Clark, Ximena, Timothy J. Hatton, and Jeffery G. Williamson. "What Explains Emigration Out of Latin America?" World Development 32 (2004): 1871-90. Web. 5 Apr. 2011.
1. EMIGRATION FROM LATIN AMERICA: THE ISSUES
What explains emigration from Latin America? What accounts for the differences in rates of
emigration from Latin America compared with those from other sending regions such as Asia and Africa?
Why do cross-border migration rates vary so much across Latin America?
Currently, Latin America has the highest emigration rate in the world (Figure 1), and the region
has risen to that position of prominence in four waves. The first wave involved small numbers leading
Iberian colonization, 500 years ago. The second wave was far bigger and it involved coerced black
Africans arriving in slave ships, especially during the 18th century and early 19th centuries before the
British navy intervened to shut that slave trade down. Slave destinations were mostly tropical, the
Caribbean and Brazil leading the way. The third wave involved ‘free’ European labor arriving during the
age of mass migration from about 1870 to 1940. European immigrant destinations were mostly Argentina,
Brazil and Venezuela.
[Figure 1 about here]
The fourth wave began after the 1960s, and it has been very different from what preceded it. First,
the past thirty or forty years have been ones of net emigration, not, as in the past four centuries, net
immigration. This experience highlights Latin American exceptionalism. Typically, countries pass
through what might be called life cycle cross-border migration experience (Hatton and Williamson 2004:
Chp. 4). Pre-industrial countries are too poor to enable their workers to find the resources to invest in long
distance migration to high-wage labor markets abroad. As development takes place at home, the
emigration constraint is gradually released: even if the sending country undergoes some catching up on
the high-wage host country, emigration surges. At some point, the constraint ceases to bind, and further
reductions in the wage gap causes emigration to fall off. Indeed, as the country graduates to a high-wage
industrial status, it begins to absorb immigrants. Symmetrically, high-wage New World countries undergo
the same life cycle, except that they begin with modest immigration rates that then surge to a peak before
falling off as sending Old World regions catch up. Latin America is an exception to these life cycle rules.
After absorbing immigrants for centuries, it is now a major source of emigrants. Second, Latin American
emigrations, like that of other sending regions, have been constrained by restrictive policy in high-wage
destinations. This was not true, of course, in the age of free migration before WWI. But it was also not
true when the US first imposed quotas after WWI. Indeed, the US Immigration Act of 1921 had a prowestern
hemisphere bias since such immigration was not subject to the quotas, perhaps because US
authorities thought there was no reason to set quotas since Latin Americans were too poor and too distant
to become candidates for US immigration. If they thought so, they were wrong since the share of US
immigrants coming from Mexico alone was already 15.1 percent in 1920-1929 (Hatton and Williamson
2004: Table 9.1), the numbers rising by 319,824 over that decade, from 135,678 in the 1910s to 455,502
in the 1920s. Equally important, Mexicans acted the way host countries always want guestworkers to act,
fleeing US labor markets when they went sour in the 1930s (75,240 Mexicans left the US during the
depression decade). Latin American immigration quotas were a post WWII phenomenon.
Third, while intra-regional migration within Latin America has never been extensive, recent
evidence suggests the potential for much more (see Tables 1 and 2). For example, the rise of world oil
prices, political events in some sending countries (e.g. coups d’état), along with explicit domestic
government policies favoring skilled workers pulled many Latin Americans to Venezuela in the 70s and
80s, in particular from Colombia and Chile.1 This increased the share of Latin American immigrants in
the Venezuelan population from 2.1 percent in 1971 to 3.7 percent in 1991 (with a peak of 4.4 percent in
1981). 2 Although recent economic and political events have undermined the attractiveness of Venezuela
as a destination country for intra-Latin American emigrants, it continues to be an important destination for
Colombians today. Out of 1 million foreign-born in 2001, Colombians represented 60 percent, and the
number of Colombians in Venezuela was bigger than in the United States. To take another intra-regional
example, Argentina has also attracted many Latin American immigrants, mainly from the rest of the
1 The Chilean population in Venezuela increased from 3,000 in 1971 to 25,000 in 1981, but decreased subsequently, reaching
15,000 in 2001. Colombians living in Venezuela increased from 177,000 in 1971 to 494,000 in 1981. These totals continued to
expand thereafter (although at a slower pace), reaching 600,000 in 2001.
2 According to the latest census, the share of Latin American immigrants in Venezuela has decreased to 3.1 percent (although the
share of Latin Americans in the total foreign-born population increased).
southern cone. During periods of good economic performance, Argentina has become a southern magnet,
effectively competing with the US for immigrants in a region where distance and the poverty constraint in
the sending countries tend to diminish the North American pull. Argentina is certainly the principal
destination for Bolivian emigrants,3 many of whom enter Argentina for the first time as an illegal. In
2001, the number of Bolivians living in Argentina (233,000) was more than four times the number living
in the US. Despite the recent economic downturn, Argentina also continues to be the main destination for
Paraguayan emigrants, that destination accounting in 2001 for 90 percent of total Paraguayans living
elsewhere in Latin America, and they accounted for 32 percent of Argentina’s 1 million Latin American
immigrants. What is common to both examples, of course, is the much closer proximity of Argentina
compared with the US, and the importance of poverty constraints on long distance moves for Bolivian
and Paraguayan emigrants.
[Tables 1 and 2 about here]
Costa Rica has been the traditional regional magnet in Central America, with Nicaraguans being
the main immigrants. Out of the 3.8 million Costa Rican residents in the 2000, roughly 8 percent were
foreign-born, of which 76 percent were Nicaraguans (226,000).4 These Nicaraguans were low-skilled and
low-schooled: 68 percent had less than 7 years of education and a significant share was working in
agriculture. Belize is another regional magnet, attracting high shares of Central Americans, especially
Guatemalans (for which Belize is the second most important Latin American destination after Mexico)
and Salvadorans. Guatemalan immigrants (roughly 15,000) accounted for 6 percent of Belize’s population
and for 43 percent of its total foreign-born. However, Belize has been considered a transit country for
Central American migrants on their way to Mexico or the US.
Although the US has never lost its importance as the principal destination for Latin American
emigrants, they have started to explore new options, such as Australia, Canada, and some European
countries (Spain, Italy and the UK). In addition, adverse economic conditions in Latin America and
3 Out of the total of Bolivians emigrants living elsewhere in Latin America in 2001, 87 percent were in Argentina.
4 The impact of this movement led Costa Rica and Nicaragua to sign an agreement in 1993 to regulate cross border migration.
improved conditions in Europe contributed to some return migration on the part of descendants of
previous European immigrants.5 In 2001, 21 percent (240,000) of all foreign-born in Spain had their
origin in just 6 Latin American countries (Ecuador, Colombia, Peru, Dominican Republic, Cuba, and
Argentina).6
In spite of their exploration of these and other host country alternatives, Latin American
emigrants have increasingly favored the US as their prime destination, and especially since the early
1970s. Of all emigrants from Latin America and the Caribbean countries, the share going to the US was
56 percent in 1960 and 1970, 65 percent in 1980 and 75 percent in 1990 (Pellegrino 2002: p. 27). By
2000, more than 55 percent of all immigrants in the US were Latin Americans, and about half of these
were Mexicans.7 This fact has led both governments to discuss seriously how to manage more orderly
migrations across their common border.
Thus, with the exception of a few countries, Latin American emigration has typically been out of
the region rather than across borders within the region. Cross-border emigration within Latin America has
always been relatively modest, and, furthermore, it has been falling sharply and continuously over the
past half century: 44 percent of all emigrants from Latin America and the Caribbean countries moved to
other countries within the region in 1960, 42 percent in 1970, 30 percent in 1980 and only 21 percent in
1990 (Pellegrino 2002: p. 27). Tables 1 and 2 add country detail for the most recent decade. To begin
with, except for traditional local migrant-magnets like Argentina and Venezuela (and more recently Costa
Rica and Belize), low-wage Latin America has far smaller foreign-born shares than does the high-wage
labor market up north. The foreign-born share for Canada and the US combined is 11.6 percent, while
even the two local Latin American magnets recorded much lower foreign-born shares – Argentina and
Venezuela each with 4.1 percent. Even Costa Rica and Chile, two relatively new local magnets, have
5 In much of Europe, and under certain conditions, children and grandchildren of previous European emigrants can become
citizens of the origin sending country.
6 It is worth noting that Ecuador and Colombia have signed an agreement with Spain regarding both regularization of the
undocumented and promotion of some labor mobility to meet labor shortages in Spain.
7 The IRCA legalization act of 1986 (which took effect principally in 1990-91) was intended to regulate illegal immigration, and
it contributed to this large increase of Mexicans in the US. In 1990 and 1991, Mexican immigrant admittances in the US reached
679,000 and 946,000 respectively, with a significant share responding to the IRCA program.
shares of only 7.8 percent and 1.3 percent respectively. The share for all of Latin America is only 1.2
percent; and that for Latin America less Argentina and Venezuela is a tiny 0.6 percent. Tables 1 and 2 can
be used in another way to document the dominance of emigration from the region relative to intra-Latin
American cross-border migration. Of Latin Americans living in the Americas but outside their country of
birth around the year 2000, 86 percent were in Canada and the US while only 14 percent were in their
home region. And the share in the US and Canada has been increasing.
In short, Latin American cross-border emigration is dominated by movements to high-wage labor
markets in North America, not by movements to low-wage labor markets in the region itself. Unless Latin
America starts some significant economic ‘catch up’ on the US over the next few decades, cross-border
migration within Latin America is unlikely to rise in relative importance. Thus, the rest of this paper will
focus on emigration out of Latin America, and on emigration to the US in particular.8
3. WHO EMIGRATES FROM LATIN AMERICA?
Table 3 presents the age structure of immigrants entering the US between 1997 and 2001. There
we focus on Latin American (LA) immigrants and compare their age structure with that of the populations
in their countries of origin. The immigrants have been divided into three age groups: 0-14, 15-64 (a range
representing the active adult population, which in turn could be used as a proxy for those in the labor
force), and 65 years and older. Table 3 suggests no significant difference between LA immigrants and the
rest of the world, showing that 76-77 percent of US immigrants fall into the 15-64 range. Now compare
the share of the LA immigrants adult with that of their sending country, 61 percent, implying that LA
immigrants entering the US were much more likely to be labor-market-oriented adults than was true of
their home populations, for a difference of more than 15 percentage points. This self-selection of laborforce-
oriented adults has been true of international migrations since the early 19th century (Hatton and
8 There is another reason for using US immigration data to explore these issues. Time series data documenting Latin American
net emigration is still of poor quality, and that documenting pair-wise movements is even worse.
Williamson 1998: pp. 11-12), and, although Table 3 does not document it, migrations have always selfselected
young adults (Williamson 2001). What is true of sending Latin America is also true of receiving
US: the US population aged 15-64 accounted for 66 percent of the total population. In short, Latin
American immigrants were much more likely to be adults active in the labor force than the populations
they left or joined: their migration clearly served to reduce labor supply in their home country and
increase it in the US. Finally, the fact that these immigrants tended to be adults (and probably young
adults) suggests that they were responding primarily to labor market forces, rather than to political
instability or violence at home. Immigrants fleeing political conditions at home tend to move as a family,
while immigrants responding to labor market signals are those of economically-active ages.
[Table 3 about here]
Having generalized about Latin American immigrants in the US, note the considerable range
within Latin America. Cuba, Dominican Republic, El Salvador, Colombia, Nicaragua and Peru are among
the Latin American countries that sent immigrants with the highest adult shares. Perhaps more relevant,
however, is the difference between the adult share of immigrants and that of the sending country. Here,
the really large numbers are for Nicaragua (a huge 35.4 percent difference), El Salvador (23.1 percent),
Guatemala (20.8 percent), Honduras (20.5 percent) and Haiti (18 percent). These five manifested
especially strong self-selection of adults. Those that had the weakest adult self-selection were Cuba and
Mexico, both recording a 12.4 percent difference between immigrant and home adult shares. We think
that this difference is likely to be explained by the combination of two events: first, whether a country’s
emigration experience started early or late (both Cuba and Mexico have the oldest US immigration
experience stretching back to the 1950s and 1960s); and second, the “family reunification effect” which
became an important part of US immigration policy from the late 1960s.9
Table 4 compares the educational composition of Latin Americans living in the US with that of
the population in their countries of origin. The self-selection by skill appears to be even more dramatic
9 The 1965 Amendments to the Immigration Act replaced the pre-existing country of origin quotas. Under the seven preference
categories, first applied to the Eastern Hemisphere only and extended to the Western Hemisphere in 1978, nearly two thirds of
than by age. The table shows the percentage of adults over 25 years having secondary or tertiary
educational levels.10 It is quite clear that Latin Americans living legally in the US have, on average,
considerably higher levels of education than is true for populations at home.11 In particular, if we consider
the percentage of people having completed at most secondary education (first and third columns), those
shares for LA in the US are more than double (sometimes triple) those in the origin country. This pattern
holds even if we assume that secondary figures for the origin countries refer only to the highest level
attained (as opposed to completed), as in the fifth column.12
[Table 4 about here]
While Latin American immigrants to the US are more educated and more likely to be
economically active than their compatriots left at home, they are less educated, on average, than the
Americans they join. To the extent that education helps predict income, the inference here is that Latin
American emigration creates more earnings inequality both at home and in the United States, but creates
less earnings inequality for the Americas as a whole.
Inference is one thing, and evidence is another. So, does emigration from Latin America reduce
or increase poverty rates and inequality there? Immigrants into the United States entered at the bottom of
the urban income distribution before World War I (Hatton and Williamson 1998: Chps. 7-11) and the
same seems to be true of Latin American immigrants today. But did they and do they leave from the
bottom of the income distribution in the sending regions? Based on this educational attainment and age
data, it looks like the answer is no.13 Indeed, we would have been surprised by any other answer since
history tells us that world migration has always been financially constrained. Whether legal or illegal,
international migration is expensive, and really poor people can’t afford it. Consider the case of Mexico, a
country that supplied almost a third of US immigration in the 1990s. Daniel Chiquiar and Gordon Hanson
the available visas were allocated to relatives of US citizens or residents. This excludes immediate relatives who did not come
under the worldwide quota until the 1990 Immigration Act.
10 Puerto Rico and Cuba were excluded from Table 4 due to unavailability of information from Barro-Lee (2000).
11 We must note however that these numbers do not take into account that more than 82 percent of illegal immigrants in the US
come from Latin America (in particular from Mexico, 69 percent) and that this fact could influence some of these patterns.
12 These educational self-selection patterns are less dramatic for tertiary educational achievement.
13 Borjas (1987) and the Roy (1951) model not withstanding.
(2002) have recently shown the following: starting with the facts we have seen in Table 4 -- Mexican
immigrants are much less educated than the average United States citizen, but they are more educated
than the average Mexican -- Mexican immigrants living in the US would have fallen mostly in the middle
and upper portions of the Mexican wage distribution, not at the bottom; and, thus, Mexican immigration
has raised income inequality at home (Chiquiar and Hanson 2002: pp. 3-4). While Mexican emigration
has certainly improved the lives of those who have moved, and while it must have served to reduce North
American inequality, it does not appear to have reduced poverty rates much in Mexico. We need to
establish whether these Mexican findings apply to the rest of Latin America.
We are aware that having lots of relatively skilled LA immigrants in the US seems to signal a
brain drain problem if these immigrants represent an important share of the skilled labor force in their
home country, if they remit at low rates, if they never return, or if their departure fails to provoke a
positive schooling response at home. But if they do remit at high rates, and/or return and/or create
business links with their home countries and/or provoke more schooling at home, a current brain drain
might well turn in to a future brain gain. One important aspect of the brain drain debate is typically
ignored, namely that successful young immigrants in the US accumulate human capital at very fast rates,
thus tending to catch-up on natives. Individuals migrate to the US for at least two reasons, to exploit the
wage gap given skills, and to accumulate more skills in a host country labor market where those skills
may be rewarded more generously than is true back home. Much depends on the size of the skill premium
in host and sending countries, something we shall have more to say about in the next section.
4. EXPLAINING EMIGRATION TO THE UNITED STATES BY SOURCE COUNTRY
Why focus on US immigration? Part of the answer lies with Table 5 where those born in Latin
America but living in other world regions is documented for the year 2000. While significant numbers are
living in Europe, the overwhelming majority (88 percent) are living in the United States. Thus, explaining
emigration out of Latin America is largely a matter of explaining emigration to the United States.
Table 6 reports annual flows of legal immigrants into the United States since 1971. The Latin American total per decade rose two and a half times between 1971-80 and 1991-2000, from 1.8 to 4.3 million. Although US
immigration from all regions increased markedly over these three decades, the numbers arriving from
Latin America far exceed those arriving from Europe and Africa. And while the numbers arriving from
Latin America in the 1970s exceeded those for Asia by only a small margin, by the 1990s they were
bigger by about 70 percent. Legal immigration is, of course, only part of the story; some of the Latin
American totals in the late 1980s and early 1990s were legalizations of previously illegal immigrants.14
Estimates for the stock of illegal immigrants in the United States (Table 7) show that these are even more
concentrated among Latin Americans and that the numbers doubled between 1990 and 2000.
[Tables 5-7 about here]
Examining the determinants of US immigration rates from a variety of countries can shed light on
what drives emigration from Latin America as compared with other source regions. It also allows us to
see whether there are important differences between the factors that drive Latin American emigration to
the US and those that operate in other parts of the world. Here we draw on our earlier research on the
determinants of US immigration.15 Our database includes 81 source countries (for which explanatory
variables can also be documented), 22 of which are in Latin America, covering the 28 years from 1971 to
1998. These 81 source countries accounted for 82.5 percent of all US immigration during this period. The
dependent variable is the number of (legal) immigrants accepted into the United States who were born in
some sending country as a proportion of that country’s population. Since this rate is bounded at zero, we
use the log of the rate in our regressions. We also use random effects regressions, a method which
exploits both the cross-section and time-series variation in the data. The results appear in Table 8.16
[Table 8 about here]
14 Under the Immigration Control and Reform Act of 1986, 2.7 million were legalized, mainly in 1989-91, most of whom were
from Latin America.
15Clark, Hatton and Williamson (2002). That previous paper also describes in detail the sources of our data and the variables
included in the data base.
16For the Latin American sample (n=22) in Table 6, the Hausmann test statistic for the random effects model is 12.49, i.e. the RE
model passes the test comfortably.
1 1
The first regression equation is estimated including all 81 source countries while the second
regression is estimated only for the 22 Latin American countries in the sample. We focus first on the
estimates for the complete, world 81-country sample. The explanatory variables are those suggested by
theories of migration, explained at greater length elsewhere (Clark, Hatton and Williamson 2002; Hatton
and Williamson 2002, 2004).
Relative income between the sending country and the US is, of course, central to the migration
decision and it is represented in Table 8 by the ratio of the source country purchasing-power-parityadjusted
GDP per capita to that of the US. The coefficient is negative as expected and highly significant.
Furthermore, the coefficient implies that a ten percent rise in US income (the leader surging ahead), or a
ten percent fall in source country income (the follower falling behind), leads to a 15 percent rise in
immigration from that country. However, the migration decision also depends on the relative return to
skills, not just income differentials, and hence the income effect must be ‘deflated’ by relative skill levels.
Here we use the ratio of the number of years of education of those aged 15 and over in the source country
relative to the US, and, as expected, this variable has a negative impact on immigration. Holding income
constant, a rise in the source country’s average education level by 10 percent (equivalent to 0.55 years of
schooling averaged across all sending countries) reduces the immigration rate by 7 percent. We are not
asserting that more education diminishes emigration rates, but rather only that GDP per capita
differentials are explained in part by schooling differentials, and that this factor must be accommodated in
the analysis. If income differentials were instead documented by earnings differentials for individuals at
common levels of schooling, then we might well get different results on the schooling variable.
Migration theory also suggests that differences across countries in the return to skills will select
migrants from different parts of the skill distribution (Borjas 1987, 1991), what has come to be called the
Roy model (Roy 1951). These Roy model effects are captured here by the ratio of the Gini coefficients, a
statistic describing the distribution of household income, source country relative to the US. If a sending
country has more inequality than the US (an unlikely case for most sending regions except Latin
America), then those at the top of the income distribution will have less incentive to emigrate, while those
at the bottom will have more. If instead a sending country has less inequality than the US, then those at
the bottom will have less incentive to immigrate, while those at the top will have more.17 If inequality is
similar in source and destination, then, provided that the source has a higher average income, there is an
incentive to emigrate throughout the income distribution. Thus, unless poverty constrains poor potential
immigrants from leaving the sending country, the immigration rate to the US should follow an inverted U
shaped function of relative inequality.
The results in Table 8 strongly support the Roy hypothesis, with the peak immigration rate
occurring at a ratio of 1.12, very close to the point where inequality in the destination and source country
are the same. We shall return to this finding below, since, in the Latin American case, it may need to be
reinterpreted in terms of the qualification in italics, namely, poverty is likely to constrain the emigration
of the very poor residing at the bottom of the income distribution in the sending country.
Unless return migration is very inexpensive,18 the discounted present value attached to any long
distance move should be higher at younger ages since the returns are spread over a longer future working
life. Thus, source countries with larger cohorts of young people should generate more migrants and higher
emigration rates. In Table 8, the coefficient on the share of population aged 15-29 is positive as expected
but it is not significant, at least for the full 81 country sample. We have no explanation for this surprising
result, especially, as we shall see in a moment, given the opposite is true for Latin America alone. Most
observers also stress what has come to be called the ‘friends and relatives effect.’ An established stock of
previous migrants from the same source generates network effects that lower the costs and reduce the
risks of migration, and, through remittances by previous migrants, may even supply the initial investment
necessary to finance the move by new migrants. Table 8 documents this ‘friends and relatives effect’ by
using the stock of those residing in the US but born in the source country, per thousand of the source
17 Perhaps the best documented Latin American example is Uruguay, an egalitarian country with a ubiquitous and deep
commitment to education. The 1996 census recorded 26.7 percent in Uruguay with schooling equal to or greater than 10 years.
However, the same figure as applied to emigrants from Uruguay was 81.2 percent in the US in 1990 (Pellegrino 2002: p. 34).
18 Under these conditions, migrations are less likely to be permanent and are more likely to repeat.
population, lagged one period.19 Since this effect is sometimes thought to be nonlinear, the squared term
is also included. The pattern of coefficients reported in Table 8 implies that the stock effect is most
powerful at low levels and that it diminishes as the stock increases. At the average stock/population ratio
in our data, the effect of raising the expatriate stock by 1000 is to generate an additional annual inflow
from the source country of about 10 immigrants per year. A very powerful influence indeed, implying
very strong historical persistence.
Other country characteristics also matter and a very important one is distance from the
destination. This gravity effect is measured in Table 8 by the great circle distance from Chicago in
thousands of miles. The coefficient indicates that an additional thousand miles between sending country
and the US reduces the immigration rate by 21 percent.20 Whether the country is landlocked also has a
large negative effect, although it is not quite significant at conventional levels. Even more important is
whether the source country is English speaking, a factor which increases the number of immigrants from
the sending country almost three fold. Political upheavals and violence, the most important source of
which are civil wars, also have a significant effect, increasing the number of immigrants to the US by
about 22 percent.
Clearly, US immigration policy also matters in determining immigrant source and Table 8
accommodates this by a series of dummies. Prior to 1978 there were separate quotas for the Western
Hemisphere (chiefly Latin America) and the Eastern Hemisphere (the rest). The dummy for 1971-1978
reflects the merging of these two quotas into a worldwide quota in 1979. The effect from 1979 onwards
seems to have been positive, especially for the Eastern Hemisphere countries, despite some decrease in
the overall quota. The legalization of illegal immigrants (which is recorded as part of total immigration)
under the Immigration Reform and Control Act of 1986 (effective 1988) is captured by a variable that
represents the estimated stock of illegal immigrants by source country in 1980. This IRCA influence is
19 The lag is introduced under the premise that the ‘friends and relatives’ effect has to be in place before the migrant makes the
move. One reason to think that the marginal effect declines is that an increment to the stock adds little to migrant networks once
the stock gets very large relative to the source population.
applied only to the years 1989 to 1991 when the program was in effect.21 This effect varies across
countries with the largest impact being on Mexico where it served to double the immigration rate in those
years. The dummy for 1992 to 1998 (for all source countries) is intended to reflect the expansion in the
overall immigration quota that took effect following the Immigration Act of 1990 (effective 1992). This
increased the number of immigrants by 13 percent (as compared with the expansion of about 20 percent
in the quota for the Western Hemisphere).
Finally, regional dummies are also included. Here the excluded region is Western Europe so that
the coefficients on the other regions reflect differences from that benchmark. For the most part, the
coefficients on these regional dummies are small, suggesting that the ‘fundamentals’ can by themselves
explain regional differences in US immigration rates. Notable exceptions are the very large negative
intercept for Africa and the fairly large positive intercept for the Middle East. Within the Americas, there
are large coefficients for the border states -- Canada and Mexico, and for the Caribbean. While these
reflect the effects of contiguity they also reflect the place in the US (Chicago) from which distance is
measured.22
5. EXPLAINING LATIN AMERICAN EMIGRATION TO THE UNITED STATES
As Figures 2-4 document, cross-border migration rates vary enormously in Latin America. What
accounts for that variance? Are the same fundamentals at work that we have isolated for the world at
large?
[Figures 2-4 about here]
20 The gravity effect can be seen in Figure 2 where the lowest emigration (or highest immigration) rates are for the southern cone
(Argentina, Brazil, Chile and Paraguay), farthest from the US. Of course, these temperate areas have also always been relatively
high-wage.
21 The reason for using the estimated number of illegals in 1980 and not some later date is that legalization was offered to those
who had been living in the country since 1982 or longer and had previous work experience.
22 If distance was measured from Los Angeles rather than from Chicago, for example, the intercept for Canada would become
less negative and the intercept for Mexico would become less positive, but that for the Caribbean would become even more
positive.
The second column in Table 8 reports estimates for the 22 Latin American countries alone. These
are remarkably similar to those in the first column for all 81 countries. Hence, it appears that in general
Latin American emigration to the United States is driven by the same forces as for US immigration as a
whole, although the forces themselves may, of course, be larger or smaller. However, there are some
differences in the magnitudes of the estimated coefficients that are worth stressing. The most notable
difference between Latin America and the rest is the large and now significant effect of the share of
population aged 15-29. This regional difference may be explained by the fact that long distance moves
from Asia and Africa may be more typically family affairs, an issue that future research needs to explore.
In any case, this result implies that an increase in the proportion of the population aged 15-29 from, say,
25 to 30 percent would increase the typical Latin American country’s US immigration rate by 20 percent.
Furthermore, the coefficients on income and schooling differences are larger for Latin America. Thus a
ten percent increase in US relative income increases immigration from the typical Latin American country
by 25 percent and a ten percent increase in US relative education reduces it by 11 percent, much bigger
effects than we saw for all sending regions combined.
The other coefficients are very similar to those estimated on the full set of countries with the
exception of the dummy for 1992-98, which is negative. This may reflect the effect of the increased
favorable weight given to skills in the 1990 Immigration Act or it may simply reflect an increasing
number of Latin Americans choosing an illegal entry into the US (Table 7).
Inequality effects are also more powerful for Latin America, but the maximum immigration rate,
where the Gini coefficient ratio is 1.26, still fairly close to one. This inverse U shape implies that
immigration to the US is lower from those Latin American countries that are very equal or very unequal
compared to the US, and higher for those in between (that is, most like the US). Note, however, that this
is not quite the same as saying that Latin American immigrants into the US came from middle income
groups within countries. Consider the following possibility: Because Latin American income distributions
are more unequal than the US, migration should select from the bottom of the sending country’s
distribution, that is, it should be mostly the very poor who move. However, there is also the ‘poverty trap’
to consider. First, the very poor are unlikely to have the resources necessary to invest in the long distance
move to the United States. Second, and in addition, roughly constant absolute costs of migration across
prospective migrants implies that these costs would be proportionately bigger the poorer the potential
migrant. If either or both of these ‘trap’ effects dominate, then higher poverty rates in the source country
should serve to diminish emigration from that country to the US. Thus, while the migration incentives
may be very great for those at the bottom of the distribution, poverty makes it impossible to move. At the
top of the income distribution, there may be no financial constraint on emigration, but there will also be
far less incentive to move. Thus, it may be those in the middle of the sending country’s distribution that
actually emigrate. As we have seen, this is exactly what Chiquiar and Hanson (2002) have found for
Mexico in the 1990s when US immigrants came from the middle and top of the Mexican wage
distribution, and thus from the middle of the income distribution. This issue is important, and it must be
resolved with more research if we want to know whether more liberal US immigration policy will help
diminish poverty rates in Latin America.
5. LATIN AMERICA AND THE ‘SOURCES’ OF EMIGRATION TO THE UNITED STATES
Latin American emigration to the United States far exceeds that of other regions. This dominance
largely reflects higher rates of Latin American emigration to the US as compared with other source areas.
As we have seen, there are also significant differences across countries. In order to explore some of these
differences, we use our estimated equations in Table 8 to decompose the observed differences in log
immigration rates into ‘sources’ emanating from differences in the fundamentals driving immigration.
We first compare Latin America emigration rates to the US with those from other sending regions
using the parameter estimates from the first column of Table 8 for the full 81 countries. The first row of
Table 9 shows the average value of the log emigration rate from Latin America to the US minus the
average value of the log emigration rate from Europe, Africa and Asia to the US, multiplied by 100. Thus,
over the 28 years from 1971 to 1998, Latin America had, on average, emigration rates to the US that were
171 log points higher than the average for Europe, 183 higher than Asia and 317 higher than Africa. What
accounts for these differences?
[Table 9 about here]
Let us start with the European comparison, where the most important variable is income. Almost
a third of the log emigration rate difference between Latin America and Europe (50.7 log points) is
explained by lower per capita income in Latin America as compared with Europe. Lower education levels
in Latin America partly offset this effect. Inequality contributes very little because, although Latin
American inequality is higher and European inequality is lower than the US, inequality in Latin America
is closer to the US level and hence nearer to the maximum of the inverted U. Demographic structure
accounts for surprisingly little of the difference as does the immigrant stock of the sending country in the
US. Most important among the other fundamentals that favor Latin America is distance, which accounts
for 46 log points, and a higher incidence of English speaking. By themselves, these two fundamentals
explain more than a third of the difference between European and Latin American emigration rates to the
US.
Latin American emigration rates to the US are higher by a massive 317 log points than those of
the 14 African countries in our sample. Per capita income and schooling now have the reverse effect since
both are considerably lower in Africa than in Latin America. The most important variables accounting for
the difference in emigration rates to the US for these two regions are distance (positive) and English
speaking (negative). While most of the difference in average emigration remains unexplained by the nine
fundamentals included in the analysis, one excluded fundamental that is likely to have played a powerful
role is poverty. In contrast with Latin America, we know that Africans are highly mobile within Africa
(Hatton and Williamson 2003). Thus, it may be that poverty levels are just too high to permit mass
migrations out of Africa over longer distances. The comparison with the 19 sampled Asian countries
(some of which are in the Middle East) is also interesting. With the exception of the immigrant stock, all
but one of the remaining fundamentals explain very little of the 183 log point difference between Latin
American and Asian emigration rates. The fundamental that matters most is distance, and it accounts for
80 percent of the 183 log point difference. Thus, with the exception of distance, these Asian countries are,
on average, very similar to those in Latin America.
6. COMPARISONS ACROSS LATIN AMERICA
Table 10 searches for causes of the wide variance in emigration rates to the US across Latin
America. Much like Table 9, this one looks for the sources of the variance in the log emigration rate.
These sources are, once again, calculated using the second regression in Table 8 that applied to Latin
American countries only. The variable being explained is the deviation in the country emigration rate
from the Latin American (unweighted) average, again in log points. Thus, the Mexican emigration rate to
the US is higher by 81 log points than the Latin American average. Similarly, the fundamentals are
calculated as country deviations from the Latin American average. For countries like Argentina,
Barbados, Mexico, Trinidad, Uruguay and Venezuela, the fact that per capita incomes are quite a bit
higher than the Latin American average help diminish their emigration rates significantly. High relative
schooling levels push up the emigration rates significantly for countries such as Argentina, Barbados,
Chile, Panama, Trinidad and Uruguay. Except for Brazil and Honduras, where their income distributions
are very unequal, relative inequality contributes little to differences within Latin America.
[Table 10 about here]
Other fundamentals matter far more, and, once again, distance is certainly one of them. The large
positive effects for Mexico, the Caribbean and Central America contrast with the large negative effects
for countries further south such as Argentina, Bolivia, Brazil, Chile, Paraguay, Peru and Uruguay.
Relatively large accumulated stocks of immigrants living in the US also help to push up emigration rates
for the small Caribbean countries. Thus, there is very strong historical persistence in Latin American
emigration to the US; countries with an immigrant past in the US tend to have an immigrant present. But
even these large magnitudes are dwarfed by the massive effect of being English speaking as is true of
Barbados, Jamaica, Trinidad and Guyana. It should be noted, however, that there are quite a few countries
for which these fundamentals only explain a modest part of the observed differences in emigration rates.
To take one small and one large country as examples, those nine fundamentals explain only 60 percent of
the Barbados +230.7 point deviation from the Latin American average and only 44 percent of the Brazil
-352.9 point deviation. Other country-specific factors clearly matter, and ‘poverty traps’ may be part of
that missing explanation.
It is also worth asking how these emigration fundamentals contributed to changes in emigration
rates over the last thirty years, although we are prepared for large residuals given that the R2 in Table 8
for the time series is so much lower than for the cross section. Table 11 offers the decomposition, again
based on the regression for the Latin American countries alone. Here we examine contributions to the
change in the emigration rate between 1971-1973 and 1996-1998 in log points times 100.23 For Mexico
the effects of changes in schooling years and inequality relative to the US, as well as increases in the
young adult cohort and the immigrant stock, far exceed the actual rise in the emigration rate. Thus,
although there has been a substantial rise in emigration from Mexico to the US, based on the
fundamentals it should have been larger still. In other cases such as El Salvador, Nicaragua and Peru,
immigration to the US rose by far more than the fundamentals predict. While all three of these countries
have experienced political instability, our civil war measure (omitted in the table) cannot by itself explain
much of these long-term increases in emigration to the US. Still, rising civil conflict did contribute some
to rising emigration rates to the US for Colombia and Peru, and to falling rates for Guatemala.
[Table 11 about here]
While individual country experience differs from that predicted by the fundamentals, it is notable
that the majority of entries in Table 11 are positive. Thus, lagging income per capita growth relative to the
US was pushing up the emigration rate for 16 of the 22 countries. Furthermore, some of these laggingincome
effects were powerful, like for Argentina, Jamaica, Nicaragua, Peru, Venezuela and Trinidad.
Symmetrically, Chile’s good performance kept potential emigrants at home. On the other hand, education
levels increased relative to the US in 20 of the countries, led by Panama and Venezuela, and this added a further push to northward migration. The effects of changing relative inequality are mixed although they
are large for some countries such as Brazil, Honduras, Jamaica and Mexico. Where this effect is positive
(Brazil, Honduras and Mexico), this is chiefly because increasing inequality in the US has narrowed the
gap between the relatively unequal Latin American countries and the increasingly unequal United States.
The demographic effects are also mainly positive, since young adult shares rose in most of Latin America.
These demographic effects were fairly small in most countries, though they mattered a great deal in one
very important emigrating country, Mexico (accounting for 65 percent of the rise in the log emigration
rate). Finally there is that powerful contribution arising from the growth of the immigrant stock in the US,
particularly for the Caribbean and Central America. These migrant stock effects are actually the result of
fundamentals in the past that have contributed to present high immigration rates. High Latin American
emigration rates to the US have been sustained over the decades in large part due to these migrant stock
effects.
7. AN AGENDA
We think that there are a number of fruitful directions that future work on Latin American crossborder
migration could go. It seems to us that the first step should be to extend the analysis in this paper
to include the role of what we have called ‘poverty traps.’ If future successful development improved the
lot of the poor in Latin America, would that serve to increase or decrease the emigration rate? If OECD
policy became more friendly towards immigration – especially Latin American immigration – would that
serve to select poor or middle-income immigrants? Would it increase or reduce inequality in the sending
regions? The second step is to understand why cross-border migration within Latin America is so small
compared to migration out of the region. This appears to be much less true for Africa and Asia, and we
need to know why. Is it simply due to the closeness of a huge high-wage country, the US? Finally, we
need to sharpen our understanding of which forces have mattered most in raising the Latin American
emigration rate since the 1960s, especially the role of policy in the receiving regions. To repeat an
23 Note that the table excludes the impact of changing US immigration policy.
opening remark, all emigrant countries have passed through similar life-cycles over the past two
centuries. Very poor countries have very low emigration rates: the high incentive to leave is constrained
by poverty. As countries start to develop, these constraints are released and emigration rates soar. At
some middle point when they reach a more mature industrial status, emigration rates peak, falling
thereafter. Where does Latin America fit in this life-cycle? Will emigration to the US rise or fall over the
next few decades? And what will be the most important underlying forces at work? Can those forces be
easily influenced by policy? We need some answers, and a better understanding of the last few decades is
likely to supply them.
No comments:
Post a Comment