## Abstract

This paper provides new insights into the relationship between structural change and the fertility transition. We exploit the spread of an agricultural pest in the American South in the 1890s as plausibly exogenous variation in agricultural production to establish a causal link between earnings opportunities in agriculture and fertility. Households staying in agriculture reduced fertility because children are a normal good, while households switching to manufacturing reduced fertility because of the higher opportunity costs of raising children. The lower earnings opportunities in agriculture also decreased the value of child labor, which increased schooling, consistent with a quantity-quality model of fertility.

## I. Introduction

THE fertility transition that countries in North America and Europe experienced during the nineteenth and twentieth centuries is regarded as one of the most important determinants of rapid and sustainable long-run growth (Guinnane, 2011). Falling fertility rates allowed the transition from a Malthusian regime, where income per capita was roughly constant, to a regime with lower population growth and higher living standards. During the same period, these countries experienced a structural transformation: a sustained shift from agriculture to manufacturing. For example, the number of children per white woman in the United States fell from around seven to two between 1800 and 2000, and real GDP per capita increased at the same time from $1,296 to$28,702. Similarly, between 1810 and 1960, the share of the US labor force working on a farm dropped from 80.9% to 8.1%, while the share of manufacturing employment increased from 2.8% to 23.2% (Lebergott, 1966; Haines & Steckel, 2000; Bolt & van Zanden 2014). While unified growth theory suggests that the structural transformation contributed to the onset of the fertility decline (Galor, 2005), empirical evidence of a causal link is lacking so far.

In this paper, we show that the structural transformation was indeed causal for the fertility transition to take place. Our analysis focuses on the fertility transition in the American South that took place during the late nineteenth and early twentieth centuries, a period that was also characterized by a sustained shift from employment in agriculture to manufacturing (see figure 1). The empirical strategy exploits the arrival of an agricultural pest, the boll weevil, which adversely affected the cotton-producing counties of the American South after the early 1890s as a quasi-experiment (Lange, Olmstead, & Rhode, 2009). Since the spread of the boll weevil was determined by geographic conditions—mainly prevailing wind and weather conditions—it provides a plausibly exogenous source of variation in agricultural production. Our estimation strategy uses two sources of county-level variation: the timing of the boll weevil's arrival and its relatively stronger impact on local economies that were more dependent on cotton cultivation. We combine this county-level variation with complete count US Census microdata to estimate the causal link between structural change and fertility.
Figure 1.

Structural Change and the Fertility Transition in the American South, 1880–1930

This figure shows the evolution of the average number of children under age 5 per 20- to 39-year-old married woman, as well as the fraction of 10- to 65-year-olds employed in manufacturing or living or working on a farm, from 1880 to 1930, for the Cotton Belt of the American South based on full-count Census data.

Figure 1.

Structural Change and the Fertility Transition in the American South, 1880–1930

This figure shows the evolution of the average number of children under age 5 per 20- to 39-year-old married woman, as well as the fraction of 10- to 65-year-olds employed in manufacturing or living or working on a farm, from 1880 to 1930, for the Cotton Belt of the American South based on full-count Census data.

We find evidence that the lower earnings opportunities in the agricultural sector decreased fertility in the American South during the 1880–1930 period via two channels: households staying in agriculture (stayers) reduced fertility due to lower income—consistent with children being a normal good (Becker, 1960)—and households that left agriculture (switchers) reduced their fertility because of the higher implicit and direct costs of raising children in the manufacturing sector. The two channels imply that there is an unambiguously negative association between lower earnings opportunities in agriculture and fertility.1

In order to provide support for the first channel, we estimate the effect of a decline in agricultural income on fertility for stayers by using the interaction between the boll weevil incident and counties' (initial) dependence on cotton production as an instrumental variable. Our instrumental variable estimates reveal that lower agricultural income led to lower fertility among agricultural households, independent of race.2 This result is compatible with the view that the opportunity cost of child rearing was relatively low for farmwork in the American South at the beginning of the twentieth century (Jones, 1985) and potentially in agrarian economies more generally. In support of the second channel, we show that lower agricultural earnings opportunities induced some households to switch to manufacturing. This shift toward manufacturing reinforced the fertility decline since manufacturing households had, on average, substantially fewer children than agricultural households did due to higher implicit and direct costs of raising children.3

To disentangle and quantify the importance of each channel, we exploit the impact of an unprecedented increase in cigarette consumption during World War I on local tobacco cultivation in the American South as a second source of exogenous variation in agricultural production. Our instrumental variable estimates reveal that the effects of the structural change for the fertility transition in the American South are substantial: the shift away from agriculture explains about 29% of the overall marital fertility decline over the sample period.

The lower agricultural earnings opportunities also reduced the value of child labor in the American South, which resulted in higher direct costs of children and a decrease in the opportunity cost of schooling.4 Consequently, we find a substantial decline in 10- to 15-year-olds working and an increase in school attendance. We show that the rise in school attendance was driven by the decline in child labor and was not a result of a potential increase in the attractiveness of schooling and the returns to education per se. This finding is consistent with a standard quantity-quality (Q-Q) framework of fertility, which predicts that an increase in the direct costs of having children induces parents to invest more in the education (“quality”) of their offspring. Our empirical findings therefore support the view that the Q-Q framework can rationalize the well-documented rise in school enrollment that went along with structural change and the fertility transition during the previous two centuries.

Our paper relates to the unified growth theory literature, which argues that the process of industrialization contributed to the onset of the fertility decline (Galor & Weil, 1999, 2000; Galor, 2005). While this theoretical literature is well developed, empirical evidence of a causal relationship is scarce due to complicated identification resulting from potential reverse causality and omitted variable bias. Our empirical model uses plausibly exogenous variation in the earnings opportunities in agriculture to address this identification problem. In line with the prediction of unified growth theory, we find evidence of a causal link between the structural transformation and the fertility transition in the American South in the late nineteenth and early twentieth centuries.

The result that stayer households experienced a decrease in income and therefore lowered fertility (the first channel) is in line with recent empirical evidence showing that when income and wealth shocks are properly identified, children are a normal good, as Becker (1960) suggested. For example, Lovenheim and Mumford (2013) exploit regional variation in the US housing market to show that family wealth positively affects fertility. Bleakley and Ferrie (2016) find that winners of the Georgia Cherokee Land Lottery of 1832 had slightly more children than lottery losers did. Lindo (2010) and Black et al. (2013) reach the same conclusion by exploiting exogenous shocks to household income. The positive relationship between household income and fertility within agricultural occupations is also consistent with the finding in some earlier literature based on cross-sectional US data that higher income leads to more children within the same occupation (Freedman, 1963).

Our finding that switcher households decreased their fertility because the implicit and direct costs of child rearing were higher in the manufacturing sector (the second channel) relates to Wanamaker (2012), who finds that industrialization was an important determinant for the fertility decline in South Carolina between 1880 and 1900. Unlike Wanamaker (2012), we find that the reduced fertility decline is not just a result of selective migration and that human capital formation also increased as a result of structural change in the American South.

We therefore also contribute to a literature that argues that human capital formation played an important role in the relation between structural change and the fertility transition (Galor, 2005, 2011). Becker (1960) and Becker and Lewis (1973) developed the idea that parents face a trade-off between the number of children and investment in child quality. This quantity-quality (Q-Q) model is supported by the data, since there is ample evidence of a negative relation between family size and child quality (Hanushek, 1992; Becker, Cinnirella, & Woessmann, 2010; Tan, 2018). More recently, a number of studies have tested the Q-Q framework of fertility by using plausibly exogenous variation in the returns to education. For example, Bleakley and Lange (2009) argue that the sudden eradication of hookworm in the American South during the 1910s led to an effective decrease in the price of child quality, particularly in areas with high pretreatment infection rates. They document fertility behavior in line with the Q-Q model. Aaronson, Lange, and Mazumder (2014) exploit a substantial decrease in the cost of education for black children due to the rollout of the Rosenwald schools in the American South during the early twentieth century. They find that affected mothers reduced fertility along the intensive margin but, in line with Q-Q preferences, were less likely to remain childless. While these studies exploit variation in the returns to education to test the existence of a Q-Q trade-off, our paper provides direct evidence that the Q-Q model can rationalize the increase in school attendance during the structural transformation.

Finally, this study contributes to a copious literature on the fertility transition in the United States and the American South in particular. Economic historians suggest various competing hypotheses to explain the US fertility decline during the nineteenth and early twentieth centuries, ranging from changes in the cost of acquiring land to increases in the default risk of children to provide old-age care for parents to economic modernization (Easterlin, 1976; Sundstrom & David, 1988; Greenwood & Seshadri, 2002).5 The importance of economic modernization for the fertility transition in the United States has been emphasized by several studies, especially for the period after the Civil War (Guest, 1981). Consistent with the economic modernization hypothesis, recent empirical studies find industrialization (Wanamaker, 2012), better access to education (Aaronson et al., 2014), and health improvements (Bleakley & Lange, 2009) to be important determinants of the southern fertility decline. Our findings add to this literature and provide further evidence that structural change led to a fertility decline in counties of the American South that relied heavily on cotton production. The lower earnings potential in the southern agricultural sector contributed to the fertility transition by accelerating the process of industrialization and increasing the demand for human capital.

## II. The Boll Weevil as a Quasi-Experiment

The boll weevil is a vermin that depends on the cotton plant, its main source of food and host of reproduction. It first appeared in the American South near Brownsville, Texas, in 1892. By 1922, almost the entire Cotton Belt region was infested (see figure A.1). Depending on prevailing wind and weather conditions, the boll weevil could cover from 40 to 160 miles per year (Hunter & Coad, 1923). Since the timing of the arrival of the weevil is determined by geography, it is plausibly exogenous to local economic conditions and can therefore be used to identify the causal effect of lower agricultural earnings opportunities on fertility.

The boll weevil's detrimental effect on the southern agricultural sector is well documented. Lange et al. (2009) combine county-level data on agricultural production with the timing of the arrival of the boll weevil for the period 1889 to 1929 and show that it decreased local cotton production by about 50% in the first five years after contact, with no sign of recovery for at least a decade. The reduced revenues from cotton production had important impacts on local economies. Lange et al. (2009) document population movements and a shift of agricultural production from cotton to corn, the main alternative crop in the Cotton Belt. Ager, Brueckner, and Herz (2017) find that in highly cotton-dependent counties, the presence of the vermin led to farm closures, a change in tenancy arrangements, removal of land from agricultural production, and a substantial decline in farm wages and female labor force participation. Other recent work shows that the boll weevil increased school enrollment rates of blacks in Georgia (Baker, 2015) and delayed marriage, especially for young African Americans, as the boll weevil infestation changed the prospects of tenant farming (Bloome, Feigenbaum, & Muller, 2017).

The findings based on disaggregated data resonate with the older economic history and social science literature that considers the boll weevil as a large negative productivity shock to southern cotton production and a disruptive element of the whole southern economy (Street, 1957; Ransom & Sutch, 2001). Between 1909 and 1935, the estimated average reduction from full yield in the American South was 10.9%, ranging from 0.8% in Missouri to 17.8% in Louisiana. In 1921, thirty years after the boll weevil entered the Cotton Belt, the estimated losses reached their peak of 31% (US Department of Agriculture, 1951, table 52). The estimated average annual loss due to the boll weevil infestation for the four years preceding 1920 was $200 million to$300 million (Hunter & Coad, 1923).

The recent evidence based on disaggregated data revises findings of scholars that questioned whether the boll weevil played an important role in the development of the southern economy as a whole (Higgs, 1976; Osband, 1985; Wright, 1986; Giesen, 2011). Proponents of this view argued that higher cotton prices completely offset the detrimental effects that the boll weevil had on local economies. For example, Wright (1986) argues that higher cotton prices kept the southern cotton economy going; farmers refrained from diversifying agricultural production at a larger scale, and so the higher prices did not lead to a shift of resources out of agriculture in the South.6

For our empirical approach, the literature based on aggregated data raises the concern that offsetting price effects might have mitigated the decline in agricultural earnings opportunities due to the boll weevil infestation. In this respect, it is important to note that our estimation strategy exclusively uses within-county variation and includes time fixed effects (see section IVA). This alleviates the concern that fertility might have responded to aggregate price effects. Our econometric model further includes state-by-time fixed effects, which implies that our variation only comes from differentially affected counties within a given state and year. Our estimates therefore take into account any potential confounding effects that occur at the state level, even when they vary over time. For example, changes in state-specific laws, such as regulating child labor and school attendance, which potentially directly affected fertility outcomes, are captured by our econometric model.

For our empirical strategy, it is also not relevant to what extent the boll weevil led to an overall decline in agricultural earnings opportunities in the Cotton Belt, only that it induced a relative decline in more cotton-dependent counties compared to less cotton-dependent ones.7 Finally, it is also sufficient that the infestation created some exogenous variation in agricultural earnings opportunities. We do not argue that the boll weevil infestation necessarily was the main source of structural change in the American South.

## III. Data

We use the recently released complete count US Census microdata from the Integrated Public Use Microdata Series (IPUMS) to construct the relevant outcome measures for fertility, occupational choices, and school attendance (Ruggles et al., 2017). The data consist of a repeated cross-section of individuals who resided in the Cotton Belt of the American South during the period 1880 to 1930.8 We use the following data sets for the empirical analysis: (a) to study fertility, we use a sample of about 13.5 million 16- to 49-year-old married women with spouse present;9 (b) to study structural change and occupational choices, we draw on a sample of about 61 million individuals of working age (10 to 65); and (c) to analyze school attendance, we use a sample of about 7.5 million 10- to 15-year-old children who are listed together with their mothers in the Census. To overcome some of the drawbacks of a purely cross-sectional analysis, we use data provided by IPUMS that link records from the 1880 complete-count database to the 1% samples of the 1900, 1910, and 1920 Censuses at various points in the empirical analysis.

Our study uses a novel measure of household income that combines various sources of agricultural income covering the decades 1880 to 1930. Farm income is based on county-level measures of farm revenues and expenditures from the US Censuses of Agriculture (Haines, Fishback, & Rhode, 2015). Wages for farm laborers are retrieved from various official sources and vary by state over time. Unpaid family workers are assumed to receive a constant fraction of the county-specific farm income. We then assign agricultural income to individuals who report an agricultural occupation in a given year. This varies across agricultural occupations—farmers, farm laborers (wage workers), and unpaid family workers—by county or state and over time, and is denoted in constant prices of the year 1900. For nonagricultural income of these households we use the occupation-based income score (OCCSCORE) from IPUMS in constant prices.10 The online data appendix provides a detailed description of how the agriculture income variable is constructed.

We then merge the microdata with county-level data on the 1899 arrival of the boll weevil and cotton production.11 County-level data on cotton acreage are from the Census of Agriculture in 1889 (Haines et al., 2015). As many counties changed boundaries during our sample period, we aggregate counties to time-consistent “multi-counties” as in Lange et al. (2009) and Ager et al. (2017). Descriptive statistics are reported in the online appendix table A.1.

## IV. Reduced-Form Evidence

In this section, we quantify the reduced-form effects that the boll weevil infestation of the southern cotton fields had on fertility. Our econometric model follows a differences-in-differences strategy, exploiting the fact that the boll weevil arrived in different counties at different times (variation over time) and that the boll weevil had a stronger impact in highly cotton-dependent counties (variation across countries); see Ager et al. (2017). Under the hypothesis that the boll weevil had a negative effect on fertility, we would expect to find the largest fertility declines in counties with a high initial intensity of cotton production after infestation.

### A. Estimation Strategy

We use our sample of married women to estimate the following reduced-form equation,
$Fertilityict=αc+αst+βBollWeevilIntensityct+ΓXict+εict,$
(1)

where $Fertilityict$ denotes mother $i$'s number of own children under age 5. Equation (1) further controls for county fixed effects, $αc$; state-by-time fixed effects, $αst$; and a set of individual control variables, $Xict$, which includes age fixed effects, indicator variables for race, and whether the mother lives in a rural area. To account for potential time-varying effects of the latter variables, we also include race-by-rural-by-time fixed effects and all potential interactions among these three variables. The main variable of interest, $BollWeevilIntensityct$, is the interaction between a dummy variable that equals 1 if county $c$ was infested by the boll weevil at time $t$ and county $c$'s acreage share of cotton planted in 1889.12 We use data from the preinfestation year 1889 to ensure that the interaction term is exogenous to fertility changes during the boll weevil infestation period. Standard errors are Huber robust and clustered at the county level.

Since fertility is highly age dependent, we also use an extended specification that allows the effect of the boll weevil on fertility to vary by age:
$Fertilityict=αc+αst+∑g=1GβgAgejg×BollWeevilIntensityct+ΓXict+εict.$
(2)

Our variable of interest, $BollWeevilIntensityct$, is now interacted with a set of dummy variables that capture mother $i$'s age cohort $g$, in Census year $t$. We differentiate between women aged 16 to 19, 20 to 24, 25 to 29, 30 to 34, 35 to 39, 40 to 44, and 45 to 49. To capture cohort-specific differences in fertility that are independent of the boll weevil infestation, this specification also includes cohort fixed effects (interacted by county and by time). Under the hypothesis that the boll weevil has a negative effect on fertility, we would expect $β<0$ in equation (1). In equation (2), we would expect $βg<0$ with a larger coefficient in absolute size for mothers in their prime childbearing years.

### B. Results

Column 1 of table 1 reports results using estimation equation (2) for our sample of married women in the Cotton Belt of the American South. The estimates reveal that 20- to 39-year-old women were the most affected. The effect for women over age 40 is not statistically significant and practically 0. This finding serves as a consistency check since we would not expect systematic fertility adjustments of older women in reaction to the boll weevil's arrival.

Table 1.
The Impact of the Boll Weevil Infestation on Fertility
 (1) (2) (3) (4) (5) Variables Number of Children under Age 5 $=$1 if Birth Age 16–19 $×$$BollWeevilIntensityct$ −0.014 (0.011) Age 20–24 $×$$BollWeevilIntensityct$ −0.060*** (0.011) Age 25–29 $×$$BollWeevilIntensityct$ −0.051*** (0.013) Age 30–34 $×$$BollWeevilIntensityct$ −0.027** (0.013) Age 35–39 $×$$BollWeevilIntensityct$ −0.023** (0.011) Age 40–44 $×$$BollWeevilIntensityct$ 0.006 (0.011) Age 45–49 $×$$BollWeevilIntensityct$ −0.005 (0.009) $BollWeevilIntensityct$ −0.041*** −0.038** −0.046*** −0.012*** (0.011) (0.016) (0.012) (0.001) $BollWeevilIntensityct$$×$ Black −0.007 (0.023) $BollWeevilIntensityct$$×$ Above Median HH Income 0.004 (0.012) Observations 13,509,865 9,730,437 9,730,437 8,760,018 62,923,755 $R2$ 0.160 0.093 0.093 0.090 0.098
 (1) (2) (3) (4) (5) Variables Number of Children under Age 5 $=$1 if Birth Age 16–19 $×$$BollWeevilIntensityct$ −0.014 (0.011) Age 20–24 $×$$BollWeevilIntensityct$ −0.060*** (0.011) Age 25–29 $×$$BollWeevilIntensityct$ −0.051*** (0.013) Age 30–34 $×$$BollWeevilIntensityct$ −0.027** (0.013) Age 35–39 $×$$BollWeevilIntensityct$ −0.023** (0.011) Age 40–44 $×$$BollWeevilIntensityct$ 0.006 (0.011) Age 45–49 $×$$BollWeevilIntensityct$ −0.005 (0.009) $BollWeevilIntensityct$ −0.041*** −0.038** −0.046*** −0.012*** (0.011) (0.016) (0.012) (0.001) $BollWeevilIntensityct$$×$ Black −0.007 (0.023) $BollWeevilIntensityct$$×$ Above Median HH Income 0.004 (0.012) Observations 13,509,865 9,730,437 9,730,437 8,760,018 62,923,755 $R2$ 0.160 0.093 0.093 0.090 0.098

This table shows the boll weevil's impact on fertility. The dependent variable is the number of own children in the household under age 5 in columns 1 to 4 and an indicator variable that is 1 if a mother gave birth in a given year $t$ in column 5. $BollWeevilIntensityct$ is the interaction between a dummy variable that equals 1 if county $c$ was infested at time $t$ and county $c$'s acreage share of cotton planted in 1889. Columns 1 to 4 include county fixed effects, time fixed effects, and state $×$ time fixed effects, and the following set of individual controls: dummies for race, rural, age fixed effects, and interactions between race, rural, and time. We interact $BollWeevilIntensityct$ with a race dummy in column 3 and a dummy indicating whether the household income is above the median in column 4. Both specifications include the mean effects for race and above-median household income, respectively (not reported). Column 5 includes fixed effects for each mother (and hence county), birth year, and state $×$ time, and controls for the mother's age at birth. Robust standard errors clustered at the county level in parentheses: ***p$<$ 0.01, **p$<$ 0.05, and *p$<$ 0.1.

Columns 2 to 4 report results using estimating equation (1), but restricting the sample to 20- to 39-year-old women. In line with our hypothesis, the coefficient on $BollWeevilIntensityct$ is negative and highly statistically significant. Quantitatively, the estimate implies that in a county with median cotton dependency, the arrival of the boll weevil led to a reduction of the number of children less than 5 years old by 0.017 (the median cotton dependency in the sample is 0.424, such that $-0.041×0.424=-0.017$). This accounts for about 2% of the total fertility decline of married 20- to 39-year-olds in the Cotton Belt between 1880 and 1930.13 Our results remain qualitatively unchanged when using a dummy whether the mother has any child under age 5 or the number of own children under age 10 as alternative measures of fertility (see columns 1 and 2 of table A.3).14 We also obtain similar results when using alternative empirical specifications such as including a quadratic of $BollWeevilIntensityct$ or using the years of duration of the infestation instead of a binary variable (available on request). The estimates reported in columns 3 and 4 reveal no significant differences for white and black women and between households below and above the median household income.15 This shows that the effects of the boll weevil are independent of race and not driven by credit-constrained households.

One drawback of using the decennial US Census data is that we observe women's fertility at a rather low frequency. An alternative way of measuring the impact of the boll weevil on fertility is to construct a flow fertility measure. Since the Census reports the age of each child in a household, it is straightforward to calculate the respective birth year.16 We use this information to construct each mother's fertility history. That is, we construct for every mother a time-varying indicator variable, which is 1 if a child was born in a given year and 0 otherwise. The sample is based on complete count Census microdata for the years 1900, 1910, and 1920 and restricted to observations where the mother's age when giving birth is between 15 and 44. Since we know the year when the boll weevil arrived in a county, we can use this data set to explore the boll weevil's effect on the probability of a woman giving birth in a given year. The estimates using this alternative approach are reported in column 5 of table 1.17 Identification comes from within-mother variation in the probability of giving birth in a given year due to differences in the timing of the boll weevil's arrival in counties with different cotton intensities. In line with our baseline results, we find a lower probability of giving birth in counties with a high initial cotton intensity after the arrival of the boll weevil. The estimated coefficient is statistically significant at the 1% level.

### C. Potential Threats to Identification

One potential threat to identification is that fertility trends in more and less cotton-dependent counties evolved differently before the boll weevil infestation. The existence of such “pretrends” would undermine our difference-in-differences strategy because it would invalidate the use of low-cotton-dependent counties as a control group. To address this concern, we conduct an event study using the mother panel sample described above. The structure of the panel allows us to calculate the average number of births by 15 to 44-year-old women in a given county and year, $Fertilityct$. Our estimating equation is
$Fertilityct=αc+αt+∑j∈TBollWeevilctτ+j×(βjmedCottonc,1889med+βjhighCottonc,1889high)+εct,$
(3)

where $T={-10,…,-2,0,…,10}$. We omit $j=-1$ (the base year) such that the posttreatment effects are relative to the year before the arrival of the boll weevil in a given county $c$. The parameter $τ$ refers to the the year in which the boll weevil entered county $c$. $BollWeevilctτ+j$ is an indicator equal to 1 when $t=τ+j$ and 0 otherwise. Also, to capture the fertility response ten and more years prior to (after) the boll weevil infestation, we define an indicator $BollWeevilctτ-10=1$ if $t≤τ-10$ ($BollWeevilctτ+10=1$ if $t≥τ+10$) and 0 otherwise. The specification also includes fixed effects for county, birth year, and interaction of birth year and state.

We differentiate between low, medium, and highly cotton-dependent counties instead of using a continuous measure of cotton intensity to facilitate the interpretation of the event study. The indicator variables $Cottonc,1889med$ and $Cottonc,1889high$ equal 1 if the cotton share in county $c$ in 1889 is “medium” (second to third quartile) or “high” (fourth quartile), respectively, while the first quartile is the omitted category. The estimated coefficients $βjmed$ and $βjhigh$ trace out the effect of the boll weevil infestation on fertility, relative to the omitted category and base year (the year before the arrival of the boll weevil). These coefficients are visualized in figure 2, and the corresponding estimates are reported in table A.4. We find that for all $j<0$$βjmed^≈0$ and $βjhigh^≈0$, which clearly supports the identifying assumption of common pretrends, while after impact, the estimated coefficients become negative and statistically significant. The effect is also relatively stronger in high compared to medium cotton-dependent counties, corroborating our baseline estimation strategy. From figure 2, it is also apparent that the fertility decline due to the boll weevil infestation was persistent, which is in line with the finding of Lange et al. (2009) that the local effects of the boll weevil infestation were long lasting.
Figure 2.

Event Study

This figure shows the dynamic effects of the boll weevil infestation on fertility. The $X$-axis measures the number of years since the boll weevil arrived in a county $c$. The solid line depicts the effect on fertility relative to the base year (the year before infestation). The left (right) panel shows the effect for medium (highly) cotton-dependent counties. Low cotton-dependent counties are the reference group. Dashed lines indicate 90% confidence intervals.

Figure 2.

Event Study

This figure shows the dynamic effects of the boll weevil infestation on fertility. The $X$-axis measures the number of years since the boll weevil arrived in a county $c$. The solid line depicts the effect on fertility relative to the base year (the year before infestation). The left (right) panel shows the effect for medium (highly) cotton-dependent counties. Low cotton-dependent counties are the reference group. Dashed lines indicate 90% confidence intervals.

To further validate our identification strategy, we conduct two additional placebo exercises. In the first exercise, we report placebo regressions that test for effects of the boll weevil prior to actual infestation. To do so, we backdate the boll weevil infestation by twenty years. For example, in a county where the weevil entered in year 1910, we now assume it would have entered in year 1890. Estimates of regression equation (1) using this placebo specification are reported in column 4 of table A.3. Reassuringly, the interaction between the backdated boll weevil incidence and the 1889 cotton share is small and not statistically different from 0. This finding is also in line with our event study results showing that there are no pretrends before infestation. In the second placebo exercise, we add the interaction between the boll weevil and the corn share planted in 1889 to estimating equation (1). Columns 5 and 6 of table A.3 show that our main results are unchanged, while the interaction effect between the boll weevil and the corn share is small and always statistically insignificant.

Lange et al. (2009) document that farmers, as a reaction to the boll weevil, shifted agricultural production from cotton to corn, the main alternative crop in the Cotton Belt. Crop-shifting might therefore have mitigated the weevil's negative effect on fertility. To analyze whether this was actually the case, we include an interaction of Boll$WeevilIntensityct$ with a measure of a county's suitability for corn cultivation in our estimating equation.18 Since crop-shifting should be especially attractive in counties where corn could easily be planted, we would expect this interaction to be positive if there was such a mitigating effect. In columns 7 and 8 of table A.3. we show that this is not the case.

One potential concern is that our results might be driven by composition bias. The arrival of the boll weevil might have triggered selective migration of households. Households that migrated as a response to the boll weevil's arrival might on average have been wealthier and have more children. To address this issue, we look at samples of households from the 1900, 1910, and 1920 Censuses, which have been linked to the 1880 Census by IPUMS (Ruggles et al., 2017). These linked samples allow us to evaluate the effect of migration on fertility. We only consider linked households where a wife of age 20 to 39 is present in the terminal period. Reassuringly, columns 1 and 2 of table A.5 show that households that migrated out of a county did not have higher fertility; in fact, they had lower fertility. As an alternative test, in columns 3 and 4 of table A.5 we replicate the specifications of columns 2 and 5 of table 1, while restricting the sample to mothers who report to reside in their state of birth. Since the estimates are similar to the baseline estimates in table 1, we can rule out that our findings are driven by interstate migration. In conclusion, the presented evidence on migration corroborates our baseline results and makes it unlikely that composition bias is of great concern.

The boll weevil might also have increased child mortality due to poorer nutrition or even starvation, although recent empirical evidence from Clay, Schmick, and Troesken (2019) suggests that this was not the case. To address this potential concern, we explore the effect of the boll weevil infestation on child mortality and stillbirths.19 Columns 1 to 3 of table A.6 show that there was no positive effect. In this context, one further potential concern is whether the arrival of the boll weevil impaired fecundity, for example, due to greater maternal stress. Since the Censuses in 1900 and 1910 list the number of children ever born, we can construct a dummy for being childless for women aged 20 to 39 who report to be married for at least two years to proxy for impaired fecundity.20 The insignificant estimate in column 4 suggests that this was not the case. Overall, the results of table A.6 support the view that the decision of households to have fewer offspring was not a result of increased child mortality or impaired fecundity.

Although we consider only married mothers in our analysis, it could be that in infested counties, mothers have fewer children because they postpone marriage (Bloome et al., 2017). To address this concern, we include age at marriage fixed effects as additional controls in estimating equation (1).21 Reassuringly, our results indicate that the fertility behavior of married women in our sample is not driven by delayed marriage in boll weevil-infested counties (see column 9 of table A.3).

Finally, our results might also be driven by differential fertility dynamics in counties where plantation farming was considered important. Large-scale plantations favored family formation and provided strong incentives for child bearing since farm allotments were determined by family size (Elman et al., 2015). In column 10 of table A.3, we show that mothers' fertility behavior in plantation counties did not respond differently compared to the rest of the sample after the boll weevil's arrival. Since these counties were also characterized by relatively high (land) inequality, this finding can be regarded as suggestive evidence that land inequality is not a main driver of the impact that the boll weevil infestation had on fertility.

### D. Case Studies

This section provides evidence from two case studies that the boll weevil's negative effect on fertility is robust to using alternative sets of control groups. In particular, we consider control counties that were either specialized in producing other main cash crops within the Cotton Belt or are located on the frontier of the boll weevil infestation in the 1920s.

The first case study focuses on Louisiana. While Louisiana was part of the Cotton Belt and engaged in cotton cultivation, some parishes, well known for specializing in sugar cultivation, formed the “Sugar Bowl” (see Rodrigue, 2001). These parishes serve as an ideal control group to study the impact of the weevil: they were highly agricultural, but cotton production played either a very minor or even no role, and the weevil infested all parishes in Louisiana at about the same time (the first parish was infested in 1903 and the last in 1909), which makes it less likely that our estimates are confounded by time-specific effects. Figure 3a shows an event study based on equation (3) that compares the effect of the boll weevil on fertility in highly cotton-dependent parishes with Sugar Bowl parishes (the estimates are reported in table A.7). Reassuringly, the results are in line with our previous findings: At the time of impact, we see a significant and persistent reduction in fertility and no pretrends before infestation.
Figure 3.

Case Studies

Panel a shows the dynamic effects of the boll weevil infestation on fertility in Louisiana. The $X$-axis measures the number of years since the boll weevil arrived in a parish $c$. The solid line depicts the effect on fertility relative to the base year (the year before infestation). The panel shows the effect for highly cotton-dependent Louisiana parishes. Parishes of the Sugar Bowl are the reference group. Dashed lines indicate 90% confidence intervals. Panel b shows the frontier of the boll weevil infestation in 1922, the year the vermin reached its maximal spread. The case study compares fertility in counties on the frontier that have not been infested (light gray) with adjacent counties that have been infested (dark gray). Counties with a high cotton dependency are marked with an “x.” We exclude Florida's boll weevil frontier from the analysis, as some border counties were established only a few years before the 1920 Census, such as Seminole County, or even after the 1920 Census, such as Hardee County, making proper identification impossible.

Figure 3.

Case Studies

Panel a shows the dynamic effects of the boll weevil infestation on fertility in Louisiana. The $X$-axis measures the number of years since the boll weevil arrived in a parish $c$. The solid line depicts the effect on fertility relative to the base year (the year before infestation). The panel shows the effect for highly cotton-dependent Louisiana parishes. Parishes of the Sugar Bowl are the reference group. Dashed lines indicate 90% confidence intervals. Panel b shows the frontier of the boll weevil infestation in 1922, the year the vermin reached its maximal spread. The case study compares fertility in counties on the frontier that have not been infested (light gray) with adjacent counties that have been infested (dark gray). Counties with a high cotton dependency are marked with an “x.” We exclude Florida's boll weevil frontier from the analysis, as some border counties were established only a few years before the 1920 Census, such as Seminole County, or even after the 1920 Census, such as Hardee County, making proper identification impossible.

In our second case study, we analyze counties on the frontier of the area infested by the boll weevil in the year 1922, when virtually the entire Cotton Belt was infested and the spread of the vermin reached its maximal extent (see figure A.1). While in our baseline analysis identification comes from varying degrees of counties' cotton dependency, in this case study we compare counties that were infested with counties that were never infested by the weevil (see figure 3b). Counties in our control group were not infested for two reasons. One group had no or only very minor cotton cultivation, while counties in the second group cultivated cotton, but adverse weather conditions such as frost and dry climate prevented infestation. Important drawbacks of this case study, besides the smaller sample size, are that the infestation of the treated sample counties occurred relatively late (circa 1920), and some counties are sparsely populated while others did not cultivate any cotton. Table 2 reports the results of this case study based on regression equation (1). We find that infested counties experienced a significant decline in fertility relative to noninfested counties, albeit the estimate is somewhat larger compared to our baseline results. In column 2, we show that distinguishing between high and low cotton-cultivating counties in the control group does not affect our estimates.

Table 2.
Case Study: Counties on the Frontier of the Boll Weevil Infestation
 (1) (2) Variables Number of Children under Age 5 $BollWeevilIntensityct$ −0.114*** −0.112** (0.041) (0.051) $BollWeevilIntensityct$$×$ Low Cotton −0.004 (0.050) Observations 1,142,806 1,142,806 $R2$ 0.089 0.089
 (1) (2) Variables Number of Children under Age 5 $BollWeevilIntensityct$ −0.114*** −0.112** (0.041) (0.051) $BollWeevilIntensityct$$×$ Low Cotton −0.004 (0.050) Observations 1,142,806 1,142,806 $R2$ 0.089 0.089

This table shows the boll weevil's impact on fertility for the subsample of counties on the frontier of the boll weevil infestation in 1922. We compare counties on the frontier that were infested with neighboring counties that were not infested by 1922; see figure 3b. The dependent variable is the number of own children in the household under age 5. The sample consists of married women age 20 to 39 for the decades 1910 to 1930. $BollWeevilIntensityct$ is the interaction between a dummy variable that equals 1 if county $c$ was infested at time $t$ and county $c$'s acreage share of cotton planted in 1889. Regressions include county fixed effects, time fixed effects, and state $×$ time fixed effects, and the following set of individual controls: dummies for race, rural, age fixed effects, and interactions between race, rural, and time. Robust standard errors clustered at the county level in parentheses: ***p$<$ 0.01, **p$<$ 0.05, and *p$<$ 0.1.

## V. Structural Change

Recent research has documented that the boll weevil had a persistent detrimental effect on cotton production (Lange et al., 2009; Ager et al., 2017). In this section, we show that the infestation led to substantial income losses for agricultural households in cotton dependent counties (section VA). We also find that a significant number of households reacted to the reduced earnings prospects by leaving the agricultural sector for manufacturing jobs (section VB). We conclude that the boll weevil constitutes a useful source of plausibly exogenous variation that can be used to identify the economic consequences of structural change in the Cotton Belt.

### A. The Boll Weevil's Effect on Agricultural Income

This section focuses on agricultural households based on the sample of married women described in section III.22 We reestimate equation (1) based on a sample of about 5.8 million households using agricultural household income as the dependent variable. Agricultural income is calculated as the sum of the wife's and husband's income, which varies over time, across agricultural occupations, and across counties for farmers or states for farm laborers (see section III and the online data appendix for further details):
$Incomeict=αc+αst+βBollWeevilIntensityct+ΓXict+εict.$
(4)

Column 1 of table 3 presents estimates for households with wives aged 20 to 39. We find a negative effect of the boll weevil on household income in more cotton-dependent counties, which is statistically significant at the 1% level. The estimates imply that households residing in a county with a median intensity of cotton production experienced a decline of agricultural income by about 8% upon arrival of the boll weevil. Part of this effect can be interpreted as households moving down the agricultural ladder, consistent with the findings of Ager et al. (2017). However, this result also reveals that agricultural households experienced a substantial loss in earnings within occupations. This is evident from estimating equation (4) using the IPUMS OCCSCORE variable as an alternative dependent variable. The estimated $β$ is −0.03 with standard error 0.01, which is substantially smaller than the estimate presented in column 1. The likely reason for obtaining a smaller coefficient is that compared to our agricultural income measure, the OCCSCORE variable varies only across, not within, occupations. In line with the recent literature discussed in section II, our results reveal that agricultural households in the more cotton-dependent counties suffered substantial and persistent income losses. We further test whether crop shifting mitigated the income losses for agricultural households by adding the interaction of the $BollWeevilIntensityct$ with corn suitability (see section IVC). While the coefficient on the interaction term is positive, it is small and statistically insignificant (available on request). This also implies that potential shifts to alternative crops in response to the boll weevil infestation as documented by Lange et al. (2009) and Ager et al. (2017) did not fully compensate for the income losses due to impaired cotton production.

Table 3.
The Boll Weevil's Effect on Agricultural Income and Industrialization
 (1) (2) (3) (4) (5) Variables Agricultural Income Works in Manufacturing Works/Lives on Farm % in Manufacturing Leaves Farm 1880–1920 $BollWeevilIntensityct$ −0.190*** 0.009*** −0.041*** 0.146*** 0.057* (0.036) (0.004) (0.015) (0.056) (0.033) Observations 5,831,000 61,089,255 61,089,255 3,572 6,140 $R2$ 0.450 0.088 0.263 0.795 0.029
 (1) (2) (3) (4) (5) Variables Agricultural Income Works in Manufacturing Works/Lives on Farm % in Manufacturing Leaves Farm 1880–1920 $BollWeevilIntensityct$ −0.190*** 0.009*** −0.041*** 0.146*** 0.057* (0.036) (0.004) (0.015) (0.056) (0.033) Observations 5,831,000 61,089,255 61,089,255 3,572 6,140 $R2$ 0.450 0.088 0.263 0.795 0.029

This table shows the impact of the boll weevil on agricultural income and industrialization. The dependent variables are the income of agricultural households; a dummy variable that indicates whether a person works in manufacturing; works or lives on a farm; the fraction of the county population working in manufacturing (in logarithmic units); and an indicator variable that is 1 if an individual left agriculture. The sample consists of married women of age 20 to 39 in agricultural households (column 1); individuals of working age (columns 2 and 3); and aggregated county-level data in column 4 for the decades 1880 to 1930. The linked sample of male household heads is used in column 5. $BollWeevilIntensityct$ is the interaction between a dummy variable that equals 1 if county $c$ was infested at time $t$ and county $c$'s acreage share of cotton planted in 1889. In column 1, the set of individual controls includes dummies for race, rural, age fixed effects, and interactions between race, rural, and time. In columns 2 and 3, the set of individual controls includes dummies for gender, race, and age fixed effects, and interactions between race and time. The specifications in columns 1 to 3 further include fixed effects for county, and state $×$ time. The county-level regression in column 4 includes county and year fixed effects. The specification in column 5 includes a dummy for race, a quadratic in age, the cotton share in 1889, and fixed effects for time and state. Robust standard errors clustered at the county level in parentheses: ***p$<$ 0.01, **p$<$ 0.05, and *p$<$ 0.1.

### B. The Boll Weevil's Effect on Industrialization

In this section, we document that the boll weevil triggered a shift from agriculture to manufacturing in the affected counties. We reestimate equation (1) for individuals of working-age (10- to 65-year-olds) residing in the Cotton Belt of the American South during the 1880–1930 period. The dependent variable is a dummy that indicates whether an individual works in manufacturing or lives or works on a farm.23 The estimating equation is
$occict=αc+αst+βBollWeevilIntensityct+ΓXict+εict.$
(5)

Since this sample consists of both men and women, we also include a dummy for gender. Columns 2 to 3 of table 3 summarize the results. Column 2 shows that individuals in boll weevil–infested counties are more likely to be employed in manufacturing. For example, individuals living in a county with a high cotton intensity (i.e., all counties above the 75th percentile of the 1889 cotton share)24 are about 0.5 percentage points more likely to be employed in manufacturing upon the boll weevil's arrival (approximately 5% of individuals are employed in manufacturing; see table A.1). Column 3 reports a significant decline of individuals living or working on a farm consistent with the findings of Ager et al. (2017). For example, in a county with a high intensity of cotton production, the farm population went down by about 2.2 percentage points. This effect is quantitatively larger if we consider only individuals reporting a gainful occupation in agriculture (available on request). Column 4 complements the microlevel results with county-level data.25 The relative increase in manufacturing activities in these counties is also in line with Ager et al. (2017), who find a substantial relative decline in the number of farms and agricultural land usage in counties with a higher initial cotton intensity after the boll weevil's arrival. Overall, the evidence presented in this section suggests that the boll weevil triggered a shift out of agriculture in more cotton-dependent counties. The estimated effects of the boll weevil infestation on structural change may not seem sizable (consistent with Wright, 1986); however, given that the average level of manufacturing employment reached at the time in the Cotton Belt was relatively low, they are quite substantial.

One potential concern is that our results might be driven by a composition effect. That is, the shift from farming to manufacturing activities might be a consequence of selective migration. Using a set of linked representative samples from the IPUMS, we show in column 5 that in a county with a high cotton intensity, the boll weevil infestation increased the probability that households moved out of the agricultural sector by 3.1 percentage points. This confirms that our estimate reported in column 3 is not likely to be driven by selective migration.

## VI. Effect of Structural Change on Fertility

In this section, we exploit plausibly exogenous variation in agricultural production to estimate the causal effect of changes in the agricultural earnings potential on fertility in the American South. Sections VIA and VIB, document two separate channels: (a) lower agricultural income reduces the fertility of stayer households, consistent with the notion that children are a normal good, and (b) switcher households reduce their fertility, potentially because working in manufacturing is less compatible with childbearing and because the direct costs of having children are higher. We then exploit a second source of exogenous variation in agricultural production—the dramatic increase of cigarette consumption during World War I on local tobacco cultivation—to disentangle the effects of the two channels on the fertility decline. In both sections, the analysis is conducted at the county level since agricultural income is observable only for households staying in agriculture. Section VIC discusses the exclusion restriction of the instrumental variable strategy.

### A. Effect of Agricultural Income on Fertility

In this section, we quantify the effect of agricultural income on fertility for households staying in agriculture. We would expect this relationship to be positive within agricultural occupations, since the income effect is likely to dominate the substitution effect when the opportunity costs of child rearing are low. To estimate the causal relationship between agricultural income and fertility for stayer households, our empirical analysis exploits exogenous variation due to the boll weevil infestation in a two-stage least squares approach. The estimating equation is
$Fertilityct=αc+αt+δIncomect+εct.$
(6)
$Fertilityct$ is the average number of children under age 5 of 20- to 39-year-old married women in agricultural households in county $c$ in year $t$. $Incomect$ is the average labor income from agricultural activities. The empirical specification controls for county fixed effects, $αc$, and time fixed effects, $αt$. Standard errors are Huber robust and clustered at the county level.
The excluded instrument in the two-stage least squares regression is the interaction between the boll weevil incidence and the initial cotton intensity. The first-stage equation is
$Incomect=αc+αt+γBollWeevilIntensityct+εct,$
(7)
where $BollWeevilIntensityct$ is defined as in section IVA. Identification in the two-stage least squares estimation comes from the differential effect that the incidence of the boll weevil had on agricultural income and fertility due to differences in the importance of (initial) cotton production in the Cotton Belt counties of the American South.

Columns 1 to 3 of table 4 present the two-stage least squares results for stayer households. The second-stage coefficient is reported in column 1 and implies that a decline in agricultural income of 10% decreases the number of children under age 5 by 0.015. Such an income reduction would explain about 3.5% of the overall decline in the number of children under age 5 between 1880 and 1930.26 The estimated first-stage coefficient $γ$ in column 2 is negative and statistically significant at the 1% level. In counties where cotton production is relatively more important, the boll weevil infestation had a larger, negative effect on agricultural income. In terms of instrument quality, the two-stage least squares estimation strategy yields a reasonable first-stage fit. For completeness, we show the reduced-form estimate in column 3.

Table 4.
Structural Change and the Fertility Transition
 (1) (2) (3) (4) (5) (6) (7) (8) (9) Variables Number of Children under Age 5 ln(Agricultural Income) Number of Children under Age 5 Number of Children under Age 5 $Δ$ln(Income) Number of Children under Age 5 ln(Agricultural Income) % in Manufacturing Number of Children under Age 5 ln(Agricultural Income) 0.156*** 0.156*** (0.036) (0.043) % in Manufacturing −0.209*** (0.042) $BollWeevilIntensityct$ −0.439*** −0.068*** −0.413*** 0.127** −0.091*** (0.050) (0.012) (0.0380) (0.0618) (0.009) Tobacco −0.0218 −0.215*** 0.042*** (0.0280) (0.0421) (0.007) Leaves Farm −0.251*** 0.242*** (0.040) (0.068) Observations 3,568 3,568 3,568 2,346 1,149 5,693 5,693 5,693 5,693 $R2$ 0.644 0.854 0.116 0.228 0.713 0.778 0.909 Kleibergen-Paap 75.66 12.52 IV-equation 2nd stage 1st stage Reduced 2nd stage 1st stage 1st stage Reduced form form $F$-statistic instrument 1 41.23 $F$-statistic instrument 2 26.58
 (1) (2) (3) (4) (5) (6) (7) (8) (9) Variables Number of Children under Age 5 ln(Agricultural Income) Number of Children under Age 5 Number of Children under Age 5 $Δ$ln(Income) Number of Children under Age 5 ln(Agricultural Income) % in Manufacturing Number of Children under Age 5 ln(Agricultural Income) 0.156*** 0.156*** (0.036) (0.043) % in Manufacturing −0.209*** (0.042) $BollWeevilIntensityct$ −0.439*** −0.068*** −0.413*** 0.127** −0.091*** (0.050) (0.012) (0.0380) (0.0618) (0.009) Tobacco −0.0218 −0.215*** 0.042*** (0.0280) (0.0421) (0.007) Leaves Farm −0.251*** 0.242*** (0.040) (0.068) Observations 3,568 3,568 3,568 2,346 1,149 5,693 5,693 5,693 5,693 $R2$ 0.644 0.854 0.116 0.228 0.713 0.778 0.909 Kleibergen-Paap 75.66 12.52 IV-equation 2nd stage 1st stage Reduced 2nd stage 1st stage 1st stage Reduced form form $F$-statistic instrument 1 41.23 $F$-statistic instrument 2 26.58

This table shows estimates of the causal impact of structural change on fertility. Columns 1 and 6 show two-stage least squares estimates based on equations 6 and 8. Columns 2, 7, and 8 report the corresponding first-stage regressions, and columns 3 and 9 report the reduced-form regressions. The two-stage least squares specifications are conducted at the county level and include fixed effects for county and time. Columns 4 and 5 use a sample of linked households from IPUMS. The method of estimation is least squares. In column 4, we restrict the sample to men with a wife of age 20-49 in the terminal year; in column 5, we restrict the sample to men of age 20 or older in 1880 and not older than 65 in the terminal year. Further controls are a dummy for race, a quadratic in age, the cotton share in 1889, and fixed effects for time and state. Robust standard errors clustered at the county level in parentheses: ***p < 0.01, **p < 0.05, and *p < 0.1.

### B. Effect of Industrialization on Fertility

In this section, we show evidence that agricultural households that switched to manufacturing reduced their fertility. We then provide causal evidence that a shift to manufacturing due to lower agricultural earnings opportunities reduces fertility.

During our sample period, 20- to 39-year-old married women in agricultural and nonagricultural households reported to have 1.08 and 0.69 children below the age of 5, respectively. While suggestive, this is not conclusive evidence that switching to manufacturing will induce a household to reduce fertility, since households with a stronger preference for children might also be more likely to work in the agricultural sector, independent of the cost of child rearing. We address this issue by showing complementary evidence based on a sample of households from the 1900, 1910, and 1920 Censuses, which have been linked to the 1880 Census by IPUMS. This allows us to compare the fertility of switcher households to that of households remaining in the agricultural sector throughout the period.27 We restrict our sample to households that were initially (in 1880) in the agricultural sector to alleviate concerns regarding the importance of selection bias. In column 4 of table 4, the estimated coefficient on the dummy variable, Leaves Farm, indicates that switcher households have around 0.25 fewer children under age 5 than stayer households. This effect is statistically significant at the 1% level. In column 5, we show that switching to manufacturing also went along with a substantial increase in income. The results in this and the previous sections are therefore consistent with the theoretical framework by Mookherjee et al. (2012), which postulates a positive wage-fertility correlation within broad occupations or human capital categories but a negative correlation between parental wages and fertility across occupations.

While compelling, the evidence discussed above does not show that industrialization had a causal effect on fertility. A challenge to identification is that the arrival of the boll weevil represents only one source of exogenous variation. We therefore cannot simultaneously use it as an instrument for structural change on the intensive margin (reduction of agricultural income for stayer households) and the extensive margin (industrialization; that is, households switching to the manufacturing sector).

In order to disentangle and quantify the importance of each channel, we therefore exploit the unprecedented increase in cigarette consumption during World War I as a second source of exogenous variation in agricultural production.28 The commander of the American Expeditionary Forces in World War I, General Pershing, regarded tobacco as essential for the morale of American soldiers in Europe and requested cigarettes be part of the daily ration of American troops in 1917 (Tate, 2000; Brandt, 2007). Following Pershing's request, the US government spent approximately $80 million (or equivalently,$1,480 million in 2015) on tobacco products between April 7, 1917, and May 1, 1919. “[Since the US] government shipped about 5.5 billion manufactured cigarettes along with enough tobacco to roll another 11 billion overseas” (Tate, 2000, p. 75) during that period, this unprecedented increase in demand stimulated tobacco cultivation in the American South and can be regarded as plausibly exogenous for local producers.

We construct the second instrument as the product of the tobacco farm price in a given year and the share of tobacco cultivated in a county in 1909 (the last Agricultural Census prior to World War I).29 The instrument is in the spirit of a so-called shift-share instrumental variable approach as it predicts local tobacco production based on the interaction of aggregated demand shocks and a predetermined distribution of tobacco cultivation at the county level. Since most of the tobacco cultivation took place outside the Cotton Belt counties, the sample for the following empirical analysis includes all counties of the state of Kentucky and the Cotton Belt states. This ensures that we include the most important tobacco producing counties of the American South in the empirical analysis.30 It is important to note that this instrument does not need to capture the main source of variation in agricultural earnings opportunities; for our identification strategy, it is sufficient that it provides some plausibly exogenous variation in agricultural production besides the boll weevil infestation.

We use the following two-stage least squares approach using two instruments,
$Fertilityct=αc+αt+κIncomect+θMfgSharect+εct,$
(8)
where $Fertilityct$ denotes the average number of children under age 5 of 20- to 39-year-old women in county $c$ at time $t$. The two endogenous variables are $Incomect$, measured as the average logarithmic income of individuals working in agriculture, and $MfgSharect$, the fraction of the county population working in manufacturing measured in logarithmic units. Equation (8) further includes county fixed effects, $αc$, and year fixed effects, $αt$. We compute standard errors that are Huber robust and clustered at the county level. The corresponding first-stage equations are
$Incomect=αc+αt+λBollWeevilIntensityct+μTobaccoct+νct,$
(9a)
$MfgSharect=αc+αt+πBollWeevilIntensityct+τTobaccoct+ξct.$
(9b)

The excluded instruments are $BollWeevilIntensityct$ and $Tobaccoct$, defined as the interaction between the farm price of tobacco in year $t$ and county $c$'s acreage share of tobacco planted in 1909. Identification comes from the differential effect that the incidence of the boll weevil and the tobacco instrument had on agricultural income, the manufacturing share, and fertility due to differences in the importance of local cotton and tobacco production.

Column 6 of table 4 presents the county-level results on the effect that industrialization in the American South had on fertility based on estimating equation (8); the corresponding first-stage and reduced-form estimates are reported in columns 7 to 9. Consistent with our previous findings, the two-stage least squares estimates show that a decline in agricultural income and a rise in the manufacturing share significantly reduced fertility. The coefficients of interest are statistically significant at the 1% level, and the Sanderson-Windmeijer first-stage F-statistic for both instruments indicates that the instrumental variable estimates are not substantially biased. A 10% increase in the manufacturing share reduces the number of children under age 5 of 20- to 39-year old mothers by about 0.02. This effect is quantitatively sizable: based on our estimates, the increase in the manufacturing share over our sample period explains about 29% of the overall marital fertility decline between 1880 and 1930.

### C. Exclusion Restriction and Sensitivity Analysis

One potential threat to identification is that our instruments might affect the fertility behavior of agricultural households through other channels than agricultural income or a shift in the manufacturing share. In section II, we discussed recent evidence that the boll weevil induced population movements and a shift of production to corn, and it changed southern agricultural labor arrangements and labor market outcomes. It follows from this literature that these effects can be regarded as a direct consequence of changed earnings opportunities in the agricultural sector and therefore do not constitute a threat to the instrumental validity. However, it still might be the case that regardless of any changes in agricultural earnings opportunities, initial differences in these attributes in affected counties might have contributed to differential changes in fertility over the sample period. We address this potential issue in table A.8. Columns 2 to 6 of panel A show that the empirical estimates are robust to controlling for preinfestation values of population, the black share, the corn share and total acres planted in crops, the tenant share, and the female labor force participation rate at the count level interacted with a full set of time fixed effects (column 1 reports the baseline for comparison). In panel B we also include a measure of initial (1880) fertility fully interacted with time fixed effects, a flexible and demanding way of controlling for any mean reverting fertility dynamics. While we find slight changes in the relative contribution of agriculture income and the manufacturing share to fertility, our results remain qualitatively unaffected.

Another potential issue is that the lower agricultural earnings opportunities might have directly incentivized parents to invest more in the education of their children and therefore reduced fertility independent of its effect through agricultural income or the manufacturing share. In section VII, we provide evidence that this was not the case. Although we report a rise in schooling, we find no evidence of an increase in the returns to education due to the presence of the boll weevil per se. To the contrary, our evidence suggests that increased schooling is exclusively driven by diminishing returns to child labor and the associated increase in the direct cost of raising children (regardless of child quality), which decreased household income.31 Moreover, if switcher households would decide to invest more in their children's education and therefore reduce fertility, this would ultimately be a consequence of the switching decision (triggered by the lower agricultural earnings opportunities) and, hence, not constitute a threat to our identification strategy.

## VII. Human Capital

In this section, we show that the lower earnings opportunities in agriculture led to a substantial decline in child labor, which stimulated human capital formation in the Cotton Belt. For the empirical analysis, we use a sample of 10- to 15-year-old children that can be linked to their mothers from the 1900 to 1930 full count US Census microdata.32 The specification in this section is identical to equation (1), except that the dependent variable is a dummy that equals 1 if a 10- to 15-year-old child reports a gainful occupation, regularly attends school,33 or is considered to be “idle,” that is, the child neither regularly attends school nor reports a gainful occupation. Our specification further accounts for potential differences in parental education levels by including dummies for father's and mother's literacy. Standard errors are Huber robust and clustered at the county level.

Table 5 summarizes the results. In columns 1 and 2, we find that in the more affected counties, the boll weevil infestation resulted in a substantial decline in child labor, while at the same time, regular school attendance of 10- to 15-year-old children increased significantly. Both effects are statistically significant at the 1% level. For a county with a high cotton intensity (ranked at the 75th percentile), the boll weevil infestation led to a decrease in the likelihood that a 10- to 15-year-old child reports a gainful occupation by more than 4 percentage points. Given that about 23% of children in the Cotton Belt worked during the period 1900 to 1930 (see table A.1), this effect is quantitatively important and also suggests that child labor at this time was relatively less valuable outside agriculture.34 At the same time, the likelihood that a child of the same age group regularly attended school increased by 2.5 percentage points.35 Compared to the effect on child labor, the schooling effect is more modest, as about two-thirds of children of this age group attended school regularly (see table A.1).

Table 5.
The Impact of the Boll Weevil Infestation on Child Labor and Schooling
 (1) (2) (3) (4) (5) (6) (7) Variables =1 if child works =1 if attends school =1 if idle Schools Teachers Schools Teachers $BollWeevilIntensityct$ −0.079*** 0.046*** 0.034*** 0.212*** 0.514*** −0.035 0.045 (0.014) (0.014) (0.008) (0.044) (0.103) (0.070) (0.180) $BollWeevilIntensityct$ 0.481*** 0.912*** $×$ Share Child Labor (0.140) (0.331) Observations 7,641,595 7,490,359 7,490,359 1,700 1,700 1,700 1,700 $R2$ 0.234 0.221 0.047 0.785 0.774 0.790 0.776
 (1) (2) (3) (4) (5) (6) (7) Variables =1 if child works =1 if attends school =1 if idle Schools Teachers Schools Teachers $BollWeevilIntensityct$ −0.079*** 0.046*** 0.034*** 0.212*** 0.514*** −0.035 0.045 (0.014) (0.014) (0.008) (0.044) (0.103) (0.070) (0.180) $BollWeevilIntensityct$ 0.481*** 0.912*** $×$ Share Child Labor (0.140) (0.331) Observations 7,641,595 7,490,359 7,490,359 1,700 1,700 1,700 1,700 $R2$ 0.234 0.221 0.047 0.785 0.774 0.790 0.776

This table shows the boll weevil's impact on child labor and schooling. The dependent variables in columns 1 to 3 are a dummy whether a child works, regularly attends school, or is idle. In columns 4 to 7, the dependent variable is the number of Rosenwald schools and teachers per 1,000 inhabitants. $BollWeevilIntensityct$ is the interaction between a dummy variable that equals 1 if county $c$ was infested at time $t$ and county $c$'s acreage share of cotton planted in 1889. $BollWeevilIntensityct$$×$ Share Child Labor is the interaction of $BollWeevilIntensityct$ with county $c$'s child labor share in 1910. Columns 1 to 3 include the following set of individual controls: dummies for gender, race, rural, parents' literacy, age fixed effects, and interactions between race, rural, and time. All specifications include fixed effects for county, and state $×$ time. Columns 4 to 7 further control for the share of black people in county $c$ at time $t$. Robust standard errors clustered at the county level in parentheses: ***p < 0.01, **p < 0.05, and *p < 0.1.

Column 3 of table 5 shows that this smaller effect on schooling is due to the fact that a substantial fraction of parents left their children idle after the cotton fields were ravaged by the boll weevil. In a county with a high cotton intensity, this increase was almost 2 percentage points, which is a sizable effect, since only about 9% of 10- to 15-year-old children were listed as idle (see table A.1). These findings indicate that reduced agricultural earnings opportunities primarily reduced the value of child labor but did not increase the attractiveness of schooling and the returns to education per se.

We can also directly assess whether fertility declined due to changes in the returns to education by investigating fertility adjustments along the extensive and intensive margins following the theoretical framework of Aaronson et al. (2014). While increases in the returns to education would imply a decline on the intensive but not the extensive margin, an increase in the direct cost of having children implies a decline of fertility along both margins. Column 2 of table A.3 shows that fertility also decreased along the extensive margin. If the boll weevil also affected fertility directly by changing the returns to education, our evidence suggests that this effect was quantitatively modest and negligible.

Finally, columns 4 to 7 of table 5 provide additional county-level evidence that the increased direct costs of children stimulated the demand for schooling in the Cotton Belt. One prominent education program at the time was the Rosenwald Rural Schools Initiative (Aaronson & Mazumder, 2011; Carruthers & Wanamaker, 2013; Aaronson et al., 2014). The objective of this program was to narrow the racial education gap that existed in the American South at that time, especially in rural areas.36 Between 1914 and 1931, the Rosenwald Program constructed about 5,000 new schools throughout the rural American South targeted to the black rural population.37 The county-level regressions reported in columns 4 to 7 control for county fixed effects, state-by-time fixed effects, and the black population share. The estimates reveal that for infested counties, a 10 percentage point higher initial cotton share implies the construction of about two more schools and five more teachers per 1,000 inhabitants. These results show that the arrival of the boll weevil had a substantial impact on where schools were constructed. In columns 6 and 7, we add the interaction of $BollWeevilIntensityct$ with the initial (1910) county child labor share. The estimates suggest that most Rosenwald schools were constructed in counties that experienced the highest decrease in the value of child labor. This is consistent with the notion that human capital formation in the Cotton Belt was triggered by the increased direct costs of children during our sample period.

Overall, the findings presented in this section are in line with the predictions of the standard Q-Q framework. The lower value of child labor increased the direct costs of children and induced parents to invest more in child quality. The Q-Q model can therefore rationalize the well-documented increase in school enrollment that went along with the structural change and the fertility decline that the American South experienced during the 1880–1930 period.

## VIII. Conclusion

A prominent hypothesis in growth and economic development is that a sustained shift from agriculture to manufacturing contributed to the historic fertility decline in today's modern societies. Empirical evidence in support of this hypothesis remains scarce because identifying a causal relationship is challenging. This paper fills the gap in the literature by using credibly exogenous variation in agricultural earnings opportunities to estimate the causal link between structural change and the fertility transition. We show that lower agricultural earnings opportunities triggered a structural change on the intensive margin (i.e., the reduction of agricultural income for stayer households) and the extensive margin (i.e., households switching to manufacturing (industrialization)). In line with the notion that children are a normal good, we find that stayer households reduced fertility as they experienced income losses, while switcher households reduced fertility because manufacturing work is generally less compatible with raising children.

This finding also implies that there are more complex mechanisms behind the well-documented negative correlation between parental income and fertility (Jones & Tertilt, 2008). In line with the theoretical framework by Mookherjee et al. (2012), we find evidence of two potentially confounding effects. Mobility across sectors or broad occupational categories changes parents' incentives to invest in child quality and usually implies a negative fertility-income relation. Within the same occupational category, however, higher income could increase fertility depending on whether the income effect dominates the substitution effect (Doepke, 2004). Identifying the effect of a wealth/income shock for households staying within the same occupation and for households switching occupations separately is therefore crucial for researchers interested in establishing a causal relationship between fertility and parental income.

We further show that lower earnings opportunities in agriculture diminished the value of child labor, which made schooling relatively more attractive. This finding is in line with the testable implications of a standard quantity-quality model of fertility (Galor, 2011), which can rationalize the well-documented increase in school enrollment that went along with structural change and the historical fertility transition in most of Europe and North America during the nineteenth and early twentieth centuries.

One limitation of our empirical analysis is that individual Census data on wages are not available before 1940. While we collected additional historical data to improve existing measures that rely on occupational income, changes in agricultural income in this study result from wage differences across agricultural occupations or spatial and temporal variation in wages of farm laborers or farmer income. This data limitation implies that we cannot perfectly identify the income effect at the individual level. One further concern is the external validity of our findings. While our study is based on full count data covering responses of more than 10 million southern households, our findings might be specific to the particular history and characteristics of the American South. Further research on the relationship between the structural transformation and falling fertility rates in other historical settings would therefore be valuable.

Overall, our study supports the view that the structural transformation was an important determinant of the southern fertility transition during the late nineteenth and early twentieth centuries. The lower earnings opportunities in agriculture contributed to a substantial fertility decline in southern households by accelerating the process of industrialization and stimulating the demand for human capital. Having the concerns discussed above in mind, this result seems not unique to historical settings. In particular, one could think of policies that reduce population pressure in developing countries by combining rigorous child labor laws with economic programs that stimulate the transition out of the agricultural sector.

## Notes

1

This also suggests that the wage-fertility relation can be positive within broad occupational categories but negative across occupational categories (see Mookherjee, Prina, & Ray, 2012).

2

This finding is in line with research that documents a positive relationship between income and fertility for preindustrial societies and predominantly agrarian economies (Clark, 2005; Clark & Hamilton, 2006).

3

For example, during our sample period, married 20- to 39-year-old women in the Cotton Belt in agricultural households reported having 1.08 children under age 5, while the number was 0.69 for nonagricultural households.

4

The idea that child labor is an important determinant of fertility behavior since it increases the value of children's time and, at the same time, raises the opportunity cost of schooling was analyzed by Rosenzweig and Evenson (1977). In line with this argument, Hazan and Berdugo (2002) and Doepke (2004) show that child labor restrictions and education policies played an important role for the fertility decline and the transition to sustained economic growth.

5

Note that most of the fertility transition in the southern region took place during the first decades of the twentieth century (Steckel, 1992); see, for example, Elman, London, and McGuire (2015) for reasons for the delay in the southern fertility transition and Bailey and Hershbein (2015) for an overview of the literature on the US fertility transition.

6

Giesen (2011) argues that thirty years after the boll weevil's arrival in the Cotton Belt, the southern cotton economy remained relatively unchanged: the South produced even more cotton in 1921 than in 1892. Osband (1985) claims that the overall effect of the boll weevil on the southern economy was modest since he finds only minimal annual revenue losses for cotton producers.

7

Even if agricultural production increased at the aggregate level, it is not clear that this leads to an increase in farmers' net income because of potentially rising input costs, such as increased cost for fertilizer (see Lange et al., 2009, note 28). Our construction of agricultural income takes input costs into account (see the online data appendix for further details).

8

The year 1890 is omitted from the analysis since the completed census forms were lost in a fire.

9

The spouse is present for about 96% of the married women in our sample.

10

The IPUMS occupation score has been used in the literature as an approximation for income over longer periods of time (Jones & Tertilt, 2008).

11

We thank Fabian Lange, Alan Olmstead, and Paul Rhode for sharing their boll weevil data.

12

The cotton share is constructed as in Ager et al. (2017, note 14).

13

The mean of $BollWeevilIntensityct$ is 0.19 (see table A.1). The weevil's effect on fertility is therefore $0.19×(-0.041)=-0.008$. The average number of children under age 5 per married 20- to 39-year-old married women in our sample fell by about 0.45 between 1880 and 1930.

14

We address the potential threat of confounding factors that vary over time at the county level by adding county-by-time fixed effects to specification (2) and using women aged 40 to 44 as a control group. That is, identification comes from within-county variation across age cohorts only. While the control group is not optimal, the estimates turn out to be similar to column 1, suggesting that it is not likely that time-varying county-specific omitted variables are driving our findings (column 3 of table A.3).

15

We also show in table A.2 that estimates are similar when the sample is split by race.

16

We restrict the sample to children younger than 15 at the time of the Census.

17

Note that county-specific effects are captured by the mother fixed effects (in case the mother stayed throughout her fertility history at her place of residence listed in the Census).

18

Data on corn suitability come from the Food and Agriculture Organization.

19

Data are retrieved from the 1900 and 1910 Censuses (see IPUMS variables CHBORN and CHSURV) and from Fishback, Haines, and Kantor (2007).

20

In the American South at that time it was not common for married women to voluntarily delay the first marital birth; see, for example, Elman et al. (2015).

21

The age at marriage is constructed using the IPUMS variables DURMARR (available for the Census years 1900 and 1910) and AGEMARR (available for the Census year 1930).

22

We consider a household to be agricultural if it resides on a farm (indicated in IPUMS by the variable FARM) or if the husband reports one of the following occupations (OCC1950 from IPUMS): 100, 123, 810–840, and 970 if the household's location is rural.

23

Based on the variable OCC1950 from IPUMS, the categories are defined as follows: manufacturing is OCC1950 500-690, and lives or works on a farm is OCC1950 100, 123, 810-840, 970 (if rural) or lives on a farm (FARM = 2) if OCC1950 >970.

24

The 1889 cotton share at the 75th percentile is 54%.

25

For a county with an initial cotton share at the 75th percentile, the arrival of the boll weevil increased the share of the population working in the manufacturing sector by approximately 8%, which is consistent with the quantitative evidence reported in column 2.

26

The decline in the number of children under age 5 in the Cotton Belt was 0.45 during the sample period.

27

As shown in section VA, the arrival of the boll weevil decreased agricultural income and led to a fertility decline. We therefore exclude households that stayed in agriculture and lived in a county where the boll weevil was present in the terminal year (1900, 1910, or 1920).

28

Tobacco was another major cash crop in the American South during the sample period; see, for example, Towne and Rasmussen (1960).

29

The tobacco share is constructed analogous to the cotton share. We consider the tobacco farm prices of Kentucky—the largest tobacco-producing state at that time—as representative of the tobacco-producing states in the American South; the corresponding data sources are listed in the online data appendix. The evolution of the tobacco farm price is shown in figure A.2.

30

Kentucky, North and South Carolina, Tennessee, and Virginia produced more than 75% of US tobacco in 1919.

31

Following the notation in Galor (2012, section 4.1), this would correspond to an increase in $τq$ (i.e., the fixed costs of child rearing regardless of child quality), while $τe$ (i.e., the costs of investing in child quality) would remain unaffected.

32

The 1880 IPUMS full count Census data do not include information on school attendance.

33

As in Bleakley and Lange (2009), regular school attendance is a dummy variable equal to 1 if a child is attending and not working. For 1900, we construct school attendance based on the IPUMS variable SCHLMNTH. For 1910, we use the IPUMS variable SCHOOL. We consider a child to be working if a gainful occupation is reported in the Census.

34

This is in line with the view that children had a comparative advantage in picking cotton (Goldin and Sokoloff, 1984).

35

This finding is in line with Baker (2015, p. 1129) and resonates with the positive long-run impact that the weevil had on educational attainment (Baker, Blanchette, & Eriksson, 2018).

36

The racial gap in schooling in the American South at the beginning of the twentieth century was substantial and is largely explained by differences in school characteristics and the lower economic status and education levels of black parents. See Margo (1990) and Collins and Margo (2006) for an overview.

37

Since the rollout of Rosenwald schools started during the 1910s, we include only Cotton Belt counties that were infested by the boll weevil after 1910. Consequently, the sample spans the period 1910 to 1930. Note that there are no county-level population data available for the year 1925. These are imputed using the mean of the total population from the 1920 and 1930 Censuses.

## REFERENCES

Aaronson
,
Daniel
,
Fabian
Lange
, and
Bhashkar
Mazumder
, “
Fertility Transitions along the Extensive and Intensive Margins,
American Economic Review
104
(
2014
),
3701
3724
.
Aaronson
,
Daniel
, and
Bhashkar
Mazumder
, “
The Impact of Rosenwald Schools on Black Achievement,
Journal of Political Economy
119
(
2011
),
821
888
.
Ager
,
Philipp
,
Markus
Brueckner
, and
Benedikt
Herz
, “
The Boll Weevil Plague and Its Effect on the Southern Agricultural Sector, 1889–1929,
Explorations in Economic History
65
(
2017
),
94
105
.
Ager
,
Philipp
,
Markus
Brueckner
, and
Benedikt
Herz
Structural Change and the Fertility Transition in the American South
,”
Australian National University, Centre for Economic History working paper
2018-01
(
2018
).
Bailey
,
Martha J.
, and
Hershbein
, “
US Fertility Rates and Childbearing, 1800–2010,
” in
Louis P.
Cain
,
Price V.
Fishback
, and
Paul W.
Rhode
, eds.,
Oxford Handbook of American Economic History
(
New York
:
Oxford University Press
,
2015
).
Baker
,
Richard B.
, “
From the Field to the Classroom: The Boll Weevil's Impact on Education in Rural Georgia,
Journal of Economic History
75
(
2015
),
1128
1160
.
Baker
,
Richard B.
,
John
Blanchette
, and
Katherine
Eriksson
, “
Long-Run Impacts of Agricultural Shocks on Educational Attainment: Evidence from the Boll Weevil
,”
NBER working paper
25400
(
2018
).
Becker
,
Gary S.
, “
An Economic Analysis of Fertility,
” in
Orley C.
Ashenfelter
and
Richard
Layard
, eds.,
Demographic and Economic Change in Developed Countries
(
Princeton, NJ
:
Princeton University Press
,
1960
).
Becker
,
Gary S.
, and
H. Gregg
Lewis
, “
On the Interaction between the Quantity and Quality of Children,
Journal of Political Economy
81
(
1973
),
279
288
.
Becker
,
Sascha O.
,
Francesco
Cinnirella
, and
Ludger
Woessmann
, “
The Trade-Off between Fertility and Education: Evidence from before the Demographic Transition,
Journal of Economic Growth
15
(
2010
),
177
204
.
Black
,
Dan A.
,
Natalia
Kolesnikova
,
Seth G.
Sanders
, and
Lowell J.
Taylor
, “
Are Children ‘Normal’?
this review
95
(
2013
),
21
33
.
Bleakley
,
Hoyt
, and
Fabian
Lange
, “
Chronic Disease Burden and the Interaction of Education, Fertility, and Growth,
this review
91
(
2009
),
52
65
.
Bleakley
,
Hoyt
, and
Joseph
Ferrie
, “
Shocking Behavior: Random Wealth in Antebellum Georgia and Human Capital across Generations,
Quarterly Journal of Economics
131
(
2016
),
1455
1495
.
Bloome
,
Deirdre
,
James
Feigenbaum
, and
Christopher
Muller
, “
Tenancy, Marriage, and the Boll Weevil Infestation, 1892–1930,
Demography
54
(
2017
),
1029
1049
.
Bolt
,
Jutta
, and
Jan Luiten
van Zanden
, “
The Maddison Project: Collaborative Research on Historical National Accounts,
Economic History Review
67
(
2014
),
627
651
.
Brandt
,
Allan M.
,
The Cigarette Century: The Rise, Fall, and Deadly Persistence of the Product That Defined America
(
New York
:
Basic Books
,
2007
).
Carruthers
,
Celeste K.
, and
Marianne H.
Wanamaker
, “
Closing the Gap? The Effect of Private Philanthropy on the Provision of African-American Schooling in the US South,
Journal of Public Economics
101
(
2013
),
53
67
.
Clark
,
Gregory
, “
Human Capital, Fertility, and the Industrial Revolution,
Journal of the European Economic Association
3
(
2005
),
505
515
.
Clark
,
Gregory
, and
Gillian
Hamilton
, “
Survival of the Richest: The Malthusian Mechanism in Pre-Industrial England,
Journal of Economic History
66
(
2006
),
707
736
.
Clay
,
Karen
,
Ethan
Schmick
, and
Werner
Troesken
, “
The Rise and Fall of Pellagra in the American South,
Journal of Economic History
79
(
2019
),
32
62
.
Collins
,
William J.
, and
Robert A.
Margo
, “
Historical Perspectives on Racial Differences in Schooling in the United States
” (
vol. 1
, pp.
107
154
), in
Eric A.
Hanushek
and
Finis
Welch
, eds.,
Handbook of the Economics of Education
(
Amsterdam
:
North-Holland
,
2006
).
Doepke
,
Matthias
, “
Accounting for Fertility Decline during the Transition to Growth,
Journal of Economic Growth
9
(
2004
),
347
383
.
Easterlin
,
Richard A.
, “
Population Change and Farm Settlement in the Northern United States,
Journal of Economic History
36
(
1976
),
45
75
.
Elman
,
Cheryl
,
Andrew S.
London
, and
Robert A.
McGuire
, “
Fertility, Economic Development, and Health in the Early Twentieth-Century US South,
Journal of Interdisciplinary History
46
(
2015
),
185
223
.
Fishback
,
Price V.
,
Michael R.
Haines
, and
Shawn
Kantor
, “
Births, Deaths, and New Deal Relief during the Great Depression,
this review
89
(
2007
),
1
14
.
Freedman
,
Deborah S.
, “
The Relation of Economic Status to Fertility,
American Economic Review
53
(
1963
),
414
426
.
Galor
,
Oded
, “
From Stagnation to Growth: Unified Growth Theory
” (
vol. 1
, pp.
171
293
), in
Philippe
Aghion
and
Steven N.
Durlauf
, eds.,
Handbook of Economic Growth
(
Amsterdam
:
North-Holland
,
2005
).
Galor
,
Oded
Unified Growth Theory
(
Princeton, NJ
:
Princeton University Press
,
2011
).
Galor
,
Oded
The Demographic Transition: Causes and Consequences,
Cliometrica
6
(
2012
),
1
28
.
Galor
,
Oded
, and
David N.
Weil
, “
From Malthusian Stagnation to Modern Growth,
American Economic Review
89
(
1999
),
150
154
.
Galor
,
Oded
, and
David N.
Weil
Population, Technology, and Growth: From Malthusian Stagnation to the Demographic Transition and beyond,
American Economic Review
90
(
2000
),
806
828
.
Giesen
,
James C.
,
Boll Weevil Blues: Cotton, Myth, and Power in the American South
(
Chicago
:
University of Chicago Press
,
2011
).
Goldin
,
Claudia
, and
Kenneth
Sokoloff
, “
The Relative Productivity Hypothesis of Industrialization: The American Case, 1820 to 1850,
Quarterly Journal of Economics
99
(
1984
),
461
487
.
Greenwood
,
Jeremy
, and
Ananth
, “
The US Demographic Transition,
American Economic Review
92
(
2002
),
153
159
.
Guest
,
Avery M.
, “
Social Structure and US Inter-State Fertility Differentials in 1900,
Demography
18
(
1981
),
465
486
.
Guinnane
,
Timothy W.
, “
The Historical Fertility Transition: A Guide for Economists,
Journal of Economic Literature
49
(
2011
),
589
614
.
Haines
,
Michael R.
, and
Richard H.
Steckel
,
A Population History of North America
(
Cambridge
:
Cambridge University Press
,
2000
).
Haines
,
Michael R.
,
Price V.
Fishback
, and
Paul W.
Rhode
, “
United States Agriculture Data, 1840–2010,
Inter-University Consortium for Political and Social Research
(
2015
).
Hanushek
,
Eric A.
, “
The Trade-Off between Child Quantity and Quality,
Journal of Political Economy
100
(
1992
),
84
117
.
Hazan
,
Moshe
, and
Binyamin
Berdugo
, “
Child Labour, Fertility, and Economic Growth,
Economic Journal
112
(
2002
),
810
828
.
Higgs
,
Robert
, “
The Boll Weevil, the Cotton Economy, and Black Migration 1910–1930,
Agricultural History
50
(
1976
),
335
350
.
Hunter
,
Walter D.
, and
Bert R.
, “
The Boll-Weevil Problem
,”
USDA: The Farmers' Bulletin
, no.
1329
(
1923
).
Jones
,
Jacqueline
,
Labor of Love, Labor of Sorrow: Black Women, Work, and the Family from Slavery to the Present
(
New York
:
Basic Books
1985
).
Jones
,
Larry E.
, and
Michele
Tertilt
, “An Economic History of Fertility in the U.S.: 1826–1960” (pp.
165
230
), in
Peter
Rupert
, ed.,
Frontiers of Family Economics
(
Bingely, UK
:
Emerald Group Publishing
,
2008
).
Lange
,
Fabian
,
Alan L.
, and
Paul W.
Rhode
, “
The Impact of the Boll Weevil, 1892–1932,
Journal of Economic History
69
(
2009
),
685
718
.
Lebergott
,
Stanley
, “Labor Force and Employment, 1800–1960” (pp.
117
204
), in
Dorothy S.
, ed.,
Output, Employment, and Productivity in the United States after 1800
(
Cambridge, MA
:
NBER Books
,
1966
).
Lindo
,
Jason M.
, “
Are Children Really Inferior Goods? Evidence from Displacement-Driven Income Shocks,
Journal of Human Resources
45
(
2010
),
301
327
.
Lovenheim
,
Michael F.
, and
Kevin J.
Mumford
, “
Do Family Wealth Shocks Affect Fertility Choices? Evidence from the Housing Market,
this review
95
(
2013
),
464
475
.
Margo
,
Robert A.
,
Race and Schooling in the South, 1880–1950: An Economic History
(
Chicago
:
University of Chicago Press
,
1990
).
Mookherjee
,
Dilip
,
Silvia
Prina
, and
Debraj
Ray
, “
A Theory of Occupational Choice with Endogenous Fertility,
American Economic Journal: Microeconomics
4
(
2012
),
1
34
.
Osband
,
Kent
, “
The Boll Weevil Versus ‘King Cotton,'
Journal of Economic History
45
(
1985
),
627
643
.
Ransom
,
Roger L.
, and
Richard
Sutch
,
One Kind of Freedom: The Economic Consequences of Emancipation
(
Cambridge
:
Cambridge University Press
,
2001
).
Rodrigue
,
John C.
,“
Labor Militancy and Black Grassroots Political Mobilization in the Louisiana Sugar Region, 1865–1868,
Journal of Southern History
67
(
2001
),
115
142
.
Rosenzweig
,
Mark R.
, and
Robert
Evenson
, “
Fertility, Schooling, and the Economic Contribution of Children of Rural India: An Econometric Analysis,
Econometrica
45
(
1977
),
1065
1079
.
Ruggles
,
Steven
,
Katie
,
Ronald
Goeken
,
Josiah
Grover
, and
Matthew
Sobek
, “
Integrated Public Use Microdata Series: Version 7.0 [data set]
” (
2017
).
Steckel
,
Richard H.
, “The Fertility Transition in the United States: Tests of Alternative Hypotheses” (pp.
351
374
), in
Claudia
Goldin
and
Hugh
Rockoff
, eds.,
Strategic Factors in Nineteenth Century American Economic History: A Volume to Honor Robert W. Fogel
(
Chicago
:
University of Chicago Press
,
1992
).
Street
,
James H.
,
The New Revolution in the Cotton Economy
(
Chapel Hill, NC
:
University of North Carolina Press
,
1957
).
Sundstrom
,
William A.
, and
Paul A.
David
, “
Old-Age Security Motives, Labor Markets, and Farm Family Fertility in Antebellum America,
Explorations in Economic History
25
(
1988
),
164
197
.
Tan
,
Hui Ren
, “
More Is Less? The Impact of Family Size on Education Outcomes in the United States, 1850–1940,
Journal of Human Resources
(
2018
),
1154
1181
.
Tate
,
Cassandra
,
Cigarette Wars: The Triumph of the Little White Slaver
(
New York
:
Oxford University Press
,
2000
).
Towne
,
Marvin
, and
Wayne
Rasmussen
, “
Farm Gross Product and Gross Investment in the Nineteenth Century
” (pp.
255
316
), in
Trends in the American Economy in the Nineteenth Century
(
Princeton, NJ
:
Princeton University Press
,
1960
).
US Department of Agriculture
, “
Statistics on Cotton and Related Data,
Statistical Bulletin
, no.
99
(
1951
).
Wanamaker
,
Marianne H.
, “
Industrialization and Fertility in the Nineteenth Century: Evidence from South Carolina,
Journal of Economic History
72
(
2012
),
168
196
.
Wright
,
Gavin
,
Old South, New South: Revolutions in the Southern Economy since the Civil War
(
New York
:
Basic Books
,
1986
).

## Author notes

The order of authors reflects the relative contributions of the authors, with Philipp Ager and Benedikt Herz contributing equally. This paper is a substantially revised version of Ager, Brueckner, and Herz (2018). Only Ager and Herz contributed to the revision and to the process yielding the publication in this journal. There was no involvement of Brueckner. We thank Hoyt Bleakley, Davide Cantoni, Greg Clark, Carl-Johan Dalgaard, James Feigenbaum, James Fenske, Oded Galor, Paola Giuliano, Casper Worm Hansen, Erik Hornung, Peter Sandholt Jensen, Jeanne Lafortune, Lars Lønstrup, Michael Lovenheim, Bob Margo, Bhashkar Mazumder, Giovanni Mellace, Omer Moav, Battista Severgnini, Uwe Sunde, Nico Voigtlaen-der, Marianne Wanamaker, David Weil, and seminar participants at Brown, Copenhagen, LMU Munich, and the ED Workshop in Washington, DC, for helpful feedback. The opinions expressed in this paper do not necessarily reflect the opinion of the European Commission.

A supplemental appendix is available online at http://www.mitpressjournals.org/doi/suppl/10.1162/rest_a_00851.