Industrialization and the 19th Century American Fertility
Decline: Evidence from South Carolina
Marianne H. Wanamaker
†
Department of Economics, College of Business Administration
The University of Tennessee
February 2011
Abstract
Economists have frequently hypothesized that industrialization and its correlates played a major
role in fertility decline in the United States after 1850. I exploit the unique circumstances surrounding
industrialization in South Carolina between 1880 and 1900 to show that fertility rates were negatively
impacted by the arrival of textile mills. Using cross-section data on rural locations in South Carolina, I
show that textile mill arrival was associated with a signicant reduction in fertility by 1900. A falsication
test shows that these patterns did not hold in the pre-industrialization period, and a dierence-indierence analysis estimates fertility reduction at 6-10%. Controlling for migration, it appears that the
in-migration of low-fertility households is responsible for most of the observed decline. Higher rates
of textile employment and child mortality for migrant females can explain part of the result, and I
conjecture that an increase in child-raising costs induced by the separation of migrant households from
their extended families may explain the remaining gap in migrant-native fertility.
∗ I am grateful to Joe Ferrie, Joel Mokyr, Chris Taber, Xavier Duran, Tim Guinnane, Matthias Doepke, Lou Cain, Celeste
Carruthers, Bill Neilson, and two anonymous referees for helpful comments and suggestions.
This paper has also beneted
from the input of the Cliometrics Conference, National Bureau of Economics Research Summer Institute, Yale University's
Economic History Seminar, Northwestern's Economic History Seminar, Northwestern's Labor Lunch, and the University of
Tennessee's Economics Brownbag. Funding for this research from Northwestern University in the form of a Graduate Research
Grant, Graduate Research Fellowship, and Presidential Fellowship are gratefully acknowledged, as is funding from the Economic
History Association's Research Travel Grant.
† Email: [email protected]
By the dawn of the 20th century, fertility rates in the United States had undergone a century of steady
decline.
In 1800, white American females could expect to bear 7.0 children on average.
By 1900, this
1 The factors behind this decline have been the subject of a lengthy literature highlighting
number was 3.6.
the importance of intergenerational bequests, the economic value of children, and the cultural context for
American family formation. Yasuba (1962) began the literature on the importance of land availability to
the bequest process, noting that those locations with a higher availability of land for bequests experienced
higher fertility rates and vice versa. The work of Richard Easterlin and his students on the targeted bequest
2 Sundstrom and David and Carter, Ransom and Sutch presented
hypothesis is another step in this direction.
alternatives to the land availability hypothesis for explaining the early 19th century decline, emphasizing
3 Sundstrom and
instead the role of economic modernization in changing parental incentives to have children.
David highlight the role of outside labor options for farm children in reducing the old-age security provided
by these children and, therefore, parental demand. Carter, Ransom and Sutch highlight the role of westward
migration in allowing children to default on their role as providers for their parents in their old age, again
4
reducing parental demand.
The empirical evidence implies that land availability issues associated with Yasuba and Easterlin were
instrumental in determining American fertility patterns prior to the 1850s, while more traditional socioeconomic variables became important thereafter. In particular, the evidence seems to indicate a role for bequest
and land availability explanations early in the century while levels of industrialization, urbanization, child
labor laws, average wages, literacy rates and education levels hold better explanatory power for the postbel-
5 But despite these conclusions, the impact of these factors on fertility rates has yet to be measured
lum era.
in anything more than a cross-sectional, observational framework. Steckel (1992), Guest (1981), Sundstrom
and David (1988) and Vinovskis (1976) all employ cross-sectional variation in state-level labor force composition, manufacturing wages, average farm values, literacy rates, school attendance, child employment, sex
ratios, presence of nancial intermediaries, and/or land availability to explain fertility rates and reach their
respective conclusions.
1 These gures represent the total fertility rate for American females as
2 See Easterlin (?), Schapiro (?), and Easterlin (?).
3 David and Sundstrom (1988); Carter, Ransom and Sutch (2002).
4 See Haines (?), pp.212-213 for a discussion of the cultural concerns.
5 See Steckel (1992) and Wahl (1992).
1
reported by Haines (2000).
This paper is motivated by the inadequacy of previous empirical work to measure the relationship between
industrialization and fertility without embedded simultaneity. An empirical observation that industrialization
and fertility are negatively correlated is not informative about causation. Falling fertility may have spurred
industrialization rather than vice versa, or external factors may have contributed to both simultaneously.
I exploit the circumstances surrounding industrialization in South Carolina between 1880 and 1900 to
identify a plausibly exogenous change in the level of industrialization. The introduction of textile mills to
South Carolina after the Civil War marked the beginning of the state's industrial era, and Section 1 argues
that the emerging geography of South Carolina's textile industry was independent of the characteristics of
the local labor force, in particular their fertility. Comparing the fertility patterns of rural townships in South
Carolina that did and did not receive textile mills over this period, I ask whether industrialization had any
measurable impact on household fertility. I nd that the arrival of textile mills negatively impacted marital
fertility rates in South Carolina between 1880 and 1900.
The introduction of textile mills reduced 1900
marital fertility by as much as 10.4% in the cross section and 6.9% using a dierence-in-dierence approach.
There are a number of mechanisms by which industrialization may have altered a household's fertility out-
K ögel
come. First,Galor and Weil (2000),
and Prskawetz (2001), Hansen and Prescott (2002), and Tamura
(2002) all present models of economic growth and fertility decline that highlight the role of human capital
in increasing the incentives of households to invest in child quality over quantity. Second, industrialization
may have induced a rise in the implicit costs of raising children. In particular, industries with high rates
of female employment increased the opportunity cost of female time. Under the assumption that the child
production process is female-time intensive, this would have reduced the incentive to bear children.
6 Third,
industrialization altered opportunities for child labor and increased the wage dierential between children
and adults. The movement away from agricultural and at-home production to centralized production, in addition to more restrictive child labor laws, may have reduced the economic return to children, again lowering
7 , 8 Fourth, industrialization was associated with increased urbanization
parental demand and fertility rates.
6 See
Lagerlöf (2002) and Galor and Weil (1996). Schultz (1985) uses an instrumental variables approach to show that an
increase in the ratio of women's to men's wages can explain up to 25% of the Swedish fertility decline between 1860 and 1910.
Brown and Guinnane (2002) and Crafts (1989) document a negative correlation between female labor force opportunities and
fertility rates in Bavaria and England/Wales, respectively.
7 Doepke
(2004) argues that child labor regulation is a critical component in explaining fertility declines during periods of
economic growth.This view is shared by Hazan and Berdugo (2002).
8 Moehling (1999), on the other hand, argues that in the United States, child labor legislation was enacted only after industry's
resistance had waned and thus had very little impact on actual child employment.
2
and the crowding that occurred may have increased the explicit costs of raising children through higher
housing and food costs without an associated increase in the benet.
9 Finally, the developing economy in
the United States witnessed decreases in child mortality rates, especially after 1880.
10 In the short-run,
increased population density and hazardous work conditions may have contributed to higher child mortality
rates. But in the long-run, rising incomes resulting from economic growth outweighed this eect by providing
families the means for better nutrition and sanitation.
11
Given the multitude of potential mechanisms linking industrialization with reduced fertility, the purpose
of this paper is threefold. First, I provide empirical evidence to support the hypothesis that industrialization
was a cause of reduced marital fertility by documenting the change in fertility resulting from a plausibly
exogenous increase in industrial activity in South Carolina between 1880-1900 as described above. Second,
I show that an increase in urbanization and the costs of raising children, including a rising opportunity cost
of female time, can explain part of the result while mechanisms relying on human capital considerations
and increases in the adult/child wage dierential are not supported by the available data.
I also nd
some evidence of dierential infant mortality among industrial workers. Finally, I propose a third potential
mechanism linking economic growth to lower fertility rates:
an increase in the costs of raising children
resulting from the separation of households from their extended families. A pronounced fertility dierence
in migrants relative to the native population in industrial locations lends support to this hypothesis.
1
Empirical Strategy
This paper will measure the impact of industrialization on fertility rates using data from rural areas of South
Carolina between 1880 and 1900. In 1880, South Carolina was home to 14 textile mills and 84,424 spindles,
a standard measure of industrial textile capacity. By 1900, the state housed 93 mills and 1,693,649 spindles,
an increase of almost 2000%.
12 There are several unique aspects of the South Carolina industrialization
process that make it an ideal laboratory in which to test hypotheses about industrialization and fertility
9 It
is an empirical regularity that urban fertility rates were lower than rural ones as early as the 19th century, but a precise
mechanism has not been established. See Vinovskis (1976) for the empirical results and Guinnane (2010) for a more detailed
discussion.
10 See Haines (2000).
11 The expected correlation
between child mortality and fertility is ambiguous. See Doepke (2005) for a negative correlate
between infant mortality and gross fertility and Sah (1991) and Kalemli-Ozcan (2003) for a positive correlate.
12 Carlton
(1982), p.40.
3
decline. First, prior to 1910 or so, textile mills provided the overwhelming majority of industrial employment
in the state. In 1900, 30% of males and 57% of females employed in manufacturing were textile workers. No
other industry came close to matching the textile share of employment, and most of the runners-up involved
manufacturing in a decentralized setting (e.g., carpentry at 7.5% of male employment; dressmaking and
seamstresses at 32.3% of female employment).
13 As a result, South Carolina industrialization in general can
be proxied by the textile industry in particular. Second, the particular locations of textile mills in rural areas
do not appear to have been driven by the demographics of the local labor force or by other characteristics
of the location that would have, in turn, been correlated with fertility.
This second point is critical for an empirical strategy which compares
ex post
fertility rates for two
neighboring townships - one that receives a mill and one that does not. In the absence of random assignment,
in order for results to be meaningful, it must be true that mill townships and non-mill townships did not
dier in other ways. In particular, they must not have diered in ways correlated with fertility.
For South Carolina between 1880 and 1900, it appears that the location of textile mills was uncorrelated
with fertility
ex ante.
The Piedmont region where textile manufacturing focused after 1880 was homogenous
in terms of its economic activity, socio-economic characteristics, and climate prior to industrialization. And
the area was homogenous in other, less tangible, ways as well.
[T]he Civil War destroyed not only slavery, but any prewar county dierences in the popular
attitude toward agrarianism vis-à-vis industrialism. Certain postwar developments, such as the
introduction of commercial fertilizers and the subsequent general spread of cotton throughout
the [Piedmont] area, the breakdown of plantations, the rise of the crop-lien merchant credit
system and the cropping system and their subsequent general introduction to all area counties,
the completion of the rail network that blanketed the entire area with a striking uniformity, and
the triumph of Solid South politics - also brought to the area a degree of homogeneity hitherto
14
virtually unknown.
This emerging uniformity was present in labor markets throughout the region. This homogeneity justi-
13 Author's
analysis of IPUMS 1900 1% sample. See Appendix A for a detailed breakdown of 1900 employment for South
Carolina workers.
14 Tang
(1958), p.64
4
es the assumption that textile mill locations were selected independent of other drivers of fertility rates,
especially after controlling for the presence of railroads and other township observables. In many ways, if
you build it they will come applied to the South Carolina rural population in this period. Rural households
were desperate to leave their farming professions. Cotton and tobacco prices plummeted in the 1890's and
farming households were adversely impacted. Mill owners could have picked virtually any rural location and
they would have found a local labor force eager to become mill workers. It seems unlikely, therefore, that
mill proprietors would have chosen one location over another in order to enjoy monopsony power in the local
labor market.
Of course, mill proprietors did, in the end, choose particular locations over others in locating their mills.
In the early part of the industrialization period, particularly prior to 1890, mills located on the banks of
rapidly-owing rivers. These location decisions were driven both by concerns about water power and by the
need for natural humidity to support the manufacturing process. Any river location could supply humidity,
so the availability of quickly owing water capable of driving a water wheel dictated location in this early
period. The demographics of the local population would have been, if anything, secondary.
The rst steam-powered mill was built in South Carolina in 1881 and, after 1890, steam power began to
dominate as a power source. Cotton mills among the cotton elds became possible, and mill proprietors
were freed somewhat from their reliance on fast-owing water.
At rst, steam and water were used in
combination to power textile mills, but by the second half of the 1890s, most mills ran on steam alone.
15
Proximity to water was still desirable for waste removal and natural humidity. But steam power also required
massive amounts of coal input, and coal was an expensive commodity to transport. As a result, proximity
to railroad lines likely replaced water ow as the primary driver of mill location by the late 1890's.
rural locations, railroad presence was a more-or-less idiosyncratic process.
15 The
In
16 Still, I control explicitly for the
Seventh Annual Report of the U.S. Commissioner of Labor on the Cost of Production: The Textiles and Glass reports
that many mills in 1890 used a combination of steam and water and a few mills operated on steam alone.
16 If railroad location was,
in turn, dependent on some aspect of local population also correlated with fertility, this exercise fails
to satisfy the requirements for OLS estimation. The historical record indicates that railroads were built in the years following
the Civil War for two reasons. First, railroads connected commercial centers in the state. In this case, their existence in rural
communities was a byproduct of these urban connections. Potential connecting routes between two larger urban centers were
dierentiated, in part, by construction costs. Rural locations lying on the low-cost route would have been traversed by railroad
track and therefore more likely to house a textile mill. Signicant commercial activity in these rural locations would have been a
dierentiator as well. Second, railroads were constructed to connect certain areas rich in natural resource to the major railroad
networks. Short-line railroad tracks were built to service lumber mills and mines (coal and otherwise) which would have, by
necessity, located near the natural resource they were transporting. Again, rural areas located between these mills/mines and
the main railroad network would have been traversed by railroad track as a byproduct of this system. In either case, once the
railroad track was laid, its existence drove the location of textile mills in these rural locations over those without railroad access.
5
existence of a railroad in the empirics of Sections 2 and 3 in order to account for the possibility that railroads
had an independent eect on fertility.
After 1900, the pre-1900 drivers of location decisions were waning, and labor market considerations were
becoming more important as South Carolina mills began producing a higher quality cloth that required
17 As a
skilled labor. At the same time, electric power was freeing rms from prior geographic constraints.
result, 1900 was chosen as the endpoint for this analysis.
For these reasons, I assume (and later test) that mill location prior to 1900 was independent of local
fertility rates. I do not expect that mills located in townships in South Carolina where fertility was already
low (or high) relative to neighboring townships. All the same, I will test whether pre-mill fertility rates in
townships that eventually housed a textile mill were similar to rates in townships that did not. These
ex ante
results and initial estimates of the impact of textile mills in 1900 on local fertility are contained in Section
2.2. I nd that townships housing a textile mill had fertility rates up to 10.4% lower than non-mill locations
by 1900.
The results are validated in Section 2.3 using a dierence-in-dierence approach.
I use township-level
fertility rates from 1880 and 1900 to generate a dierence-in-dierence estimate for the impact of industrialization on fertility. The industrialization period in South Carolina began in the 1880s, and only 14 mills
were extant in 1880, making it a reasonable year in which to measure
was destroyed by re.)
ex ante
fertility. (The 1890 Census
The dierence-in-dierence results conrm the OLS conclusions and generate an
estimated fertility reduction of between 6 and 10%.
In Section 3, I use household-level Census data to add controls for migration and race. The results indicate
that the in-migration of low fertility households was an important driver of the fertility result. Given this
migration result, in Section 4 I evaluate the potential mechanisms at play and nd that higher rates of textile
employment and child mortality characterized the migrant experience. In addition, I hypothesize that the
loss of an extended family raised the implicit costs of child-rearing for these households as well. I further
argue that other mechanisms for the relationship between fertility and economic growth are not supported
See Atack et al (2010) for evidence that railroad presence led industrial activity in the Midwest. In the end, the nal route
taken by a particular railroad line is the outcome of an idiosyncratic process involving political forces and nancial incentives,
in addition to construction costs and natural resource locations as discussed above. See Stover (1955).
17 The
rst electrically-powered mill, Orr Mills, was built in Anderson, South Carolina in 1899. Electric adoption was slow.
By 1920 only about 50% of extant mills ran on electric power. (Source: Author's analysis of 1919-1920 Davison's Guide.)
6
for this particular historical episode, and Section 5 concludes.
2
Township Results
2.1 Township Data
Demographic and fertility data comes from the manuscript returns of the United States Census. The U.S.
Census was taken at the individual level.
Individuals were grouped into households which were, in turn,
grouped into townships. Townships were grouped into counties and then into states. Using data from the
online genealogy tool Ancestry.com, I assemble age structure and marital status data for townships in both
1880 and 1900 to measure a township-level impact of industrialization on fertility.
18
I focus on marital
fertility rates as almost all fertility occurs within marriage and Sanderson (1979) indicates that declines in
marital fertility (rather than declines in marriage rates) contributed most to the U.S. decline in this period.
In each year (t=1880, 1900), I calculate F5 fertility rates:
F 5jt =
# of children < 5 years of age in township j in year t
M arried, divorced, and widowed f emales aged 18 to 42 in township j in year t
I also perform sensitivity analysis in Section 2.2 using F3 fertility rates and eliminating widowed and
divorced women from the denominator.
19
Industrialization data comes from Davison's Blue Book, a directory of textile mills in the United States.
The directory was printed biannually beginning in 1888. I use the 1902-1903 edition to generate a timeline of
20 Davison's gives an establishment date for each indexed
mill establishment in South Carolina prior to 1900.
mill. From this data, I generate two industrialization indicators:
mill by 1895 and
Ij,1900 = 1
Ij,1895 = 1
if township j housed a textile
if township j housed a textile mill by 1900 (but not by 1895).
Additional explanatory variables are available as well: a binary variable for whether the township was a
county seat in 1900 (CT Y
18 There
SEAT1900 )
and an indicator for the presence of a railroad in the township in 1890
are 484 South Carolina townships in 1900 and 461 in 1880. In order to perform dierence-in-dierence analysis in
Section 2.3, I must create consistent township barriers between years. I do this using the county boundary descriptions from
the two census enumerations.
The process is not a precise one and requires some judgement calls.
To generate consistent
boundaries, I must consolidate some townships.
19 There
is no particular justication for using F5, but a sensitivity test using F3 generates similar results.
Preston et al
(1992) document a tendency of African American women to legitimate an out-of-wedlock birth by claiming a marital status of
married, widowed, or divorced when the female was actually single. This should serve to make my estimate more accurate as
it ensures that even out-of-wedlock fertility is appropriately captured.
20 An
accessible copy of the 1900-1901 edition has not been located. The possibility remains that mills opened and closed
prior to 1902, impacting fertility but not appearing in my industrialization measure. This would tend to bias the estimated
impact of industrialization towards zero.
7
21
(RR1890 ).
2.2 OLS Results - 1900
To test whether industrialization in South Carolina between 1880 and 1900 had an impact on townshiplevel fertility rates, I rst run a simple OLS regression of F5 in 1900 (the number of children under age 5
as a percentage of the fertile married, widowed, or divorced population) on township-level characteristics,
including the presence of a textile mill. The estimating equation is:
F 5j,1900 = α + δ1Ij,1895 + δ2Ij,1900 + βXj,1900 + j,1900
and the estimation is taken over 341 South Carolina townships in 1900.
(1)
I control separately for the
industrialization of the county in 1895 and in 1900 in order to account for the fact that the F5 fertility
measure encompasses ve years of fertility. (Ij,1900
1896 and 1900.
Ij,1895 = 1
= 1 if a textile mill was established in township j between
if a mill was established before 1895. Because
F 51900
is measured in 1900 and
represents ve years of fertility outcomes, a textile mill built in 1898 may not have the same impact on
F 51900
as one built in 1892, and controlling separately for the two time periods accounts for this.)
In the baseline specication, components of
Xj,1900
include
CT Y SEAT1900
and
RR1890 .
Variable means
and standard deviations are located in Table 1.
The universe of townships included in the estimation are those without a mill established prior to 1880,
(98.2% of all South Carolina townships), those with a town population of fewer than 2,500 individuals in
1900 (95.9% of the remaining townships), and those located outside of the state's coastal counties. Coastal
counties were not as heavily dependent on agriculture prior to the arrival of the mills, did not participate
in the textile industry, and were arguably dierent from the Piedmont counties in other ways as well. The
restriction on the town population is a recognition that, in larger towns, industrialization came in forms other
than textile mills, and the industrialization proxies used herein are inappropriate. The United States Census
Bureau, in 1910, dened an urban location to be one with greater than 2,500 inhabitants. The Bureau went
21 The
siasts:
railroad data comes from a combination of printed maps of railroads and indexes constructed by railroad enthu-
Rand McNally and Company's Indexed County and Railroad Pocket Map of South Carolina, 1892 edition, and
http://www.carolana.com/SC/Transportation/railroads/home.html.
I use these two sources to generate an indicator for
whether a South Carolina township was traversed by a railroad line in 1890. 1890 was chosen as an intermediate year. The
process for generating a railroad variable for each township is extremely labor intensive. The variable does not signicantly
alter the results in this section and was not recreated for 1880 and 1900.
8
on to note that in most regions, densely populated areas of 2,500 or more are set o from rural territories
22
and incorporated as municipalities (cities, towns, villages, boroughs, etc.). I simply follow this standard.
Results are located in Table 3 and indicate that industrialization, as proxied by textile mills, was accompanied by lower fertility. In Column (1),
weaker, negative correlation with
I1900 .
F 51900
exhibits a strong, negative correlation with
I1895
and a
The presence of a textile mill built prior to 1896 is associated with
a reduction in 1900 township fertility rates of 10.0% (=0.123/1.22). As expected, the presence of a textile
mill built after 1895 has a smaller impact of 7.5% (=0.091/1.22).
23
The presence of a textile mill in a township is correlated with lower fertility rates in 1900. What observable
characteristics of textile locations might explain this correlation?
In Column (2), I add controls for the
percentage of a township's population living in a census-dened town (T OW N P OP OF T OT ALP OP )
and the percentage of the township's population with a non-white reported race (P CT N ON W HIT E ). The
coecient on
I1895
(the indicator for pre-1895 industrialization) increases slightly (in absolute value) as a
I1900
result and remains statistically signicant but that on
(the indicator for post-1895 industrialization) is
reduced substantially (in absolute value).
Thus, results in the rst panel of Table 3 indicate that industrialization and marital fertility are negatively
correlated in the cross-section, even after controlling for racial composition and urbanization. But could it
be that low fertility locales were more likely to receive textile mills, implying reverse causation? In other
words, can the impact of industrialization, as proxied by a textile mill, be detected prior to the arrival of the
mill itself ? If so, that calls into question whether the location of these textile mills was independent of local
fertility rates. In Column (3) of Table 3, I undertake a falsication test. I estimate the impact of a textile mill
built between 1881 and 1900 on the marital fertility rate of townships in 1880, before the mills were built.
24 The results in Column (3) show only weak evidence
Control variables are the same as those in Column (1).
of dierences in fertility
22 Sensitivity
ex ante.25 F 51880
exhibits weak correlation with the post-1880 industrialization of
tests (not shown) to setting the cuto at 1,000, 2,000, 3,500, and 5,000 do not show any remarkable dierence
in conclusions relative to the baseline.
23 There
may be some concern that the 1900 town population is endogenously determined by the existence of the textile mill
and its ability to draw citizens to central locations. Limiting the data to those townships with a town population of less than
2,500 residents in 1880 (not shown) gives slightly more negative estimates for these coecients and higher levels of signicance
as well.
24 Using CT Y SEAT
1880 generates similar
P CT N ON W HIT E as control variables.
25 There
conclusions,
as
does
the
addition
of
T OW N P OP OF T OT ALP OP
AND
is, however, a signicant dierence in pre-industrial marriage rates in townships that were eventually industrialized.
See Appendix B for a matching estimator that accounts for this dierence.
9
a township for those townships with a town population of less than 2,500 by 1900. The point estimate for
δ1,
-0.037, or 2.6%, has a p-value of 0.26. While this estimate is not statistically signicant, I validate the
OLS results by estimating a dierence-in-dierence specication in the next section.
2.3 Dierence-in-Dierence Results
In order to account for the possibility that townships that industrialized between 1880 and 1900 and those
that did not were dierent in some way
ex ante,
I incorporate a dierence-in-dierence estimator for the
impact of industrialization on fertility. I dene two treatment eects for township j as follows:
Tj,t,1900 = Ij,1900 ∗ 1(t = 1900)
and
Tj,t,1895 = Ij,1895 ∗ 1(t = 1900)
where
1(t = 1900)
is an indicator equal to 1 in 1900 and 0 in 1880 and
Ij,t
was dened previously.
The dierence-in-dierence estimates for the eect of industrialization on fertility are the coecients
and
δ2
δ1
in the estimating equation below:
F 5jt = c + µ1Ij,1895 + µ2Ij,1900 + δ1Tj,t,1895 + δ2Tj,t,1900 + βXjt + λj + θ1(t = 1900) + jt
where
Ij , Tjt ,
and
Xjt
were dened previously.
1900-specic year eect, and
jt
c
is a constant,
λj
is a township-specic xed eect,
θ
is the
26 The estimation is done over t=1880, 1900.
is a random error term.
The rst-dierence estimating equation is:
∆F 5j = δ1Ij,1895 + δ2Ij,1900 + β∆Xjt + θ + ˜j
where
˜j = j,1900 − j,1880 .
treatment variables,
(2).
λj
Tj,t,1895
and
The industrialization proxies,
Tj,t,1900 ,
equal
Ij,1895
and
(2)
Ij,1900
0 in 1880 and are replaced by Ij,1895
dierence out and the
and
Ij,1900
in Equation
also dierences out of the equation.
26 CT Y SEAT , T OW N P OP OF T OT ALP OP ,
coverage variable is not.
AND
P CT N ON W HIT E
variables are available for 1880, but the railroad
Excluding the railroad variable from the specications in Columns (1) and (2) of Table 3 has no
notable impact on the estimates, and I exclude it from this specication entirely.
10
Column (4) of Table 3 contains the estimates from Equation (2). The estimated coecient on industrializaton prior to 1895 (δ1) is reduced by approximately 30% relative to Column (1), but remains statistically
signicant at the 5% level and indicates a fertility reduction associated with pre-1895 industrialization
of 6.9%.
In Column (5), I again ask whether this result can be explained by changes in the township's
racial composition and population density.
P CT N ON W HIT E .
I control for changes in
T OW N P OP OF T OT ALP OP
and
These variables can explain part of the reduction in fertility in industrialized town-
ships, but a signicant portion of the eect remains unexplained.
In Columns (6) and (7) of Table 3, I undertake two robustness checks on the dierence-in-dierence result.
In Column (6), I limit the denominator to married females (excluding widowed and divorced females). Again,
there is little change in the results. To determine the sensitivity of the results to the fertility measure used,
I repeat the estimation on
3
F 3j,1900 .
There is little change in coecients as reported in Column (7).
Controlling for Migration
The advantage of township data is that townships span the 20 year industrialization period, thereby allowing
a dierence-in-dierence estimate. But townships are not composed of an immobile set of individuals and
the fertility pattern documented above could simply be a result of selective migration without changing the
fertility outcomes of households
per se.
In this section, I generate a new sample of households that allows
me to control for migration patterns.
3.1 Household Data
In order to evaluate the impact of migrating households on the township-level results documented previously,
the location of the household at two points in time is needed.
The 1900 Census contains information on
an individual's state of birth, but not their county or township.
As the vast majority of South Carolina
residents in 1900 were born in South Carolina, this variable has limited power to identify migrants.
I remedy this problem using the genealogy tool Ancestry.com to link individuals between the 1900 Census
and the 1880 Census and use their township location in both to infer migration. I begin by using the 100%
sample of the 1880 U.S. Census from the North American Population Project to gather information on
284,412 South Carolina resident males aged 0 to 20 in 1880. Of those, 34,841 are uniquely matched to the
11
1900 U.S. Census using the search function of Ancestry.com and the individual's full name, race, and age
27 , 28 Of those, 12,999 are married with their spouse present in the household
(± 1 year) to make a match.
and reside in South Carolina townships with fewer than 2,500 town residents in 1900 and outside of the
coastal counties.
This subsample of married South Carolina males, aged 19 to 41 in 1900 with a spouse
present in the household, is the basis for analysis in this section. The 1900 Census return as transcribed by
Ancestry.com contains a full list of household members, their name, relationship to the head of household,
age, race, marital status, marital duration, and birthdate. (Occupation data also exists but has not been
transcribed in the Ancestry.com database.) The location of the male head at two points in time, 1880 and
1900, gives me a measure of household migration.
29
Two caveats are in order, both stemming from the absence of the 1890 Census. First, without the 1890
data, linkages must be made across 20 years instead of ten, and the tendency of individuals to misreport
their name and/or age increases with time. This reduces the matching success rate and makes successful
30 Second, because the objects of interest are young, fertile households in 1900, I can
matches less accurate.
only link the male head as he is the only member of the household who would have been alive and with the
same last name in 1880. (Females would have been enumerated under their maiden name in 1880.) That
presents the possibility that even though the male head did not migrate between 1880 and 1900, his wife did
(or vice versa), and I will not capture that in the subsequent analysis.
31
3.2 Migration Results
To measure the impact of migration on the results previously reported, I estimate an equation of the following
form:
F 5HH
i,j = c + δ1Ij,1900 + δ2Ij,1895 + θYi + βXj + αM OV ERi + i,j
where
F 5HH
i,j
27 Ancestry.com
(3)
is the (discrete) number of children less than age 5 in household i in township j, analogous
is a subscription genealogy database. Every individual in the 1880 and 1900 U.S. Census has been digitized
on this site and is searchable by a subset of household characteristics, including name, race, and age. Some information has not
been transcribed, including occupation.
28 Failed
matches result both from locating no matching individuals in 1900 (75% of failed matches) and from locating more
than one matching individual in 1900 (25% of failed matches).
29 The
data generating process yields an out-of-township migration rate of 67% (including out-of-county and out-of-state
migration), an out-of-county migration rate of 50% (including out-of-state migration), and an out-of-state migration rate of
16% for married household heads between 1880 and 1900. These rates will reect both legitimate moves and mis-matched data
pairs.
30 This will serve to bias any impact of migration on
31 The size or direction of this bias is unknowable.
fertility towards zero as I will be falsely classifying stayers as movers.
12
to the F5 fertility rate from Section 2. The vector of township characteristics,
Yi
Xj ,
is dened in Section 2.1.
contains the household head's age, age squared, wife's age, age squared (all contained in
and a race indicator for the household head (RACE ).
M OV ERi
AGEV ECT OR),
is an indicator for whether household i
remained in the same township between 1880 and 1900. The coecient
α
measures dierential fertility for
township movers, those who change townships between 1880 and 1900.
Table 2 gives sample means for variables contained in the household-level data. As a consistency check,
F¯5
from Table 1 and
HH
F¯5
from Table 2 are similar (1.22 in both cases).
An ordered probit estimator is used for Equation (3) as values for
standard errors for
32 Coecients and
are discrete.
δ1, δ2, and α are reported in Table 4, but no longer have a marginal eect interpretation.
Instead, I compute the marginal eect on the expected value of
E(F 5HH )
F 5HH
F 5HH
implied by
δ1,
i.e., the change in
resulting from the arrival of a textile mill by 1895.
I rst determine whether the township and household data give generally consistent results in estimation.
The two datasets are not the same in the limit as restrictions on the male household head (aged 19 to 41
in 1900, born in South Carolina, etc.)
not weighted by population.
33
are not present in the township data and township results were
With that caveat, in Column (1) of Table 4, I estimate the impact of
industrialization on fertility in the household data controlling only for variables available in the township
dataset. The results indicate a negative correlation between
F 5HH
and pre-1895 industrialization of about
4.1% in the household sample compared to 10.0% in the township data (6.9% in the dierence-in-dierence
results).
34
Column (2) of Table 4 adds additional controls for race and age but does not yet control for migration.
The implied reduction in fertility corresponding to pre-1896 industrialization (estimated by
In Column (3) the
M OV ERi
δ1)
indicator is included in Equation (3) and the estimate for
α
is 4.1%.
is negative
and signicant. But bisecting the sample into movers (Column (4)) and non-movers (Column (5)) clearly
shows that, among migrants, residence in an industrial township, proxied by
I1895 , is associated with reduced
fertility. Movers in Column (4) show a 6.9% reduction in fertility when moving to a township that housed a
32 See Appendix C for details. Poisson estimation does not remarkably alter the results.
33 Weighting the results in Table 2 by township population does not remarkably alter the
results. It is not possible to restrict
the datasets such that the household data is a subset of the township data because age of male spouse is not an available
variable in the township data.
34 The
data exclude households with a female spouse less than 15 years of age and greater than 50 years of age as these entries
are more likely data errors than actual spousal matches.
13
textile mill by 1895. On the other hand, non-movers in those same townships exhibited increased fertility of
1.1% (not statistically signicant). This is a striking result as it indicates that households who remained in
textile locations between census enumerations exhibited no change in fertility relative to those households who
remained in a non-textile location (represented by the
δ1
coecient in Column (5)). The industrialization
process was correlated with low fertility, as earlier results have indicated, but migrating households were
responsible for the negative correlation.
There are (at least) three potential explanations for this result.
First it is possible that migrants to
textile townships were disproportionately from low-fertility native townships (urban or otherwise) relative
to other migrants and brought these fertility patterns with them. But adding control variables for average
fertility and population, either in 1880 (Column (6) of Table 4) or 1900 (Column (7)), of the migrant's home
township does not reduce the estimate of
δ1
35 These conclusions indicate
coecient relative to Column (4).
that the nativity of movers to textile townships cannot explain their low fertility behavior.
A second potential explanation is that movers to textile locations were negatively selected and had low
fertility rates relative to their native townships. Perhaps these low-fertility households were attracted to the
labor markets in industrial locales and brought their low fertility with them. But, if so, we would expect
migrants to townships where mills were built late in the period to have low fertility as well, and that is
not the case. In every specication of Table 4 involving mover households (Columns (4), (6), and (7)) the
estimate for
δ1 is large and negative while that for δ2 is insignicantly dierent from zero.
Migrants exhibited
lower fertility than non-migrants in textile locations, but only in locations where they had been exposed to a
textile mill for ve years or longer. Migrants to townships where mills were relatively new (i.e., built between
1896 and 1900, represented by
I1900 = 1)
did not have lower fertility rates than their peers. This points not
to selective migration of households, but to a dierent fertility response to residence in a textile locale.
This gives rise to a third explanation. In the next section, I document that female migrants to textile
townships exhibited occupational selection that would have contributed to a higher opportunity cost of
female time and manifested itself in higher rates of child mortality. In addition, I postulate that the loss of
proximity to extended family would have also characterized the migrant experience and served to increase
35 The
variables
HOM ET W P F1880
and
HOM ET W P F1900
F 5 marital fertility of a household's home
HOM ET W P P OP1880 and HOM ET W P P OP1900 represent
represent the average
township measured in 1880 and 1900, respectively. The variables
the average town population of a household's home township measured in 1880 and 1900, respectively.
14
36
the cost of raising children as well.
4
Evaluating the Mechanisms Behind Fertility Decline
Township and household data indicate a reduction in fertility in textile locations driven by low-fertility
migrants. In this section, I evaluate the ability of various mechanisms for the relationship between fertility
and economic growth, as laid out in the introduction, to explain the result.
4.1 Migration, Occupational Choice, and Child Mortality
In Section 3, I identied migrating households as the driver of reduced fertility in textile locations and argued
that fertility of migrants diered from natives after their arrival in textile locations, but not neccesarily before.
In other words, industrialization seems to have impacted township movers more strongly than stayers.
What might explain the dierential impact of industrialization on movers and stayers?
choice seems a natural place to look for dierences between these two groups.
Occupational
But occupation data for
the 14,360 individuals contained in the household data set used in Table 4 has not been transcribed by
Ancestry.com. I transcribed the occupation of the male and female household heads for the 2,342 households
residing in townships in 1900 that were industrialized between 1880 and 1900 (i.e., where
I1900 = 1)
from the original census enumerations.
I1895 = 1
or
I divide occupations into three categories: Farmers &
37 Proportions of the sample in each category is shown in
Farm Laborers, Textile Mill Workers, and Other.
Table 5. It is clear that movers had a higher textile employment rate than did stayers. 4% of working men
classied as stayers in a textile township were employed in textiles while 11% of men classied as movers
were so-employed. Among women, the numbers were 3% and 7%, respectively. Movers also had a higher
rate of employment in non-farm occupations in general.
Adding occupation data to the regressions indicates that these two occupation categories, Textile and
Other employment, are signicantly, negatively correlated with fertility. A simple ordered probit estimation
of Equation (3) on households in industrial townships conrms, in Column (1) of Table 6, that fertility is
strongly correlated with migrant status in industrial locales. In Column (2), I add a vector of dummies to
36 In
addition, it does not appear that migrants' fertility rates were mis-measured due to leaving children behind, the so-called
missing children hypothesis. (See Moehling (2002).) There is no discernable dierence in the ratio of children born to children
in the household for young migrants compared to young stayers in industrial townships.
37 The
Other category is comprised mostly of service occupations.
15
control for the occupation of the household heads, both male and female. (Farmer is the omitted variable
for males, and no occupation is the omitted variable for females.) Occupations are strongly correlated with
fertility, especially for females employed in the textile industry. Conditional on occupation, the correlation
between migration and fertility is reduced by approximately 25% (from 9.8% to 7.2%). But the estimated
impact of industrial location on fertility (α) remains signicant, even after controlling for occupation.
Occupational choice, then, can explain part of the dierence in fertility between movers and stayers in
textile locations. In addition, the combination of occupational choice and migrant status had a signicant
impact on infant and child mortality rates in industrialized locales, which may have had an additional
negative impact on observed migrant fertility rates.
The 1900 Census asked married women how many children they had borne and how many were surviv-
38 Taking the sample of married females from the sample of the 1900 Census described in Section 3.1,
ing.
it is straightforward to calculate a child survival rate as the ratio of the number of children surviving to the
number ever born, and then to examine the relationship between that number and textile mill location.
39
I estimate the relationship between a married female's reported child survival rate and the presence of a
textile mill in the household's 1900 township location, conditional on other characteristics of the household.
The estimating equation is:
Sij = τ Dij + αM OV ER + βXi + θYj + i,j
where
Si,j
is the observed child survival rate for household i in township j in 1900.
(4)
In contrast to
the fertility results in Sections 2 and 3, the survival measure is not limited to the ve years prior to the
1900 Census enumeration.
Instead, the measure will reect cumulative infant and child mortality over
the entirety of a female's childbearing years. The
instead I construct a variable,
I1895
and
I1900
variables are no longer appropriate, and
Dij , that represents the proportion of household i's marital duration for which
40 The same set of control variables are used to estimate Equation (4) as in
township j housed a textile mill.
38 Unfortunately,
standardized infant and child mortality statistics for South Carolina townships do not exist for this time
period.
39 A
caveat is in order: this statistic will reect the cumulative number of deaths of the household's children (in infancy or
otherwise) over the entire span of the female's childbearing. It is impossible to determine from this data whether these deaths
occurred in years before or after the introduction of a township's textile mills.
40 For
example, if township j received a textile mill in 1896 and household i was married in 1894, this number is 0.67, etc. To
the extent that households moved between the date of marriage and 1900, this exposure measure will be mis-measured.
16
Equation (3) above, and the same town population and age restrictions apply.
Table 2 gives the mean and standard deviation of
Sij
and
Dij .
Results, located in Table 7, indicate
no discernible dierence in infant and child mortality rates between residents of industrialized and nonindustrialized townships as evidenced by Column (1). Restricting the sample to younger women or newly
married couples does not change the result (not shown).
41 These are small numbers; the
τ
coecient in
Column (1) represents a reduction in survival rates of 0.9%.
Column (2) of Table 7 adds a control for migration status and the results are unaected. In Column (3),
I add an interaction between mover status and the exposure measure,
Dij .
The resulting estimates indicate
that infant and child mortality was signicantly higher among movers to industrial locations than among
their peers. A mover to a township with a textile mill present for the entirety of their marriage duration
experienced a reduction in child survival rates of 3.7% (0.031/0.84) relative to non-movers or movers to
non-industrial locales. Adding controls for occupation in Column (4) indicates that the occupationl choice
of migrants can explain part of their adverse child mortality consequences. Female employment in a textile
mill had a large, negative correlation with child mortality rates, while employment of the male had a more
muted correlation. Conditional on occupational status, there is no signicant impact on survival rates of
living in a textile location or of being a migrant.
Higher levels of child mortality and dierences in occupational choice may explain lower observed fertility rates among migrants in industrial locales.
In the next section, I evaluate the remaining potential
explanations for the link between industrialization and fertility and propose an additional explanation for
the observed reduced fertility among township movers.
4.2 Evaluating Potential Mechanisms
The introduction to this paper listed several ways in which economic growth may have aected fertility rates.
Increases in the return to human capital, increases in the costs of raising children (including the opportunity
cost of women), altered opportunties for child labor, increased urbanization, and changes in infant and child
mortality were all presented as possibilities. The available evidence indicates that increases in the costs of
raising children, heightened child mortality, and increased urbanization were all signicant factors in inducing
41 Replacing D
ij
with the industrialization proxies employed previously generates the same conclusion.
17
lower fertility in South Carolina textile locations while human capital and child labor considerations were
not at play.
Changes in the costs of raising children emerge as an explanation for the results in this paper for two
reasons. First, female occupation is a strong predictor of household fertility in the sample, and low migrant
fertility can be explained in part by higher rates of female employment in textile and service industries.
Fertility is most strongly associated with textile employment, but there is a signicant impact among service
workers as well. African American women employed in the service industry, who were themselves disproportionately migrants, exhibited signicantly lower fertility than their peers. (Results not shown.) Thus,
industrialization in South Carolina had both a direct and an indirect eect on the cost of female time.
Women were employed by the mill directly, and their mill employment would have been incompatible with
child-bearing. Unlike the farm where pregnant and nursing mothers could integrate childbearing and rearing,
the mill setting would have prevented such an integration. Data on the employment of married females in the
textile mills from the 1% IPUMS sample of the 1900 Census indicates that, of self-described cotton mill operatives, 43% were female and, of that number, 18% were married females.
42 In addition, the industrialization
process in South Carolina created a tertiary service sector within local economies. The arrival of a textile
mill increased demand for female-provided services such as cooking, laundering, growing small amounts of
crop and livestock, and boarding. These occupations were dominated by African American females.
Second, the observed dierence in fertility for migrants relative to natives is consistent with a dierent cost
of children explanation. As noted in the last section, dierences in occupational choice and infant and child
mortality can explain some of the dierence in fertility rates between migrants and natives. The dierence
between estimates of
δ1
and
δ2
in Column (4) of Table 4 indicate that the reduced fertility of migrant
households occurred after their arrival in industrialized townships and increased with their tenure. What
other aspect of living in an industrialized township would have dierentially aected migrants relative to
natives? One potential answer: the costs of child care. For natives in industrialized locales, extended family
could have served as a backstop for childcare.
But for migrants this option was not available.
Childcare
in the mills was sporadically supplied, and most families would have relied on household members for this
42 United
States Bureau of Labor: Report on Condition of Woman and Child Wage-Earners in the United States, 1910
18
service. The lack of extended family members to perform this task would have resulted in an increase in the
cost of raising children for migrants relative to natives and may have been another cause of the observed
fertility dierential.
43
An increase in child mortality was also highlighted in the last section as a potential reason for lower
migrant fertility in the textile townships of South Carolina. The available data also suggests an increase in
population density was an important factor in inducing lower fertility. In the township results of Table 3,
controlling for the population density of a township (T OW N P OP OF T OT ALP OP ) reduces the coecient
on the industrialization proxy in Column (5) by 29% for townships industrialized after 1895 and by 18%
for townships industrialized before that date. Thus, the increase in urbanization resulting from textile mill
44
establishment can explain some of the fertility impact of industrialization.
On the other hand, increases in the return to human capital and reductions in the labor opportunities
of children are unlikely drivers of this result. Unlike in other industries, there is little evidence that textile
manufacturing increased the return to human capital or incentivized parents to invest in child quality over
quantity. Textile mills relied on low-skilled operatives, and there was no obvious increase in the return to
45 Alternatively, a variant of the original quality/quantity hypothesis
education, even in spillover industries.
focuses on the potential impact of an increase in income on parental preference for quality over quantity.
But this also has little support in the data as indicators for male occupational status (a proxy for household
income) have a muted correlation with fertility in comparison to those for female occupational status (a
proxy for both household income and the opportunity cost of female time). (See Table 6.) This seems to
disfavor a child human capital or quality/quantity tradeo explanation.
46
In addition, a decrease in the labor opportunities of children is an improbable mechanism. The potential
43 This hypothesis is further supported by the fact that all migrants,
in industrial locales and otherwise, exhibited lower fertility
rates than their peers, and the absence of an extended family would have aected migrants no matter their nal destination.
But migrants to textile locations exhibited even lower fertility rates and may also have been dierentially aected by the loss
of extended family. Their increased likelihood of being engaged in occupations which were incompatible with child-rearing (i.e.,
non-agricultural occupations outside of their own home) would have increased their reliance on child care relative to migrants
to other locations. This hypothesis is testable. Using information on the location of parents of the household-level data sample
in 1880 and 1900, I can measure the relationship between extended family location and fertility for migrants. I reserve this as
an avenue for future research.
44 Adding a quadratic term for population density does not change this result.
45 See Becker et al (2009) for evidence that the textile industry in Prussia exhibited
low returns to education relative to metal,
rubber, and other industries.
46 The
result is even more striking if you consider the females were likely to underreport their employment status in textile
mills. In that case, male occupation may also be proxying for female occupation and this may explain part of the signicant,
negative coecient on male textile employment.
19
here is that as families moved from farm to factory, they faced a tradeo in the form of higher incomes for
adult members of the family, but a reduced ability of younger children to participate in production. But
available sources indicate that children represented a large proportion of workers in Southern mills, even
when mills self-reported their employment numbers.
Statistics provided by the mills themselves indicate
that up to 25% of their operatives were under the age of sixteen in the earliest years of the 20th century.
47 Child labor legislation was not passed in South Carolina until 1903,
18% were below the age of fourteen.
well after the period examined in this paper.
If changes in the labor opportunities of children drove the
industrialization results, it must be that mill employment, extensive as it was, was less well renumerated
than their previous employment, generally as farm laborers.
This is a questionable assumption, in part
because children, with their smaller and more dextrous hands, were actually better suited for some mill
tasks than were adults.
I conclude that increases in child mortality and in the opportunity cost of female time and other costs of
raising children, including those resulting from increased population density, are the most likely explanations
for lower fertility in industrial locations.
5
Conclusion
The evidence presented in this paper gives support to the hypothesis that the process of economic growth
preceded declines in marital fertility in the United States. I exploit the fact that South Carolina's industrial
experience in general can be proxied by the textile industry in particular and that the location of textile
mills was independent of local fertility behavior.
I evaluate the impact of the arrival of a textile mill
on rural, marital fertility rates in South Carolina between 1880 and 1900 and document a strong, negative
relationship. The estimated impact is reduced somewhat using a dierence-in-dierence estimator to account
for dierences in marital fertility rates prior to the arrival of mills, but the impact remains signicant.
In order to evaluate potential mechanisms to explain the result, I build a household-level data set.
I
use the location of male heads of household at two points in time to measure migration and show that the
in-migration of low-fertility households to industrial townships is a large driver of the observed dierences
in marital fertility. Households who remained in textile mill locations before and after the arrival of the mill
47 Thompson
(1906), p.223. This latter statistic is for North Carolina.
20
experienced no reduction in household fertility.
I use the migration result and the particulars of the South Carolina industrialization process to distinguish
between many potential explanations for the relationship between economic growth and fertility. I argue that
increases in the returns to human capital and decreases in the labor opportunities of children are unlikely
drivers of the observed fertility reduction. Instead, an increase in the costs of raising children, including a
heightened opportunity cost of female time, adverse child mortality outcomes among migrating industrial
workers, and increases in urbanization appear the most likely explanations.
21
6
Tables
Sample Mean
Standard Deviation
1.22
0.18
1.42
0.14
industrialized by 1895)
0.06
0.25
industrialized between 1896 and 1900)
0.05
0.22
Town population as % of total population in 1900
0.069
0.12
Town population as % of total population in 1880
0.032
0.22
0.06
0.24
0.05
0.22
0.60
0.49
F 51900
F 51880
I1895 =1(Township
I1900 =1(Township
CT Y SEAT1900
CT Y SEAT1880
RR1890
Table 1:
Township Data Variable Summary - All Townships with Town Population
< 2, 500
Corresponds to Table 3
Sample Mean
Standard Deviation
1.22
1.00
0.11
0.31
0.07
0.26
0.10
0.29
0.63
0.48
0.49
0.50
MALEAGE
30.49
5.7
FEMALEAGE
26.94
6.1
0.57
0.49
0.84
0.25
0.11
0.30
F 51900 = Number of children < 5
I1895 =1(Township industrialized by 1895)
I1900 =1(Township industrialized between 1896
CT Y SEAT1900
RR1890
RACE (1 = BLACK )
M OV ERa
Sij = Survival Rate of children ever born
Dij = % of marriage duration for which textile
Table 2:
and 1900)
mill is present
1900 Variable Summary for Household Data
Corresponds to Tables 4, 6, and 7
a) Conditional on remaining in South Carolina.
22
23
Estimation Results, Equations (1) and (2)
341
2.58
Y
Y
Y
6.0%
(0.046)
-0.056
(0.042)
-0.073*
ALL
∆F5
(5)
341
2.40
Y
10.0%
(0.035)
-0.059*
(0.033)
-0.071**
ALL
a) In Column (7),
I1895
becomes
I1897
and
I1900
represents only those mills built between 1898 and 1900.
341
2.40
Y
7.3%
( 0.048)
-0.079
(0.044)
-0.097**
ALL
F3,1900
M ARR
F5,1900
a
(7)
(6)
Di-in-Di
*** indicates signicance at 1% level; ** indicates signicance at 5% level; * indicates signicance at 10% level
Point estimates of coecients. Standard errors in parentheses.
Table 3:
341
341
341
0.04
Y
2.25
341
0.13
Y
Y
Y
6.9%
(0.045)
-0.074
(0.041)
-0.084**
ALL
∆F5
(4)
Number of Observations
0.06
Y
Y
Y
Y
2.6%
(0.035)
-0.012
(0.033)
-0.037
ALL
F5,1880
(3)
1880 Fertility:
Falsication Test
F-statistic
Adjusted R-squared
∆CT Y SEAT
∆P CT N ON W HIT E
∆T OW N P OP OF T OT ALP OP
RR1890
CT Y SEAT1900
T OW N P OP OF T OT ALP OP1900
P CT N ON W HIT E1900
Other Control Variables
10.4%
(0.044)
-0.047
(0.040)
-0.127***
ALL
Y
10.0%
δ1
Percent reduction in fertility
implied by
(0.043)
I1900
(Townships industrialized after 1895)
the coecient on
-0.091**
-0.123***
δ2,
I1895
(0.040)
the coecient on
ALL
F5,1900
F5,1900
(Townships industrialized by 1895)
δ1,
Townships
Dependent Variable
(2)
(1)
1900 Fertility
OLS Baseline
Township Sample Results
24
-0.050
I1895
δ1,
the coecient on
M OV ERi
Y
12,999
0.03
7,452
Estimation Results, Equation (3)
12,999
Y
Y
Y
Y
-6.9%
(0.048)
-0.033
(0.042)
5,547
0.03
Y
Y
Y
Y
+1.1%
(0.063)
0.051
(0.054)
0.014
M OV ER = 0
M OV ER = 1
-0.094**
F 5HH
(5)
F 5HH
(4)
Point estimates of coecients. Standard errors in parentheses.
Table 4:
12,999
0.03
Y
Y
0.03
Y
Y
0.00
Y
Y
Y
-3.7%
Y
-4.1%
(0.019)
-0.061***
(0.038)
0.001
(0.033)
-0.050
ALL
Y
-3.7%
(0.038)
-0.004
(0.033)
-0.055*
ALL
F 5HH
(3)
7,452
0.03
Y
Y
Y
Y
Y
Y
-7.7%
(0.048)
-0.036
(0.042)
-0.097**
M OV ER = 1
F 5HH
(6)
F 5HH
(7)
7,452
0.03
Y
Y
Y
Y
Y
Y
-7.3%
(0.048)
-0.035
(0.042)
-0.094**
M OV ER = 1
*** indicates signicance at 1% level; ** indicates signicance at 5% level; * indicates signicance at 10% level
Number of Observations
Pseudo R-squared
CT Y SEAT1900
RR1890
RACE
AGEV ECT OR
HOM ET W P F1880
HOM ET W P F1900
HOM ET W P P OP1880
HOM ET W P P P OP1900
Other Control Variables
Marginal eect of I1895
on E(F 5HH ) implied by δ1
(as a percent of F¯5HH )
α,
(0.038)
-0.003
(Townships industrialized after 1895)
I1900
δ2,
the coecient on
(0.032)
(Townships industrialized by 1895)
the coecient on
ALL
F 5HH
F 5HH
Households
Dependent Variable
(2)
(1)
1900 Household Sample - Ordered Probit Results
25
0.60
0.60
0.65
Proportion of all Employed Movers
Total
0.74
Proportion of all Employed Stayers
Proportion of all Movers
0.73
Proportion of all Stayers
Table 5:
0.08
0.11
0.11
0.04
0.04
0.26
0.29
0.29
0.22
0.22
Other
Farmers and
0.13
0.58
0.01
0.12
0.01
0.62
−−−
−−−
0.14
Farm Laborers
0.01
Unemployed
Household Occupation Data
Textile
Workers
Farmers and
Farm Laborers
Male
0.01
0.07
0.02
0.03
0.01
Workers
Textile
0.07
0.35
0.07
0.35
0.08
Other
Female
0.79
−−−
0.79
−−−
0.77
Unemployed
1900 Household Sample - Occupation Results
Dependent Variable
α,
(2)
F 5HH
F 5HH
I1895 = 1
Households
δ2,
(1)
the coecient on
the coecient on
I1900
M OV ERi
Marginal eect of M OV ERi
on E(F 5HH ) implied by α
(as a percent of F¯5HH )
or
I1900 = 1
I1895 = 1
or
I1900 = 1
0.059
0.052
(0.048)
(0.048)
-0.127***
-0.095**
(0.047)
(0.047)
-9.8%
-7.2%
Occupation Variables
1(Wife=Farmer or Farm Laborer)
0.048
(0.073)
1(Wife=Textile Worker)
-1.639***
(0.316)
1(Wife=Other Worker)
-0.100
(0.092)
1(Husband=Textile Worker)
-0.323***
(0.056)
1(Husband=Other Worker)
-0.047
(0.092)
Other Control Variables
CT Y SEAT1900
RR1890
RACE
AGEV ECT OR
Pseudo R-squared
Number of Observations
Table 6:
Y
Y
Y
Y
Y
Y
Y
Y
0.03
0.04
2,349
2,349
Estimation Results, Equation (3)
Point estimates of coecients. Standard errors in parentheses.
*** indicates signicance at 1% level; ** indicates signicance at 5% level; * indicates signicance at 10%
level
26
Infant and Child Mortality Results
Dependent Variable
Households
τ
- coecient on
α,
Dij
(1)
(2)
(3)
(4)
Si,j
Si,j
Si,j
Si,j
ALL
ALL
ALL
I1895 = 1 OR
I1900 = 1
-0.0085
-0.0079
0.013
-0.011
(0.0078)
(0.0078)
(0.013)
(0.023)
-0.0049
-0.0010
-0.024
(0.005)
(0.005)
(0.021)
the coecient on MOVER
MOVER x
Dij
Percent change in
survival rate implied by τ
1.0%
0.9%
-0.031**
0.0045
(0.016)
(0.027)
1.5%
0.8%
Occupation Variables
1(Wife=Farmer or Farm Laborer)
-0.021
(0.017)
1(Wife=Textile Worker)
-0.140**
(0.058)
1(Wife=Other Worker)
-0.015
(0.021)
1(Husband=Textile Worker)
-0.076***
(0.018)
1(Husband=Other Worker)
-0.037
(0.013)
Other Control Variables
CT Y SEAT1900
RR1890
RACE
AGEV ECT OR
Adjusted R-squared
Number of Observations
Table 7:
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
0.02
0.02
0.02
0.03
11,188
11,188
11,188
2,284
Estimation Results, Equation (4)
Point estimates of coecients. Standard errors in parentheses.
*** indicates signicance at 1% level; ** indicates signicance at 5% level; * indicates signicance at 10%
level
27
7
Appendix A: The Occupation of South Carolinians in 1900
The 1900 employment of the population by industry is given in Table 8, tabulated by gender.
The data
source is the IPUMS 1% sample of the 1900 U.S. Census. Labor force participation rates for the two groups
are, of course, dierent. For South Carolinians between the ages of 15 and 40, 91.4% of males were labor
force participants while 40.5% of females were. Among the
married
population in this age range, the gures
48 The vast majority of both male and female workers were employed in
are 97.8% and 22.7%, respecively.
agriculture (66% of males and 61% of females).
Outside of agriculture, manufacturing and services held
employment for roughly 25% of the remaining male population and 39% of remaining females. Trade and
transport sectors employed the remaining 9% of men and almost no women.
Occupation Category
Percentage of Employed Males Percentage of Employed Females
Farming and Agriculture
66.3%
60.5%
Services
11.7%
29.8%
9.0%
0.6%
12.9%
9.1%
Trade and Tranport
Manufacturing
Percent of Manufacturing in:
Textiles
Dressmaking
Carpentry
Saw and Planing Mills
Table 8:
30.5%
57.0%
0%
32.3%
7.5%
0%
6.8%
0%
1900 Occupation Statistics for South Carolina
Source: Author's Analysis of 1900 IPUMS 1% Sample (Ruggles et al, 2008)
Within manufacturing, the blossoming textile industry in 1900 was the employment leader. More than
30% of males employed in manufacturing were in the textile industry. Carpentry (7.5%) and sawmills (6.8%)
came in a distant second and third.
For females, the proportion of manufacturing jobs in textiles was
even higher at 57%. Most of the remaining women in manufacturing were employed in dressmaking and as
seamstresses (32.3%), an occupation likely centered in the operative's home rather than in an central locale.
It is clear from Table 8 that industrial employment (taking workers from their home to a centralized
production facility) in South Carolina in this time period was limited almost exclusively to textile mills.
All other employment was either agricultural or decentralized (dress-making, carpentry, etc.).
The sole
exception is the small level of employment among men in the region's saw mills. But as only 0.8% of males
48 Source:
Author's analysis of 1900 IPUMS 1% sample. (Ruggles et al (2008))
28
were so-employed, the contribution of saw milling to industrial employment will be ignored.
8
Appendix B: A Propensity Score Matching Estimator
Township Sample: Matching Estimates
1900 Fertility
OLS
Di-in-Di
(1)
(2)
(3)
(4)
F5,1900
F5,1900
∆F5
∆F5
Model
LLM
KM
LLM
KM
Townships
ALL
ALL
ALL
ALL
-0.139***
-0.090**
-0.100**
-0.066*
(0.050)
(0.045)
(0.044)
(0.039)
-0.061**
-0.056*
-0.052
-0.046
(0.029)
(0.031)
(0.034)
(0.028)
11%
7.4%
8.2%
5.4%
Dependent Variable
δ1,
the coecient on
I1895
(Townships industrialized by 1895)
δ2,
the coecient on
I1900
(Townships industrialized after 1895)
Percent reduction in fertility
implied by
δ1
Table 9:
Propensity Score Matching Estimates
Point estimates of coecients. Standard errors in parentheses.
*** indicates signicance at 1% level; ** indicates signicance at 5% level; * indicates signicance at 10%
level
Although the falsiciation test in Table 3, Column (3) indicated no signicant pre-industrialization difference in marital fertility rates, the same cannot be said for marriage rates and, thus, crude fertility rates.
Future mill townships had signicantly lower marriage rates for women in the fertile age range in 1880 than
did non-mill towns. Given this dierence, I generate a propensity score matching estimate of the impact
of industrialization on fertility in order to account for
ex ante
dierences in characteristics of industrial
townships. The estimate is performed in two steps. First, I generate a propensity score based on observables
of a township in 1880: its 1880 population, county seat status, town population, marriage rate among
49
fertile females (age 18 to 42), and the percentage of the population that is non-white.
49 As
a
robustness
test,
I
also
generate
a
propensity
29
score
that
includes
the
I run a probit
population
density
estimator of both pre-1895 industrialization (I1895 ) and post-1895 industrialization (I1900 ) on these variables
and generate, for each township, the probability of housing a textile mill in these two time periods.
Next, I use a local linear matching (LLM) and and a kernel matching (KM) estimator, both with bootstrapped standard errors, to generate a measure of the average treatment eect of industrialization on the
treated townships. See Todd (2008) for details. Results are located in Table 9. Both models generate results
consistent with those detailed in Table 3.
9
Appendix C: The Ordered Probit Estimator
F 5HH :
An ordered probit model for
Assume a latent variable:
HH
F5ij
∗
=
δ1Ij,1895 + δ2Ij,1900 + βYi + θXj + αM OV ERi + i,j
where variable denitions are given in Section 3.2.
Let
ω0 ≺ ω1 ≺ ω2 ≺ ... ≺ ω5
be unknown threshold points and dene
F 5HH = 0 if F 5HH ∗ ≤ ω0
F 5HH = 1 if ω0 ≺ F 5HH ∗ ≤ ω1
F 5HH = 2 if ω1 ≺ F 5HH ∗ ≤ ω2
etc.
F 5HH = 6 if ω5 ≺ F 5HH ∗
Estimation of this model in Section 3.2 is performed using maximum likelihood.
In order to evaluate the marginal eect of
mean of all variables except
I1895
I1895
on
F 5HH ,
I calculate the expected value of
F 5HH
at the
and calculate:
E(F 5HH |I1900 , Yi , Xj , M OV ERi , I1895 = 1) − E(F 5HH |I1900 , Yi , Xj , M OV ERi , I1895 = 0)
as the marginal eect of
of
I1895
on
F 5HH .
The results in Table 4 report this marginal eect as a percentage
HH
F¯5 .
(T OW N P OP OF T OT ALP OP1880 ) as an explanatory variable.
30
References
Atack, Jeremy, Fred Bateman, Michael Haines, and Robert A. Margo.
2010. Did Railroads
Induce of Follow Economic Growth?: Urbanization and Population Growth in the American Midwest
1850-1860.
Social Science History, 34: 171197.
Becker, Sascha O., Erik Hornung, and Ludger Woessmann. 2009. Catch Me If You Can:
and Catch-up in the Industrial Revolution.
Education
Working Paper.
Brown, John C., and Timothy W. Guinnane. 2002. Fertility Transition in a Rural, Catholic Population: Bavaria 1880-1910.
Carlton, David L.
1982.
Population Studies, 56(1): 3549.
Mill and Town in South Carolina: 1880-1920.
Baton Rouge:Louisiana State
University Press.
Carter, Susan B., Roger L. Ransom, and Richard Sutch.
2002. Family Matters: The Life-Cycle
Transition and the Unparallelled Antebellum American Fertility Decline.
Economic Growth, Technology, and Demographic Change,
History Matters: Essays on
, ed. Timothy W. Guinnane, William A.
Sundstrom and Warren Whatley. Stanford:Stanford University Press.
Crafts, N.F.R. 1989. Duration of marriage, fertility, and women's employment opportunities in England
and Wales in 1911.
Population Studies, 43(2): 32535.
David, Paul, and William Sundstrom.
Family Fertility in Antebellum America.
1988. Old-age Security Motives, Labor Markets and Farm
Explorations in Economic History, 25(2): 16497.
Davison Publishing Company. multiple years. Davison's Textile Blue Book.
Ridgewood, N.J.
Doepke, Matthias. 2004. Accounting for Fertility Decline During the Transition to Growth. Journal of
Economic Growth, 9: 347383.
Doepke, Matthias.
Facts?
2005. Child Mortality and Fertility Decline: Does the Barro-Becker Model Fit the
Journal of Population Economics, 18(2): 337366.
31
Galor, Oded, and David N. Weil. 1996. The Gender Gap, Fertility, and Growth. American Economic
Review, 86(3): 37587.
Galor, Oded, and David N. Weil. 2000. Population, Technology, and Growth:
nation to the Demographic Transition and Beyond.
From Malthusian Stag-
American Economic Review, 90(4): 806828.
Guest, Avery M. 1981. Social Structure and U.S. Inter-State Fertility Dierentials in 1900. Demography,
18: 465486.
Guinnane, Timothy W. 2010. The Historical Fertility Transition and Theories of Long-Run Growth:
Guide for Economists.
A
Yale University Working Paper.
Haines, Michael. 2000. The White Population of the United States, 1790-1920. A Population History of
North America,
, ed. Michael Haines and Richard Steckel, 305369. Cambridge:Cambridge University
Press.
Hansen, Gary D., and Edward C. Prescott. 2002. Malthus to Solow. American Economic Review,
92(4): 12051217.
Hazan, Moshe, and Binyamin Berdugo.
2002. Child Labor, Fertility, and Economic Growth.
The
Economic Journal, 112(482): 81028.
Kalelmi-Ozcan, Sebnem. 2003. A Stochastic Model of Mortality, Fertility, and Human Capital Investment.
Journal of Development Economics, 70(1): 10318.
Kögel, Tomas, and Alexia Prskawetz. 2001. Agricultural Productivity Growth and the Escape from
the Malthusian Trap.
Lagerlöf, Nils-Petter.
Journal of Economic Growth, 6: 33757.
2002. Gender Equality and Long-Run Growth.
Journal of Economic Growth,
8(4): 40326.
Leuven, Edwin, and Barbara Sianesi.
2003. PSMATCH2:
Stata module to perform full Maha-
lanobis and propensity score matching, common support graphing, and covariate imbalance testing.
http://ideas.repec.org/c/boc/bocode/s432001.html. Version 1.2.3.
32
Moehling, Carolyn M. 1999. State Child Labor Laws and the Decline of Child Labor. Explorations in
Economic History, 36(1): 72106.
Moehling, Carolyn M.
2002. Broken Homes: The `Missing' Children of the 1910 Census.
Journal of
Interdisciplinary History, 33(2): 205233.
North Atlantic Population Project and Minnesota Population Center.
Count Microdata. NAPP Version 2.0 [computer les].
2008.
NAPP: Complete
Minneapolis, MN:Minnesota Population Center
[distributor].
Preston, Sameul H., Suet Lim, and S. Philip Morgan. 1992. African-American Marriage in 1910:
Beneath the Surface of Census Data.
Demography, 29(1): 115.
Ruggles, Steven, Matthew Sobek, Trent Alexander, Catherine A. Fitch, Ronald Goeken, Patricia Kelly Hall, Miriam King, and Chad Ronnander. 2008. Integrated Public Use Microdata
Series: Version 4.0 (Machine-Readable Database). Minneapolis, MN:Minnesota Population Center.
Sah, Raaj K. 1991. The Eects of Child Mortality Changes on Fertility Choice and Parental Welfare.
Journal of Political Economy, 99(3): 582606.
Schultz, T. Paul. 1985. Changing World Prices, Women's Wages, and the Fertility Transition:
1860-1910.
Sweden,
Journal of Political Economy, 93: 112649.
Steckel, Richard H. 1992. The Fertility Transition in the United States:
Tests of Alternative Hypotheses.
Strategic Factors in Nineteenth Century American Economic History,
, ed. Claudia Goldin and Hugh
Rocko, 351374. Chicago:University of Chicago Press.
Stover, John R. 1955. The Railroads of the South, 1865-1900. Chapel Hill:North Carolina University Press.
Tamura, Robert. 2002. Human Capital and the Switch from Agriculture to Industry. Journal of Economic
Dynamics and Control, 27: 20742.
Tang, Anthony M.
1958.
Economic Development in the Southern Piedmont, 1860-1950: Its Impact on
Agriculture. Chapel Hill, NC:University of North Carolina Press.
33
Thompson, Holland. 1906. From the Cotton Field to the Cotton Mill: A Study of the Industrial Transition
in North Carolina. New York:The MacMillan Company.
Todd, Petra E. 2008. Matching Estimators. The New Palgrave Dictionary of Economics, , ed. Steven N.
Durlauf and Lawrence E. Blume. New York:Palgrave Macmillan.
United States Bureau of Labor. 1910. Report on Condition of Woman and Child Wage-Earners in the
United States. Washington, D.C.:Government Printing Oce.
United States Commissioner of Labor. 1892. Cost of Production: The Textiles and Glass. Washington:Government Printing Oce.
United States of America, Bureau of the Census.
1880.
Tenth Census of the United States, 1880.
Washington, DC.
United States of America, Bureau of the Census. 1900. Twelfth Census of the United States, 1900.
Washington, DC.
Vinovskis, Maris. 1976. Socio-economic Determinants of Interstate Fertility Dierentials in the United
States in 1850 and 1860.
Journal of Interdisciplinary History, 6: 37596.
Wahl, Jenny Bourne. 1992. Trading Quantity for Quality:
in the Nineteenth Century.
Explaining the Decline in American Fertility
Strategic Factors in Nineteenth Century American Economic History, , ed.
Claudia Goldin and Hugh Rocko, 37597. Chicago:University of Chicago Press.
Wright, Gavin. 1979. Cheap Labor and Southern Textiles before 1880. The Journal of Economic History,
39(3): 655680.
Yasuba, Yasukichi. 1962. Birth Rates of the White Population in the United States, 1800-1860: An Economic Study. Baltimore:Johns Hopkins University Press.
34