0.21 0.49
1 . -0.21 . 0.59 0.79 . 0.59 1.13
2 . -0.35 . 0.55 0.69 . 0.41 0.61
Germany, No Trend
0 -0.02 0.00 0.91 0.55 0.77 0.32 0.15 0.24
1 . -0.01 . 0.67 0.78 . 0.25 0.31
2 . -0.02 . 0.72 0.78 . 0.26 0.27
Germany, Trend
0 0.00 0.19 0.92 0.57 0.83 0.33 0.17 0.38
1 . 0.11 . 0.70 0.85 . 0.27 0.35
2 . 0.42 . 0.66 0.80 . 0.28 0.38

Average bias of coefficient; proportion of 95 per cent confidence intervals containing the true coef-
ficient of zero; and average standard error. Simulations are based on the coordinates of 41 English
counties, 150 African tribal areas, and 100 German counties, where the variables do or do not
contain spatial trends. SP denotes a semiparametric regression with a thin plate spline in latitude
and longitude; OLS is ordinary least squares with robust standard errors; Conley uses a Conley
kernel correction. Degree is the polynomial degree of longitude and latitude variables added to
each regression: no variables, linear, and quadratic.
Table A.2: Monte Carlo estimates of bias and 95 per cent coverage probabilities for semi-
parametric and OLS regressions over three sets of geographical coordinates.

31
84130–84190. Millwrights 84120. Watch and instrument makers 84220–84290. Sheet metal-
workers 87330–87390.24 These are expressed per 100,000 men over 60 who were working,
retired, or unemployed.
Textile employment in 1831 is from Marshall (1833, 10–11). Apprenticeship fees for
the London Watchmakers Company are from Moore (2003).
Wages of agricultural labourers for the 1760s and 1833 are taken from Hunt (1986)
(with the entry for Nottingham corrected to match the original source) and population
data from Wrigley (2009).
Water flow for each square kilometer of England is based on the area that drains into
it, multiplied by the tan of its slope; both from the USGS Hydro1k database. Each county
is assigned a value equal to the 98th percentile of the flow across its squares. Coal distance
is the distance of the centre of each county to the nearest county with a coal field. Counties
with a coalfield were assigned a distance of 20 kilometers.
Literacy is the percentage of male convicts from each county around 1800 that were
literate from Nicholas and Nicholas (1992, Table 3) and height is the height of army vol-
unteers from 1788 to 1805 from Floud (1986).
Market potential is the sum of aggregate income (1760s wage times 1750 population)
of each county weighted by the inverse squared distance to the centre of the county. Book-
sellers is the number of booksellers in 1751 measured by Dowey (2016) relative to county
population in 1750. Lawyers are the number of attorneys in 1730 relative to county popu-
lation from Aylett (1987) and country banks in 1796 are from Brunt (2006).

References
Allen, Robert C. 2009a. The British Industrial Revolution in Global Perspective. Cambridge:
Cambridge University Press.

Allen, Robert C. 2009b. “The Industrial Revolution in Miniature: The Spinning Jenny in
Britain, France, and India.” Journal of Economic History 69:901–927.

Anderson, Michael. 1971. “Urban migration in Nineteenth Century Lancashire: Some

Insights into Two Competing Hypotheses.” Annales de Démographie Historique pp. 13–26.

Armstrong, John. 1987. “The Role of Coastal Shipping in UK Transport: an Estimate of

Comparative Traffic Movements in 1910.” Journal of Transport History 8:164–178.

Armstrong, John and Philip S. Bagwell. 1983. Coastal Shipping. In Transport in the Industrial
Revolution, ed. Derek H Aldcroft and Michael J. Freeman. Manchester University Press.

Aylett, Philip. 1987. “A Profession in the Marketplace: The Distribution of Attorneys in

England and Wales 1730–1800.” Law and History Review 5:1–30.
24
See http://historyofwork.iisg.nl/major.php for details of HISCO codes.

32
Baines, Edward. 1835. History of the Cotton Manufacture in Great Britain. London: H. Fisher,
R. Fisher, and P. Jackson.

Belhoste, Jean-Francois and Denis Woronoff. 2005. "The French Iron and Steel Industry
during the Industrial Revolution". In The Industrial Revolution in Iron: The Impact of British
Coal Technology in Nineteenth-Century Europe, ed. Chris Evans and Göran Rydén. London:
Routledge.

Ben Zeev, Nadav, Joel Mokyr and Karine van der Beek. 2017. “Flexible Supply of Appren-
ticeship in the British Industrial Revolution.” Journal of Economic History 77:208–250.

Bensimon, Fabrice. 2011. “British Workers in France, 1815–1848.” Past and Present 213:147–
189.

Berg, Maxine. 1994. The Age of Manufactures, 1700-1820: Industry, Innovation and Work in
Britain. Second ed. London: Routledge.

Berg, Maxine and Pat Hudson. 1992. “Rehabilitating the Industrial Revolution.” Journal of
Economic History 45:24–50.

Bester, C. Alan, Timothy G. Conley and Christian B. Hansen. 2011. “Inference with De-
pendent Data Using Cluster Covariance Estimators.” Journal of Econometrics 165:137–151.

Bester, C. Alan, Timothy G. Conley, Christian B. Hansen and Timothy J. Vogelsang. 2016.
“Fixed-b Asymptotics for Spatially Dependent Robust Nonparametric Covariance Mat-
rix Estimators.” Econometric Theory 32:154–186.

Bogart, Dan. 2005. “Turnpike Trusts and the Transportation Revolution in 18th Century
England.” Explorations in Economic History 42:479–508.

Broadberry, Stephen, Bruce M. S. Campbell, Alexander Klein, Mark Overton and Bas van
Leeuwen. 2015. British Economic Growth, 1270–1870. Cambridge: Cambridge University
Press.

Brunt, Liam. 2006. “Rediscovering Risk: Country Banks as Venture Capital Firms in the
First Industrial Revolution.” Journal of Economic History 66:74–102.

Canay, Ivan M., Joseph P. Romano and Azeem M. Shaikh. 2017. “Randomization Inference
under an Approximate Symmetry Assumption.” Econometrica 85:1013–1030.

Chapman, Stanley D. 1970. “Fixed Capital Formation in the British Cotton Industry, 1770-
1815.” Economic History Review 23:235–266.

Christie, John R. R. 2018. Chemical Glasgow and its Entrepreneurs. In Compound Histories:
Materials, Governance and Production, 1760–1840, ed. Lissa L. Roberts and Simon Verrett.
Leiden: Brill.

33
Clapham, John. 1939. An Economic History of Modern Britain Volume 1: The Early Railway
Age 1820–1850. Cambridge: Cambridge University Press.

Clark, Gregory. 2011. Average Earnings and Retail Prices, UK, 1209-2010. Working paper
University of California, Davis, Department of Economics.

Clark, Gregory. 2014. The Industrial Revolution. In Handbook of Economic Growth, Volume
2, ed. Philippe Aghion and Steven N. Durlauf. New York: Elsevier.

Coleman, D. C. 1983. “Proto-Industrialization: A Concept Too Many.” Journal of Economic

History 36:435–448.

Conley, Timothy. 1999. “GMM Estimation with Cross Sectional Dependence.” Journal of
Econometrics 92:1–45.

Cookson, Gillian. 2018. The Age of Machinery: Engineering in the Industrial Revolution, 1770-
1850. London: Boydell and Brewer.

Crafts, N. F. R. 1977. “Industrial Revolution in England and France: Some Thoughts on

the Question, ’Why was England First?’.” Economic History Review 30:429–441.

Crafts, N. F. R. 1986. British Economic Growth during the Industrial Revolution. Oxford: Ox-
ford University Press.

Crafts, N. F. R. 2021. “Understanding Productivity Growth during the Industrial Revolu-

tion.” Economic History Review 74:307–377.

Crafts, Nicholas and C. K. Harley. 1992. “Output Growth and the British Industrial Revolu-
tion: A Restatement of the Crafts-Harley View.” Economic History Review 45:703–730.

Crafts, Nicholas and Nikolaus Wolf. 2014. “The Location of the UK Cotton Textiles In-
dustry in 1838: a Quantitative Analysis.” Journal of Economic History 74:1103–1139.

Cummins, Neil and Cormac Ó Gráda. 2022. “Artisanal Skills, Watchmaking, and the In-
dustrial Revolution: Prescot and Beyond.” Northern History . Forthcoming.

de la Croix, David, Matthias Doepke and Joel Mokyr. 2018. “Clans, Guilds, and Mar-
kets: Apprenticeship Institutions and Growth in the Pre-Industrial Economy.” Quarterly
Journal of Economics 133:1–70.

Desaguliers, John Theophilus. 1744. A Course of Experimental Philosophy, Volume 2. London:

W. Innys.

Dowey, James. 2016. Mind over Matter: Access to Knowledge and the British Industrial
Revolution PhD thesis London School of Economics, Department of Economic History.

34
Eden, Frederic Morton. 1797. The State of the Poor. London: J. Davis.

Engle, Robert F., C. W. J. Granger, John Rice and Andrew Weiss. 1986. “Semiparametric
Estimates of the Relation Between Weather and Electricity Sales.” Journal of the American
Statistical Association 81:310–320.

Feinstein, Charles H. 1998. “Pessimism Perpetuated: Real Wages and the Standard of Liv-
ing in Britain during and after the Industrial Revolution.” Journal of Economic History
58:625–658.

Floud, Roderick. 1986. Long-term Changes in Nutrition, Welfare and Productivity in Bri-
tain. SN 2131 UK Data Service.

Gneiting, Tilmann and Peter Gutthorp. 2010. Continuous Parameter Stochastic Process
Theory. In Handbook of Spatial Statistics, ed. Alan E. Gelfand, Peter Diggle, Peter Guttorp
and Montserrat Fuentes. Boca Raton: CRC Press.

Harris, J. R. 1976. “Skills, Coal and British Industry in the Eighteenth Century.” History
61:167–182.

Harris, J. R. 1988. The British Iron Industry 1700–1850. London: Macmillan.

Harris, J.R. 1998. Industrial espionage and technology transfer: Britain and France in the eight-
eenth century. Aldershot: Ashgate.

Hausman, William J. 1987. “The English Coastal Coal Trade, 1691–1910: How Rapid was
Productivity Growth?” Economic History Review 40:588–596.

Herrendorf, Berthold, Richard Rogerson and Akos Valentinyi. 2014. Growth and Struc-
tural Transformation. In Handbook Of Economic Growth, ed. Philippe Aghion and Steven
Durlauf. Vol. 2B New York: Elsevier.

Horrell, Sara and Deborah Oxley. 2012. “Bringing Home the Bacon? Regional Nutrition,
Stature, and Gender in the Industrial Revolution.” Economic History Review 65(4):1354–
1379.

Humphries, Jane. 2013. “The Lure of Aggregates and the Pitfalls of the Patriarchal Per-
spective: A Critique of the High Wage Economy Interpretation of the British industrial
Revolution.” Economic History Review 66:693–714.

Humphries, Jane and Benjamin Schneider. 2019. “Spinning the Industrial Revolution.”
Economic History Review 72:126–155.

Hunt, E. H. 1986. “Industrialization and Inequality: Regional Wages in Britain, 1760-1914.”

Journal of Economic History 46:935–66.

35
Hyde, Charles K. 1977. Technological Change and the British Iron Industry, 1700–1870. Prin-
ceton: Princeton University Press.

Ibragimov, Rustam and Ulrich K. Müller. 2010. “t-Statistic Based Correlation and Hetero-
geneity Robust Inference.” Journal of Business and Economic Statistics 28:453–468.

James, Gareth, Daniela Witten, Trevor Hastie and Robert Tibshirani. 2021. An Introduction
to Statistical Learning. New York: Springer. 289–310.

Jones, E. L. 1968. “Agricultural Origins of Industry.” Past and Present (40):58–71.

Jones, Eric L. 2010. Locating the Industrial Revolution: Inducement and Response. World Sci-
entific.

Kanefsky, John W. 1979. The DIffusion of Power Technology in British Industry, 1760–1870.
PhD thesis University of Exeter.

Kelly, Morgan and Cormac Ó Gráda. 2016. “Adam Smith, Watch Prices, and the Industrial
Revolution.” Quarterly Journal of Economics 131:1727–1752.

Kelly, Morgan and Cormac Ó Gráda. 2021. “Connecting the Scientific and Industrial Re-
volutions: The Role of Practical Mathematics.” Journal of Economic History . Forthcoming.

Kelly, Morgan, Joel Mokyr and Cormac Ó Gráda. 2014. “Precocious Albion: A New Inter-
pretation of the British Industrial Revolution.” Annual Reviews of Economics .

Kennedy, John. 1819. “Observations on the Rise and Progress of the Cotton Trade in Great
Britain.” Memoirs of the Literary and Philosophical Society of Manchester 3:115–137.

Leunig, Tim, Chris Minns and Patrick Wallis. 2012. “Networks in the Premodern Economy:
The Market for London Apprenticeships, 1600-1749.” Journal of Economic History . Vol.
71, No. 2 (June), pp. 413-41.

Marshall, John. 1833. A Digest of all the Accounts . . . presented to Parliament during the Last
Thirty-five Years. London: J. Haddon.

McCloskey, Deirdre N. 2010. Bourgeois Dignity: Why Economics Can’t Explain the Modern
World. Chicago: University of Chicago Press.

McCloskey, Deirdre N. 2016. Bourgeois Equality: How Ideas, Not Capital or Institutions, En-
riched the World. Chicago: University of Chicago Press.

Mokyr, Joel. 1999. Editor’s Introduction: The New Economic History and the Industrial
Revolution. In The British Industrial Revolution: An Economic Perspective, ed. Joel Mokyr.
Boulder: Westview Press.

36
Mokyr, Joel. 2009. The Enlightened Economy: An Economic History of Britain 1700–1850. New
Haven: Yale University Press.

Mokyr, Joel. 2019. The Economics of Apprenticeship. In Apprenticeship in Early Modern

Europe, ed. Maarten Prakand and Patrick Wallis. Cambridge University Press pp. 20–43.

Mokyr, Joel. 2021. “’The Holy Land of Industrialism’ Rethinking the Industrial Revolu-
tion.” Journal of the British Academy 9:223–247.

Mokyr, Joel, Assaf Sarid and Karine van der Beek. 2021. “The Wheels of Change: Human
Capital, Millwrights, and Industrialization in Eighteenth-Century England.” Economic
Journal forthcoming.

Moore, Dennis. 2003. British Clockmakers and Watchmakers Apprentice Records 1710–1810.
Ashbourne: Mayfield Books.

Moscona, Jacob. 2019. Agricultural Development and Structural Change, Within and
Across Countries. Working paper MIT.

Müller, Ulrich K. and Mark W. Watson. 2021. Spatial Correlation Robust Inference. Work-
ing paper. Department of Economics Princeton University.

Musson, A. E. and Eric Robinson. 1969. Science and Technology in the Industrial Revolution.
Manchester: Manchester University Press.

Neal, Larry. 1994. The Finance of Business during the Industrial Revolution. In The Eco-
nomic History of Britain Since 1700. Volume 1: 1700-1860, ed. Roderick Floud and Deirdre
McCloskey. Cambridge: Cambridge University Press.

Nicholas, Stephen J. and Jacqueline M. Nicholas. 1992. “Male Literacy, Deskilling, and the
Industrial Revolution.” Journal of Interdisciplinary History 23(1):1–18.

Ogilvie, Sheilagh. 2019. The European Guilds: an Economic Analysis. Princeton: Princeton
University Press.

Pressnell, Lionel. 1956. Country Banking in the Industrial Revolution. Oxford: Clarendon
Press.

Redford, Arthur. 1964. Labour Migration in England, 1800–1850. Revised ed. Manchester
University Press.

Roll, Eric. 1930. An Early Experiment in Industrial Organization: History of the Firm of Boulton
and Watt 1775-1805. Abingdon: Frank Cass.

Samuel, Raphael. 1977. “Workshop of the World: Steam Power and Hand Technology in
Mid-Victorian Britain.” History Workshop Journal 3:6–72.

37
 Satchell, Max. 2017. Navigable waterways and the Economy of England and Wales 1600-
1835. In The Online Historical Atlas of Transport, Urbanization and Economic Development in
England and Wales c.1680-1911, ed. L. Shaw-Taylor, D. Bogart and A. E. M. Satchell. The
Cambridge Group for the History of Population and Social Structure.

Smiles, Samuel. 1863. Industrial Biography: Iron-workers and Tool-makers. London: John
Murray.

Smiles, Samuel. 1865. Lives of Boulton and Watt. London: John Murray.

Stephenson, Judy Z. 2018. “’Real’ Wages? Contractors, Workers and Pay in London Build-
ing Trades, 1650–1800.” Economic History Review 71:106–132.

Styles, John. 2007. The Dress of the People: Everyday Fashion in Eighteenth Century England.
New Haven: Yale University Press.

Szostak, Rick. 1991. The Role of Transportation in the Industrial Revolution: A Comparison of
England and France. Montreal: McGill-Queen’s University Press.

Tann, Jennifer. 1970. The Development of the Factory. London: Cornmarket.

Tann, Jennifer. 1974. “The Textile Millwright in the Early Industrial Revolution.” Textile
History 5(1):80–89.

Temin, Peter. 1997. “Two Views of the Industrial Revolution.” Journal of Economic History
57(63–82).

von Tunzelmann, G. N. 1978. Steam Power and British Industrialization to 1860. Oxford
University Press.

Wood, Simon N. 2003. “Thin Plate Regression Splines.” Journal of the Royal Statistical Society.
Series B 65:95–114.

Wood, Simon N. 2017. Generalized Additive Models: An Introduction with R. Second ed. Boca
Raton: CRC Press.

Wrigley, E. Anthony. 1988. Continuity, Chance and Change. Cambridge: Cambridge Univer-
sity Press.

Wrigley, E. Anthony. 2009. “Rickman Revisited: The Population Growth Rates of English
Counties in the Early Modern Period.” Economic History Review 62:711–735.

Wrigley, E. Anthony. 2010. Energy and the English Industrial Revolution. Cambridge: Cam-
bridge University Press.

38

View publication stats