Industrialization and urbanization: Did the steam engine contribute to the growth of cities in the United States ?

Explorations in
Economic History
Explorations in Economic History 42 (2005) 586–598
www.elsevier.com/locate/eeh
Industrialization and urbanization: Did the
steam engine contribute to the growth of cities
in the United States? q
Sukkoo Kim *
Washington University in St. Louis, USA
NBER, USA
Received 11 June 2004
Available online 6 April 2005
Abstract
Industrialization and urbanization are seen as interdependent processes of modern eco-
nomic development. However, the exact nature of their causal relationship is still open to con-
siderable debate. This paper uses ?rm-level data from the manuscripts of the decennial
censuses between 1850 and 1880 to examine whether the adoption of the steam engine as
the primary power source by manufacturers during industrialization contributed to urbaniza-
tion. While the data indicate that steam-powered ?rms were more likely to locate in urban
areas than water-powered ?rms, the adoption of the steam engine did not contribute substan-
tially to urbanization.
Ó 2005 Elsevier Inc. All rights reserved.
Keywords: Industrialization; Urbanization; Factory; Steam power; Water power
q I am grateful to Bob Margo, the editor, for his extensive comments and suggestions which greatly
improved the paper. I also thank Henry Overman, Jan De Vries, Jan Pieter Smits, two anonymous
referees, and participants of the 2004 Conference on Growth and Development in Global Perspectives at
Academia Sinica in Taipei, 2004 RSAI meetings in Seattle, and workshops at the St. Louis Federal
Reserve and Washington University for their comments. Financial support from Washington University
through a faculty research grant is gratefully acknowledged.
* Fax: +1 19354156.
E-mail address: soks@wueconc.wustl.edu.
0014-4983/$ - see front matter Ó 2005 Elsevier Inc. All rights reserved.
doi:10.1016/j.eeh.2005.03.001

---

S. Kim / Explorations in Economic History 42 (2005) 586–598
587
1. Introduction
Industrialization and urbanization are seen as interdependent processes of modern
economic development. In the United States, industrialization began in the early nine-
teenth century as manufacturing re-organized from artisanal shops to non-mechanized
factories in a handful of industries; however, in the second half of the nineteenth cen-
tury, manufacturing activity rose in scale, became more mechanized and spread to
numerous industries. The rise of a manufacturing sector, especially in the second half
of the nineteenth century, coincided with a signi?cant growth in urban population. As
the domestic labor force in manufacturing doubled from 10 to 20% between 1850 and
1880, so too did the share of the population in urban places, from 15 to 30%.
One of the major developments associated with industrialization was the shift in
primary power sources for manufacturing from hand and water power to steam
power, particularly in large factories. This shift towards greater use of steam power
by manufacturers is believed to be explained by a sharp decline in the relative user cost
of steam compared to other power sources. According to AtackÕs (1979) estimates, the
annual costs per horsepower of steam fell below that of water power in the 1840s. By
1870, steam power capacity in manufacturing was greater than that of water-power
(Fenichel, 1979 and Rosenberg and Trajtenberg, 2004). While rigorous comparisons
of hand and steam power costs do not exist, the relative cost of using hand power may
have risen as wages were increasing over this period (see Margo, 2000).
In a recent paper, Rosenberg and Trajtenberg (2004) argue that the adoption of the
steam engine by manufacturers, and in particular the Corliss engine, was responsible
for the rapid rise in urbanization.1 By releasing ?rms from the locational limitations
of topography and climate and o?ering them the freedom to locate in cities, they argue
that the deployment of the Corliss steam engine served as a catalyst for the relocation of
?rms from rural locations to cities. While the idea that the steam engine contributed to
urbanization is not new, and indeed many early promoters of the steam engine pro-
claimed locational freedom as one of its major bene?ts, Rosenberg and TrajtenbergÕs
paper represents the ?rst serious empirical estimate of this hypothesis.2 In sum, they
1 Rosenberg and Trajtenberg believe that the Corliss steam engine was a general purpose technology that
was responsible for triggering economic growth in the late nineteenth century. They argue that the steam
engine, by fostering urbanization, allowed the economy to capture signi?cant bene?ts of agglomeration
economies. However, previous studies that use the growth accounting framework suggest a limited role of
general purpose technologies on economic growth. For example, Crafts (2004) and Crafts and Mills (2004)
?nd that the impact of the steam engine on UK economic growth was rather modest. A social savings
calculation suggests that the steam engineÕs contribution to growth was about 0.05% per year between
1870 and 1910.
2 For example, the Scienti?c American, in their May 12, 1849 issue, wrote: ‘‘A water-mill is necessarily
located in the country afar from the cities, the markets, and the magazines of labor, upon which it must be
dependent. Water appears to run very cheaply, but it always rents for a high price, and the [capital] cost of
dams, races, water wheels etc. is on the average quite as great as that of a steam engine and equipage. . . A
man sets down his steam-engine where he pleases—that is, where it is most to his interest to plant it, in the
midst of the industry and markets, both for supply and consumption of a great city—where he is sure of
always having hands near him, without loss of time in seeking for them, and where he can buy his raw
materials and sell his goods, with adding the expense of double transportation.’’ See Hunter (1985, p.104).

---

588
S. Kim / Explorations in Economic History 42 (2005) 586–598
show that counties which adopted a higher stock of Corliss steam-engines as of 1870
exhibited faster population growth in subsequent decades.
This paper evaluates the role of the steam engine, as well as other primary power
sources, on the location of manufacturing ?rms using establishment-level data from
the manuscript censuses of manufactures for the period between 1850 and 1880.
Since the data from the manuscript censuses contain information on power sources
of establishments and whether these establishments were in rural or urban places, it
is possible to estimate more precisely whether the adoption of the steam engine con-
tributed to urbanization. Although the census data do not distinguish between Cor-
liss and other types of steam engines, there is little reason to believe that the
locational impact of steam engines was isolated to the Corliss type.
The analysis of the manuscript census data shows that steam-powered employees
were on average almost ?ve and a half times more likely to locate in cities than
water-powered employees between 1850 and 1880. However, when compared to
employees that did not use inanimate power as their primary power source, such
as hand-power, steam-powered employees were on average 0.55 times less likely to
locate in cities. At any given point in time, the di?erences in predicted probabilities
of urbanization for steam-powered employees were negligible. However, when the
calculations factor into account the changes in the distribution of primary power
sources over time, the shift in power source from hand and water-power into
steam-power may have contributed to about 8–10% increase in the rate of
urbanization.
Rather than the adoption of steam-power, the shift from artisanal to factory orga-
nization of production in manufacturing may have been the single most important
contributor to urbanization in the second half of the nineteenth century. Factory
workers, who did not use inanimate power, as well as steam-powered and water-
powered factory workers were on average two to three times more likely to locate
in cities than their non-factory counterparts. Factory production is estimated to have
increased urbanization by about 27% at any given point time. Finally, employees in
?rms with a higher intensity of female labor were much more likely to locate in
urban areas whereas employees with a higher intensity of child labor were less likely
to do so.
2. Data
This paper uses the Atack–Bateman–Weiss (ABW) sample of manufacturing
?rms drawn from the manuscripts of the decennial censuses for 1850, 1860, 1870,
and 1880 to examine whether the adoption of the steam engine contributed to urban-
ization. In addition to the typical census type information on output, raw materials,
capital, labor, and wages, the data provide information on primary power source.
For 1850–1870, the census enumerators collected information on whether an estab-
lishment used one of ?ve types of power sources: water, steam, hand, animal, and
combination. However, information on the level of horsepower was only sporadi-
cally reported for a meaningful analysis of this variable. For 1880, the census sched-

---

S. Kim / Explorations in Economic History 42 (2005) 586–598
589
ules provide information on the level of primary horsepower generated by water and
steam, but do not contain information on other types of power sources.
The ABW data contain information on the location of ?rms at the county level
and whether its location is urban or rural. An area was de?ned as urban if it was
an incorporated town or city which contained a population of at least 2500. How-
ever, it is important to note that, unlike the modern de?nition which uses the county
as the smallest geographic unit of analysis for determining whether a place is rural or
urban, the most likely unit of geographic observation was the minor civil division
during this period. Thus, it is possible to infer from the data whether an establish-
ment was located in the rural or urban part of the county. Finally, the establishments
are categorized by standard industrial code (sic) at the 3-digit industry level.
To eliminate potential outliers in the data, the samples were restricted to establish-
ments with positive values of output, employment, and capital. Firms with capital-la-
bor ratios below $50 per worker were omitted. The data were restricted to
manufacturing industries de?ned by modern 3-digit industries ranging from 200 to
399; however, 351 (steam engine) was deleted. For the 1880 data, the so-called ‘‘special
agent’’ industries are under-represented in the random sample. In a number of indus-
tries, such as cotton, wool, silk, iron and steel, etc., experts rather than regular enumer-
ators were chosen to gather information. However, these manuscript schedules
collected by ‘‘special agents’’ have never been located. Several strategies are deployed
to ensure that the data analysis is robust to the under-enumeration of these industries.3
Table 1 demonstrates that the manufacturing employment of the establishments
in the ABW sample became increasingly more urbanized between 1850 and 1880.
Over this period, the share of employment in urban locations rose from 40 to
71%. However, the growth in urbanization of the manufacturing labor force was
not monotonic as the share of urban employment fell between 1860 and 1870. Unfor-
tunately, it is di?cult to know whether this decline was caused by the Civil War or
whether it is related to sampling problems with the 1870 data, such as an under-enu-
meration of ?rms in the South. From a regional perspective, little discernible pat-
terns emerge in the ABW data except for the possible emergence of north-south
divergence in urbanization rates by 1880.
Table 2 reports the share of urban manufacturing employment by 2-digit indus-
tries. The ABW data indicate that the share of manufacturing employment in urban
locations varied signi?cantly by industries. Employees in tobacco, apparel, printing,
and miscellaneous industries were much more likely to locate in cities whereas those
in food and lumber-wood industries were more likely to reside in rural places. For
employment in some industries such as textiles and primary metals, locational pat-
terns shifted somewhat over time and became more concentrated in cities by 1880.
The industry data also indicate that the drop in urbanization between 1860 and
1870 was heavily concentrated in a few industries such as apparel, lumber-wood, pa-
per, and chemicals.
3 The articles by Atack et al. (2002, 2003, 2004) carefully address many of the important data issues,
including those related to ‘‘special agent’’ industries, that pertain to the ABW data.

---

590
S. Kim / Explorations in Economic History 42 (2005) 586–598
Table 1
Share of urban manufacturing employment by region: data from the census manuscripts (percentage)
1850
1860
1870
1880
United States
40.4%
51.5%
44.2%
71.0%
New England
38.7
39.6
29.6
68.7
Middle Atlantic
43.9
67.2
60.4
78.4
East North Central
31.5
35.5
33.7
71.0
West North Central
50.8
43.1
37.1
64.8
South Atlantic
41.8
58.4
13.2
62.5
East South Central
33.0
27.5
58.0
48.5
West South Central
7.3
15.4
24.4
29.2
Mountain
—
—
—
—
Paci?c
72.7
30.4
52.2
51.3
Sources. See Atack and Bateman (1999).
Table 2
Share of urban manufacturing employment by industry: data from the census manuscripts (percentage)
1850
1860
1870
1880
All Manufactures
40.4%
51.5%
44.2%
71.0%
20 Food
35.3
48.8
45.1
64.0
21 Tobacco
73.5
71.3
44.5
88.0
22 Textiles
31.9
47.3
38.7
81.1
23 Apparel
72.3
82.3
69.2
89.4
24 Lumber
18.8
24.4
15.5
31.7
25 Furniture
50.3
54.1
53.0
69.9
26 Paper
42.1
67.9
27.6
75.1
27 Printing
83.7
82.3
72.3
98.2
28 Chemicals
35.3
54.2
22.1
52.1
29 Petroleum
—
—
—
—
30 Rubber
—
—
—
—
31 Leather
36.7
41.9
34.9
68.8
32 Stone
48.0
52.5
44.4
49.5
33 Primary
12.3
55.7
49.0
73.3
34 Fabricated
51.9
63.3
52.7
72.5
35 Machinery
78.3
63.8
69.4
65.9
36 Electrical
—
—
—
—
37 Transportation
30.2
42.3
29.5
83.0
38 Instruments
44.0
—
—
95.5
39 Miscellaneous
81.6
76.0
71.3
88.5
Sources. Atack and Bateman (1999).
3. Did the steam engine contribute to the growth of cities?
To assess whether the adoption of the steam engine led ?rms to locate in urban
areas, we select a discrete choice model where the dependent variable takes on a va-
lue of 1 if a ?rm is located in an urban area and 0 if it is located in a rural area. More

---

S. Kim / Explorations in Economic History 42 (2005) 586–598
591
speci?cally, the regression estimates are based on the logit model of the following
form:
ln½P i=ð1 À P iÞ? ¼ a þ b Location
Industry þ b Factory
1
i þ b2
i
3
i
þ b Women
Children
Steam
Water
4
i þ b5
i þ b6
i þ b7
i
þ b Steam à Factory þ b Water à Factory þ u
8
i
9
i
i;
ð1Þ
where Pi is the probability that a ?rm i is located in an urban area and (1 À Pi) is the
probability that it is located in a rural area. To estimate the impact of primary power
sources on location, we construct dummy variables for steam-powered and water-
powered ?rms; the omitted category is dominated by hand-powered ?rms, but also
include a small number of ?rms that use some combination of various power
sources.
To control for other factors that in?uence ?rm location decisions, we include as
independent variables a factory dummy variable, which equals one if a ?rm em-
ployed more than 15 laborers, the share of female employees, the share of children
employees (for 1880 only), and interactions between steam and water power dummy
variables with the factory dummy variable.4 In addition, to account for unmeasured
industry and location speci?c e?ects, the regressions contain 3-digit industry and
county-level locational ?xed-e?ects, respectively.
The regression sample means are reported in Table 3. Between 1850 and 1880, the
proportion of workers in factories rose steadily from 61.3 to 77.9%. Over the same
time period, the share of steam-powered employees rose from 20.7 to 52.2% whereas
the share in water-power fell from 31.7 to 6.3%. Similarly, the share of steam-pow-
ered factory workers rose from 17.0 to 46.2% whereas that of water-powered factory
workers fell from 21.4 to 3.1%. The intensity of female labor in manufacturing de-
clined from 23.3 to 16.6%, but this decline may be an artifact of changes in reporting
criterion. In 1850 and 1860, the data contain female workers of all ages; however, in
1870 and 1880, the data only include female workers over 15 years of age. In 1880,
about 4% of the manufacturing employment in the ABW sample was comprised by
children.
The logistic regressions reported in this paper are weighted by employment. How-
ever, there are di?ering strategies for interpreting the logistic regression coe?cients.
One standard strategy involves reporting the coe?cients in odds-ratios, (P/(1 À P)),
by simply taking the exponent or anti-logarithm of the logit regression coe?cients.
The odds-ratio has the simple interpretation in that a coe?cient greater (less) than
one means that manufacturing employment is more (less) likely to locate in an urban
rather than a rural location. The second strategy involves translating the e?ects on
logged odds into the e?ects on probabilities. This latter strategy provides a more con-
crete estimate of the independent variableÕs e?ect on urbanization.
Table 4 reports the logistic regression coe?cients in terms of odds-ratios. The re-
sults indicate that steam-powered employees were more likely to locate in urban
4 Employment is de?ned as one plus men, women and child employees.

---

592
S. Kim / Explorations in Economic History 42 (2005) 586–598
Table 3
Regression sample means weighted by employment, 1850–1880
1850
1860
1870
1880
Urban
0.404
0.515
0.442
0.710
Factory
0.613
0.644
0.702
0.779
Women
0.233
0.222
0.164
0.166
Children
—
—
—
0.043
Steam
0.207
0.251
0.415
0.522
Water
0.317
0.269
0.189
0.063
Steam * Factory
0.170
0.187
0.342
0.462
Water * Factory
0.214
0.188
0.143
0.031
Number of Firms
4402
4700
3912
5654
Number of Emp.
44,375
51,709
47,552
87,599
Sources. Atack and Bateman (1999).
Table 4
Determinants of urban location of manufacturing employment, 1850–1880 (logit regression reported in
odds-ratio with z-statistics in parentheses)
1850 (1)
1860 (2)
1870 (3)
1880 (4)
1880+ (5)
Factory
2.132*
2.020*
3.296*
1.732*
2.237*
(24.6)
(25.3)
(36.7)
(20.7)
(37.0)
Women/labor
0.993
2.606*
1.314*
9.590*
7.891*
(À0.2)
(21.7)
(5.7)
(47.9)
(44.0)
Children/labor
—
—
—
0.931
0.719*
(À0.8)
(À3.7)
Steam
0.664*
0.581*
0.659*
0.290*
0.749*
(À7.1)
(À12.5)
(À8.4)
(À33.4)
(À26.8)
Water
0.076*
0.167*
0.122*
0.074*
0.719*
(À38.8)
(À35.4)
(À21.1)
(À39.5)
(À27.5)
Steam * Factory
0.635*
1.662*
1.189*
3.533*
1.183*
(À6.8)
(9.6)
(3.1)
(29.6)
(12.6)
Water * Factory
2.826*
1.163*
2.196*
3.643*
1.032
(14.1)
(2.5)
(7.5)
(16.3)
(1.7)
Fixed e?ects
County
Yes
Yes
Yes
Yes
Yes
Industry
Yes
Yes
Yes
Yes
Yes
Pseudo R2
0.13
0.11
0.08
0.15
0.14
Number of Est.
4351
4679
3833
5535
5533
Number of Emp.
42,004
51,059
46,256
83,849
83,547
* P < 0.05.
locations than water-powered employees, but were less likely to do so compared to
employees in the omitted category.5 Compared to water-powered employees, steam-
powered employees were 8.7 times more likely to locate in cities in 1850, but that
5 The omitted category consists of establishments that used hand, animal, and combination of power
sources as well as those that failed to report their sources of primary power.

---

S. Kim / Explorations in Economic History 42 (2005) 586–598
593
?gure declined to an average of 4.3 for the decades between 1860 and 1880. On the
other hand, when compared to the employees in the omitted category, steam-pow-
ered employees were on average 0.64 times less likely to locate in cities between
1850 and 1870. In 1880, the odds fell even more sharply to 0.29, but this decline
may be in part due to a change in reporting procedure for primary power data.6
Factory employees, those who worked for establishments with greater than 15
employees, were on average twice as more likely to locate in cities than those in smal-
ler establishments. Except for 1850, steam-powered factory employees were also
more likely to locate in cities. The interaction between steam-power and factory
dummy variables indicates that employees in steam-powered factories were 1.7
and 1.2 times more likely to locate in cities in 1860 and 1870, and 3.5 times more
likely in 1880. Surprisingly, employees in water-powered factories, except for 1860,
were even more likely to locate in cities than those of steam-powered factories. In
1850, employees in water-powered factories were 2.8 times more likely to locate in
cities than those in other ?rms; the odds-ratio fell to 1.2 in 1860, but rose again to
2.2 and 3.6 in 1870 and 1880, respectively.
For 1880, a second logistic regression, Eq. (5), was estimated using levels of horse-
power of steam and water power and their interactions with a factory dummy. For
this speci?cation, the odds of locating in cities declined to 0.75 and 0.72 for a unit
increase in steam and water power per worker, respectively; however, their factory
interaction was signi?cant for steam but not for water power. The lack of signi?-
cance may be due the positive correlation between horsepower intensities in steam
and water power and factory organization.
Employees in ?rms with a higher intensity of women workers were much more
likely to locate in cities. Except for 1850, when the coe?cient was not signi?cant,
an increase in the intensity of women workers increased the probability of locating
in cities from 1.3 to 9.6 times between 1860 and 1880. On the other hand, employees
of ?rms that utilized children more intensely in 1880 were more likely to locate in
rural locations.
While the odds-ratios provide an intuitively useful way of interpreting the logit
model, an alternative strategy of estimating out-of-sample predictions in probabili-
ties provides a means of estimating the changes in the probability of the dependent
variable associated with a change in the independent variable. However, since the
relationships between the independent variables and the dependent variable in prob-
ability are non-linear and non-additive, the simple partial derivative, especially for
dichotomous dummy variables, is a poor estimate of the e?ects on probability. In-
stead, to derive a more accurate estimate, it is necessary to calculate the predicted
di?erences in probabilities associated with a dummy variable group or a standard
6 In order to determine whether the results of the paper are sensitive to the under-enumeration of
‘‘special agent’’ industries for 1880, the analysis was repeated accordingly. First, ‘‘special agent’’ industries
were re-weighted to match the published aggregates. Second, the ‘‘special agent’’ industries were omitted
from the sample. In both instances, the logit regression estimates were essentially identical to those
reported in the paper.

---

594
S. Kim / Explorations in Economic History 42 (2005) 586–598
deviation change of a continuous independent variable (see Pample (2000), Hamil-
ton, 2004).
This paper employs two types of counter-factual experiments based on predicted
probabilities. The ?rst method is analogous to using the standard partial equilibrium
framework where predicted changes in probability of the dependent variable are esti-
mated for a change in an independent variable holding all other variables constant.
This method estimates the contribution of the steam engine (or other independent
variables) on urbanization at any given point in time assuming that all other factors
are held constant. The second method takes into account the changing distribution
of primary power sources over time. As shown in Table 3, there has been a system-
atic shift in primary power sources from hand and water to steam power over time.
This method estimates the changes in the level of urbanization over time that can be
attributed to the changing distribution of primary power sources.
To estimate the predicted probabilities based on the ?rst method outlined above,
it is necessary to select a starting point for calculating the out-of-sample predictions.
While there is no standard choice, a useful one is the probability associated with the
sample mean. For example, the sample mean of the share of urban manufacturing
employment in 1880 as reported in Table 3 is P0 = 0.71; the logit for this probability,
L0, is equal to 0.8954 (ln (P0/(1 À P0)) = ln (0.71/0.29)). To compute the impact of an
independent variable on urbanization, the logistic regression coe?cient in logit (log-
arithm of the odds-ratios reported in Table 3) is added to the initial logit value. If the
independent variable is a dummy variable, then L1 = L0 + bi; if the independent var-
iable is continuous, then L1=L0 + bi * SDi, where SD is the standard deviation. The
probability associated with this new logit L1 is then equal to P1 = 1/(1 + eÀL1). The
di?erence in predicted probabilities, P1 À P0, measures the independent variableÕs
impact on urbanization.
The computations on predicted probabilities indicate that the adoption of factory
production and female intensive labor force contributed positively to urbanization,
but that the adoption of inanimate primary power sources, both steam and water
power, led to a decrease in urbanization. While the reported estimates are based
on the 1880 logit coe?cients, similar results are obtained for other years. Factory
production based on animate power sources increased the share of urban manufac-
turing employment by 10%, but if steam and water-powered factories are included,
then the impact of factories increased to 27%. Like factory production, one standard
deviation increase in the intensity of the female labor force increased urbanization by
about 11%.
Surprisingly, the adoption of the steam engine contributed to a signi?cant decline
in the probability of urban employment. The steam power dummy variable contrib-
uted to a decline in urban employment by 29.5%. However, the net e?ect of the steam
engine, when one includes the e?ect of steam-powered factories, was negligible. The
adoption of water power led to an even more signi?cant decline in the share of urban
employment. The water power dummy variable accounted for a 55.6% decline in the
share of urban employment; however, if the impact of water-power factories was
added, water-power on net contributed to a 31.3% decline in urban manufacturing
employment. When the exercise was repeated for Eq. (5) using levels of horsepower,

---

S. Kim / Explorations in Economic History 42 (2005) 586–598
595
a standard deviation increase in horsepower per worker led to a decline in urban
employment by 9 and 11% for steam and water, respectively.
The second counter-factual experiment asks how much of the changes in urban-
ization over time can be attributed to the change in the distribution of power sources
from hand and water-power to steam-power. Using the 1850 estimated logit coe?-
cients, it is possible to estimate the predicted changes in the rate of urbanization if
the 1880 mean distribution of employment in water-powered shops, steam-powered
shops, non-powered factory, water-powered factory, and steam-powered factory is
used, holding all other variables at the 1850 sample means. By this estimation, the
shift in the distribution of power sources, which also takes into account the changes
in the organization of production, may have increased urbanization by about 8% be-
tween 1850 and 1880. Alternatively, if the 1880 logit coe?cients are used on the 1850
employment distribution, then the change in the rate of urbanization is about 10%.
Despite the lengthy set of controls, the logit coe?cient estimates on the steam en-
gine variable are likely to be biased upwards due to omitted factors that are posi-
tively correlated with use of steam and location in an urban area. If the choice of
power sources is endogenous, then there may be an endogeneity bias in the regres-
sion estimates since urban ?rms were more likely to adopt steam engines rather than
water wheels as their primary power source. On the other hand, measurement error
in the use of steam may bias the results in the opposite direction, although this bias is
likely to be mitigated by the use of dummies rather than horsepower. Finally, to
determine whether there may be an additional bias caused by the fact that establish-
ments in some industries may reside solely in urban or rural locations within a
county, the regressions were repeated without county ?xed-e?ects and the results
were essentially unchanged.
3.1. Decomposing industry ?xed-e?ects
The data presented in Table 2 indicate that some industries were much more likely to
locate in urban areas than others. Table 5 examines the role of industry ?xed-e?ects by
estimating logit regressions based on dummy variables for 2-digit industries. The omit-
ted industry was stone, clay, and glass (sic 32), as well as a few other manufacturing
industries (sic 29, sic 30, sic36, sic 38) whose sample sizes were very small. While there
were considerable variations in the data, the logit regressions show that ?rms in some
industries such as lumber and wood, chemicals, leather, and transportation were rela-
tively more likely to locate in rural areas; on the other hand, ?rms in printing, miscel-
laneous, and apparel manufactures were generally more likely to locate in urban areas.
The examination of the industrial patterns of urbanization shows that industries
that were more likely to locate in rural areas were generally intensive in raw materials
derived from agriculture and forests. On the other hand, the industries that were
more likely to locate in urban areas were mostly labor intensive such as printing, mis-
cellaneous, and apparel, and to a lesser extent tobacco and textiles industries. It is
also interesting to note that some industries such as lumber and wood that relied
on intensive use of inanimate power were more likely to locate in rural rather than
in urban areas.

---

Document Outline

• Industrialization and urbanization: Did the steam engine contribute to the growth of cities in the United States?
  • Introduction
  • Data
  • Did the steam engine contribute to the growth of cities?
    • Decomposing industry fixed-effects
  • Conclusion
  • References

---