Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Closing Heaven’s Door: Evidence from the 1920s U.S.
Immigration Quota Acts∗
Philipp Ager Casper Worm Hansen
Abstract
The introduction of immigration quotas in the 1920s fundamentally changed U.S. immigration
policy. We develop an empirical model that captures the missing immigrants of this policy change
to estimate the economic consequences of immigration restrictions for the U.S. economy. The
implementation of the quota system led to long-lasting declines in the foreign-born share and
population growth in areas with more missing immigrants per capita. This effect was largely
driven by the policy-restricted supply of immigrants from quota-affected nationalities and reduced
fertility. Our main finding is that while native white workers faced sizable earnings losses, black
workers benefited from the quota system and improved their relative economic status within the
more affected areas. We also find that productivity in the manufacturing sector declined as a result
of the new immigration policy.
Keywords: Immigration, population dynamics, local labor markets, racial wage gap, productivity growth
JEL Codes: J31, J61, N31, O15
∗Philipp Ager, University of Southern Denmark; [email protected]. Casper Worm Hansen, University of Copenhagen;
[email protected]. We thank Leah Boustan, Brian Cadena, Nathan Nunn, and Marianne Wanamaker for
detailed comments and suggestions, as well as Martha Bailey, Hoyt Bleakley, Michael Clemens, Carl-Johan Dalgaard,
Dave Donaldson, James Fenske, Nicola Gennaioli, Paola Giuliano, Walker Hanlon, Stephan Heblich, Peter Sandholt
Jensen, Jeanne Lafortune, Ethan Lewis, Bob Margo, Petra Moser, Paul Rhode, Daniel Sturm, Jose Tessada, Zach Ward,
Jeff Williamson, Asger Moll Wingender, Niko Wolf and seminar participants at the AEA Meetings 2018 in Philadelphia,
the University of Copenhagen, Humboldt University, the Second Workshop on Immigration, Health, and Well-Being
in Oxford, and the Economic Demography Workshop in Denver for valuable feedback. We gratefully acknowledge
financial support from the Danish Research Council grant No. DFF – 4182-00043. This paper was previously entitled
“National Immigration Quotas and Local Economic Growth” (University of Copenhagen Discussion Papers Department
of Economics No.16-11, November, 2016).
1 Introduction
The issue of how immigration restrictions influence labor market conditions and demographic out-
comes is highly policy relevant. Recent events such as the European migrant crises or the discussion
about increased border security in the United States have further stimulated the ongoing debate about
how rich countries should respond to higher immigration. In general, labor economists have investi-
gated the effect of immigration on a range of economic outcomes, such as native wages, employment,
and productivity,1 but, despite its importance, still relatively little is known about the economic effects
of immigration restrictions, especially at the sub-national level.
This paper contributes to the understanding of how immigration restrictions have influenced the
U.S. economy from a historical perspective. The debate on imposing immigration restrictions to
prevent entry of certain ethnic groups into the United States is not a new one (Hutchinson, 1981;
Higham, 2002). In the 1920s, the United States changed its open door policy for European immigrants
by introducing immigration quotas based on national origins (King, 2000). While approximately 30
million immigrants from Europe arrived in the United States between 1850 and 1914 (Abramitzky
et al., 2014), the implementation of the quota system curtailed European immigration to the United
States from 4.5 million between 1910 and 1914, to less than 800,000 between 1925 and 1929 (U.S.
Department of Commerce, 1931, Table 99). This fundamental change in immigration policy in the
1920s is unprecedented in U.S. history and potentially can be used to evaluate the causal effects of
immigration restrictions on the U.S. economy (Abramitzky and Boustan, forthcoming).
We exploit this regime change in U.S. immigration policy—the passage of the Emergency Quota
Act of 1921 and the Immigration Act of 1924, which abruptly ended the era of unrestricted immi-
gration from Europe to the United States (Goldin, 1994)—to study the economic consequences of
immigration restrictions for the U.S. economy during the period 1900-1940.2 The implementation
of the quota system provides two sources of variation which we exploit in our empirical analysis.
First, the quota system restricted immigration from some nations more than from others. This allows
us to calculate the missing immigrants due to the quota system for each nationality by comparing
the expected number of immigrants under free migration to the corresponding quota numbers. The
missing immigrants are assigned to each local area (county or city) in the United States based on
their nationality share in 1920. To measure the local effect of immigration this assignment rule builds
1See, for example, Card (1990, 2001, 2009), Borjas (2003, 2006, 2014), Peri (2012), Lewis and Peri (2015), Fogedand Peri (2015), Cadena and Kovak (2016), and Dustmann et al. (2016).
2We refer to Section 2 for a detailed description of the immigration quotas.
1
on an established observation in the migration literature that newly arriving immigrants tend to settle
in areas where previous immigrants of the same nationality live (Bartel, 1989; Card, 2001; Mun-
shi, 2003). It provides us with an immigration rate of missing immigrants which varies across the
United States, and we refer to it as quota exposure. Our estimation strategy combines the quota ex-
posure with the second source of variation arising from the timing of the policy change that permits
a before-and-after quota comparison of the outcomes of interest. With the two sources of variation
at hand, we can estimate the effects of a quota supply-driven decrease in immigration due to miss-
ing migrants which, due to variation in the 1920 nationality shares, is unequally distributed across
local economies.3 We implement our estimation strategy using three different samples: U.S. Census
county level data (1900–1940); a repeated cross-section of U.S. Census microdata (1900–1940); and
hand-collected data from the U.S. Census of Manufactures at the city level (1909–1929).
Our empirical analysis provides a rigorous quantitative assessment of the economic consequences
of the quota system for the U.S. economy focusing on local economies. The first section of the
empirical analysis establishes that the quota system led to a long-lasting relative decline in the foreign-
born share, and lower population growth. Our results reveal that the foreign-born share reduces by
1.6 percentage points, and the 10-year population growth rate declines by 6.7 percentage points in
counties with one additional missing immigrant per-100-inhabitants-per-year.4 A simple accounting
exercise demonstrates that the direct effect of missing immigrants due to the quota system explains
about 68 percent of the total population growth effect. One additional factor that contributed to the
decline in population growth was that women of childbearing age had fewer children after the quota
system was in place. The corresponding decline in marriage rates in the more affected counties
suggests that young women adjusted fertility to changing economic conditions.
In the second section of the empirical analysis we study the effect of the quota system on local
labor market outcomes. The regime change in U.S. immigration policy had a sizable effect on the
earnings of native workers.5 Our results reveal that native workers living in counties that were more
exposed to the quota system were pushed into lower-wage occupations. This effect, however, differs
substantially by gender and race. For men, the implementation of the quota system led to substantial
3While this strategy has some similarities to the more traditional shift-share instrumental variable approach (Bartik,1991), we focus on isolating a specific policy-driven variation in the immigration flow to the United States (see Section3.2 for more details).
4One additional missing immigrant per-100-inhabitants-per-year amounts to 10 percent (out of the total population)fewer immigrants arriving over a decade.
5Due to the lack of individual wage data before 1940, we use occupation and industry-based earnings scores in theempirical analysis to evaluate the effect of the immigration quota acts on the earnings of native workers. We refer toSection 3.1 for further details.
2
earnings losses for white workers, while black workers benefited from it. Our estimates imply that
white workers living in counties with one additional missing immigrant per-100-inhabitants-per-year
experienced a reduction in earnings of about 2.4 percent, while black workers, on average, increased
their earnings by about 2.3 percent each decade from 1920 to 1940. Black female workers also expe-
rienced significant gains in the more quota affected counties, while white women were not affected at
all. Consistent with this finding, we show that after the quota restrictions were in place native workers
mostly filled up the low-skilled jobs of the missing immigrants (which meant some economic gains
for black workers, who were, on average, at the bottom of the earnings distribution at that time, and
losses for white workers). After the quota system was implemented, young white adults (age 15-21)
in the more affected counties also stayed longer in school compared to their black peers, suggesting
local labor market conditions in these counties improved for black workers, at least in relative terms.
In order to evaluate to what extent the implementation of the quota system reduced the black-
white earnings gap, we additionally exploit the within-county-level variation of the quota exposure
by race. We find a stronger decline of the black-white earnings gap in the more affected counties. For
the average level of quota exposure (i.e., 0.69 missing immigrants per 100 inhabitants in a county per
year), the earnings of black-male workers increased up to 3.1 percent, relative to white-male workers
in the post-quota period (1920-1940). This suggests that in comparison to black workers, European
immigrant labor at that time had a lower elasticity of substitution to white native workers.6 Our
finding that the quota acts increased the relative economic status of black workers in more affected
areas before World War II also contributes to the debate on the evolution of black-white income
differences throughout U.S. history (e.g., Myrdal, 1944; Smith, 1984; Smith and Welch, 1989; Margo
2016).7 Overall, we find that the relative earnings of black workers increased modestly after the
introduction of the quota system which is consistent with the fact of some, albeit weaker, black-white
income convergence, compared to after 1940 (Margo, 2016).
The last section of the empirical analysis evaluates whether labor productivity in the manufactur-
ing sector changed after the quota system was in place. Our estimates reveal that the quota system led
to a decline in labor productivity in manufacturing at the city level. Between 1919 and 1929, every
6The insight that there are “winners and losers from immigration” relates to a recent paper by Borjas (forthcoming),who reassesses the wage impact of the Mariel boatlift and shows that the increased number of low-skilled immigrantsfrom Cuba lowered the wage of native low-skilled groups.
7For a linked sample of black men, Collins and Wanamaker (2014) show that the Great Migration contributed substan-tially to the decline in the black-white earnings gap before World War II. Other studies primarily focus on the evolution ofblack economic progress during the period 1940–1970 (e.g., Donohue and Heckman, 1991; Malony, 1994; Margo, 1995;Collins, 2000; Boustan, 2009; Carruthers and Wanamaker, 2017).
3
additional missing immigrant in the urban population (per-100-inhabitants-per-year) decreased value
added-per-worker and wages-per-worker by 4.6 percent and 3.5 percent, respectively. The productiv-
ity decline is not driven by changes in the capital intensity or the capital-output ratio of manufactures
in the more quota affected cities, as we find no robust evidence that firms made adjustments along
these margins in response to the implementation of the quota acts.8 Our interpretation is that agglom-
eration externalities, and the degree of substitutability between native and immigrant workers in the
manufacturing sector, could be the driving forces behind this result.
The results of this paper advance the understanding of the economic consequences of immigration
restrictions for the U.S. economy. Clemens et al. (forthcoming) is one of the relatively few studies
that investigates the labor market effects of a change in U.S. immigration policy—the exclusion of
the so-called bracero workers on December 31, 1964. Using a differences-in-differences approach,
Clemens et al. show that despite half a million seasonally employed farm workers from Mexico be-
ing excluded from the labor force, domestic farm laborers did not experience a rise in real wages or
employment. They argue that this null finding is the result of employers adopting less labor-intensive
technologies and changes in crop production. Chen (2015) investigates how the Chinese Exclusion
Act of 1882 affected the occupational standing of Chinese immigrants, and finds that the average oc-
cupational income score of Chinese compared to Japanese immigrants significantly declined after the
Exclusion Act of 1882. Greenwood and Ward (2015), Massey (2016), and Ward (2017) examine how
the U.S. quota laws during the 1920s changed migration behavior. Greenwood and Ward (2015) show
that emigration rates declined significantly after the introduction of immigration quotas, especially
from unskilled occupations and farming, which Ward (2017) argues is driven by a lower rate of un-
planned return migration during the 1920s. While Massey (2016) examines how the enactment of the
Emergency Immigration Act of 1921 affected migrant selection, and finds that the average skill level
of immigrants increased after the Emergency Immigration Act.9 While our paper complements these
studies, our focus is a different one. Because we are interested in the broader economic implications
of immigration restrictions for local economies, we show that the implementation of the quota system
resulted in more affected areas to substantial population and labor productivity declines, and earnings
8Within the agricultural sector, Lew and Cater (2018) show for a restrictive set of counties/census divisions locatedin the Northern Great Plains along the US-Canadian border that the more restrictive U.S. immigration policy during the1920s increased American farmers adoption of tractors compared to Canadian farmers. For a more general discussionabout the historical evolution of capital-skill complementarity in the United States we refer the reader to Goldin and Katz(1998), Atack et al. (2004), Katz and Margo (2014), and Lafortune et al. (2015).
9More information on classical immigration topics such as immigrant selection and assimilation can be found in thesurveys of Borjas (1994), Kerr and Kerr (2011), Abramitzky and Boustan (forthcoming), and Hatton and Ward (2018) forexample.
4
losses for native-born workers relative to less affected areas.10
Our analysis of the 1920s immigration quota acts contributes to the immigration debate in the
United States from a historical perspective (e.g., Borjas, 1994, 1999; Card, 1990, 2005; Cortes,
2008; Saiz, 2003). Many American economic historians have evaluated how immigration affected
the U.S. economy during the 19th and early 20th centuries, although ours is the first study to eval-
uate the broader economic impacts of the quota system for the U.S. economy with a focus on local
economies.11 We provide estimates of well-identified effects of immigration restrictions on popu-
lation, labor productivity, and earnings of native workers from this unprecedented change in U.S.
immigration policy. In particular, our estimation approach exploits only within-county or within-city
variation of the data to study trends in outcomes (related to quota exposure) before and after the imple-
mentation of the quota system, which can be used to evaluate whether the policy change is plausibly
exogenous. Our findings are robust to a number of sensitivity checks; most importantly, we demon-
strate that our results are not driven by shocks to immigration due to World War I and the Literacy
Act of 1917.
This research also complements empirical studies that have investigated the long-term impact of
(historical) immigration flows on local economies in the United States.12 A recent county-level study
by Sequeira et al. (2017) finds positive long-term effects of immigration from Europe to the United
States during the Age of Mass Migration.13 The historical inflow of European migrants still benefits
counties today in terms of productivity gains in the agricultural and manufacturing sector, increased
urbanization, and higher rates of patenting during the Age of Mass Migration.14 Peri (2012) uses
state-level panel data to analyze the long-run effects of immigration on employment and productivity
10Since the first circulation of (an older version) of this paper other researchers have also started to evaluate the impactof the quota system on several socio-economic and political outcomes (Abramitzky and Boustan, 2017; Tabellini 2017;Xie, 2017; Doran and Yoon, 2018). These related studies investigate how the 1920s immigration restrictions affectednatives’ decision to migrate and their investment in human capital (Abramitzky and Boustan, 2017), political outcomesand public spending (Tabellini, 2017), manufacturing wages and black migration (Xie, 2017) and inventions (Doran andYoon, 2018).
11See, for example, Abramitzky et al. (2012, 2013, 2014, 2016); Bandiera et al. (2013); Collins (1997); Dunlevyand Hutchinson (1999); Hatton and Williamson (1998); Ferrie (1999); Greenwood (2008); Kim (2007); Timmer andWilliamson (1998). For further details and references we refer to the surveys on immigration in American history ofAbramitzky and Boustan (forthcoming), Carter and Sutch (1997), and Hatton and Williamson (1998).
12A growing literature in cultural economics investigates the long-term consequences of immigration for the U.S.economy (e.g., Fernandez and Fogli, 2009), in particular showing that the cultural composition of US counties matters foreconomic development (e.g., Ager and Brueckner, 2013, 2018; Burchardi et al., 2016.)
13Likewise, Rodriguez-Pose and von Berlepsch (2014, 2015) also find that historical settlement patterns of migrantsstill matter for local economic development in the United States today.
14Further evidence on immigrants’ positive impact on science and innovation inthe United States is from Akcigit et al.(2017), Moser et al. (2014), and Hunt and Gauthier-Loiselle (2010).
5
for the period 1960-2006. Peri finds no evidence that immigrants crowed out employment, however
he reports significant productivity gains from the net inflows of immigrants for the receiving states.
Our paper adds to the findings of this literature by providing evidence that the immigration restric-
tions established during the 1920s led—in the more affected areas—to a significant decline of labor
productivity in the manufacturing sector, and to sizable earnings losses for native (white) workers.
2 Historical Background
Opposition to immigration has a long history in the United States (Hutchinson, 1981; Higham, 2002).
The rise of the Know-Nothing party—an anti-immigration movement—during the 1850s as a response
to increased Catholic immigration from Germany and Ireland is such an example. The first laws that
restricted immigration to the United States at a larger scale were the Page Act of 1875 and the Chinese
Exclusion Act of 1882. While the Page Act of 1875 forbade indentured labor from “China, Japan,
or any Oriental countries”, immigration of alien convicts, and women for the purpose of prostitution,
the Exclusion Act of 1882 barred immigration from China to the United States in general. European
immigration remained virtually unrestricted until the passage of the Immigration Act of 1917 (also
referred to as the Literacy Act).15 The 1917 act required from immigrants over sixteen years of
age to pass a literacy test by reading 30-40 lines in any language of their choice. It further banned
immigration from the so-called Asiatic Barred Zone, which included most parts of Asia and the Pacific
Islands.16 The official statistics document that the total number of annual immigrants admitted to
the United States dropped from circa 300,000 in the years just preceding the Immigration Act of
1917 act to around 100,000 immediately afterwards. Yet, the 1917 act effectively failed to reduce
immigration from Europe to the United States on a larger scale, because literacy rates in Europe were
rising rapidly during this time period. Already in 1920 annual immigration rose above 400,000 and
exceeded 800,000 in 1921 (U.S. Department of Commerce, 1924, Table 65).
The first immigration policy effectively restricting the number of immigrants from Europe to the
15Various interest groups, such as the American Federation of Labor, were behind the passage of the Immigration Actof 1917. Goldin (1994) argues that some labor unions and native-born rural Americans were against free immigration andnearly succeeded in implementing immigration restrictions in the 1890s. However, what changed from 1890 to 1917 andbecame decisive for the vote in favor of the 1917 act was the position on free immigration among the older immigrationgroups living predominantly in the Midwestern areas, and rural-natives living in the American South, who initially had amore liberal view on immigration. We refer the reader to Hutchinson (1981), Goldin (1994), and Higham (2002) for moredetails on the political economy behind the passage of the Immigration Act of 1917.
16The Asiatic Barred Zone did not include Japan because the Gentlemen’s Agreement of 1907—an informal agreementbetween the governments of the United States and Japan—already limited immigration of Japanese workers to the UnitedStates.
6
United States was the Emergency Quota Act of 1921.17 This law restricted the annual number of
aliens of any nationality to 3 percent of the number of foreign-born persons of such nationality listed
in the U.S. Census of 1910. The implementation of nationality quotas reduced the total number of
immigrants from 805,228 in 1921 to 309,556 in 1922 (U.S. Department of Commerce, 1929, Ta-
ble 100).18 Exempted from the quota system were immigrants born in Canada, Mexico, and South
America (Massey, 2016). The 1921 act asymmetrically affected immigration from different Euro-
pean regions: Immigration from the southeastern part of Europe was now severely restricted, while
immigration from the northwestern part of Europe was still considered desirable (King, 2000). For
example, the number of Italian immigrants was reduced from 222,260 in 1921 to the 1922-quota
number of 42,057, while the number of Swedish immigrants in 1921 was 9,171 and the 1922-quota
number was 20,042 (U.S. Department of Commerce, 1924, Tables 71, 79).
The Immigration Act of 1924 (also known as the Johnson-Reed Act) replaced the Emergency
Quota Act and made the quota system permanent (King, 2000). The 1924 act involved two signif-
icant changes: First, the ceiling was reduced from 3 percent to 2 percent of the foreign-born stock.
Second, the modified formula for the national origins quotas was based on the foreign-born stock of
1890 instead of 1910, which meant that the annual quota number was reduced from 357,083 in 1924
to 164,667 in 1925 (U.S. Department of Commerce, 1929, Table 111). This change almost com-
pletely prevented immigration from Southern and Eastern Europe. For example, the annual quota for
Russia dropped from 24,405 to 2,248 immigrants (U.S. Department of Commerce, 1929, Table 111).
Further, the 1924 act completely banned immigration from Asia. Until July 1, 1927, the nationality
quotas were based on the 1890 Census. After July 1, 1927, the annual quota was fixed to a total of
150,000 immigrants, and the 1924 act was substituted for a national origins plan, which based the
quota allocation by country on the national origins of the white population in the United States in the
1920 Census (King 2000, pp.204-212).19 The national-origins-based quota system went into effect,
with some delay, on July 1, 1929, and remained in place, apart from some minor modifications, until
1965 when it was replaced by the Immigration and Nationality Act.
17There were no serious considerations in the United States Congress to implement quotas or any other blanket restric-tions to restrict immigration from Europe to the United States before 1920 (Goldin, 1994).
18Total immigration was limited to 357,000 per annum (King, 2000, p.200).19The 1927 immigration quotas by country are depicted in King (2000, Table 7.1). On top of limiting immigration from
southeastern European countries, the national origins system shifted the quotas towards immigration from Britain whilereducing quotas from other northwestern European countries, such as Germany, Ireland, or Sweden.
7
3 Data and Estimation Strategy
3.1 Data
Our empirical analysis evaluates how the foreign-born share, population, individual outcomes, such
as fertility and native earnings, and manufacturing activity responded to the introduction of the quota
acts in the 1920s. The U.S. Census collected county-level data on demographic, economic, and so-
cial variables for every decade since 1790, including information on population and the number of
foreign-born. These data are retrieved from the ICPSR 2896 data file for the sample period 1900–
1940 (Haines, 2010).20 The essential ingredients for constructing the quota-exposure measure are the
1920 county/city-level population data and the number of foreign-born by nationality based on the
full-count U.S. Census microdata from IPUMS (Ruggles et al., 2017), data on annual immigration
(1899-1930), and the annual quota numbers by nationality (1922–1930).21 Annual immigration data
are reported in Willcox (1929, Table III, pp.389-393) for the years 1899–1924. This time series is
extended for the years 1925-1930 using the annual immigration data from the Statistical Abstract of
the United States (U.S. Department of Commerce, 1929, Table, 106; 1931, Table 99). Data on the
annual quota number are collected from the Statistical Abstract of the United States (U.S. Department
of Commerce, 1924, Table, 79; 1931 Table 104). We refer to Section 3.2 for further details how the
quota measure is constructed.
The individual data consist of a repeated cross-section of individuals based on the 1-percent na-
tional random IPUMS samples of the population for the Censuses 1900 to 1940. We employ a sample
of 15- to 65-year-old workers to study how the quota acts affected the earnings of natives, and a sample
of women aged 15-49, to study fertility behavior and marriage. One limitation of the individual data
is that there is no information on individual income before 1940. Since individual earnings are not
reported, economic historians rely on so-called “earning scores” based on occupations or industries to
study, e.g, the returns to migration in the United States before 1940 (Abramitzky et al., 2012; Collins
and Wanamaker, 2014). We follow this literature and use an occupation-based earnings score in our
empirical analysis. This measure is based on the annual earnings data reported in Lebergott (1964,
Table A-18), which are available for a broad category of industries over the period 1900-1960.22
20More information about the data set, such as scope of study, data collection, and data source can be found at http://www.icpsr.umich.edu/icpsrweb/ICPSR/studies/02896.
21The immigration data correspond to the fiscal year, which ended on June, 30. For example, the immigration year of1922 refers to immigration inflows between July 1, 1921 and June 30, 1922.
22These broader industry categories are agriculture; manufacturing; mining (incl. subcategories); construction; trans-portation (incl. subcategories); communication and utilities (incl. subcategories); trade; service (incl. subcategories);
8
We assign the annual earnings for a given census year (1900, . . . , 1940) and industry category to
any individual reporting an industry code in IPUMS (“ind1950”), which corresponds to the broader
industry classification.23 Since the Lebergott data refer to annual earnings of full time employees,
we adjust the earnings of workers by race, gender, and region of residence (South/non-South) for
each industry category based on their actual earnings in the 1940 Census as described in Collins and
Wanamaker (2014, Appendix A2). The adjustment factor in 1940, Adfactorirgl;1940, is obtained by
dividing the earnings, Earningsirgl, of race r, gender g, in region l for every industry, i, in 1940 by
the average earnings in the 1940 Census in that industry, Earningsi.24 For every worker in our sam-
ple, we multiply the “Lebergott earnings score” with the 1940 adjustment factor, Adfactorirgl;1940, to
obtain a race, gender, and location specific earnings score by industry. As in Collins and Wanamaker
(2014, p.246), we also modify the agricultural earnings reported in Lebergott to account for differ-
ences between hired farm labor and farm operators’ income.25 The Lebergott earnings score varies
across occupations but, compared to the more frequently used IPUMS occupation score (“occscore”),
it also captures variation within occupations over time.26 Our results are robust to using alternative
occupation-based earnings scores (see the Appendix and Section 4.2 for further details). The micro-
data are then merged with the quota-exposure measure.
For the city-level analysis we digitized data from the Census of Manufactures for the years 1909,
1914, 1919, 1925, and 1929. The manufacturing outcome variables are value added-per-worker and
wages-per-worker to measure labor productivity, and horsepower-per-worker and horsepower-per-
output to proxy for capital intensity and the capital-output ratio, respectively.27 The data constraints
resulted in the creation of a balanced panel of 246 cities with more than 10,000 inhabitants in 1909.
Table 1 reports the summary statistics.
government (incl. subcategories); and finance, insurance, and real estate.23For example, individuals reporting “ind1950” codes 306-499 in IPUMS work in manufacturing and obtain the cor-
responding manufacturing earnings value from Lebergott (1964). We base the assignment on the “ind1950” code fromIPUMS, as it is considered to be comparable across different Census years.
24The average 1940 earnings in industry i are based on 15-65-year-old workers reporting an occupation in that industry.25Farm operators’ net income is retrieved from the Major Statistical Series of the USDA, which contains information
about farm operators’ net income for the years 1910-1956 (U.S. Department of Agriculture 1957; Volume 3, Table 17).There is no information on farm operators’ income for the Census year 1900. We obtain the 1900 value of farm operators’income by multiplying the ratio of farm operators’ income to earnings of hired labor in agriculture in 1910 with theearnings of hired labor in agriculture in 1900 reported in Lebergott (1964).
26The Lebergott earnings score is denoted in constant prices using 1900 as the reference year. We usedhttps://www.measuringworth.com/uscompare/ to convert the Lebergott earnings score into constant prices.
27Note that information on horsepower for the 1925 Census of Manufacturers is not available. We use a linear interpo-lation to obtain the 1925 values.
9
3.2 Estimation Strategy
The variable of interest, Quota exposurec, captures the extent to which a given area (e.g., a county or
city) was affected by the quota system. We construct Quota exposurec as:
Quota exposurec =100
Pc,1920
N∑n=1
max(Mn,22−30 −Qn,22−30, 0
) FBnc,1920
FBn,1920
. (1)
The variable Mn,22−30 denotes the predicted average number of immigrants of nationality n that were
supposed to arrive per year during the post-quota period 1922–1930 if the quota system had not been
in place. This prediction is based on the following regression model: Mnt = β1 ln t + β2 (ln t)2 +
εit,28 where Mnt refers to the actual immigration of nationality n and year t for the period 1900–
1914 reported in Willcox (1929, Table III, pp. 389-393). Since immigration to the Untied States
was significantly interrupted by World War I, the predictions are only based on the pre-WWI annual
immigration flows (see, e.g., the dashed lines in Figure 1).29 We use the estimated coefficients β1
and β2 from the prediction model to obtain the counterfactual immigration flows for nationality n for
the years 1922–1930. The variable Qn denotes the average quota number for nationality n per year
during the period 1922–1930. We interpret the difference between Mn,22−30−Qn,22−30 as the average
number of missing immigrants of nationality n per year due to the quota system. We set the number
of missing immigrants for any nationality n to zero if the expression Mn,22−30 − Qn,22−30 < 0. This
is always the case for countries without legal quotas (Qn,22−30 ≡ ∞), such as Mexico and Canada.
28The regression model allows for potential non-linearities in the immigration flows. Alternative functional forms thatalso capture such non-linearities yield similar results (not reported). We further set the prediction for nationality n for anyyear t to zero if Mnt < 0.
29One complication using the Willcox data is that Table III only reports the number of immigrants from Czechoslovakia,Finland, Poland, and Yugoslavia since 1920 (these countries either obtained their independence (Poland, Finland) orwere newly formed states (Czechoslovakia, Yugoslavia) during or right after World War I). Willcox, however, reportsdetailed tables of sending countries (e.g., from Germany, Austria-Hungary, Russia, and Turkey) that contain migrants fromthese newly established countries before 1920. The number of immigrants from Poland for the years 1899-1919 consistsof Polish immigrants from Austria-Hungary (see Table XXIIIa and XXIIIb, pp.483-484); Germany (see Table XXVII,p.486); and from Russia (see Table XXXII, p.488). The number of Finnish immigrants was included together with Russia(see Table XXXII, p.488); the number of immigrants from Czechoslovakia were included together with Austria-Hungary(see Table XXIIIa and XXIIIb, pp.483-484), and the number of immigrants from Yugoslavia were included together withAustria-Hungary (see Table XXIIIa and XXIIIb, pp.483-484) and Turkey (see Table XXXV, p.489). With this informationat hand we constructed separate time-series for those four countries for the years 1899-1919 and deducted the numbersfrom the sending countries that contained these numbers originally. For Hungary separate immigration numbers existsince 1905. For the years 1899-1904, 75% of Germans, 20% Hebrew, and all Magyar and Romanian immigrants fromAustria-Hungary were assigned to Hungary based on the racial distribution reported in Tables XXIIIa and XXIIIb for1910, the first year that separates immigrants from Austria and Hungary by “principle races”. For 1905–1909, the numberof Polish, Czechoslovakian, and Yugoslavian immigrants were deducted from the totals of Austria and Hungary accordingto the racial distribution of these two countries in 1910.
10
Equation (1) distributes these missing immigrants locally according to the share of immigrants
of nationality n living in county c prior to the quota system in 1920 (i.e., FBnc,1920/FBn,1920).30
We assign the missing immigrants to the nationality shares in 1920 for two reasons: First, the U.S.
government assigned separate quota numbers for the newly established European countries such as
Czechoslovakia, Finland, Poland, and Yugoslavia. The 1920 Census is the first where nationality
shares for those countries can be constructed systematically (see also footnote 29). Second, the 1920
Census is the last before the introduction of the quota system and thus a natural reference point.31
Our assignment rule draws on a well-established fact in the migration literature that newly arriving
immigrants tend to settle in areas where previous immigrants of the same nationality live, and that
settlement patterns of immigrants varied by nationalities across the United States during the Age of
Mass Migration (e.g., Ager and Brueckner, 2013). The basic intuition is that we would expect more
missing immigrants after the introduction of the quota acts in the 1920s in areas with larger pre-
existing immigrant communities of affected nationalities. Summing over all foreign nationalities and
multiplying by the factor 100Pc,1920
provides the total number of missing immigrants in the 1920s per-100-
inhabitants-per-year in area c. Conceptually, Quota exposurec can be interpreted as the annual number
of missing immigrants during the 1920s per-100-inhabitants in a given area c due to the introduction
of the quota system. It captures the regional effects of a fundamental change in national immigration
policy and is, for example, conceptually similar to recent work in the trade literature which analyzes
the effects of trade liberalization on local economies (Kovak, 2013; Dix-Carneiro and Kovak, 2017).
Figure 1 illustrates our procedure of predicting national immigration flows (i.e., Mn,22−30) for
the following four sending countries: Russia, Italy, Ireland, and Sweden. The predicted immigration
flows are the black dashed lines. For example, Panel A shows that the number of predicted Russian
immigrants in 1923 is around 137,000, while under the Immigration Act of 1921, the annual Russian
quota number was close to 30,000 (solid red line), which implies that during 1923 circa 117,000
Russian immigrants were missing because of the quota system. These missing Russian immigrants
are subsequently allocated across counties in the United States according to FBnc,1920/FBn,1920 in
equation (1). Figure 1 also shows that while the quota system resulted in missing Italian immigrants
30We construct the share of immigrants by nationality based on information from the IPUMS variable “BPL” for thefollowing birthplaces outside the United States: U.S. outlying areas/territories including Hawaii, Canada, Mexico, Centraland Latin America, Caribbean, Denmark, Finland, Norway, Sweden, United Kingdom, Belgium, France, Netherlands,Switzerland, Albania, Greece, Italy, Portugal, Spain, Austria, Bulgaria, Czechoslovakia, Germany, Hungary, Poland,Romania, Yugoslavia, Russian Empire, Asia, Africa, and Oceania.
31Similar results are obtained if we use nationality shares and county population in 1910 instead of 1920, albeit theestimated coefficients are, as expected, somewhat smaller (see Appendix A for further details).
11
(Panel B), the predicted immigration flows for Irish and Swedish immigrants were lower than their
quota numbers (Panels C and D), implying no missing immigrant of these nationalities.32
Figure 2 shows the total number of missing immigrants for the post-quota period 1922–1930 by
year. We see that the annual number of missing immigrants under the Immigration Act of 1921 is
approximately 720,000, while the Immigration Act of 1924 elevates this number to about 860,000.
Based on the total US population in 1920, this gives an annual average of 0.8 missing immigrants per-
100-inhabitants-per-year or, alternatively, 8 percent fewer immigrants (over the total U.S. population)
during a decade compared to a counterfactual scenario of free immigration.
Panel A of Figure 3 visualizes Quota exposurec at the county level and reveals that the well-known
immigrant clusters in the Northeast, the Great Lakes region, and the West Coast are generally more
affected by the quota system compared to counties located in the American South and the southern
parts of the Midwest. Since our baseline empirical specifications exploit only county-level variation
within the same state, Panel B of Figure 3 displays Quota exposurec conditional on state fixed effects
and reveals significant variation in quota exposure within states as well.
Appendix A discusses how sensitive our main findings are to modifications of the treatment mea-
sure. At first we compare our estimates to a more crude specification of quota treatment that resembles
what Clemens et al. (forthcoming) used to evaluate the effect of the bracero exclusion. The following
expression Quota Nationalitiesc =∑N
n=1 q1920nc illustrates the construction of this measure which is
then interacted with an indicator for the post-quota period. The fraction of the quota affected popula-
tion in 1920 for a given area c is∑N
n=1 q1920nc ≡ (
∑Nn=1 FBnc,1920× In)/Popc,1920, where the indicator
In is one if nationality n is subject to the 1920s quota acts and zero otherwise. This implies that
any nationality subject to the 1920s quota acts is being treated independently of their actual quota
number. For example, pre-existing Swedish and German communities are assumed to be treated
just like Russian or Italian communities, even if their corresponding quota number would exceed the
predicted inflow of migrants for the years 1922-1930 (i.e., there are no missing migrants from Ger-
many and Sweden). Since Quota exposurec in equation (1) explicitly exploits this additional variation
across quota affected nationalities, we consider it as the preferred treatment measure for the empirical
analysis. Second, it is worthwhile to note that while the baseline quota-exposure variable is based
on predicting immigration by nationality, we obtain qualitatively similar results when replacing the
predicted immigration inflow with the average annual number of immigrants from 1910–1914 by na-
tionality, or following the approach of Greenwood and Ward (2015) which captures the “bite” of the
32The prediction for Swedish migrants is set to zero for the years it would yield to negative values.
12
quota system (see Appendix A for further details).33 Third, we show in Appendix A that our results
are robust to controlling for the direct effect of the most important non-quota affected nationalities
(Canadians and Mexicans).
The quota-based “experiment” we have mind is the following: Firms located in areas A and B
plan to employ X workers in period t based on past immigration flows. In period t immigration to
area A is unexpectedly reduced due to the implementation of the quota system, while immigration
to area B remains unregulated. Now, we can observe how labor market outcomes change in area A
as firms readjust to the new policy relative to the labor market outcomes in area B where firms do
not need to make such readjustments.34 More formally, our econometric model follows a differences-
in-differences (DiD) approach to identify the effects of the quota system. The variable of interest,
Quota exposurec, measures the differential impact of the quota system in terms of the immigration
rate of missing immigrants during the 1920s across counties (or cities).35 More quota-affected areas
are expected to have more missing immigrants per capita relative to areas less affected by the quota
system (compared to the counterfactual of free immigration to the United States).
Our DiD estimation strategy resembles the more classical shift-share instrumental-variable ap-
proach (e.g., Bartik, 1991) in the sense that we rely on local past settlement locations of the affected
immigrant groups and aggregate shocks to migration flows. Since the quota system induced a sudden
shift of immigration inflows at the national level which are plausibly exogenous to local economies,
it increases the general validity of our and the more classical shift-share instrumental-variable ap-
proach (Jaeger et al., 2018). However, one crucial difference between the two approaches is that our
estimation strategy properly isolates the policy-driven variation in the aggregate flow of immigrants
to the United States from the implementation of the quota system. Because the classical shift-share
approach would exploit all immigration shocks (domestic and abroad) that occurred over the period
1900-1940 it is therefore not well-suited to cleanly identify the effects of the quota system which is
the main purpose of this study.36
33Data on the actual number of pre-WWI immigrants admitted over the period 1910–1914 are collected from theStatistical Abstract of the United States (U.S. Department of Commerce, 1931, Tables 99).
34We implicitly assume no spillovers between the two areas. However, in reality, the exclusion of immigrants to areaone could increase the number of immigrants coming to area two. The empirical implication of this issue is that weare only able to estimate the relative effects of the quota system (i.e., the effect between treatment and control areas).In Appendix A we also report results that take such potential spillovers into account in case Mexican and Canadianimmigration patterns changed in response to the implementation of the quota system.
35The concept of “missing immigrant” is just our way of interpreting the cross-sectional treatment measure in theDiD strategy, where the actual time variation comes from comparing outcomes before and after 1920. Conceptually, ourapproach is similar to the DiD strategies applied in Bleakley (2007) or Hornbeck and Naidu (2014).
36See, for example, the discussion on empirical approaches to identify the causal effects of immigration on local
13
For the county-level analysis, we estimate the following event-study model for the period 1900-
1940:
yct =1940∑j=1900
αj Quota exposurec × Ijt + X′ctΓ + λc + µt + φst + εct, (2)
where yct is either the foreign-born share or (log) population in county c in year t, Quota exposurec
is interacted with a full set of time fixed effects (Ijt ), where the omitted time period of comparison is
1920. This implies that α1930 quantifies the effect of the quota system on the outcome in 1930 relative
to 1920 independently from its effect in 1940, for example. In order to take into account potential
mean reversion due to initial level differences, the control vector X′ct includes (log) county population
sizes in 1910 interacted with a full set of time fixed effects, as well as controls for the (potential)
immigration shocks caused by World War I and the Literacy Act of 1917 (see Appendix A for more
details). All of our specifications control for county-fixed effects (λc) and time-fixed effects (µt). The
baseline specification also includes state-by-time-fixed effects (φst), such that the variation used to
identify the effects of the missing migrants is based on the comparison of outcomes between more
and less (or non-) quota affected counties within the same state over time. We cluster the error term,
εct, at the county level.
While our analysis does not provide any instrumental variable estimates, one could think of equa-
tion (2) as a corresponding “first-stage regression” when the foreign-born share is the outcome vari-
able. If Quota exposurec is really capturing variation of missing immigrants, one would expect to see
a break in the relationship between the foreign-born share and Quota exposurec after the implemen-
tation of the quota system. The effect on population is more likely to be muted or amplified by other
demographic changes such as fertility adjustments and in- and out-migration that were set in motion
after the quota system was in place. If the direct effect is not fully offset, one should see a negative
effect of Quota exposurec on population growth after the introduction of the quota system. We also
estimate different variants of equation (2), which are explained in detail in Section 4.
4 Empirical Results
4.1 Population Dynamics
As a first step, we provide some prima facie evidence that more affected areas experienced a (relative)
decline in the overall inflow of immigrants. We derived a retrospective measure on the number of
economies in Lewis and Peri (2015).
14
immigrants at the county-by-year level based on immigrants’ year of arrival and their current place
of residence from the IPUMS full-count census data in 1920 and 1930. This information is then used
to construct a location-based measure of immigration flows aggregated by quota exposure for every
year from 1900 to 1930.37
Figure 4 graphs the (spline of the) annual average inflow of immigrants by Quota exposurec. We
group counties such that if their quota-exposure value is above the sample median and below sample
median, they are assigned to the treatment (solid line) and control group (dashed line), respectively.
The reported average inflow of immigrants in Panel A is conditional on county fixed effects, while
in Panel B we also control for World War I and the Literacy Act of 1917.38 Before World War I,
counties in the so-called “treatment group” received on average more immigrants, but the trend in
the flows is relatively similar to counties in the “control group”.39 After the outbreak of World War
I, both treatment and control counties experienced significant declines in immigrant inflows, but the
average decline seems to be somewhat stronger among counties in the treatment group. There is an
immediate reversion in immigration inflows in the aftermath of World War I close to pre-war levels.
Following the Immigration Act of 1924, the average inflow of immigrants into treatment counties
reduces significantly compared to the control counties. This reveals that the more affected areas, in
fact, experienced a relative decline in immigration after the implementation of the quota acts compared
to the years before the outbreak of World War I. Panel B demonstrates that this conclusion is robust
to controlling for the possible confounding effects of the immigration shocks caused by World War I
and the Immigration Act of 1917. In the empirical analysis we always control for these two additional
immigration shocks.
Figure 5 depicts the event-study estimates of equation (2), controlling for county and state-by-time
fixed effects, as well as the “missing immigrants” due to World War I, a control for the Literacy Act
37The use of the variable YRIMMIG from IPUMS for this purpose is not without problems. One main concern is “yearheaping”, since some immigrants might not exactly remember the date of arrival at the time of the Census enumerationand just use the nearest round number as year of arrival (e.g., 1900, 1910, . . . ). Another concern is that the immigrationflows are downward biased due to death and emigration before the year of enumeration, since we measure year of arrivalretrospectively (i.e., in 1920 and 1930). We have compared the year of arrival measure with the Willcox (1929, Table III)immigration time series, which is based on the admission of immigrant aliens by year and country of last residence for aselected group of nationalities, to check whether measurement issues might confound our findings. While the two timeseries move very much in the same direction, the Willcox immigration flows are larger than the immigration flows derivedfrom IPUMS, which is expected, as the latter measure reflects immigrants who stayed in the United States and did not dieor return to their home country before the Census enumeration (results available upon request).
38Appendix A explains the construction of the controls for World War I and the Literacy Act of 1917 in detail.39There is a steep decline in the average inflow of immigrants from 1900 to the following years, which is not visible in
the splines reported here. However, one could regard this as evidence of “year heaping”. This decline, however, is similarbetween treatment and control counties, suggesting that this type of measurement error is less of a concern in our case.
15
of 1917, and log county population size in 1910 interacted with time fixed effects. Panel A reveals
that the share of foreign-born residents in counties more exposed to the quota system increased prior
to quota system, which is consistent with the evidence in Figure 4. This upward trend in the foreign-
born share is completely reversed after the quota system is in place. For example, a one-unit increase
in Quota exposurec (or equivalently, having one immigrant less per-100-inhabitants-per-year in a
county) translates into a decline in the foreign-born share of 2.9 percentage points by 1930 and 4.3
percentage points by 1940.40 Panel B of Figure 5 shows how the missing immigrants influenced the
county-population growth path during the sample period. Counties more exposed to the quota system
experienced more rapid population growth in the pre-quota period, while there is a clear indication of
a trend break in the post-quota period, suggesting that the quota system reduced population growth.
Table 2 reports the DiD estimates for the foreign-born share (columns 1-3) and population growth
(columns 4-6). The estimation equation is (2) and the method of estimation is least squares. Compared
to the event-study, Quota exposurec is now interacted with an indicator variable which is one in the
post-quota period and zero otherwise. Column 1 controls for county- and time-fixed effects, column
2 adds controls for the immigration shocks coming from World War I and the Literacy Act, and log
county population in 1910 (interacted with time-fixed effects), and column 3 replaces the time-fixed
effects with state-by-time-fixed effects. We regard column 3 as our baseline specification. Columns
4-6 follow the same sequence, but we additionally control for county-specific linear trends, which
give us the effect of the quota system on population growth since the outcome is log population.
The DiD estimates for the foreign-born share are all negative and statistically significant at the
1-percent level, confirming the insights from the event-study (Panel A of Figure 5) that the quota
system reduced the foreign-born share substantially.41 For our baseline specification (column 3), a
one-unit increase in Quota exposurec, (or equivalently having one less immigrant per-100-inhabitants-
per-year in a county), led to a 1.6 percentage points decline in the share of foreign-born residents in
the post-treatment period. The coefficient can be interpreted as a lower-bound estimate, since Panel
A of Figure 5 revealed a positive pre-quota trend in the foreign-born share which would bias the
DiD estimate towards zero. Including county-specific linear trends to the foreign-born specification
of column 3 increases the DiD estimate, αDiD to −0.054 with a standard error of 0.004. Thus, the
county-specific linear trends eliminate pre-quota trends in the foreign-born share and, although we
40One additional missing immigrant per-100-inhabitants-per-year in a county translates into 10 percent fewer immi-grants arriving over each decade from 1920 to 1940. The average quota exposure is 0.25 missing immigrants per-100-inhabitants-per-year and one standard deviation is equal to 0.43 (see Table 1).
41Comparing columns 1-3, it is worth noting that the magnitude becomes smaller as we add further controls.
16
do not include these trends in our baseline specification for the labor market outcomes (in Section
4.2), our findings are robust to adding such controls. County-specific linear trends are included in
columns 4-6 where log population is the outcome variable, which is equivalent to studying how the
quota system influenced population growth. Results show a negative and significant effect of Quota
exposurec on population growth. The coefficients are rather stable and our baseline estimate, reported
in column 6, shows that one additional missing immigrant per-100-inhabitants-per-year lowered the
10-year population growth rate by 6.7 percentage points.
In the following, we outline a simple accounting exercise to evaluate how much of the negative
population growth effect is directly attributable to the missing immigrants. This exercise is based on
the so-called fundamental demographic equation:
Pct+1 = Pct +Bct −Dct + Ict −Xct, (3)
where Pct+1 is population size at time t + 1 in county c, Bct is the number of live births, Dct is
the number of deaths, Ict is the number of immigrants (and in-migrants), and Xct is the number of
emigrants (and out-migrants). Equation (3) can be rearranged such that:
pct+1 = bct − dct + ict − xct, (4)
where pt+1 ≡ (Pct+1 − Pct) /Pct denotes population growth, bct ≡ Bct/Pct the crude birth rate, dct ≡Dct/Pct the crude death rate, ict ≡ Ict/Pct the immigration rate, and xct ≡ Xct/Pct the emigration
rate. We now assume that all variables in equation (4) are functions of the quota system (Qt):
pct+1(Qct) = bct(Qct)− dct(Qct) + ict(Qct)− xct(Qct). (5)
Differentiating this equation with respect to the quota system (Qt) yields the following expression:
∂pct+1
∂Qct
=∂bct∂Qct
− ∂dct∂Qt
+∂ict∂Qt
− ∂xct∂Qt
, (6)
where the arguments have been omitted for simplicity. Given that Quota exposurec in itself is an
immigration rate (of missing immigrants), the estimate in column 6 of Table 2 directly reveals how
much of the negative population growth effect can be directly explained by the missing immigrants
(i.e., ∂ict/∂Qt = α). However, two minor adjustments are needed: First, the immigration rate is
denoted per 100 inhabitants, so we first need to multiply the coefficients by 100. Second, population
growth is measured over a 10-year period, while the immigration rate is measured annually, implying
17
that the coefficients also should be divided by 10. Taking these two adjustments together results in
αDiD = −0.68. According to equation (6), had the estimated coefficient αDiD been equal to unity,
the effect would be fully accounted for by the missing immigrants. Our estimate suggests that about
68% of the negative population growth effect can be accounted for by missing migrants. Taking
uncertainty into account (the 95% confidence interval is [−1.09;−0.027]) reveals that most of the
reduction in population growth is directly attributable to reduced immigration.
Next, we present results that reveal a negative effect of the quota system on fertility (i.e., ∂bct/∂Qt <
0), which also contributes to the overall negative population growth effect. Table 3 provides evidence
that the implementation of the quota system substantially reduced fertility and led to a decline in mar-
riage rates. The estimating equation is (2) and the method of estimation is least squares. Compared
to the data on the foreign-born share and population, marital status and fertility are measured at the
individual level and connected to Quota exposurec via the county of residence of an individual at
the time of the census enumeration. As in Table 2, Quota exposurec is interacted with an indicator
which is one in the post-quota period and zero otherwise. Besides the baseline controls of equation
2 (time-fixed effects, state-by-time-fixed effects, county-fixed effects, 1910 (log) county population
interacted by time, the missing immigrants from World War I, and the exposure to the Literacy Act of
1917), the individual regressions also include controls for the place of residence (rural/urban), race,
sex, and fixed effects for age and birthplace.
The first three columns of Table 3 summarize the effects on fertility for the period 1900–1940.
Fertility is measured by the number of own children under age 5 in the household. The sample
includes 15- to 49-year-old white and black women, such that the fertility variable corresponds to the
general fertility rate measured over a five-year period.42 Column 1 reports a negative and statistically
significant effect of quota exposure on fertility at the 1-percent level.43 The magnitude of the estimated
coefficient implies that for one additional missing immigrant per-100-inhabitants-per-year in a county,
the general fertility rate declines by 0.04. This number can be compared to the aggregate decline in
the general fertility rate from the pre- to the post-quota period of 0.12, suggesting an important role of
the quota system for the fertility transition over the sample period. Columns 2 and 3 of Table 3 show
that the negative effect on fertility is not exclusively driven by women born in the United States (and
both parents born in the United States) or first- and second-generation immigrant women.
One channel consistent with the observed fertility decline is that marriage-market conditions wors-
42This fertility measure has, for example, previously been used by Bleakley and Lange (2009).43Results remain unchanged if we include women of all other races to the sample (not reported).
18
ened after the quota system was in place. The effects on marriage rates presented in column 4 of Table
3 support this hypothesis. The dependent variable is a dummy variable whether a young women (age
15-35) was married at the time of the census enumeration. Column 4 shows that after the implemen-
tation of the quota system there was a statistically significant decline of marriage rates in the more
affected counties. The point estimate implies that having in a county, one less immigrant per-100-
inhabitants-per-year, reduced the likelihood of sampling a married woman by 1.4 percentage points
in the post-treatment period. This finding is consistent with the notion that changing economic con-
ditions in the more affected counties reduced the likelihood of women getting married after the quota
system was in place. Figure 6 depicts event-study estimates for the specifications corresponding to
columns 1 and 4 of Table 3 which confirm the main results of this subsection.44
The remaining two columns present separate results for young women born in the United States
(and both parents born in the United States) and first-and second-generation immigrant women. While
U.S.-born women were less likely to marry in the more quota affected counties, we find no effect
for first- and second- generation immigrants. This result is somewhat surprising as one might have
expected that the quota system would mainly worsen the local marriage-market conditions among
first- and second-generation immigrants (e.g., Angrist, 2002; Lafortune, 2013). On the other hand,
Abramitzky et al. (2016) find that during this time period, 39 percent first-generation immigrants
(arriving to the U.S. unmarried) married outside their own nationality, suggesting that marriage market
conditions for natives also worsened. Consistent with our marriage rates results, we show in the next
subsection that after the implementation of the quota system, economic conditions for native workers
deteriorated in the more affected counties.
4.2 Labor Market Outcomes of Native Workers
Table 4 examines how the quota-policy-driven reduction of immigrant labor affected native earnings
(we refer to Appendix B for a theoretical motivation of the empirical results in this section). We use
the Lebergott earnings score to proxy the economic status of native workers. The dependent variable
is measured at the individual level and relates to Quota exposurec via the county of residence of the
individual. The estimating equation is (2) and the method of estimation is least squares. As in Tables 2
and 3, Quota exposurec is interacted with an indicator taking on the value one in the post-quota period,
and zero otherwise. The specifications of Table 4 include, besides the baseline controls of equation
44Panels A and B reveal no evidence of pre-quota trend differences (indirectly supporting our main identifying assump-tion), while after the introduction of the quota system there are immediate declines in marriage and fertility.
19
(2), controls for marital status, place of residence (rural/urban), race, sex, literacy, and fixed effects
for age and birthplace. The sample spans the period 1900–1940 and is restricted to 15-to 65-year-old
U.S.-born workers reporting an occupation in the Census.45
Table 4 summarizes the results for the Lebergott earnings score. As described in Section 3.1,
this measure varies across broad industry categories and over time. Overall, native workers experi-
enced a substantial decline in their earnings after the implementation of the quota system in more
affected counties, but this effect varies substantially by gender and race. The estimate reported in
column 1 indicates that native workers living in a county with one additional missing immigrants per-
100-inhabitants-per-year experienced a 2.1-percent decline in earnings in the post-treatment period.
Alternatively, a one standard deviation increase in quota exposure, which is equivalent to having 0.78
fewer immigrants per 100 inhabitants arriving per year, translates into a 1.6-percent decrease in native
earnings. In column 2 in order to understand the importance of population sorting as a response to
the shock, we exclude workers born outside the current state of residence. For example, one could
argue that part of the baseline estimate is driven by in-migration of lower-skilled native workers from
less quota-exposed counties in order to substitute for the missing immigrant labor in the more quota-
exposed counties. The result presented in column 2 suggests that interstate migration is not driving
our finding, since the coefficient of interest remains statistically significant at the 1-percent level and
similar in magnitude. Column 3 also shows that interstate migrants in more affected counties experi-
enced earning losses relative to interstate migrants in less affected counties. The estimates reported in
columns 2 and 3 are also relatively similar in magnitude. Still, we need to acknowledge that with the
data at hand we cannot rule out whether native workers relocated within the same state as a response to
the implementation of the quota system. We refer the reader to Appendix C for more results on (black)
in-migration which are consistent with the findings of Collins (1997) that European immigration de-
layed the Great Migration before World War I. The estimates presented in columns 4 and 5 reveal that
the negative effect on native earnings varies substantially by gender. Only male workers experienced
significant earnings losses after the implementation of the quota system in more affected counties,
which seems plausible since European immigration was strongly male biased (Angrist, 2002).
The remaining four columns summarize results by gender and race. For these specifications, we
add a triple-interaction term, Quota exposurec×Ipostt ×blacki, to estimating equation (2), where blacki
is an indicator variable whether an individual i is listed as black in the census. These specifications
45The sample includes only native white and black workers, who contain 99 percent of all native workers during thesample period.
20
further include race-by-time and race-by-county fixed effects to absorb any race-specific trends and
location-specific effects by race that occurred between 1900 and 1940. For native male workers, the
DiD estimates reported in columns 6 and 7 reveal heterogeneous effects by race which are striking.
Both Quota exposurec and the triple-interaction term, turn out to be statistically significant at the 1-
percent level but with opposite signs. The shortage of immigrant labor substantially decreased the
earnings of white males in counties with more missing immigrants per 100 inhabitants, whereas the
DiD estimate for black males is αall + αblack = 0.023, with a p-value of 0.117 (and αall + αblack =
0.044, with a p-value of 0.007 when omitting state-by-time fixed effects), which suggests that for
one additional missing immigrants per-100-inhabitants-per-year, the earnings of black male workers
increased between 2.3 to 4.4 percent in the post-treatment period.
Both earning losses of white workers and the economic gains of black workers contributed to a
narrowing of the black-white earnings gap in the more affected counties over the time period 1900-
1940. This can be seen more directly in column 7, which only exploits the within-county variation
of the data (i.e., compared to column 6 this specification also includes county-by-time fixed effects).
Conceptually, this specification compares the earnings score between black and white workers living
within the same county who are equally exposed to the quota system. The estimated coefficient on
the triple-interaction term, Quota exposurec × Ipostt ×blacki, is positive and statistically significant at
the 1-percent level (note, that the main effect is absorbed by the county-by-time fixed effects). The
point estimate implies that for the average level of quota exposure (i.e., 0.69 missing immigrants per-
100-inhabitants-per-year; see Table 1), the earnings score of black males increased by 3.1 percent
relative to white males over the post-quota period. Since the quota system reduced the number of
immigrants by 8 percent of the total U.S. population each decade (see Figure 2), this would suggest
that the introduction of the quota system increased relative earnings of black male workers by 1.8
to 3.5 percent over the post-treatment period (based on the estimates of column 6). For comparison,
estimates by Smith (1984) show that black-white male income ratios based on occupation status (for
ages 20-64) increased by almost 7 percent from 1900 to 1940, although according to Smith’s estimates
most of the convergence occurred before 1920.46 Our results also resonate with Margo (2016), who
finds some steady black-white income convergence before 1940, based on revised black-white income
46One important difference is that Smith (1984) uses national average race-specific income-occupation weights basedon the 1970 federal Census to evaluate changes in the black-white income ratio for the years 1890 to 1980, while weadjust the Lebergott data for race, location, and gender based on information about the worker’s actual earnings reportedin the 1940 Census (see Section 3.1 for the details). Further limitations of Smith’s study are highlighted in Margo (2016,pp.309-311) and refer, for example, to the problem of using national based weights for our sample period which likely ledSmith to underestimate the relative black-white income convergence before World War II.
21
estimates for the pre-World War II period.
Columns 8 and 9 summarize the results for native female workers by race. In column 8, Quota
exposurec is statistically insignificant and literally zero indicating no earnings losses for white females
in the more quota affected counties, whereas the triple-interaction term turns out to be positive and
statistically significant at the 1-percent level. The point estimate for black females is αall + αblack =
0.036, with a p-value of 0.002 (and αall + αblack = 0.026, with a p-value of 0.001 when omitting
state-by-time-fixed effects), suggesting that for every missing immigrants per-100-inhabitants-per-
year in a county, the earnings of black female workers increased between 2.6 to 3.6 percent in the
post-treatment period (relative to the pre-treatment period).
The black-white earnings gap in the more affected counties also narrowed for female workers,
which can be observed when we only exploit the within-county variation of the data in column 9. The
coefficient of the triple-interaction term is positive and statistically significant at the 1-percent level
implying that for the average level of quota exposure, the earnings score of black females increased
by 1.9 percent relative to white females over the post-quota period. Finally, the event-study estimates
of Figure 7, which correspond to columns 1 and 7 of Table 4, confirm the presented results. There
are no significant pre-quota trends, while the sign of DiD estimates materializes after 1920, when the
quota system was in place. We further obtain similar insights when using the occupational income
score or the 1940 earnings score as an alternative outcome variable (we refer to Appendix D for more
details).
The results of Table 4 suggest that the quota system pushed the average native white worker
into lower paid occupations, suggesting some degree of complementarity between native white and
immigrant labor. Black workers, on the other hand, were to a large extent substitutes for (unskilled)
immigrant workers and filled some of the vacant jobs due to the reduced supply of immigrant labor
in the more quota-affected counties, which is consistent with Collins (1997). We also find that in
the more affected counties young white adults (age 15-21), compared to their black peers, were more
likely to stay longer in school, suggesting that labor market conditions for black workers in these
counties improved (in relative terms) after the quota system was implemented. The point estimate for
young white adults is αall = .0156, with a p-value of 0.04 while the effect for their black peers is
αall + αblack = −.0187, with a p-value of 0.055.
The effect on native earnings presented in Table 4 is also consistent with the pattern that we
observe when looking at some specific occupations of native workers after the quota system was in
place. Using the 1920 full count census, we can identify the most common unskilled and skilled
22
occupations where immigrants of the most affected nationalities (above average of missing migrants)
worked before the quota system was introduced.47 The dependent variable is an indicator variable
that is one, if a native worker reports to work in what we classified in 1920 as the most common
unskilled or skilled occupations of immigrants from the most affected nationalities. The specifications
presented in Table 5 include the same set of controls as in column 1 of Table 4, columns 3-4 of Table
5 further include race-by-time and race-by-county fixed effects. Our estimates reveal that after the
implementation of the quota system, native workers in the more affected counties were more likely to
work in jobs that we classified as the most common low skilled occupations of the missing immigrants
of the most affected nationalities. For black workers who were at the bottom of the conditional
earnings distribution in the pre-quota period (1900-1920), this meant some improvements of their
economic status suggesting that the fundamental change in U.S. immigration policy contributed to a
black-white income convergence in the more affected counties over the 1900-1940 period.48
Appendix Table 7 provides further evidence on the degree of complementarity between native and
immigrant workers. While missing migrants from new sending countries (such as Italy or Russia)
reduced the earnings of native workers, they benefited from missing migrants from old sending coun-
tries (such as Great Britain). This result indicates that immigrants from old sending countries with
occupational skills similar to native workers were much closer substitutes in the production process.
Even across new sending countries, our estimates reveal substantial variation in the degree of com-
plementary between native and immigrant workers. For example, Finns, presumably because of their
higher literacy levels, were much closer substitutes to native workers than Russian immigrants.
4.3 Estimating the Effect of the Quota System on the Manufacturing Sector
In this subsection, we evaluate the economic effects of the quota system on labor productivity and
capital intensity in the manufacturing sector at the city level. The dependent variables are (log) value
added-per-worker and (log) wages-per-worker in order to measure labor productivity, and horsepower-
per-worker and horsepower-per-output to proxy for capital intensity and the capital-output ratio, re-
47The most common low skilled occupations of this foreign-born group were mine operatives and laborers (7.97%), op-erative and kindred workers (13.12%), and laborers (20.76%). The most common high skilled occupations were managers,officials, and proprietors (8.61%) and salesmen and sales clerks (3.51%). The numbers in parenthesis refer to the occu-pation distribution of nationalities with above average missing migrants and are based on the 1920 full count census fromIPUMS. Occupations refer to IPUMS code OCC1950 and are listed at https://usa.ipums.org/usa-action/variables/OCC1950#codes_section.
48The conditional earnings distribution in the pre-quota period shows that white native workers were at the top, immi-grants in the middle, and black workers at the bottom of the earnings distribution (available upon request).
23
spectively. Quota exposurec is now calculated at the city level as outlined in Section 3.2. We have
a balanced sample of 246 cities for the years 1909, 1914, 1919, 1925, and 1929, where the average
population size in 1910 was around 113,000. The city-level regressions control for city-fixed effects,
state-by-time-fixed effects, (log) city population size interacted with time-fixed effects, and the im-
migration shocks related to WWI and the Literacy Act of 1917. The method of estimation is least
squares. Standard errors account for arbitrary heteroskedasticity and are clustered at the city level.
Figure 8 displays the event-study estimates for the four outcomes and Table 6 presents the cor-
responding DiD estimates. Panels A and B of Figure 8 reveal that labor productivity was increasing
during the pre-quota period in cities more exposed to the quota system, which indicates that the DiD
estimates reported in Table 6 are possibly biased towards zero (those could be interpreted as lower
bounds of the actual productivity losses). The point estimates displayed in Panel A imply that a one
mean increase in Quota exposurec (equivalently to 1.4 more missing immigrants per-100-inhabitants-
per-year) decreased labor productivity in 1925 by 11.1 percent and in 1929 by 10.4 percent.49 Panel B
reveals productivity losses of 3.5 percent by 1925 and 4.6 percent by 1929, for a one additional miss-
ing immigrant per-100-inhabitants-per-year. Overall, our estimates suggest that the implementation
of the quota system amounted in more affected cities to substantial productivity losses in the manu-
facturing sector. Panels C and D of Figure 8 summarize the results on whether the implementation of
the quota system led firms to adjust their capital intensity or the capital-output ratio. We use machine
horsepower-per-worker as proxy for capital intensity and machine horsepower-per-output as a proxy
for the capital-output ratio. In contrast to the productivity results, the event studies of Panels C and
D show no clear pre-quota trend which would bias the DiD estimates towards zero. The estimates
indicate that, if anything, manufacturing establishments in the more affected cities decreased capital
intensity after the quota system was in place, but these results are not very precisely estimated. There
is also no effect on the capital-output ratio. Overall, our findings indicate that manufacturers in more
affected cities did not adjust their capital intensity, nor the capital-output ratio relative to less affected
cities as response to the quota system.
49The average quota exposure in the city sample is significantly higher than in the county sample (see also Table 1).This is not unexpected as immigrants tend to settle in more populous areas, and the average city-population size (in 1910)was around 110,000 inhabitants, whereas the average county-population size (in 1910) was close to 30,000 inhabitants.
24
5 Conclusion
The 1920s were marked by a fundamental change in U.S. immigration policy. The passage of the
Emergency Quota Act of 1921 and the Immigration Act of 1924 ended the era of unrestricted immi-
gration from Europe to the United States. While economic historians have investigated the political
economy of immigration restrictions in the United States (Goldin, 1994; Timmer and Williamson,
1996; 1998) and explored how the quota acts affected migrant selection and return migration (Green-
wood and Ward, 2015; Massey 2016), a rigorous quantitative assessment of the broader economic
impacts of these immigration restrictions for the U.S. economy focusing on local economies has been
missing so far. This paper aims to fill this gap in the literature. The passage of the quota acts in the
1920s implied that some nationalities—mainly from Europe—were to a different extent affected by
these laws, while for other nationalities, such as Canadians or Mexicans, immigration to the United
States still remained open without any notable restrictions. Our empirical analysis exploited this
policy-driven variation along with the spatial distribution of different pre-existing nationality net-
works across the United States to evaluate how key drivers of economic growth, such as population
and labor productivity, responded to the missing immigrants due to the implementation of the quota
system.
We found that the implementation of the quota system had a negative effect on population growth.
More affected counties experienced a long-lasting decline in population growth relative to less af-
fected counties. We demonstrated that this decline is mainly due to the asymmetric effect of reduced
immigration inflows from quota-affected nationalities across counties. Women also reduced fertility
in the more affected counties, which reinforced the negative effect the quota system had on population
growth. In the more affected cities, the manufacturing sector experienced a substantial decline in la-
bor productivity growth. Since immigrants predominantly settled in urban areas and thereby increased
the density of economic activity (e.g., Ciccone and Hall, 1996), the decline in labor productivity in
the more affected cities might be a result of agglomeration externalities due to the shrinking of the
manufacturing sector, or because firms did not adjust their capital intensity after the introduction of
the quota system.
The shutdown of large-scale immigration from Europe to the United States during the 1920s also
significantly reduced the earnings of an average native worker. This overall negative effect of the
missing immigrants on native earnings turned out to differ substantially by race and gender. After the
implementation of the quota system, white male workers experienced sizable losses in earnings in the
25
more affected counties. This finding could indicate that those workers, and immigrant labor, were to
some extent complements in the production process at that time. On the other hand, black workers
(males and females), who were closer substitutes to (unskilled) immigrant labor, benefited from the
immigration restrictions and increased their relative economic status in the more affected counties
during the post-quota period. Our finding suggests that the quota system, which had a differential im-
pact on the supply of immigrant labor across counties, might have triggered some black-white income
convergence before the 1940s. Overall, we conclude that the fundamental change in immigration
policy during the 1920s generated winners and losers, but probably not the ones the policymakers at
that time had hoped for. Clearly, whether native workers benefited or lost from this policy change
depended on how much they stood in competition with immigrant workers.
ReferencesAbramitzky, Ran and Leah Platt Boustan, “Adaptation of Native Labor and Capital to Mass Mi-
gration: Evidence from the Immigration Act of 1924,” 2017. Unpublished.
and , “Immigration in American History,” Journal of Economic Literature, forthcoming.
, , and Katherine Eriksson, “Europe’s tired, poor, huddled masses: Self-selection and economicoutcomes in the age of mass migration,” The American Economic Review, 2012, 102 (5), 1832–1856.
, , and , “Have the poor always been less likely to migrate? Evidence from inheritance prac-tices during the Age of Mass Migration,” Journal of Development Economics, 2013, 102, 2–14.
, , and , “A nation of immigrants: Assimilation and economic outcomes in the age of massmigration,” Journal of Political Economy, 2014, 122 (3), 467–506.
, , and , “Cultural assimilation during the age of mass migration,” 2016. NBER Working Paper22381.
Ager, Philipp and Markus Brueckner, “Cultural diversity and economic growth: Evidence from theUS during the age of mass migration,” European Economic Review, 2013, 64, 76–97.
and , “Immigrants’ Genes: Genetic Diversity and Economic Development in the US,” EconomicInquiry, 2018, 56 (2), 1149–1164.
Akcigit, Ufuk, John Grigsby, and Tom Nicholas, “Immigration and the Rise of American Ingenu-ity,” 2017. NBER Working Paper 23137.
Angrist, Josh, “How do sex ratios affect marriage and labor markets? Evidence from America’ssecond generation,” The Quarterly Journal of Economics, 2002, 117 (3), 997–1038.
26
Atack, Jeremy, Fred Bateman, and Robert A. Margo, “Skill intensity and rising wage dispersionin nineteenth-century American manufacturing,” The Journal of Economic History, 2004, 64 (1),172–192.
Bandiera, Oriana, Imran Rasul, and Martina Viarengo, “The making of modern America: Migra-tory flows in the age of mass migration,” Journal of Development Economics, 2013, 102, 23–47.
Bartel, Ann P., “Where do the new US immigrants live?,” Journal of Labor Economics, 1989, 7 (4),371–391.
Bartik, Timothy J., The debate over state and local economic development policies, WE UpjohnInstitute for Employment Research, 1991.
Bleakley, Hoyt, “Disease and development: evidence from hookworm eradication in the AmericanSouth,” The quarterly journal of economics, 2007, 122 (1), 73–117.
and Fabian Lange, “Chronic disease burden and the interaction of education, fertility, andgrowth,” The Review of Economics and Statistics, 2009, 91 (1), 52–65.
Borjas, George J., “The Economics of Immigration,” Journal of Economic Literature, 1994, 32 (4),1667–1717.
, Heaven’s door: Immigration Policy and the American Economy, Princeton, NJ: Princeton Univer-sity Press, 1999.
, “The labor demand curve is downward sloping: Reexamining the impact of immigration on thelabor market,” The Quarterly Journal of Economics, 2003, 118 (4), 1335–1374.
, “Native internal migration and the labor market impact of immigration,” Journal of Human Re-sources, 2006, 41 (2), 221–258.
, Immigration economics, Harvard University Press, 2014.
, “The wage impact of the Marielitos: A reappraisal,” ILR Review, forthcoming.
Boustan, Leah Platt, “Competition in the promised land: Black migration and racial wage conver-gence in the North, 1940–1970,” The Journal of Economic History, 2009, 69 (3), 755–782.
Burchardi, Konrad, Thomas Chaney, and Tarek A. Hassan, “Migrants, Ancestors, and ForeignInvestments,” 2016. NBER Working Paper 21847.
Cadena, Brian C. and Brian K. Kovak, “Immigrants equilibrate local labor markets: evidence fromthe Great Recession,” American Economic Journal: Applied Economics, 2016, 8 (1), 257–290.
Card, David, “The impact of the Mariel boatlift on the Miami labor market,” ILR Review, 1990, 43(2), 245–257.
, “Immigrant inflows, native outflows, and the local labor market impacts of higher immigration,”Journal of Labor Economics, 2001, 19 (1), 22–64.
27
, “Is the new immigration really so bad?,” The Economic Journal, 2005, 115 (507), 300–323.
, “Immigration and Inequality,” American Economic Review, 2009, 99 (2), 1–21.
Carruthers, Celeste K. and Marianne H. Wanamaker, “Separate and unequal in the labor market:human capital and the jim crow wage gap,” Journal of Labor Economics, 2017, 35 (3), 655–696.
Carter, Susan B. and Richard C. Sutch, Historical perspectives on the economic consequences ofimmigration into the United States, National Bureau of Economic Research Cambridge, Mass.,USA, 1997.
Chen, Joyce J., “The Impact of Skill-Based Immigration Restrictions: The Chinese Exclusion Act of1882,” Journal of Human Capital, 2015, 9 (3), 298–328.
Ciccone, Antonio and Robert Hall, “Productivity and the Density of Economic Activity,” AmericanEconomic Review, 1996, 86 (1), 54–70.
Clemens, Michael A., Ethan G. Lewis, and Hannah M. Postel, “Immigration Restrictions as ActiveLabor Market Policy: Evidence from the Mexican Bracero Exclusion,” forthcoming.
Collins, William J., “When the tide turned: Immigration and the delay of the great black migration,”The Journal of Economic History, 1997, 57 (3), 607–632.
, “African-American economic mobility in the 1940s: a portrait from the Palmer Survey,” TheJournal of Economic History, 2000, 60 (3), 756–781.
and Marianne H. Wanamaker, “Selection and economic gains in the great migration of AfricanAmericans: new evidence from linked census data,” American Economic Journal: Applied Eco-nomics, 2014, 6 (1), 220–252.
Cortes, Patricia, “The effect of low-skilled immigration on US prices: evidence from CPI data,”Journal of Political Economy, 2008, 116 (3), 381–422.
Dix-Carneiro, Rafael and Brian K. Kovak, “Trade liberalization and regional dynamics,” AmericanEconomic Review, 2017, 107 (10), 2908–46.
Donohue III, John J. and James Heckman, “Continuous versus Episodic Change: The Impact ofCivil Rights Policy on the Economic Status of Blacks,” Journal of Economic Literature, 1991, 29(4), 1603–43.
Doran, Kirk and Chungeun Yoon, “Immigration and Invention: Evidence from the Quota Acts,”in “The Role of Immigrants and Foreign Students in Science, Innovation, and Entrepreneurship,”University of Chicago Press, 2018.
Dunlevy, James A. and William K. Hutchinson, “The impact of immigration on American importtrade in the late nineteenth and early twentieth centuries,” The Journal of Economic History, 1999,59 (4), 1043–1062.
28
Dustmann, Christian, Uta Schonberg, and Jan Stuhler, “Labor supply shocks, native wages, andthe adjustment of local employment,” The Quarterly Journal of Economics, 2017, 132 (1), 435–483.
Fernandez, Raquel and Alessandra Fogli, “Culture: An empirical investigation of beliefs, work,and fertility,” American Economic Journal: Macroeconomics, 2009, 1 (1), 146–177.
Ferrie, Joseph P., Yankeys now: Immigrants in the antebellum US 1840-1860, Oxford UniversityPress, 1999.
Foged, Mette and Giovanni Peri, “Immigrants’ effect on native workers: New analysis on longitu-dinal data,” American Economic Journal: Applied Economics, 2016, 8 (2), 1–34.
Goldin, Claudia, “The political economy of immigration restriction in the United States, 1890 to1921,” in Claudia Goldin and Gary Leibcap, eds., The regulated economy: A historical approachto political economy, University of Chicago Press, 1994, pp. 223–258.
and Lawrence F. Katz, “The origins of technology-skill complementarity,” The Quarterly Journalof Economics, 1998, 113 (3), 693–732.
Greenwood, Michael J., “Family and sex-specific US immigration from Europe, 1870–1910: Apanel data study of rates and composition,” Explorations in Economic History, 2008, 45 (4), 356–382.
and Zachary Ward, “Immigration quotas, World War I, and emigrant flows from the United Statesin the early 20th century,” Explorations in Economic History, 2015, 55, 76–96.
Haines, Michael R., “ICPSR. Historical, Demographic, Economic, and Social Data: The UnitedStates, 1790-2002,” 2010.
Hatton, Timothy J. and Jeffrey G. Williamson, The age of mass migration: Causes and economicimpact, Oxford University Press, 1998.
and Zachary Ward, “International Migration in the Atlantic Economy 1850-1940,” 2018. CEHDiscussion Papers 02, Centre for Economic History, Research School of Economics, AustralianNational University.
Higham, John, Strangers in the land: Patterns of American nativism, 1860-1925, Rutgers UniversityPress, 2002.
Hornbeck, Richard and Suresh Naidu, “When the levee breaks: black migration and economicdevelopment in the American South,” American Economic Review, 2014, 104 (3), 963–90.
Hunt, Jennifer and Marjolaine Gauthier-Loiselle, “How much does immigration boost innova-tion?,” American Economic Journal: Macroeconomics, 2010, 2 (2), 31–56.
Hutchinson, Edward P., Legislative history of American immigration policy 1798-1965, Universityof Pennsylvania Press, Philadelphia, United States, 1981.
29
Jaeger, David A, Joakim Ruist, and Jan Stuhler, “Shift-share instruments and the impact of immi-gration,” 2018. NBER Working Paper 24285.
Katz, Lawrence F. and Robert A. Margo, “Technical change and the relative demand for skilledlabor: The united states in historical perspective,” in “Human capital in history: The Americanrecord,” University of Chicago Press, 2014, pp. 15–57.
Kerr, Sari P. and William R. Kerr, “Economic Impacts of Immigration: A Survey,” Finnish Eco-nomic Papers, 2011, 24 (1), 1–32.
Kim, Sukkoo, “Immigration, industrial revolution and urban growth in the United States, 1820-1920:Factor endowments, technology and geography,” 2007. NBER Working Paper 12900.
King, Desmond, Making Americans: Immigration, race, and the origins of the diverse democracy,Harvard University Press, 2000.
Kovak, Brian K., “Regional effects of trade reform: What is the correct measure of liberalization?,”American Economic Review, 2013, 103 (5), 1960–76.
Lafortune, Jeanne, “Making yourself attractive: Pre-marital investments and the returns to educationin the marriage market,” American Economic Journal: Applied Economics, 2013, 5 (2), 151–178.
, Jose Tessada, and Ethan G. Lewis, “People and Machines: A Look at the Evolving RelationshipBetween Capital and Skill In Manufacturing 1860-1930 Using Immigration Shocks,” 2015. NBERWorking Paper 21435.
Lebergott, Stanley, Manpower in economic growth: The American record since 1800, New York,McGraw-Hill, 1964.
Lee, Jong-Wha and Hanol Lee, “Human capital in the long run,” Journal of Development Eco-nomics, 2016, 122, 147–169.
Lew, Byron and Bruce Cater, “Farm mechanization on an otherwise featureless plain: tractors on theNorthern Great Plains and immigration policy of the 1920s,” Cliometrica, 2018, 12 (2), 181–218.
Lewis, Ethan G. and Giovanni Peri, “Immigration and the Economy of Cities and Regions,” Hand-book of Regional and Urban Economics, 2015, 5, 625–685.
Maloney, Thomas N., “Wage compression and wage inequality between black and white males inthe United States, 1940–1960,” The Journal of Economic History, 1994, 54 (2), 358–381.
Margo, Robert A., “Explaining black-white wage convergence, 1940–1950,” ILR Review, 1995, 48(3), 470–481.
, “Obama, Katrina, and the Persistence of Racial Inequality,” The Journal of Economic History,2016, 76 (2), 301–341.
Massey, Catherine G., “Immigration quotas and immigrant selection,” Explorations in EconomicHistory, 2016, 60, 21–40.
30
Moser, Petra, Alessandra Voena, and Fabian Waldinger, “German Jewish emigres and US inven-tion,” The American Economic Review, 2014, 104 (10), 3222–3255.
Munshi, Kaivan, “Networks in the modern economy: Mexican migrants in the US labor market,”The Quarterly Journal of Economics, 2003, 118 (2), 549–599.
Myrdal, Gunnar, An American dilemma, Volume 2: The Negro problem and modern democracy,Vol. 2, Transaction Publishers, 1944.
Peri, Giovanni, “The effect of immigration on productivity: Evidence from US states,” Review ofEconomics and Statistics, 2012, 94 (1), 348–358.
Rodrıguez-Pose, Andres and Viola Von Berlepsch, “When migrants rule: the legacy of mass mi-gration on economic development in the United States,” Annals of the Association of AmericanGeographers, 2014, 104 (3), 628–651.
and Viola von Berlepsch, “European Migration, National Origin and Long-term Economic De-velopment in the United States,” Economic Geography, 2015, 91 (4), 393–424.
Ruggles, Steven, Katie Genadek, Ronald Goeken, Josiah Grover, and Matthew Sobek, “Inte-grated Public Use Microdata Series: Version 7.0 [dataset],” 2017.
Saiz, Albert, “Room in the kitchen for the melting pot: Immigration and rental prices,” Review ofEconomics and Statistics, 2003, 85 (3), 502–521.
Sequeira, Sandra, Nathan Nunn, and Nancy Qian, “Migrants and the Making of America: TheShort-and Long-Run Effects of Immigration during the Age of Mass Migration,” 2017. NBERWorking Paper 23289.
Smith, James P., “Race and human capital,” The American Economic Review, 1984, 74 (4), 685–698.
and Finis R. Welch, “Black economic progress after Myrdal,” Journal of Economic Literature,1989, 27 (2), 519–564.
Tabellini, Marco, “Gifts of Immigrants, Woes of Natives: Lessons from the Age of Mass Migration,”2017. Unpublished.
Timmer, Ashley S. and Jeffrey G. Williamson, “Racism, xenophobia or markets? The politicaleconomy of immigration policy prior to the Thirties,” 1996. NBER Working Paper 5867.
and , “Immigration policy prior to the 1930s: Labor markets, policy interactions, and globaliza-tion backlash,” Population and Development Review, 1998, 24 (4), 739–771.
U.S. Department of Agriculture, Major Statistical Series of the U.S. Department of Agriculture,Washington, DC: Government Printing, 1957.
U.S. Department of Commerce, Statistical Abstract of the United States, United States GovernmentPrinting Office, 1924.
, Statistical Abstract of the United States, United States Government Printing Office, 1929.
31
, Statistical Abstract of the United States, United States Government Printing Office, 1931.
U.S. Immigration Commission, Reports of the Immigration Commission: Statistical Review of Im-migration, 1820-1910; Distribution of Immigrants, 1850-1900, U.S. Government Printing Office,1911.
Ward, Zachary, “Birds of passage: Return migration, self-selection and immigration quotas,” Ex-plorations in Economic History, 2017, 64, 37–52.
Willcox, Walter F., International Migrations Volume I: Statistics, New York, National Bureau ofEconomic Research, 1929.
Xie, Bin, “The Effect of Immigration Quotas on Wages, the Great Black Migration, and IndustrialDevelopment,” 2017. Unpublished.
32
Figure 1: Missing immigrants by nationality
Notes: This figure shows actual immigration, predicted immigration, and the quota numbers from the
countries: Russia, Italy, Ireland, and Sweden. We refer the reader to Section 3.2 for further details.
Figure 2: Total missing quota immigrants
Notes: This figure shows the (calculated) total number of missing immigrants due to the quota system by
year. We refer the reader to Section 3.2 for further details.
Figure 3: Quota exposure
Notes: This figure shows baseline quota exposure (at the county level). Panel A without controls, and Panel
B with state fixed effects. We refer the reader to Section 3.2 for further details.
Figure 4: Total inflow of immigrants by median quota exposure
Notes: This figure shows (the splines of) the county average of the log number of immigrants by median
quota exposure. The solid/dashed line shows the (log) average immigrants for counties above/below median
quota exposure. We refer the reader to Section 4.1 for further details.
Figure 5: Event-study estimates for foreign-born share and log population
Notes: This figure reports event-study estimates for the foreign-born share (Panel A) and log population
(Panel B). The omitted year of comparison is 1920. We refer the reader to Section 4.1 and Table 2 for further
details
Figure 6: Event-study estimates for marriage and fertility
Notes: This figure reports event-study estimates for marriage (Panel A) and fertility (Panel B). The omitted
year of comparison is 1920. We refer the reader to Section 4.1 and Table 3 for further details.
Figure 7: Event-study estimates for log Lebergott Earnings Score
Notes: This figure reports event-study estimates for the (log) Lebergott Earnings Score. The omitted year
of comparison is 1920. We refer the reader to Section 4.2 and Table 4 for further details.
Figure 8: Event-study estimates for the manufacturing sector (city level)
This figure reports event-study estimates for the manufacturing sector at the city level. The omitted year of
comparison is 1920. We refer the reader to Section 4.3 and Table 6 for further details.
(1) (2) (3) (4) (5)N mean sd min max
County levelShare of foreign-born 14,060 0.071 0.092 0 0.655Ln population 14,060 9.800 1.043 1.386 15.220Quota exposure 14,060 0.249 0.435 0 5.193
Individual levelNumber of children below age 5 1,368,322 0.368 0.708 0 7== 1 if married 937,527 0.530 0.499 0 1Ln Lebergott earnings score 1,569,620 6.004 0.655 3.548 7.058==1 if works in most common low skilled occ for immigrants 1,612,045 0.176 0.381 0 1==1 if works in most common high skilled occ for immigrants 1,612,045 0.102 0.303 0 1==1 if works in most common agricultural occ for immigrants 1,612,045 0.150 0.357 0 1
City levelln value added per worker 1,258 7.698 0.529 5.123 12.46ln wages per worker 1,260 6.798 0.441 3.891 11.38ln horsepower per worker 1,260 0.990 0.543 -1.547 5.549ln horsepower per output 1,260 -7.527 0.542 -10.62 -4.997Quota exposure 1,260 1.412 1.170 0.0730 6.924
Table 1: Summary statistics
Notes: See Section 3.1 for further details. Quota exposure is defined in equation (1) of Section 3.2.
(1) (2) (3) (4) (5) (6)
Quota Exposure x Ipost -0.0578*** -0.0288*** -0.0161*** -0.0738*** -0.0876*** -0.0681***(0.00371) (0.00304) (0.00259) (0.0144) (0.0173) (0.0210)
ln Population size 1910 x time FE No Yes Yes No Yes YesWWI x Ipost No Yes Yes No Yes YesLiteracy Act x Ipost No Yes Yes No Yes YesCounty FE Yes Yes Yes Yes Yes YesTime FE Yes Yes Yes Yes Yes YesState-by-time FE No No Yes No No YesCounty-specific linear trend No No No Yes Yes Yes
Observations 14,060 14,060 14,060 14,060 14,060 14,060R-squared 0.893 0.921 0.959 1.000 1.000 1.000
Table 2: Population dynamics -- The effect of the immigration quotas on the share of foreign-born and population
Dependent Variable
Share of foreign-born ln(population)
Notes: The table reports DiD estimates. The observations are at the county level for the period 1900 to 1940. The outcome variable is the share offoreign-born in columns (1)-(3) and ln(population) in columns (4)-(6). Quota exposure is defined in equation (1) of Section 3.2. All specificationsinclude county and time fixed effects. Additional controls are ln population size in 1910 interacted with a full set of time fixed effects, WWI, and theLiteracy Act of 1917 in columns (2)-(3) and (5)-(6), a county-specific linear trend in columns (4)-(6), and state-by-time fixed effects in columns (3) and(6). Appendix A details the construction of the controls for WWI and the Literacy Act of 1917. The total sample includes all counties for which thereexist data for all the years. Constants are not reported. Standard errors (in parentheses) account for arbitrary heteroskedasticity and are clustered atthe county level. *** p<0.01, ** p<0.05, * p<0.1.
(1) (2) (3) (4) (5) (6)
Quota Exposure x Ipost -0.0406*** -0.0196*** -0.0223*** -0.0139*** -0.0124*** -0.00365(0.00656) (0.00583) (0.00822) (0.00337) (0.00476) (0.00567)
ln Population size 1910 x time FE Yes Yes Yes Yes Yes YesWWI x Ipost Yes Yes Yes Yes Yes YesLiteracy Act x Ipost Yes Yes Yes Yes Yes YesIndividual controls Yes Yes Yes Yes Yes YesCounty FE Yes Yes Yes Yes Yes YesTime FE Yes Yes Yes Yes Yes YesState-by-time FE Yes Yes Yes Yes Yes Yes
Observations 1,368,322 967,321 318,232 937,527 677,442 200,338R-squared 0.153 0.151 0.184 0.337 0.333 0.366
Table 3: Population dynamics -- The effect of the immigration quotas on fertility and marriage
Dependent Variable
Number of children below age 5 ==1 if married
Notes: The table reports DiD estimates. The observations are at the individual level for the period 1900 to 1940. The sample spans 15-49-year-oldwomen in columns (1)-(3) when the number of children below age 5 is the outcome variable and 15-35-year-old women in columns (4)-(6) when theoutcome variable is a dummy whether being married. Quota exposure is defined in equation (1) of Section 3.2. In columns (2) and (5) the sample isrestricted to women born in the United States. (and both parents born in the United States), while in columns (3) and (6) the sample includes only first-and second-generation immigrant women. All regressions include county fixed effects, time fixed effects, state-by-time fixed effects, ln population sizein 1910 interacted with a full set of time fixed effects, and controls for WWI and the Literacy Act of 1917 (Appendix A details the construction of thecontrols for WWI and the Literacy Act of 1917). The set of individual controls includes a dummy for race (black/white), place of residence(rural/urban), age fixed effects, and birthplace fixed effects. Constants are not reported. Standard errors (in parentheses) account for arbitraryheteroskedasticity and are clustered at the county level. *** p<0.01, ** p<0.05, * p<0.1.
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Quota Exposure x Ipost -0.0208*** -0.0198*** -0.0205*** -0.0215*** 0.000299 -0.0243*** -0.00480(0.00502) (0.00531) (0.00588) (0.00598) (0.00481) (0.00601) (0.00507)
Quota Exposure x Ipost x black 0.0479*** 0.0453*** 0.0408*** 0.0275**(0.0119) (0.0103) (0.0140) (0.0125)
ln Population size 1910 x time FE Yes Yes Yes Yes Yes Yes Yes Yes YesWWI x Ipost Yes Yes Yes Yes Yes Yes Yes Yes YesLiteracy Act x Ipost Yes Yes Yes Yes Yes Yes Yes Yes YesIndividual controls Yes Yes Yes Yes Yes Yes Yes Yes YesCounty FE Yes Yes Yes Yes Yes Yes Yes Yes YesTime FE Yes Yes Yes Yes Yes Yes Yes Yes YesState-by-time FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
Sample AllLives in state
of birthLives outside state of birth
Men Women Men Men Women Women
Observations 1,569,620 1,114,087 455,514 1,205,563 364,037 1,205,335 1,205,975 363,812 362,945R-squared 0.642 0.659 0.583 0.572 0.754 0.578 0.591 0.765 0.783
Table 4: Immigration quotas and their impact on the earnings of native workers
Dependent Variable
ln(Lebergott Earnings Score)
Notes: The table reports DiD estimates. The observations are at the individual level for the period 1900 to 1940. The sample spans 15-65-year old native born workers. The outcomevariable is the ln Lebergott earnings score (see Section 3.1 for further details). Quota exposure is defined in equation (1) of Section 3.2. All regressions include county fixed effects,time fixed effects, state-by-time fixed effects, ln population size in 1910 interacted with a full set of time fixed effects, and controls for WWI and the Literacy Act of 1917 (AppendixA details the construction of the controls for WWI and the Literacy Act of 1917). The set of individual controls includes the following indicator variables: marital status, place ofresidence (rural/urban), gender, race (black/white), literacy. We further include birthplace and age fixed effects. Columns (6)-(9) further include quota exposure, WWI, and theLiteracy Act of 1917 each interacted with a dummy for race (black), race-by-time fixed effects, and race-by county fixed effects. County-by-time fixed effects are added to thespecifications in column (7) and (9). Constants are not reported. Standard errors (in parentheses) account for arbitrary heteroskedasticity and are clustered at the county level. ***p<0.01, ** p<0.05, * p<0.1.
==1 if works in most common low skilled occ for immigrants
==1 if works in most common high skilled occ for immigrants
==1 if works in most common low skilled occ for immigrants
==1 if works in most common high skilled occ for immigrants
(1) (2) (3) (4)
Quota Exposure x Ipost 0.0138*** -0.00410** 0.0136*** -0.00373**(0.00354) (0.00159) (0.00351) (0.00158)
Quota Exposure x Ipost x black 0.0113** 0.00326**(0.00561) (0.00153)
ln Population size 1910 x time FE Yes Yes Yes YesWWI x Ipost Yes Yes Yes YesLiteracy Act x Ipost Yes Yes Yes YesIndividual controls Yes Yes Yes YesCounty FE Yes Yes Yes YesTime FE Yes Yes Yes YesState-by-time FE Yes Yes Yes Yes
Observations 1,612,045 1,612,045 1,611,835 1,611,835R-squared 0.074 0.048 0.082 0.050
Table 5: Probablility of natives working in the most common immigrant occupationsDependent Variable
Notes: The table reports DiD estimates. The observations are at the individual level for the period 1900 to 1940. The sample spans15-65-year old native born workers. The outcome variable is an indicator variable which is one if a native worker reports to workin what we classified in 1920 as the most common unskilled (column 1) or skilled (column 2) occupations of immigrants from themost affected nationalities (see footnote 47 for further details). Quota exposure is defined in equation (1) of Section 3.2. Allregressions include county fixed effects, time fixed effects, state-by-time fixed effects, ln population size in 1910 interacted with afull set of time fixed effects, and controls for WWI and the Literacy Act of 1917 (Appendix A details the construction of the controlsfor WWI and the Literacy Act of 1917). The set of individual controls includes the following indicator variables: marital status, placeof residence (rural/urban), gender, race (black/white), literacy. We further include birthplace and age fixed effects. Columns (3)-(4) further include quota exposure, WWI, and the Literacy Act of 1917 each interacted with a dummy for race (black), race-by-timefixed effects, and race-by county fixed effects. Constants are not reported. Standard errors (in parentheses) account for arbitraryheteroskedasticity and are clustered at the county level. *** p<0.01, ** p<0.05, * p<0.1.
ln(va per worker) ln(wages per worker) ln(hp per worker) ln(hp per value added)
(1) (2) (3) (4)
Quota Exposure x Ipost -0.0452* -0.0127 -0.0315 -0.00857(0.0258) (0.00861) (0.0231) (0.0365)
ln Population size 1910 x time FE Yes Yes Yes YesWWI x Ipost Yes Yes Yes YesLiteracy Act x Ipost Yes Yes Yes YesCity fixed effects Yes Yes Yes YesTime fixed effects Yes Yes Yes YesState-by-time fixed effects Yes Yes Yes Yes
Observations 1,228 1,230 1,230 1,230R-squared 0.850 0.931 0.858 0.826
Table 6: Immigration quotas and their impact on the manufacturing sector
Dependent Variable
Notes: The table reports DiD estimates. The observations are at the city level for the years 1909, 1914, 1919, 1925, and 1929.The outcome variable is ln value added per worker (column 1), ln wages per worker (column 2), horsepower per worker(columns 3) and horsepower per value added (columns 4). Quota exposure is defined as in equation (1) of Section 3.2, butcalculated at the city level. All regressions include, city fixed effects, time fixed effects, state-by-time fixed effects, and lnpopulation in 1910 interacted with a full set of time fixed effects. Linear interpolation was used to construct horsepower atthe city level for the year 1925. Constants are not reported. Standard errors (in parentheses) account for arbitraryheteroskedasticity and are clustered at the county (city) level. *** p<0.01, ** p<0.05, * p<0.1.
For Online Publication
Appendix
A. The Construction of Alternative Quota Exposure Measures and Controls for
World War I and the Literacy Act of 1917
Appendix A describes the construction of the World War I and the Literacy Act of 1917 controls and
explores how robust our results are to alternative assumptions when constructing the quota-exposure
variable. The control for the World War I immigration shock is constructed in the same spirit as Quota
exposurec:
WWIc =100
Pc,1910
N∑n=1
(Mn,15−19 −WWIn,15−19
) FBnc,1910
FBn,1910
, (7)
where Mn,15−19 is the predicted average annual number of immigrants of nationality n, supposed to
arrive during WWI (1915–1919).50 The predictions are based on the same method as outlined in
Section 3.2. WWIn,15−19 is the average number of immigrants of nationality n per year that actually
arrived to the United States during the war years. Analogously to the construction of the missing
immigrants from the quota system, Mn,15−19 − WWIn,15−19 denotes the missing immigrants from
World War I. These missing immigrants are then distributed according to the nationality share in 1910,
FBnc,1910/FBn,1910 following the same logic as outlined in Section 3.2. The total number of missing
immigrants in each area (county/city) is then scaled by the area population in 1910, 100/Pc,1910. For
the empirical analysis, we interact this cross-sectional variable with an indicator variable which is one
after 1910 and zero otherwise. While nationality shares and population are measured in 1920 when
constructing Quota exposurec, they are measured in 1910 in equation (7). This is motivated by the
timing of the two events. However, as we will show in this Appendix, similar results are obtained
when using nationality shares and population in 1910 to construct Quota exposurec instead.
With the data at hand it is not possible to construct the Literacy Act control in the spirit of
“missing-immigrants”. Instead, this measure takes on the following form:
Literacy Actc =N∑n=1
Literacyn,1915FBnc,1910
Pc,1910, (8)
50Since the immigration year ends on June, 30, the year 1915 refers to immigration from July 1, 1914 to June 30, 1915.See also footnote 20 in the main text.
44
where Literacyn,1915 is the share of people with no schooling in the age group 15-64, measured in
1915 in the home country of nationality n (Lee and Lee, 2016), and FBnc,1910/Pc,1910 is the foreign-
born share of nationality n in county c in 1910. The basic idea of this measure is that the share of
people with no schooling in the sending country is supposed to capture the share of people failing the
literacy test from this nationality when immigrating to the United States. For the empirical analysis,
we interact Literacy Actc with an indicator which equals one after 1910 and zero otherwise.
Next, we report the robustness of our main findings to including controls for the 1920 county pop-
ulation share of the two most important non-quota affect nationalities Canada and Mexico (interacted
with a set of time dummies) and using alternative methods to construct the quota-exposure variable.
The first alternative measure is already described in section 3.2 in detail and follows the spirit of the
treatment bracero exposure variable in Clemens et al. (forthcoming). This alternative measure is
constructed as:
Quota Nationalitiesc =N∑n=1
q1920nc , (9)
which is then interacted with a post-quota indicator. The fraction of the quota affected population in
1920 for a given area c is∑N
n=1 q1920nc ≡ (
∑Nn=1 FBnc,1920 × In)/Popc,1920, where indicator In is one
if nationality n is subject to the 1920s quota acts and zero otherwise.
The second alternative is constructed as:
Quota exposureA2c =
100
Pc,1920
N∑n=1
(Mn,22−30 −Qn,22−30
) FBfnc,1920 + FBs
nc,1920
FBfn,1920 + FBs
n,1920
, (10)
where FBfnc,1920 (FBs
nc,1920) is the number of first (second) generation immigrants of nationality n
living in area c in 1920, and FBfn,1920 (FBs
n,1920) is the number of first (second) generation immigrants
of nationality n living in the United States in 1920. The remaining variables are defined as in equation
(1). Compared to Quota exposurec, this measure exploits the extended local nationality network
consisting of both first- and second generation immigrants.
The third alternative replaces 1920 nationality shares and population by the ones reported in 1910:
Quota exposureA3c =
100
Pc,1910
N∑n=1
(Mn,22−30 −Qn,22−30
) FBnc,1910
FBn,1910
, (11)
where the remaining variables are define as outlined in equation (1).
45
The forth alternative is constructed as:
Quota exposureA4c =
100
Pc,1920
N∑n=1
(IF n,10−14 −Qn,22−30
) FBnc,1920
FBn,1920
, (12)
where the predicted immigrant inflow (Mn,22−30) for the period 1922-1930 is replaced by the average
immigration inflow from 1910 to 1914, IF n,1910−1914.
The final alternative follows Greenwood and Ward (2015) and is constructed as:
Quota exposureA5c =
N∑n=1
(Mn,22−30 −Qn,22−30
Mn,22−30
)FBnc,1920
Pc,1920, (13)
where 0 ≤(Mn,22−30 −Qn,22−30
)/Mn,22−30 < 1 captures the bite of the quota for nationality n.
Appendix Figure 1 shows the relationship between our baseline quota-exposure variable and the
alternatives quota exposure measures A2-A5, which are all highly correlated. The results presented in
Appendix Tables 1-4 show that overall qualitatively similar results are obtained for the foreign-born
share, population growth, marriage, fertility, the Lebergott earnings score, and the manufacturing out-
comes, when controlling for the direct effect of Mexican and Canadian immigration, using Quota Na-
tionalitiesc (except for the effect on population growth which is negative but statistically insignificant),
or Quota exposureA2c −Quota exposureA5
c . Column 1 of all tables reported in this Appendix replicates
our baseline finding for convenience and column 2 reports results where the county-population shares
of Mexicans and Canadians in 1920 interacted with a full set of time dummies are added to the
baseline estimating equation (results for the alternative quota exposure measures A1-A5 remain un-
affected when adding the county-population shares of Mexicans and Canadians in 1920 interacted by
time (available upon request)).
B. Theoretical Motivation
Appendix B presents a simple theoretical model, which is used to clarify how the introduction of
the quota system could have influenced native wages. It also motivates the earnings score estimation
equation presented in Section 4.2.
Our model assumes that the total production in a county c and year t is described by a Cobb-
Douglas production function:
Yct = ActKαctL
1−αct , with 0 < α < 1 and Kc0 given, (14)
46
where Act is total factor productivity, Kct is the total capital stock, and Lct is a CES aggregate of two
types of labor:
Lct =(η(Lhct)σ−1
σ + (1− η)(Llct)σ−1
σ
) σσ−1
, 0 < η < 1, σ > 0 and σ 6= 1. (15)
One can think of these two types of labor inputs as high skilled (Lhct) and low skilled (Llct) labor, and
σ denotes the elasticity of substitution between these two factors. For simplicity, we assume that all
high skilled workers (Lhct) are natives and all low skilled workers (Llct) are immigrant workers (i.e.,
the stock of foreign-born workers).
The next assumptions are that goods and factor markets are perfectly competitive and that no
migration of workers between counties takes place (i.e., local labor markets). In the short-run, real
wages are then given by:
whct = (1− α) ηAct
(Kct
Lct
)αL1−σ−1
σct
(Lhct)σ−1
σ−1, (16)
wlct = (1− α) (1− η)Act
(Kct
Lct
)αL1−σ−1
σct
(Llct)σ−1
σ−1. (17)
It can be shown that when the capital stock is fixed in the short run, native wages will respond to a
marginal increase in the stock of foreign-born workers in the following way:
∂whct∂Llct
= (1− α) (1− η)ηAct
(1− α− σ − 1
σ
)(Kct
Lct
)α×
L−σ−1
σct
(Lhct)σ−1
σ−1L
σσ−1−1
ct (1− η)(Llct)σ−1σ−1. (18)
One can easily verify that the effect on the native wage is ambiguous. For example, if 1 − α < σ−1σ
the effect is negative. For the case where foreign-born and native workers are perfect substitutes
(i.e., σ → ∞), the native wage rate decreases as the number of foreign-born increases because of
diminishing returns to labor. While if foreign-born and native workers are perfect complements (i.e.,
σ → 0), the effect turns out to be positive. In other words, the negative effect coming from diminishing
returns to labor can be more than offset by the complementarity between the two labor types, which
demonstrates one of our main points in the paper.
Next, we motivate the earnings score estimation equation used in Section 4.2. To this end, we
assume that capital is perfectly mobile across counties, which implies that through import/export of
47
capital, the county-level capital intensity (kt ≡ Kct/Lct), adjusts in each period to accommodate:51
k = (Act)1
1−α
(α
rt
) 11−α
, (19)
where rt is the national real interest rate, which is allowed to vary over time. Taking equation (16) in
logs and substituting condition (19) in for capital k, we obtain the following expression:
lnwhct = ln (1− α) η +1
1− αlnAct +
α
1− αlnα
rt+(
1− σ − 1
σ
)ln Lct +
(σ − 1
σ− 1
)lnLhct.
(20)
Next, we denote (20) in first-differences:
lnwhct+1 − lnwhct =1
1− α([lnAct+1 − lnAct] + α [ln rt − ln rt+1]) +
(1− (σ − 1) /σ)([
ln Lct+1 − ln Lct
]−[lnLhct+1 − lnLhct+1
]), (21)
and using ∆ lnxct ≡ lnxct+1 − lnxct and that the CES aggregate for the two labor types can be
rewritten as Lct =(η + (1− η)
(Llct/L
hct
)σ−1σ
) σσ−1
Lhct:
∆ lnwhct =1
1− α∆ lnAct +
α
1− α∆ ln rt + (1− σ
σ − 1)∆ lnFBct, (22)
where FBct ≡(η + (1− η)
(Llct/L
hct
)σ−1σ
) σσ−1
. We note that in the case that native and immigrant
workers are perfect substitutes, immigration has no (direct) effect on wages, as capital flows balance
out the negative effect coming from diminishing returns. Next, we see that ∆ lnFBct is driven by
changes in the number of foreign-born relative to native workers and, therefore, a function of the
quota system (i.e., missing flow of immigrants per capita), such that ∆ lnFBct = F (Quota expo-
surec × Ipostt ). Let’s assume for simplicity that ∆ lnFBct = Quotaexposurec × Ipostt and insert this
relationship into equation (22). This yields the following expression:
∆ lnwhct =1
1− α∆ lnAct + µt + βQuota exposurec × I
postt , (23)
where µt ≡ α1−α∆ ln rt is a time fixed effect and β ≡ (1− σ
σ−1). Since the growth rate of total factor
51One could alternatively assume no capital mobility at all and end up with similar predictions. This would, however,change the transitions dynamics. Therefore, to avoid this complication and to illustrate our point, we simply assumeperfect capital mobility across counties. Notice, we also assume that labor markets are completely local (i.e., no in andout-migration), which is also a theoretical simplification. All we need in the empirical model is that labor is not perfectlymobile.
48
productivity (∆ lnAct) could be influenced by the quota system as well, our framework generally does
not allow us to identify the elasticity of substitution (via estimating β). Instead, we end up estimating
the reduced-form effect which, according to equation (23), works either through the degree of com-
plementarity of native and immigrant labor and/or productivity. Estimation equation (23) suggests
that changes in (log) wages should be regressed on Quota exposurec × Ipostt , which captures the flow
of (missing) immigrants. However, the empirical specification used in this paper regresses the level
of (log) wages on Quota exposurec × Ipostt instead. Nevertheless, controlling for county fixed effects
makes these two approaches similar in spirit.
The main reason for this particular choice is that the earnings-score specifications use repeated
cross-sectional data at the individual level, which makes it impossible to construct ∆ lnwhct at the
individual level. In addition, if one believes that ∆ lnAct is influenced by county-specific factors
which are time-invariant and correlated with Quota exposurec × Ipostt , this equation motivates why it
could be important to control for county fixed effects, even if the outcome is in (log) changes. In our
case, this is then equivalent to adding county-specific linear trends to the estimating equation. Adding
county-specific linear trends to the estimating equation reported in column 1 of Table 4 returns an
estimate equal to −0.015 with a p-value of 0.035, which indicates that the findings on native earnings
are generally robust to this extension.
C. The Effect of the Quota System on In-migration
Appendix C summarizes how the quota system influenced (black) in-migration. Appendix Table
5 shows how the black population and the black-population share (at the county-by-time level) re-
sponded to the implementation of the quota system. The estimation equation is (2), where (for sim-
plicity) Quota exposurec is interacted with an indicator which is one in the post-quota period and zero
otherwise. For comparison, column 1 reports the DiD estimate for county population growth where
the sample has been restricted to counties that report a black population larger then zero. As in Table
2, the estimate is negative and statistically significant. For the same sample of counties, column 2
shows a positive but statistically insignificant estimate for black population growth, suggesting that
more quota-exposed counties did not experience any significant increases in black population due to,
for example, in-migration. The estimates reported in columns 1 and 2 of Appendix Table 5 would sug-
gest that the (log) black-population share increased more in counties with more missing immigrants,
which is confirmed in column 3. Importantly, this increase is not due to black in-migration into
49
more quota-exposed areas, but rather driven by a general population decline (i.e., the denominator) as
outlined in section 4.1.
Appendix Figure 2 documents whether the implementation of the quota system triggered inter-
state migration based on event-study estimates (DiD estimates available upon request). The outcome
variable is an indicator variable for whether a 15- to 65-year-old native worker lives outside the state
of birth. The estimating equation is (2), but also including a set of individual level controls (see notes
to Appendix Figure 2 for details). Panel A shows no evidence of interstate migration being system-
atically related to quota exposure. This is also the case for Panel B, where the triple-interaction term
Quota exposurec × Ipostt ×blacki is added to the estimating equation in order to capture differences in
internal population movements by race.
Overall, the evidence in Appendix C indicates that the quota system was not an important trig-
ger of internal (interstate) population movements. However, there are two important caveats to this
non-finding: First, all specifications control for immigration shocks related to World War I and the
Literacy Act of 1917, and consistent with the hypothesis that European immigration delayed the Great
Migration (Collins, 1997), the World War I shock variable is positively related to the black-county
population share and interstate migration (not reported). Second, the indicator-variable used for Ap-
pendix Figure 2 does not capture intrastate migration. In fact, exploiting the linked sample of Collins
and Wanamaker (2014) for the period 1910-1930 reveals that black migrants were more likely to opt
for places exposed to the quota system. However, this relationship is not robust to the WWI migration-
shock variable, which generally supports our findings reported in this section (not reported).
D. Alternative Occupation-based Earnings Scores
We use two alternative occupation-based earnings scores to check the robustness of the results pre-
sented in Table 4 of Section 4.2. The first measure is the so-called occupational income score (“occ-
score”) from IPUMS.52 The occupational income score captures variation in individual’s earnings if
individuals change their occupations after the quota system was in place. It is important to note that
the occupational income score only captures variation across occupations, but not within occupations
over time. We also need to assume that relative incomes of different occupations were constant over
time.52The “occscore” is based on a 1956 special report from the Census on occupational characteristics. It reports median
incomes by occupations, which reflect the relative economic standing of occupations in 1950. The IPUMS then assignedto every individual in the Census reporting an occupation (“occ1950”) the respective value of their occupation.
50
The second measure follows the spirit of the IPUMS occupation score, but we use the information
about wages and salaries of workers from the IPUMS 1940 full count sample instead. Compared to the
IPUMS occupation score, our measure has the advantage that we can construct an earnings score that
differs by gender, race, census region (Northeast, Midwest, South, and West), and occupation.53 The
“1940 earnings score” captures variation in individuals’ wage income that arose from the implemen-
tation of the quota acts in the 1920s if individuals changed their occupation or kept their occupation
but moved to a different region. However, compared to the Lebergott earnings score this measure is
time invariant and the ranking of occupations is based on the terminal year of the sample. A further
drawback is that the 1940 Census did not report any business and farm income. To overcome this is-
sue in part, we assign farm operators (occ1950 codes 100 and 123) their net income in 1940 from the
Lebergott earnings score adjusted by race, gender, and location (South/Non-South); however, we do
not have such income information on other businesses than farming. As for the occupational earnings
score, we further need to assume that the relative standing of occupations was constant over time.
Because of these limitations, we consider the Lebergott earnings score to be our preferred measure.
The results of Appendix Table 6 confirm the findings of Table 4 when using the occupational in-
come score (Panel A) and the 1940 earnings score (Panel B) as alternative outcome variable (although
the estimates for female workers by race are not statistically significant for the 1940 earnings score).
53We use the following broad occupation categories from IPUMS variable “occ1950”: professional, technical; farmers;managers, officials, and proprietors; clerical and kindred; sales workers; craftsmen; operatives; service workers (privatehouseholds); service workers (not household); farm laborers; and laborers.
51
Appendix Figure 1: Quota exposure
Notes: This figure shows the relationship between baseline quota exposure and alternative measures of quota
exposure. We refer the reader to Appendix A for further details.
Appendix Figure 2: Event-study estimates for out-of-state migrant status
Notes: This figure reports event-study estimates for the dummy variable which indicates whether the sampled
individual is born out of state. The omitted year of comparison is 1920. We refer the reader to Appendix C
for further details.
(1) (2) (3) (4) (5) (6) (7)
Panel A:
Quota Exposure x Ipost -0.0161*** -0.0148*** -0.0148*** -0.0155*** -0.0159*** -0.240***(0.00259) (0.00250) (0.00202) (0.00163) (0.00213) (0.0286)
Quota Nationalites x Ipost -0.290***(0.0174)
ln Population size 1910 x time FE Yes Yes Yes Yes Yes Yes YesShare Mex/Can 1920 x time FE No Yes No No No No NoWWI x Ipost Yes Yes Yes Yes Yes Yes YesLiteracy Act x Ipost Yes Yes Yes Yes Yes Yes YesCounty FE Yes Yes Yes Yes Yes Yes YesTime FE Yes Yes Yes Yes Yes Yes YesState-by-time FE Yes Yes Yes Yes Yes Yes Yes
Quota measure Baseline Baseline Alternative 1 Alternative 2 Alternative 3 Alternative 4 Alternative 5
Observations 14,060 14,060 14,060 14,060 14,060 14,060 14,060R-squared 0.959 0.976 0.961 0.959 0.959 0.959 0.959
Panel B:
Quota Exposure x Ipost -0.0681*** -0.0727*** -0.0599*** -0.0399*** -0.0608*** -0.848***(0.0210) (0.0210) (0.0194) (0.0135) (0.0184) (0.255)
Quota Nationalites x Ipost -0.0191(0.167)
ln Population size 1910 x time FE Yes Yes Yes Yes Yes Yes YesShare Mex/Can 1920 x time FE No Yes No No No No NoWWI x Ipost Yes Yes Yes Yes Yes Yes YesLiteracy Act x Ipost Yes Yes Yes Yes Yes Yes YesCounty FE Yes Yes Yes Yes Yes Yes YesTime FE Yes Yes Yes Yes Yes Yes YesState-by-time FE Yes Yes Yes Yes Yes Yes YesCounty-specific linear trend Yes Yes Yes Yes Yes Yes YesMQuota measure Baseline Baseline Alternative 1 Alternative 2 Alternative 3 Alternative 4 Alternative 5
Observations 14,060 14,060 14,060 14,060 14,060 14,060 14,060R-squared 1.000 1.000 1.000 1.000 1.000 1.000 1.000
Appendix Table 1: Population dynamics -- alternative measures of quota exposure
Dependent Variable
Share of foreign born
ln(population)
Notes: The table reports DiD estimates. The observations are at the county level for the period 1900 to 1940. The outcome variable is the share of foreign born in PanelA and ln(population) in Panel B. Column (1) shows the baseline estimates of Table 2 for comparison. The set of controls included are identical to Table 2 column (3) forPanel A and Table 2 column (6) for Panel B, except for column (2), which additionally includes the 1920 county-population shares of Mexicans and Canadians interactedby time. Quota exposure in column (1) is constructed according to equation (1) of Section 3.2. Column (3) replaces quota exposure with the share of quota affectednationalities ("Quota Nationalities") as described in equation (9) of Appendix A. Columns (4)-(7) present results for different versions of quota exposure as described inequations (10)-(13) of Appendix A. Constants are not reported. Standard errors (in parentheses) account for arbitrary heteroskedasticity and are clustered at the countylevel. *** p<0.01, ** p<0.05, * p<0.1.
(1) (2) (3) (4) (5) (6) (7)
Panel A:
Quota Exposure x Ipost -0.0406*** -0.0376*** -0.0472*** -0.0254*** -0.0377*** -0.643***(0.00656) (0.00635) (0.00586) (0.00366) (0.00600) (0.0771)
Quota Nationalites x Ipost -0.414***(0.0492)
ln Population size 1910 x time FE Yes Yes Yes Yes Yes Yes YesShare Mex/Can 1920 x time FE No Yes No No No No NoWWI x Ipost Yes Yes Yes Yes Yes Yes YesLiteracy Act x Ipost Yes Yes Yes Yes Yes Yes YesIndividual Controls Yes Yes Yes Yes Yes Yes YesCounty FE Yes Yes Yes Yes Yes Yes YesTime FE Yes Yes Yes Yes Yes Yes YesState-by-time FE Yes Yes Yes Yes Yes Yes Yes
Quota measure Baseline Baseline Alternative 1 Alternative 2 Alternative 3 Alternative 4 Alternative 5
Observations 1,368,322 1,361,012 1,368,322 1,368,322 1,368,488 1,368,322 1,368,322R-squared 0.153 0.153 0.153 0.153 0.153 0.153 0.153
Panel B:
Quota Exposure x Ipost -0.0139*** -0.0137*** -0.0154*** -0.00913*** -0.0125*** -0.199***(0.00337) (0.00346) (0.00333) (0.00227) (0.00287) (0.0438)
Quota Nationalites x Ipost -0.100***(0.0305)
ln Population size 1910 x time FE Yes Yes Yes Yes Yes Yes YesShare Mex/Can 1920 x time FE No Yes No No No No NoWWI x Ipost Yes Yes Yes Yes Yes Yes YesLiteracy Act x Ipost Yes Yes Yes Yes Yes Yes YesCounty FE Yes Yes Yes Yes Yes Yes YesTime FE Yes Yes Yes Yes Yes Yes YesState-by-time FE Yes Yes Yes Yes Yes Yes YesCounty-specific linear trend Yes Yes Yes Yes Yes Yes Yes
Quota measure Baseline Baseline Alternative 1 Alternative 2 Alternative 3 Alternative 4 Alternative 5
Observations 937,527 932,700 937,527 937,527 937,633 937,527 937,527R-squared 0.337 0.338 0.337 0.337 0.337 0.337 0.337
Appendix Table 2: Population dynamics -- alternative measures of quota exposure
Dependent Variable
Number of children below age 5
==1 if married
Notes: The table reports DiD estimates. The observations are at the individual level for the period 1900 to 1940. The sample spans 15-49-year-old women in Panel Awhen the number of children below age 5 is the outcome variable and 15-35-year-old women in Panel B when the outcome variable is a dummy for being married.Column (1) shows the baseline estimates of Table 3 for comparison. The set of controls included are identical to Table 3 column (1) for Panel A and Table 3 column (4)for Panel B, except for column (2), which additionally includes the 1920 county-population shares of Mexicans and Canadians interacted by time. Quota exposure incolumn (1) is constructed according to equation (1) of Section 3.2. Column (2) replaces quota exposure with the share of quota affected nationalities ("QuotaNationalities") as described in equation (9) of Appendix A. Columns (3)-(6) present results for different versions of quota exposure as described in equations (10)-(13) ofAppendix A. Constants are not reported. Standard errors (in parentheses) account for arbitrary heteroskedasticity and are clustered at the county level. *** p<0.01, **p<0.05, * p<0.1.
(1) (2) (3) (4) (5) (6) (7)
Quota Exposure x Ipost -0.0208*** -0.0182*** -0.0222*** -0.00786*** -0.0229*** -0.344***(0.00502) (0.00494) (0.00504) (0.00300) (0.00424) (0.0645)
Quota Nationalites x Ipost -0.173***(0.0450)
ln Population size 1910 x time FE Yes Yes Yes Yes Yes Yes YesShare Mex/Can 1920 x time FE No Yes No No No No NoWWI x Ipost Yes Yes Yes Yes Yes Yes YesLiteracy Act x Ipost Yes Yes Yes Yes Yes Yes YesIndividual Controls Yes Yes Yes Yes Yes Yes YesCounty FE Yes Yes Yes Yes Yes Yes YesTime FE Yes Yes Yes Yes Yes Yes YesState-by-time FE Yes Yes Yes Yes Yes Yes Yes
Quota measure Baseline Baseline Alternative 1 Alternative 2 Alternative 3 Alternative 4 Alternative 5
Observations 1,569,620 1,560,009 1,569,620 1,569,620 1,569,913 1,569,620 1,569,620R-squared 0.642 0.643 0.642 0.642 0.642 0.642 0.642
Appendix Table 3: Earnings of native workers -- alternative measures of quota exposure
Dependent Variable
ln(Lebergott Earnings Score)
Notes: The table reports DiD estimates. The observations are at the individual level for the period 1900 to 1940. The sample spans 15-65-year old native bornworkers.The outcome variable is the ln Lebergott earnings score (see Section3.1 for further details). Column (1) shows the baseline estimates of Table 4 for comparison.The set of controls included are identical to Table 4 column (1), except for column (2), which additionally includes the 1920 county-population shares of Mexicans andCanadians interacted by time. Quota exposure in column (1) is constructed according to equation (1) of Section 3.2. Column (2) replaces quota exposure with the shareof quota affected nationalities ("Quota Nationalities") as described in equation (9) of Appendix A. Columns (3)-(6) present results for different versions of quotaexposure as described in equations (10)-(13) of Appendix A. Constants are not reported. Standard errors (in parentheses) account for arbitrary heteroskedasticity andare clustered at the county level. *** p<0.01, ** p<0.05, * p<0.1.
(1) (2) (3) (4) (5) (6)
Panel A:
Quota Exposure x Ipost -0.0452* -0.0584** -0.0173* -0.0428* -0.995**(0.0258) (0.0264) (0.00989) (0.0228) (0.495)
Quota Nationalites x Ipost -0.504(0.368)
Quota measure Baseline Baseline Alternative 1 Alternative 2 Alternative 3 Alternative 4
Observations 1,228 1,228 1,228 1,228 1,228 1,228R-squared 0.850 0.851 0.849 0.849 0.850 0.850
Panel B:
Quota Exposure x Ipost -0.0127 -0.0181** -0.0106*** -0.00594 -0.267(0.00861) (0.00832) (0.00297) (0.00792) (0.188)
Quota Nationalites x Ipost -0.0762(0.151)
Quota measure Baseline Baseline Alternative 1 Alternative 2 Alternative 3 Alternative 4
Observations 1,230 1,230 1,230 1,230 1,230 1,230R-squared 0.931 0.931 0.931 0.931 0.931 0.931
Panel C:
Quota Exposure x Ipost -0.0315 -0.0369 -0.0187* -0.0256 -0.799*(0.0231) (0.0234) (0.0109) (0.0197) (0.437)
Quota Nationalites x Ipost -0.413(0.343)
Quota measure Baseline Baseline Alternative 1 Alternative 2 Alternative 3 Alternative 4
Observations 1,230 1,230 1,230 1,230 1,230 1,230R-squared 0.858 0.859 0.858 0.858 0.858 0.858
Panel D:
Quota Exposure x Ipost -0.00857 -0.00916 -0.00814 -0.00676 -0.406(0.0365) (0.0371) (0.0135) (0.0305) (0.698)
Quota Nationalites x Ipost -0.365(0.512)
Quota measure Baseline Baseline Alternative 1 Alternative 2 Alternative 3 Alternative 4
Observations 1,230 1,230 1,230 1,230 1,230 1,230R-squared 0.826 0.826 0.826 0.826 0.826 0.826
Controls Panel A-Dln Population size 1910 x time FE Yes Yes Yes Yes Yes YesShare Mex/Can 1920 x time FE No Yes No No No NoWWI x Ipost Yes Yes Yes Yes Yes YesLiteracy Act x Ipost Yes Yes Yes Yes Yes YesCity fixed effects Yes Yes Yes Yes Yes YesTime fixed effects Yes Yes Yes Yes Yes YesState-by-time fixed effects Yes Yes Yes Yes Yes Yes
Notes: The table reports DiD estimates. The observations are at the city level for the years 1909, 1914, 1919, 1925, and 1929. The outcome variableis ln value added per worker (Panel A), ln wages per worker (Panel B), horsepower per worker (Panel C) and horsepower per value added (Panel D).Column (1) shows the baseline estimates of Table 6 for comparison, except for column (2), which additionally includes the 1920 county-populationshares of Mexicans and Canadians interacted by time. The set of controls included are identical to Table 6. Quota exposure in column (1) isconstructed according to equation (1) of Section 3.2, but calculated at the city level. Column (2) replaces quota exposure with the share of quotaaffected nationalities ("Quota Nationalities") as described in equation (9) of Appendix A. Columns (3)-(5) present results for different versions ofquota exposure as described in equations (10)-(13) of Appendix A. Constants are not reported. Standard errors (in parentheses) account forarbitrary heteroskedasticity and are clustered at the county level. *** p<0.01, ** p<0.05, * p<0.1.
Appendix Table 4: Manufacturing -- alternative measures of quota exposure
Dependent Variable
ln(value added per worker)
ln(wages per worker)
ln(horsepower per worker)
ln(horsepower per value added)
ln(population) ln(black population) ln(share black) share blacks
VARIABLES (1) (2) (3) (4)
Quota Exposure x Ipost -0.0709*** 0.0112 0.345*** 0.00638***(0.0218) (0.0686) (0.0522) (0.00143)
ln Population size 1910 x time FE Yes Yes Yes YesWWI x Ipost Yes Yes Yes YesLiteracy Act x Ipost Yes Yes Yes YesCounty FE Yes Yes Yes YesTime FE Yes Yes Yes YesState-by-time FE Yes Yes Yes YesCounty-specific linear trend Yes Yes No No
Observations 13,198 13,198 13,198 14,060R-squared 1.000 0.997 0.963 0.991
Appendix Table 5: Population dynamics -- The effect of the immigration quotas on the black population
Dependent Variable
Notes: The table reports DiD estimates. The observations are at the county level for the period 1900 to 1940. The outcomevariable is ln population in column (1); ln black population in column (2); and the (ln) share of blacks in columns (3)-(4). Quotaexposure is defined in equation (1) of Section 3.2. All specifications include county and time fixed effects. Additional controlsare ln population size in 1910 interacted with a full set of time fixed effects, WWI, and the Literacy Act of 1917, state-by-timefixed effects, and a county-specific linear trend in columns (1)-(2). See Appendix C for further details. Constants are notreported. Standard errors (in parentheses) account for arbitrary heteroskedasticity and are clustered at the county level. ***p<0.01, ** p<0.05, * p<0.1.
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Panel A:
Quota Exposure x Ipost -0.0159*** -0.0137*** -0.0133*** -0.0171*** -0.00981 -0.0180*** -0.0162*(0.00417) (0.00508) (0.00468) (0.00411) (0.00964) (0.00414) (0.00905)
Quota Exposure x Ipost x black 0.0381*** 0.0367*** 0.0663*** 0.0636**(0.00653) (0.00699) (0.0250) (0.0261)
Sample AllLives in state
of birthLives outside state of birth
Men Women Men Men Women Women
Observations 1,612,045 1,143,685 468,342 1,240,295 371,731 1,240,075 1,240,748 371,509 370,677R-squared 0.363 0.376 0.329 0.382 0.378 0.387 0.400 0.392 0.423
Panel B:
Quota Exposure x Ipost -0.0151*** -0.0146*** -0.0168*** -0.0188*** -0.00162 -0.0207*** -0.00658(0.00401) (0.00453) (0.00391) (0.00390) (0.00794) (0.00379) (0.00676)
Quota Exposure x Ipost x black 0.0248*** 0.0217*** 0.0104 0.00872(0.00562) (0.00551) (0.00639) (0.00700)
Sample AllLives in state
of birthLives outside state of birth
Men Women Men Men Women Women
Observations 1,611,505 1,143,377 468,109 1,239,815 371,671 1,239,595 1,240,268 371,449 370,621R-squared 0.651 0.668 0.593 0.592 0.712 0.599 0.609 0.730 0.746
Controls Panel A-Bln Population size 1910 x time FE Yes Yes Yes Yes Yes Yes Yes Yes YesWWI x Ipost Yes Yes Yes Yes Yes Yes Yes Yes YesLiteracy Act x Ipost Yes Yes Yes Yes Yes Yes Yes Yes YesIndividual controls Yes Yes Yes Yes Yes Yes Yes Yes YesCounty FE Yes Yes Yes Yes Yes Yes Yes Yes YesTime FE Yes Yes Yes Yes Yes Yes Yes Yes YesState-by-time FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
Appendix Table 6: Alternative occupation-based earnings scoresDependent Variable
ln(Occupational Income Score)
ln(1940 Earnings Score)
Notes: The table reports DiD estimates. The observations are at the individual level for the period 1900 to 1940. The sample spans 15-65-year old native born workers. Theoutcome variable is the ln occupational income score from IPUMS in Panel A and the ln 1940 earnings score in Panel B (see Appendix D for further details). Quota exposure isdefined in equation (1) of Section 3.2. All regressions include county fixed effects, time fixed effects, state-by-time fixed effects, ln population size in 1910 interacted with a full setof time fixed effects, and controls for WWI and the Literacy Act of 1917 (Appendix A details the construction of the controls for WWI and the Literacy Act of 1917). The set ofindividual controls includes the following indicator variables: marital status, place of residence (rural/urban), gender, race (black/white), literacy. We further include birthplaceand age fixed effects. Columns (6)-(9) further include quota exposure, WWI, and the Literacy Act of 1917 each interacted with a dummy for race (black), race-by-time fixed effects,and race-by county fixed effects. County-by-time fixed effects are added to the specifications in column (7) and (9). Constants are not reported. Standard errors (in parentheses)account for arbitrary heteroskedasticity and are clustered at the county level. *** p<0.01, ** p<0.05, * p<0.1.
(1) (2) (3) (4) (5) (6)
Quota Exposureold x Ipost 0.0756*(0.0427)
Quota Exposurenew x Ipost -0.0225***(0.00518)
Quota Exposurej x Ipost 0.713*** -0.109*** -0.0295 -0.0363 0.191*(0.193) (0.0141) (0.0195) (0.0712) (0.105)
Quota Exposure-j x Ipost -0.0264*** -0.00527 -0.0190*** -0.0195** -0.0217***(0.00526) (0.00493) (0.00635) (0.00763) (0.00509)
Country j Great Britain Russia Italy Austria FinlandControls Table 4 Yes Yes Yes Yes Yes Yes
Observations 1,569,620 1,569,620 1,569,620 1,569,620 1,569,620 1,569,620R-squared 0.642 0.642 0.642 0.642 0.642 0.642
Appendix Table 7: Missing specific immigration groups and their impact on the earnings of native workers
Dependent Variable
ln(Lebergott Earnings Score)
Notes: The table reports DiD estimates. The observations are at the individual level for the period 1900 to 1940. The sample spans 15-65-yearold native born workers.The outcome variable is the ln Lebergott earnings score (see Section3.1 for further details). The set of controls includedare identical to Table 4 column (1). Quota exposure in column (1) is constructed according to equation (1) of Section 3.2, but split into old andnew immigration groups. Quota Exposureold refers to missing immigrants from Northern and Western Europe, while Quota Exposurenew refersto immigrants from Southern and Eastern Europe. The classification follows the "Dillingham Report" (U.S. Immigration Commission, Table 6).Columns (2)-(6) present results by specific immigration groups. Quota Exposurej denotes missing immigrants from country j, where j = {GreatBritain, Russia, Italy, Austria, and Finland}. Quota Exposure-j denotes missing immigrants from all other countries except j . Constants are notreported. Standard errors (in parentheses) account for arbitrary heteroskedasticity and are clustered at the county level. *** p<0.01, **p<0.05, * p<0.1.