Modeling the age and age composition of late 19th century U.S. immigrants from Europe
Michael J. Greenwood*
Department of Economics, University of Colorado
Campus Box 256, Boulder, CO, 80309
*E-mail address: [email protected]
2
Abstract
Using panel data on 12 European source countries that are followed for 26 years (1873-1898), this paper studies agespecific emigration rates and the age composition of U.S. immigration. Two age groups are the focus of attention,
15-40 and over 40. Emigration-rate models and compositional models that satisfy adding-up conditions are
estimated by the Hausman-Taylor Instrumental Variable approach. Younger migrants responded more strongly to
job opportunities than to wage differentials, whereas older migrants responded more strongly to wage differentials.
Both age groups tended to follow recent past migrants to the U.S. Relatively many younger (and relatively fewer
older) migrants came from countries with higher percentages of their work forces in agriculture. Higher sourcecountry birthrates discouraged younger migrants, presumably by raising the cost of family migration.
Keywords: Age composition; U.S. immigration; Late nineteenth century; Europe
3
Introduction
Much research has concerned the volume and/or rate of migration from Europe to the U.S. during the late 19th
and early 20th centuries when institutional impediments to international migration from Europe were low or
nonexistent and movement costs had fallen substantially due to the availability of the steamship (Gould, 1979;
Hatton and Williamson, 1998). Hatton and Williamson (p. 11) ask a much less studied question: “Who were the
emigrants?” “Who” could refer to sex, age, skills, and numerous other migrant characteristics. This paper focuses
specifically on the age of late 19th century U.S. immigrants from Europe and essentially asks in the context of an
econometric model “why were the emigrants who they were?”
The age composition of 19th century immigration from Europe varied considerably across source countries
and by source country over time, where age composition refers to the fraction of immigrants in various age classes.
Why were U.S. immigrants from certain source countries younger, whereas those from others were older, and why
did the age composition of the flows from the various source countries change in systematic ways? The goal of this
paper is to use panel data and a model that satisfies adding-up conditions to address such questions. The panels
consist of 12 source countries that are followed annually from 1873 to 1898. Age refers to three classes: under 15,
15-40, and over 40, with primary focus on the latter two groups.
Age at entry of U.S. immigrants was potentially important for both the U.S. and the immigrants’ source
countries. Younger migrants have more years to contribute to the U.S. economy. Studies of contemporary U.S.
immigration suggest that age at migration importantly shapes how well the immigrants perform upon entering the
U.S. labor force (Friedberg, 1991). Other studies show similar results for Canada (Schaafsma and Sweetman, 2001).
Younger immigrants assimilate more rapidly, in part because they learn English more quickly, which in turn
augments their labor productivity and earnings and facilitates their social and cultural assimilation. The same
process of assimilation appears to have characterized 19th century immigrants (Hatton and Williamson, 1998).
Immigrant age in combination with sex potentially shapes the rate of increase of the second generation
immigrant population. Birth rates are highest for the same ages for which migration propensities are highest, and
young female immigrants in their peak child-bearing years contribute to the population potential of the receiving
country. Through the marriage market, even young male migrants may influence subsequent population growth.
Both personal and societal costs of migration are tied to the age of the migrant. Education, skills, and
experience are all functions of age. Specific occupational skills accrue to more experienced workers and thus with
4
age. Eichengreen and Gemery (1986) show that during the 1890s skilled immigrants were about 2 years older than
unskilled immigrants. If such education, skills, and experience accumulate in a non-English language environment,
they are less readily transferred to the U.S. Although associated losses of human capital may be overcome with
investments like learning English, the losses are real, and they are tied in part to the age of the migrants. Moreover,
younger immigrants impose costs on public education that are not recovered for some years. Older immigrants who
do not participate in the labor force may constitute a burden, though this type of transfer was unlikely to be as
important during the 19th century as it is today.
The earliest detailed descriptive analysis of migration is Ravenstein (1885), who focuses on such topics as
sex and distance moved, but completely ignores age. Yet 50 years later Dorothy Swaine Thomas in her famous
Research Memorandum on Migration Differentials (1938) could claim that “one generalization about migration
differentials which can be considered definitely established…is the following: there is an excess of adolescents and
young adults among the migrants…compared with the non-migrating or the general population” (p.11). In
concluding her chapter on “age differentials,” Thomas (p.54) argues that “of all the gaps in our knowledge of the
operation of age-selective migration, the most important are:
(1) Our lack of precise information as to the operation of communities of varying economic and social
structure upon age selection…
(2) Our lack of precise information as to the operation of distance as a factor limiting or extending the
range of age-selective migration…
(3) Our lack of any information at all as to the operation of the upswings and downswings in economic
conditions upon age-selective migration.”
Perhaps surprisingly, to the present day little analytical research has been conducted on any of the issues noted by
Thomas. One of the few exceptions that treat historical data is Hatton and Williamson (1998), who study sex- and
age-specific emigration from Irish counties, 1881-1911.1 The goal of this paper is to address the issues raised by
Thomas, but in the context of historical U.S. immigration from Europe, where countries rather than communities
provide the spatial dimension.
The paper is organized as follows. Section 2 discusses the data that underlie the study, including the 12
European source countries that form the data base. Section 3 develops the models of age-specific emigration and of
age composition, and discusses the methodology used to study composition. Section 4 describes econometric
strategy, and Section 5 presents empirical results. The final section provides a summary and conclusions.
2.
Data
The number of countries in the data set is critical because it sets the degrees-of freedom constraint in the
5
estimated models at n-1, where n is the number of countries. Based upon the availability of data on both the
dependent and independent variables, only 12 countries are included in the study.2 These are the same 12 frequently
used in studies of historical U.S. immigration (e.g., Hatton and Williamson, 1998, p. 54): Belgium, Denmark,
France, Germany, Ireland, Italy, Netherlands, Norway, Portugal, Spain, Sweden, and the U.K. (less Ireland).
U.S. immigration data were first collected and reported on a systematic basis for fiscal year 1820. From then
until 1898 three age groups were reported: under 15, 15-40, and over 40. Beginning in 1899 the groupings were
under 14, 14-44, and over 44. The age groupings were first reported by country of origin for 1873, and annual data
for 1873-1898 are compiled in The Bureau of Statistics (1903). Except for the earliest decade reported (1821-1830)
and the latest (1891-1898), the aggregate age shares did not vary appreciably. For the 1891-1898, a significant
increase occurred in the share 15-40, whereas fairly sizable decreases occurred for the under 15 and over 40 groups.
Of course, the 1890s reflect a significant shift of immigrant sources from Western and Northern Europe to Southern
and Eastern Europe. Sex ratios (ratios of male to female immigrants) increased greatly, in part because many
migrants from the new source countries moved temporarily and without their families.
The aggregate figures conceal appreciable differences in immigrant age composition across source
countries, as well as appreciable changes over time for specific source countries. For 1873-1898, for the 12
European source countries that serve as the database, the percentage of children was 22.5% in 1873, peaked at
24.5% in 1884, fell to a low of 10.5% in 1895, and was 15.3% in 1898.3 The percentage 15-40 was at a low of
63.5% in 1874, rose to a high of 84.8% in 1895, and fell to 72.3% in 1898. Finally, the percentage of immigrants
over 40 was at a high of 15.0% in 1874, fell to a low of 4.4% in 1894, and was 12.4% in 1898. The primary focus of
this paper is on the two oldest age groups because they consist mostly of adult decision makers. Each group
displays enough variation in the aggregate to provide sufficient interest for study.
Far more variability is observed when the data are examined for specific source countries. Two noteworthy
observations are evident. First, wide variations exist across countries for a given year (Table 1). In 1873, 74.5% of
the immigrants from Spain were 15-40, whereas only 46.7% of those from Norway were in this age class. In 1898,
these percentages ranged from a high of 88.7% (Ireland) to a low of 58.9% (Netherlands). The oldest age class also
reflects considerable variation. In 1873, 29.7% of the immigrants from Norway were over 40, but only 9.3% of
those from Ireland were over 40. In 1898, the range was from 21.5% (Spain) to 6.2% (Ireland). Second, for many
countries of birth considerable changes occurred over time in the age composition of their U.S.-bound migrants.
6
Based on the earliest and latest data points, certain countries had large declines in the percentage 15-40 (Belgium,
Portugal), others had substantial increases (Denmark, France, Germany, Ireland, Netherlands, Norway, Sweden),
and some had changes of less than five percentage points (Italy, Spain, Great Britain). Again based on the earliest
and latest data points, the oldest age group also reflects changes over time, though for many countries these changes
were modest. For Belgium the percentage over 40 increased from 9.6% to 18.4%, for Norway it fell from 29.7% to
10.2%, for Spain it increased from 13.5% to 21.5%, for Sweden it fell from 15.6% to 7.6%, and for Portugal it fell
from 16.7% to 12.6%. Of course, based on the maximum and minimum for each country, the differences are greater.
Over the 26 years studied here, the 12 countries accounted for both a declining number and a declining
share of U.S. immigrants as countries of Southern and Eastern Europe joined those from Western and Northern
Europe as important sources (Table 1). The number declined from 381,103 in 1873 to 138,774 in 1898.4 The
percentage of total U.S. immigrants attributed to the 12 countries fell from 82.9% in 1873 to 60.5% in 1898. Both
because the 12 countries are the only ones for which panel data are available or can reasonably be constructed for
the period, and because even at the end of the period they are representative of migration from Europe to the U.S.,
the 12 countries form the data base.
Due to the very broad age classes reported in official immigration statistics, this study focuses on children
(under 15), on older persons (over 40), and on individuals in their prime ages (15 to 40). The migration of children
is primarily a function of the migration of their parents. Although the number of children in the flow is important
because they will be raised and schooled in the United States and subsequently enter the labor force, the main
concern of this paper is with the two older classes because they are more or less independent decision makers.
Unfortunately, the age data are very broad, which limits the empirical analysis to the upper tail of the age
distribution (over 40) and the next youngest grouping (15-40). On average, over the 26 years studied here, the upper
tail accounted for 13.4% of the immigrants 15 and over, which is not a trivial percentage.
With respect to the independent variables, where annual data were not available for 1873-1898, synthetic
series were generated for each country or panel. The main data sources are Maddison (2003), Mitchell (1992), and
Williamson (1995). For a detailed discussion of the construction of these series, see Greenwood (2005). Table 2
provides means and standard deviations for all variables included in this study.
3. The models
The two models are specified below, one that concerns age-specific emigration rates from the 12 European
7
source countries to the U.S. and one that specifically focuses on the age composition of these flows. The models
focus on factors that are likely to have influenced 19th century migrant flows and for which empirical measures are
available. These models contain three vectors of variables, one to reflect the relative economic advantage of country
i compared to the U.S., one to reflect the costs of transferring accumulated occupational skills to the U.S., and one
that includes “control” variables. The general form of each model is the same, but the compositional model entails
an adding-up restriction (discussed below) not inherent in the emigration-rate model. The form of the models is
M iat (or Siat ) = βi + δ t + λ1v1it + λ2v2it + λ3v3it + ε iat ,
(1)
where
M iat = the rate of migration of age group a from country i to the U.S. in year t;
S iat = a measure of the age composition of immigrants from country i to the U.S. in year t;
β i = fixed country effects;
δ t = fixed time effects;
v1it = a vector of variables that reflects the differential economic advantage between country i and the U.S.;
v2 it = a vector of variables that reflects relative migration costs including both direct entry
costs and skill transferability costs associated with moving from country i to the U.S.;
v3it = a vector of control variables;
λi = vectors of unknown parameters for i = 1 ,…, 3; and
εi = random errors.
The v’s contain certain variables that are time-invariant, such as distance from country i to the U.S. One objective of
this study is to estimate the coefficients of such variables.
The expected returns to and costs of migrating from country i to the U.S. are relevant considerations for the
potential migrant in i considering such a move, and these returns and costs clearly are functions of the age at which
migration occurs (Morgan and Robb, 1981; Millington, 2000). The near universal observation that young adults
have higher propensities to migrate than older persons reflects higher returns to and perhaps lower costs of migration
at earlier ages. The returns to migration decline with age because by postponing a potentially profitable move to an
older age the migrant foregoes those future returns that are discounted least (Rij in Fig. 1). Moreover, older persons
8
develop more family and community ties, including more community-specific human capital, which increase the
costs of migrating (Cij in Fig. 1) (Hatton and Williamson, 1998). These general relationships ought to be reflected in
the emigration-rate regressions reported below.
In themselves, these relationships have no implications for the age composition of migration. That is, they do
not allow predictions regarding which types of countries (e.g., low-income versus high-income) would be the source
of relatively younger compared to relatively older U.S. immigrants. However, the concept of the “marginal
migrant” provides a framework that yields predictions regarding the age composition of U.S.-bound emigrants from
various countries. For the marginal migrant, the returns to migration equal the costs of migration, so the net benefit
to migration is zero (α0 in Fig. 1). The returns to migration are the migrant’s U.S. benefits less foregone homecountry benefits, both appropriately discounted. If we allow the ages of source-country residents to differ while for
convenience holding constant their occupational skills, the equilibrium condition for the marginal migrant is:
T
Σ(wus – wi) K*(1 + r)-(T-t) = Ci ,
t=α
(2)
where α is the age of migration, T- α is the number of post-migration years of work, wus and wi are the respective
U.S. and country-of origin wage per unit of occupational skill, K* is the index of occupational skill, and r is a
discount rate. The left side of Equation (2) represents the discounted total returns accruing to the marginal migrant
who moves from country i to the U.S. at age α; the right side represents total migration costs, which include direct
entry costs and skill losses associated with less-than-perfect international skill transferability.
In this model, α is a choice variable because the marginal migrant is assumed to adjust to factors that change
either the returns to or costs of migration (i.e., factors that cause shifts of Rij or Cij in Fig. 1). “Adjusting” refers to
migrating at a younger or older age. By choosing a different age of migration, the migrant influences his returns to
and costs of migration. Adjustments would occur if different returns or costs, apart from age itself, were tied to the
migration decision. As the age of the marginal migrant changes, the age composition of emigrants from i would
correspondingly change. Higher returns would cause the marginal migrant to be older. If higher migration costs are
not tied to any age-specific cost differences, then they would cause the marginal migrant to move at a younger age.
However, the direct costs of migration may not be uniform for all age groups. High birthrates, for example, may
exert different influences on younger and older migrants, since young children and larger families are likely to
disproportionately raise the cost of migration for younger individuals.
Thus, shifts in Rij or Cij in Fig. 1 have implications for the age of the marginal migrant and therefore for
9
immigrant composition. For example, for lower income source countries Rij is higher (R΄ij in Fig. 1) and the age of
the marginal migrant is correspondingly older (α1). Consequently, in the compositional regressions, the anticipation
is that older migrants will tend to originate in lower-income countries. For countries for which migration to the U.S.
is more costly (C΄ij in Fig. 1), the age of the marginal migrant is expected to be lower (α2). Thus, more distant and
non-English speaking source countries ought to be the origins of younger migrants.
The methodology used to study age composition has long been employed by economists to analyze systems of
consumer demand and expenditure equations (Leser, 1961; Pollak and Wales, 1969; Parks, 1969; and Barten, 1977)
and systems of cost-share equations (Berndt and Wood, 1975). Greenwood and McDowell (1999) and Greenwood,
McDowell, and Wierman (2003) employ a similar methodology to study the age composition of contemporary U.S.
immigration. As above, let a index age classes. Two groups of immigrants are the primary focus of attention—
younger, or 15-40 (y), and older, or over 40 (o). Let v be the column vector of all explanatory variables in Equation
1 so that v΄ = [v΄1, v΄2, v΄3] and also define λα = [λ΄1, λ΄2, λ΄3] for all
a = y,o. Now the estimated model may be
represented as:
Sita = β ia + δ ta + λa vit + ε ita
for
Sity + Sito = 1
for all i and t;
β iy + β io = 1
for all i;
δ ty + δ to + ε ity + ε ito = 0
for all i and t; and
a = y,o,
(3)
where
λ y + λo = 0,
where 0 is the zero vector (Greenwood, McDowell, and Waldman,1996). The conditions above indicate that the age
shares ( S ita ) must sum to one across the two age equations. Moreover, the constant terms also must sum to one, and
the coefficients on each independent variable must sum to zero. Thus, if each independent variable were set to zero,
the age shares would sum to one, as they must logically. Any change in an independent variable that increases one
share must decrease the other share, so that the shares continue to sum to one.
Four variables are included in vector v1 to reflect differential economic opportunities of the various source
countries compared to the U.S.: (1) relative real wage rate in year t-1, defined as source-country index relative to
10
U.S. index (Williamson, 1995); (2) relative growth of GDP over the prior three years, measured as average annual
growth in the U.S. relative to that in the source country (Maddison); (3) percentage of the economically active
population of i that was engaged in manufacturing (Mitchell, 1992); and (4) percentage of the economically active
population in i that was engaged in agriculture (Mitchell).
Given migration costs and other characteristics of country i, the higher the potential migrants’ wage in
country i, the higher must be the expected wage in the U.S. to induce a move to the U.S. Thus, higher relative
wages should discourage migration from i to the U.S.5 As suggested by Hatton and Williamson (1998), because
younger individuals (15-40) had more to gain due to their longer expected working lives, they ought to have been
more sensitive to economic differentials than older persons (over 40). Similarly, more rapid growth of GDP in the
U.S. relative to that in country i ought to reflect job opportunities and be attractive to migrants, especially those with
more expected years of work. The two industrial composition variables ought to take opposite signs with higher
percentages in manufacturing reflecting better opportunities at home and higher percentages of agriculture reflecting
fewer opportunities there. Again, such variables should be more relevant for individuals in prime working ages.
Five variables are included in v2 , the vector for the cost of moving from country i to the U.S.: (1) total
migration from i to the U.S. over the prior 2 years6; (2) birthrate in i (Mitchell)7; (3) a dummy variable for countries
where English is the official language (equals one for Ireland and for the rest of the U.K. and otherwise zero); (4) a
dummy variable that takes a value of one for Southern European countries (Spain, Portugal, Italy) and is otherwise
zero; and (5) distance from i to the U.S. (the distance between the principal debarkation point and New York City).
Numerous historical and contemporary migration studies find that past migration is a critical determinant of
current migration (Greenwood, 1969; Dunlevy and Gemery, 1977; Hatton and Williamson, 1998). Past migrants not
only provided potential current migrants with information about the U.S., but also they frequently paid the
transatlantic fares of their relatives and friends. For example, for 1910, 1911, 1912, and 1913, 25%, 33%, 36%, and
32%, respectively, of the immigrants reported that they were assisted with their fares, and these figures are almost
certainly low.8 Past migrants also provided current migrants with food and shelter, specific job-related information,
and language, religious, and cultural familiarity that eased the transition to the new country.
Higher birthrates in i should have discouraged younger individuals from migrating, especially females with
young children. Children also raised the cost of family-unit migration. Higher birthrates should have a lesser
impact on the potential migration of older individuals. Particularly later in the period studied here, migrant sex
11
ratios varied considerably across source countries. More so than migrants from elsewhere in Europe, those from
Southern Europe tended to move temporarily to the U.S., leaving their families behind with the intention of
returning to their source countries or later reuniting with their families in the U.S. Higher migrant sex ratios and
relatively younger flows probably went hand-in-hand. The Southern Europe dummy variable ought to pick up this
influence, along with the increased flows from Southern Europe later in the period.
In place-to-place migration models, distance serves as a proxy for the money costs of travel and for the
nonmonetary costs of moving farther away from relatives and friends. During the late 19th century, a one-way fare
from Liverpool to New York was about $17.50, whereas that from Naples, Italy, to New York was about $25.9 Such
fares appear to reflect distance. Thus, distance ought to have discouraged migration to the U.S. Based on the theory
of the marginal migrant, younger migrants are expected to have come from more distant countries. Finally,
knowledge of English should have facilitated the transfer of skills to the U.S., as well as social and cultural
assimilation, and therefore have encouraged migration. Due to the lower costs of transferring accumulated skills to
the U.S. from English-speaking countries, relatively older migrants are expected to originate in them.
4.
Estimation Strategy
Consider the following relationship
M iat = ϕvit + ε iat ,
where
M iat
(4)
is country i’s emigration rate to the U.S. of age class a in year t. (To simplify the notation, the
subscript a is dropped.) Now partition vit such that vit = [ xit | zi ], where xit is a kx1 vector of variables that measure
characteristics of country i, year t, and zi is a gx1 vector of time-invariant variables for country i. Now write (4) as
M it = βi + δ t + ωxit + γzi + ε it ,
(5)
where βi is unobserved country-specific effects, δt is unobserved time-specific effects, ω and γ are vectors of
unknown parameters, and εit is random disturbances. The Hausman-Taylor Instrumental Variable (IV) approach
(Hausman and Taylor, 1981) is used to estimate Equation (5). In this approach the x’s are partitioned into (x1 | x2)
and the z’s are partitioned into (z1 | z2), where the k1 elements of x1 and the g1 elements of z1 are uncorrelated with
the country effects, but the k2 elements of x2 and the g2 elements of z2 are correlated with the country effects. All
variables are twice mean differenced (through time and across countries), so βi and δt drop out of the estimation.
However, mean differencing and using (cross-sectional and temporal) dummy variables are algebraically equivalent
12
(but mean differencing allows the estimation of temporally invariant variables).
A decision must be made regarding which variables may be correlated with unobserved country-specific
effects and which are exogenous. The distinction is based on the idea that any variable that could reflect a taste for
the U.S. may be correlated with the unobservables. For example, the number of recent migrants from i to the U.S.
may reflect a taste for the U.S. that is transmitted to potential migrants and therefore may be correlated with the
unobservables. A similar argument holds for treating the English-language dummy as endogenous. In the tables
below that report the Hausman-Taylor IV estimates, variables are identified as belonging to x (temporally variable)
or z (temporally invariant) and as exogenous (x1, z1) or endogenous (x2, z2). For each regression, a test is performed
for the exogeneity of the HT instruments. In no case can exogeneity be rejected.
5.
Empirical Results
Table 3 reports two types of empirical findings, emigration-rate regressions for both the 15-40 and the over
40 age classes and the age-share relationship for only the 15-40 age class (because the over 40 group is the mirror
image of the 15-40 group, as described in Equation 3). Consider first the emigration-rate regressions. Some
noteworthy contrasts are evident between younger and older migrants.10 Unexpectedly, the older group responded
more strongly to economic incentives as defined by relative wages. In fact, although it takes the expected negative
sign, the coefficient on the relative wage variable is not significant in the regression for the 15-40 group. This
finding seems consistent with Baines’s (1991) observation that “differences in wage rates between two
countries…do not predict emigration very well” (p. 26). He argues that migrants may have had little information
regarding wages, but were more likely to have responded to job opportunities. Indeed, the younger group responded
positively and significantly to U.S. growth, whereas the response of the older group was positive but not significant.
A higher percentage of the economically active population in the manufacturing sector of the source country
significantly discouraged the older group from migrating, presumably because better employment opportunities
were available in the home country. Moreover, whereas the older group was discouraged from leaving countries
with a high fraction of the economically active population in agriculture, the younger group departed such countries
for the U.S. This finding is consistent with the Hatton and Williamson (1994) hypothesis regarding the difficulty of
young people acquiring land in Europe.
In the vector for the costs of migrating, prior migration stands out as a variable that clearly attracted
migrants to the U.S. Per thousand migrants over the prior two years, about 60 younger and 12 older persons moved
13
to the U.S. Both younger and older migrants tended to come from Ireland and Great Britain (as reflected in the
English dummy variable), but the response of the younger migrants was considerably greater. Relatively few
migrants of each age came from Southern Europe. Higher birthrates discouraged younger individuals from
migrating, but older migrants tended to originate in these same countries.
Clearly, the share of younger migrants was significantly greater for relatively low-wage source countries,
which is consistent with the theory regarding the marginal migrant. Relatively more young migrants came from
countries with higher fractions of their economically active populations in agriculture and in manufacturing. With
respect to agriculture this finding is again consistent with the Hatton and Williamson hypothesis. The result
regarding manufacturing may suggest that relatively high concentrations of manufacturing jobs facilitated the
transfer of skills of younger persons from European source countries to the U.S., but the exact mechanism is unclear.
Presumably by raising the cost of family migration, higher birthrates significantly shifted migration away from the
younger group and toward the older group. Neither relative growth of GDP nor prior migration from i to the U.S.
played a role in shaping the age composition of U.S. immigrants.
6.
Summary and Conclusions
This paper provides models of age-specific U.S. immigration and the age composition of U.S. immigration
from Europe during the late 19th century. In this way, the paper focuses on “who” migrated as opposed to “how
many” migrated, where who refers to age groups. As noted in the introduction, we have long known that migration
propensities are closely related to age, peaking in the early to mid-twenties and thereafter declining. However, as
stressed by Thomas (1938) in the 1930s, we knew little about the factors that caused differences in age patterns of
migration for places of differing characteristics. With considerable validity, this same claim can be made today.
Thomas notes a lack of “precise information as to the operation of communities of varying economic and
social structure upon age selection” (1938, p.54). Focusing on late 19th century U.S. immigration, this paper
substitutes countries for Thomas’s communities and shows that the industrial structure of employment helped shape
age selection. Older potential migrants were discouraged by relatively well-developed manufacturing sectors in
European source countries. Agricultural orientation also helped shape the age selection of migration. Younger
migrants tended to originate in European economies that were more agricultural, and older migrants tended to come
from less agriculturally oriented countries.
Distance is another factor that Thomas noted, arguing that we lack “precise information as to the operation
14
of distance as a factor limiting or extending the range of age-selective migration” (1938, p.54). Although the
empirical results suggest that older persons tended to come from distant countries, they also suggest that younger
persons were not much influenced by distance. One possibility for this latter finding is that the distances were all
great. One standard deviation from the mean (3,670 miles) amounted to 315 miles. Perhaps more importantly,
current migrants had a strong tendency to follow past migrants. If past migrants moved to the U.S. from less distant
countries, then the variable for prior migration would pick up the effects of distance (Greenwood, 1969).
Finally, Thomas decries “our lack of any information at all as to the operation of the upswings and
downswings in economic conditions upon age-selective migration” (1938, p.54). On this point the empirical
evidence is quite clear. Both younger and older individuals responded to economic incentives, but in different ways.
Whereas older individuals were significantly discouraged from leaving high-wage countries for the U.S., younger
persons responded significantly to job opportunities. Hatton and Williamson (1998), building on a substantial body
of earlier work by Williamson, show that “economic fundamentals” importantly influenced the rates of migration
from various European source countries to the U.S. This study demonstrates that some of these same fundamentals
helped shape the age structure of such migrations.
During the period studied here, the trend was unmistakably in the direction of relatively more younger and
relatively fewer older U.S. immigrants. For 15 to 40 year-old immigrants, the percentages (of all immigrants from
the 12 source countries) are as follows: 66.6% for 1873-1879; 68.1% for 1880-1889; and 76.7% for 1890-1898.
Corresponding percentages for those over 40 are 13.9%, 10.4%, and 8.9%. To what extent do the findings reported
above contribute to a better understanding of these trends? Hatton and Williamson (1998) attribute the declining
numbers of European emigrants from the original sources to narrowing wage differentials between these countries
and the U.S. Such “fundamentals” also appear to have influenced the age composition of the flows. Older migrants
tended to come from lower-wage source countries, and as the wage gap narrowed, their relative numbers declined.
Moreover, continuing declines in agricultural employment significantly affected younger workers, releasing more of
them for migration to the U.S. and elsewhere. Declining birthrates in European source countries also may have
reduced the costs of migrating relatively more for younger individuals than for older persons.
Available data significantly limit the scope of this study of migrant age composition. More detailed age
groups would be helpful in the analysis. Moreover, information on marital status and family migration would be
helpful. However, without a substantial reworking of the historical migration data, which would truly be a
15
Herculean task, none of these studies is possible at the present time.
16
References
Baines, D., 1991. Emigration from Europe 1815-1930. Macmillan, London.
Barten, A.P.,1977. The systems of consumer demand functions approach: a review. Econometrica 45 (1), 23-51.
Berndt, E.R., Wood, D.O., 1975. Technology, prices and the derived demand for energy. Review of Economics and
Statistics 57 (3), 259-268.
Bureau of Statistics, 1903. Monthly Summary of Commerce and Finance of the United States, No. 12, Series 19021903 (June), USGPO, Washington, D.C.
Dunlevy, J.A., Gemery, H.A., 1977. The role of migrant stock and lagged migration in the settlement patterns of
nineteenth century immigrants. Review of Economics and Statistics 59 (2), 137-144.
Easterlin, R.A., 1961. Influences in European overseas emigration before World War I. Economic Development and
Cultural Change 9 (3), 331-351.
Friedberg, R.M., 1991. The labor market assimilation of immigrants in the United States: the role of age at arrival.
Unpublished manuscript.
Gould, J.D., 1979., European inter-continental emigration 1815-1914: patterns and causes. Journal of European
Economic History 8 (3), 593-679.
Greenwood, M.J., 1969., An analysis of the determinants of geographic labor mobility in the United States. Review
of Economics and Statistics 51 (2), 189-194.
Greenwood, M.J., McDowell, J.M., and Waldman, D.M., 1996. A model of the skill composition of US
immigration. Applied Economics 28 (1), 299-308.
Greenwood, M.J., McDowell, J.M., 1999. Legal U.S. Immigration: Influences on Skill Composition. W.E. Upjohn
Institute for Employment Research, Kalamazoo.
Greenwood, M.J., 2005., The sex composition of U.S. immigration from Europe, 1820-1929: a panel data study.
Unpublished manuscript, University of Colorado at Boulder.
Hatton, T.J., Williamson, J.G., 1994. International migration 1850-1939: an economic study. In Hatton, T.J.,
Williamson, J.G., Migration and the International Labor Market, 1850-1939. Routledge, London, 3-32.
Hatton, T.J., Williamson, J.G., 1998. The Age of Mass Migration: Causes and Economic Impact. Oxford University
Press, New York.
Hausman, J.A., Taylor, W.E., 1981. Panel data and unobservable individual effects. Econometrica 49 (6), 13771398.
Leser, C.E.V., 1961. Commodity group expenditure functions for the United Kingdom, 1948-1957. Econometrica 29
(1), 24-32.
Maddison, A., 2003. The World Economy: Historical Statistics. Organization for Economic Co-Operation and
Development, Paris.
Millington, J., 2000. Migration and age: the effect of age on sensitivity to migration stimuli. Regional Studies 34
(6), 521-533.
Mitchell, B.R., 1992. International Historical Statistics, 1850-1988. Stockton Press, New York.
17
Morgan, J.N., Robb, E.H., 1981. The impact of age upon interregional migration. Annals of Regional Science 15
(3), 31-45.
Parks, R.W., 1969. Systems of Demand Equations: An Empirical Comparison of Alternative Functional Forms.
Econometrica 37 (4), 629-650.
Pollak, R.A.,Wales T.J., 1969. Estimation of the linear expenditure system. Econometrica 31 (4), 611-628.
Ravenstein, E.G., 1885. The laws of migration. Journal of the Statistical Society 48, part II, 167-227
Schaafsma, J., Sweetman, A., 2001. Immigrant earnings: age at immigration matters. Canadian Journal of
Economics 34 (4), 1066-1099.
Thomas, D.S., 1938. Research Memorandum on Migration Differentials. Social Science Research Council, New
York.
U.S. Department of Labor, Bureau of Immigration, 1914. Annual Report of the Commissioner General of
Immigration for the Fiscal Year Ended June 30, 1913. USGPO, Washington, D.C.
U.S. Treasury Department, Immigration Bureau, 1892. Letter from the Secretary of the Treasury, Transmitting a
Report of the Commissioners of Immigration upon the Causes which Incite Immigration to the United States.
USGPO, Washington, D.C.
Williamson, J.G., 1974. Migration to the new world: long term influences and impact. Explorations in Economic
History 11 (4), 357-389.
Williamson, J.G., 1991. Did England’s cities grow too fast during the industrial revolution. In Higonet, P., Landes,
D.S., Rosovsky, H., Favorites of Fortune: Technology, Growth, and Economic Development since the Industrial
Revolution. Harvard University Press, Cambridge, 359-394.
Williamson, J.G., 1995. The evolution of global labor markets since 1830: background evidence and hypotheses.
Explorations in Economic History 32 (2), 141-196.
18
Footnotes
* I am grateful to John P. Greenwood and Fred Ziel for assistance in compiling the data used in this study and to the
Editor of this journal and two reviewers for a number of helpful suggestions. Jeffery Williamson kindly provided
the real wage data used in the study.
1. This study pools Irish census data for 32 counties for 1881, 1891, 1901, and 1911. In this sense the
approach is considerably different than that taken in the present study.
2. The econometric approach used in this paper requires a “between estimator,” which is based on group
means. Thus, the maximum number of independent variables in any regression is limited to 11.
3. Bureau of Statistics (1903), 4358-4360.
4. Immigrant numbers for years toward the end of the 26-year period were low in general.
5. Ideally, we would like to have data on the relative wage of younger compared to older individuals, but in the
absence of such data, country-wide measures must be employed.
6. Following Greenwood (1969) and Dunlevy and Gemery (1977), I also included a migrant stock variable in
the estimated regressions (an estimate of the number of individuals born in country i and enumerated in the U.S.
census in year t, measured both in absolute terms and relative to the source-country population, as in Hatton and
Williamson (1998)), but this variable was never statistically significant.
7. The rate of natural increase lagged 20 years was never significant and therefore is not included in the
model. Following Easterlin (1961), this variable has frequently been used in studies of transatlantic migration.
8. U.S. Department of Labor (1914), p.18.
9. U.S. Treasury Department (1892), p.183 and p.220.
10. In an effort to determine whether potential migrants in Southern Europe were more responsive to wage
differences and/or job opportunities than those elsewhere in Europe, I interacted the Southern Europe dummy with
both the relative wage variable and the GDP growth variable. With one exception these interactions were not
significant. The exception is in the emigration-rate regression for the over 40 group, where the interaction is
positive, suggesting that this group in Southern Europe was less deterred by wage differences than its counterparts in
other parts of Europe.
19
Table 1
Age Composition of U.S. immigration from 12 source countries, 1873 and 1898
_________________________________________________________________________________________________________
Total Immigrants
Percent aged 15-40
Percent aged over 40
Country
1873-98
1873
1898
1873
1898
1873
1898
Belgium
43,651
1,176
695
72.9
63.3
9.6
18.4
Denmark
161,252
4,931
1,946
69.0
79.1
11.4
9.4
France
142,810
14,798
1,990
64.8
72.7
19.1
15.9
Germany
2,455,519
149,671
17,112
61.6
70.2
12.9
12.3
Ireland
1,304,275
77,344
25,128
73.2
88.7
9.3
6.2
Italy
834,202
8,757
58,613
67.4
64.2
15.8
15.4
Netherlands
96,392
3,811
767
50.8
58.9
15.3
15.4
Norway
330,054
16,247
4,938
46.7
79.6
29.7
10.2
Spain
14,515
541
577
74.5
71.9
13.5
21.5
Sweden
682,767
14,303
12,398
62.8
84.0
15.6
7.6
Great Britain
1,502,759
89,482
12,893
62.8
67.1
14.3
15.9
Portugal
22,759
24
1,717
70.8
63.1
16.7
12.6
Source: Bureau of Statistics (1903, 4358-4360).
20
Table 2
Means and Standard Deviations
______________________________________________________________________________________
Variable
Mean
Std. dev.
Relative real wage, t-1 (i/US)
0.424
0.152
Relative growth of GDP, avg. t-1 to t-3 (US/i)
1.020
0.035
Percent manufacturing in i
0.237
0.068
Percent agriculture in i
0.475
0.188
Total migration prior 2 years (x103)
24.751
35.789
Birthrate in i (per thousand)
31.612
4.465
English spoken in i
0.167
0.373
Southern Europe
0.250
0.434
Distance from i to U.S. (x103)
3.670
0.315
Population of i, 15-39 (x106)
6.192
6.084
Population of i, over 39 (x106)
4.593
4.528
Percent pop. 15 and over that was 15-39 (x102)
0.379
0.013
Migration rate, 15-40 (x10-3)
5.210
7.490
Migration rate, over 40 (x10-3)
0.844
1.158
Percent of immigrants 15 and over that was 15-40
85.771
6.073
Sources: Relative (international) real wage (Williamson, 1995); relative growth of GDP (Maddison, 2003); percent
manufacturing in i (Mitchell, 1992); percent agriculture in i (Mitchell); total migration prior 2 years (Bureau of
Statistics, 1903); birthrate in i (Mitchell, 1992); population of i, and also by age (Mitchell, 1992); migration rate, by
age (Bureau of Statistics, 1903; Mitchell, 1992). Annual data were developed by linear interpolation and, where
necessary, by (backward) extrapolation.
21
Table 3
Emigration rates and the age composition of U.S.-bound migrants 15 to 40 and over 40, 1873-1898:
Hausman-Taylor instrumental variable estimates
M i g r a t i o n R a t e s
Age Composition
Variable
15-40
Over 40
15-40
Differential econ. Opportunity
Relative real wage, t-1 (x1)
-6.391
-4.657
13.653
(1.249)
(4.836)
(2.008)
Relative growth of GDP (x1)
13.422
1.405
-6.473
(1.960)
(1.120)
(0.746)
Percent manufacturing in i (x1)
2.602
-6.975
68.107
(0.264)
(3.710)
(4.995)
Percent agriculture in i (x1)
12.820
-4.593
24.112
(2.472)
(4.399)
(3.068)
Cost of Migrating
Total migration prior 2 years (x2)
0.060
0.012
0.005
(5.718)
(6.179)
(0.410)
Birthrate in i (x2)
-0.632
-0.016
-1.106
(4.698)
(0.587)
(5.537)
English spoken in i (z2)
9.061
1.480
-4.011
(2.378)
(1.601)
(0.424)
Southern Europe (z1)
-5.928
-1.156
3.829
(2.185)
(1.835)
(0.626)
Distance from i (z1)
4.655
1.585
-1.789
(1.072)
(1.516)
(0.167)
Control variables
Population 15-39 (over 39) of i (x1)
-0.416
-0.129
0.060
(2.597)
(2.530)
(0.446)
Constant (z1)
-8.521
0.236
99.117
(0.514)
(0.060)
(2.521)
Test for exogeneity of HT instruments
Note. Absolute t values in parentheses.
0.570
0.999
0.829
22
Expected Returns (R)
and Costs (C)
C΄ij
Cij
R΄ij
Rij
0
α2
α0
α1
Age at Migration (α)
Fig. 1. The marginal migrant—The expected effects
of returns and costs on age at migration.