WPS4925
P olicy R eseaRch W oRking P aPeR 4925
International Migration, Transfers
of Norms and Home Country Fertility
Michel Beine
Frédéric Docquier
Maurice Schiff
The World Bank
Development Research Group
Trade Team
May 2009
Policy ReseaRch WoRking PaPeR 4925
Abstract
This paper examines the relationship between original and detailed data on migration. The results
international migration and source country fertility. The provide evidence of a significant transfer of fertility
impact of international migration on source country norms from migrants to their country of origin: a one
fertility may have a number of causes, including a percent decrease in the fertility norm to which migrants
transfer of destination countries' fertility norms and an are exposed reduces home country fertility by about 0.3
incentive to acquire more education. It provides provide percent for origin countries.
a rigorous test of the diffusion on of fertility norms using
This paper--a product of the Trade Team, Development Research Group--is part of a larger effort in the department to
understand the impact of international migration on migrants' countries of origin. Policy Research Working Papers are
also posted on the Web at http://econ.worldbank.org. The contact person's email is msewadeh1@worldbank.org.
The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.
Produced by the Research Support Team
International Migration, Transfers of
Norms and Home Country Fertility
Michel Beinea , Frédéric Docquierb and Maurice Schi¤c
a
University of Luxemburg and CES-Ifo
b
FNRS and IRES, Université Catholique de Louvain, IZA-Bonn and CReAM-London
c
World Bank, University of Chile, and IZA-Bonn
May 2009
Abstract
1
Abstract. This paper examines the relationship between international migration
and source country fertility. The impact of international migration on source country
fertility may have a number of causes, including a transfer of destination countries'
fertility norms and an incentive to acquire more education. We provide a rigorous test
of the di¤usion of fertility norms using original and detailed data on migration. Our
results provide evidence of a signi...cant transfer of fertility norms from migrants to
their country of origin: a one percent decrease in the fertility norm to which migrants
are exposed reduces home country fertility by about 0.3 percent for origin countries.
JEL classi...cation: J13, J61, O11.
Keywords: International migration, endogenous fertility, human capital, social
norms.
Acknowledgement. This paper was written while Michel Beine was visiting the
World Bank, DECRG Trade Unit. Thanks are due to Chris Parsons for providing
us with the data and to Assaf Razin, Hillel Rapoport, Caglar Ozden, Ramón Lopez,
Paola Conconi, Antonio Spilimbergo and participants of the OECD-CEPII conference
on Brain Drain in Paris and of the TOM conference in Louvain for their useful
comments. The views in this paper are those of the authors and do not necessarily
ect
re those of their respective institutions, the World Bank Executive Directors and
the governments they represent. The second author acknowledges ...nancial support
from the Belgian Federal Government (PAI grant P6/07 Economic Policy and Finance
in the Global Equilibrium Analysis and Social Evaluation) and from the Marie-Curie
research and training network TOM (Transnationality of migrants).
2
1 Introduction
With population forecasts indicating rapid population growth in developing countries
and slow growth in developed ones, international migration is likely to continue to play
an (increasingly) important role in the global economy. A world of rapid population
growth and increasing pressure on natural resources would greatly bene...t from South-
North migration if the latter resulted in a reduction in source countries'fertility rates.
Migration may a¤ect the fertility of migrants living in the host country, of mi-
grants'households back home, and of the home country population as a whole. This
paper focuses on the latter, that is, on the impact of migration on the fertility of
the population in migrants' country of origin. It presents a theoretical model and
s
empirically tests the model' predictions. In particular, we test whether international
migration results in a transfer of behavioral norms regarding fertility from host to
migrants'home countries. Although the literature on diaspora externalities is grow-
ing rapidly, it has so far not provided any robust evidence of migration externalities
of this sort. An exception is Spilimbergo (2008) who shows that foreign-trained indi-
viduals promote democracy in their home countries, but only if foreign education is
acquired in democratic countries. To justify why foreign-educated individuals bring
about changes in domestic democracy, Spilimbergo puts forward several mechanisms.
Foreign-educated leaders give more con...dence to foreign investors, are better mo-
tivated to keep up with the more developed countries where they studied, share a
common identity with the international democratic community, etc. Most of the rele-
vant mechanisms involve post-training returns of foreign-educated individuals to their
home country. In this paper, we use a similar conceptual empirical framework and
investigate whether it also matters for fertility behavior. An important di¤erence
with Spilimbergo is that transferring fertility norms does not imply return migra-
tion. Communications with migrants abroad, growing interest by the source country
residents in foreign culture and habits, or better media coverage of migrants'destina-
tion are reasonable channels through which the diaspora abroad induces behavioral
changes at origin. If such a transfer of fertility norms can be detected, this opens
the door to many other behavioral transfers between migrants'receiving and sending
countries (in terms of education choices, consumption habits, votes, etc.) and a deep
renewal of the literature on diaspora externalities. Indeed, contrary to the traditional
literature (whose e¤ects are conveyed through decreases in communication barriers,
transaction and information costs), behavioral transfers involve changes in the very
preferences of those left behind.
Our main ...nding is that international migration results in a transfer of behavioral
norms regarding fertility from host to migrants'home countries, resulting in a decrease
(increase) in home country fertility rates if they are higher (lower) than host country
rates. This result is robust to the instrumentation of control variables or the average
fertility at destination itself. In our best speci...cation, a one percent decrease in the
fertility norm to which migrants are exposed reduces home country fertility by about
3
0.3 percent for origin countries.
The paper is organized as follows. Section 2 presents a selected review of the
literature on the impact of migration on fertility rates in the three groups mentioned
above. Section 3 develops a theoretical model where alternative hypotheses regard-
ing the impact of international migration on source country fertility are examined.
Section 4 presents the econometric speci...cation, describes data sources and reports
the empirical results. Section 5 concludes.
2 Selected Literature Review
A necessary condition for migration to result in source countries' adoption of host
countries'behavioral norms is that they can be adopted by the migrants themselves.
Similarly, one would expect these norms to be adopted by migrant households since
they would most likely obtain the relevant information on host countries'norms before
the rest of the home country population, and in a more direct and detailed manner.
s
Examining what the literature says about migration' impact on the fertility of both
migrants and home country migrant households is thus important for understanding
whether and through what channels migration impacts source country fertility. Stud-
ies on the impact on migrants and migrant households are reviewed in Section 2.1,
and on the source country population in Section 2.2.
2.1 Fertility Impact on Migrants and Migrant Households
s
The bulk of the research has dealt with migration' impact on migrants' fertility.
Several hypotheses have been examined in this literature, including socialization,
adaptation, and selection.1 According to the socialization hypothesis, migrants are
socialized by early childhood experiences and post-migration fertility levels remain
similar to those in source areas or countries. Early studies on US internal migra-
tion ...nd support for this hypothesis, with Goldberg (1959, 1960) and Freedman and
Slesinger (1961) showing that rural-urban migrants exhibit higher fertility rates than
urban natives. However, they do not examine changes in migrants'fertility rates over
time. Moreover, ...ndings of later studies are generally consistent with the adaptation
rather than the socialization hypothesis.
According to the adaptation hypothesis, the impact of host (home) country values
and norms on migrants' behavior increases (decreases) with the length of the mi-
gration, with migrants'fertility rates converging to those of natives over time. This
hypothesis has received wide support in the literature, both for internal (rural-urban)
and international migration.
1
A fourth hypothesis is that of disruption whereby migrants show low fertility levels immediately
following migration. However, this hypothesis does not tell us much if anything about migration' s
impact on completed (lifetime) fertility, which is the subject of this paper.
4
Internal migration studies that examine the fertility impact of rural-urban migra-
tion have found support for the convergence of migrants'fertility rates to those of na-
tives. Studies on internal migration in developing countries include Myers and Morris
(1966) on Puerto Rico, Goldstein (1973) on Thailand, Martine (1975) on Colombia,
Park and Park (1976) on Costa Rica, Hiday (1978) on the Philippines, Faber and Lee
(1984) on Korea, Hervitz (1985) on Brazil, Lee and Pol (1993) on Mexico, Brockero¤
(1995) on thirteen African countries, Umezaki and Ohtsuka (1998) on Papua New
Guinea and Kulu (2003) on Estonia. Convergence results are also obtained in studies
of international migration, including Stephen and Bean (1992) and Lindstrom and
Giorguli Saucedo (2002) for women of Mexican origin living in the US.
Convergence of migrants' fertility rates might be due to selection rather than
adaptation. Migrants do not constitute a random sample of the home population
and they might exhibit lower fertility rates than the overall population. The studies
described above do not control for potential selection e¤ects, with some ...nding that
selection e¤ects played an important role in explaining the change in fertility associ-
ated with migration, while others did not or found that the presence or absence of
selection e¤ects depends on the type of migration examined. Studies by Goldstein
(1973), Hervitz (1985) and Kulu (2003) examined the various hypotheses and found
strong support for the adaptation rather than the selection hypothesis, though White
s
et al.' (1995) found limited support for the latter in a study on internal migrants in
2
Peru.
The impact of migration on the fertility of households back home has also been
analyzed, though by a much smaller number of studies. One hypothesis examined
is that the in uence of host countries'fertility norms persists after migrants return
home and thus results in a decrease in fertility. For instance, Lindstrom and Giorguli
Saucedo (2002) ...nd that Mexico-US temporary migration of women reduces long-
term household fertility. Another hypothesis is that migration reduces fertility while
the migrant is away and raises it when the migrant returns, a hypothesis con...rmed in
the case of male migration (e.g., Hervitz 1985). The results have been interpreted as
being due to interruption and catching up of fertility, with no clear long-term fertility
impact.
Lindstrom and Muñoz-Franco (2005) examine the impact of migration on women' s
modern contraceptive knowledge and use -- and thus on their fertility -- in rural
Guatemala. They ...nd that contraceptive use increases and fertility falls with vari-
ables such as having family members in urban or international destinations, living in
a community where urban migration is common, having social ties to urban or inter-
national migrants, and having an urban migration experience. They also ...nd that
these variables become non-signi...cant once they control for their knowledge-di¤usion
impact, concluding that it is through the knowledge acquired from urban migration
experiences, contacts with urban or international migrants, or living in a community
where such migration is prevalent, that contraceptive knowledge and use increases
2
They ...nd that education and having fewer children are positively related to geographic mobility.
5
and fertility declines.
s
Thus, most studies on migration' fertility impact have con...rmed that migration
to low-fertility countries (regions) reduces migrants' fertility in the home country
(region), and that the reduction in fertility is due to adaptation of migrants'fertil-
ity behavior to the norms of the host countries (regions). The studies also obtain
similar results with respect to home countries'migrant household fertility behavior.
Moreover, the latter are associated with a transfer of norms from the host country or
region to the migrant household or community, either because of return migration or
because of information obtained from migrants.
2.2 Fertility Impact on Home Country Population
Another question is whether migration results in a change in fertility rates of the
population in migrants' countries of origin. Since migrants' behavioral norms tend
to converge to those of their host countries, it is not unreasonable to assume that
migrants might serve as channels for the transmission of such norms and might a¤ect
the behavior of natives in their countries of origin, including their fertility behavior.
In such a case, the positive spillover e¤ect of migration in terms of reduced population
pressure would be vastly greater than if the decline in fertility rates only a¤ected the
migrants.
It is important to note that the impact of international migration on fertility
rates in migrants' home country may operate through several channels. The ...rst
channel consists of migrants' direct communication with their family, friends and
community. Second, migration typically triggers an increase in interest by source
countries' population about the situation in host countries as well as that of their
s
country' migrants living there. This tends to be re ected, inter alia, in an increase
in media coverage of both the host country and of the migrants living there.
Third, media attention is also likely to focus on the situation of return migrants,
including their economic performance, views and behavioral modes, and how they
might di¤er from those of natives. Fourth, a number of studies have found that
migration and migrant networks result in increased trade between host and source
countries (Gould 1994, Rauch 2001, Rauch and Trindade 2002) and in increased
investment from the former to the latter (Kugler and Rapoport 2006, Javorcik et
al., 2006). Thus, increased business-related contacts with migrants' host countries
is likely to constitute another channel through which the latter'norms are di¤used
to source countries' natives. Finally, fertility and other behavioral norms that are
di¤used through these various channels are likely to be further di¤used to those who
do not have direct access to them through word-of-mouth.3
The issue of international migration as a channel for the di¤usion of fertility norms
3
Those with direct access to information on fertility norms in host countries or to returning or
visiting migrants may learn about them indirectly through others who do have direct access to such
information.
6
has not been systematically studied. The only research we are aware of that examines
the link between international migration and source country fertility is Fargues (2007).
His analysis is based on fertility behavior in three source countries, namely Morocco,
Turkey and Egypt. Migration from Morocco and Turkey over the period 1960-2000
was essentially to the low-fertility countries of Western Europe while that of Egypt
was essentially to the high-fertility countries of the Persian Gulf. Fargues shows that
fertility rates in these countries were a¤ected by the rates prevailing in their migrants'
host countries, with rates declining in Morocco and Turkey and increasing in Egypt.
He also ...nds that the degree to which the demographic transition has been attained
increases with migration rates across regions of Morocco and Turkey and decreases
with migration rates across regions of Egypt.
Fargues posits that the impact of host countries' fertility rates on those in mi-
grants'home countries is due to the transfer of behavioral norms from host to source
country. However, he does not subject his hypothesis to rigorous testing or consider
alternative ones.4
This paper provides a rigorous econometric analysis of the relationship between
international migration and source country fertility. The impact of the former on
the latter may have a number of causes, one of which is the transfer of host country
fertility norms. These causes are examined theoretically and the hypotheses derived
from the model are tested empirically. The econometric analysis is based on a new
database of international bilateral migration for the year 2000 (Parsons et al. 2007)
that covers all countries and territories. We ...nd that fertility in migrants' home
countries decreases (increases) in the case where it is higher (lower) than fertility in
the host countries.
3 Theory
As South-North migration can a¤ect fertility decisions in the South through multi-
ple channels, assessing the e¤ect of di¤usions of norms requires controlling for the
other mechanisms at work. A stylized theoretical model is helpful to derive testable
predictions.
The main mechanisms we envisage here are the following. By a¤ecting the ex-
pected return to higher education, migration prospects impact on adults' human
capital investments, which in turn, determine the opportunity cost of raising children.
As a more educated child has a higher probability to emigrate to a rich country, expec-
tations about o¤spring migration impact on the ' quantity-quality'tradeo¤. Through
remittances, past migrations impact on adults' income and a¤ects the demand for
4
Ebanks et al. (1975) for Barbados and Lee and Farber (1985) for Korea compute the impact of
migration on fertility in the home country. However, their calculations are unrelated to the impact
of migration on fertility behavior back home. Rather, they calculate what the fertility in the home
country would have been had migrants stayed home by assuming that migrants'fertility rates are
equal to those of observably similar non-migrants.
7
children. Let us ...rst describe the general model including all these mechanisms and
then solve di¤erent variants of the model focusing on each particular channel. Each
variant generates speci...c testable predictions which should be accounted for in our
empirical analysis.
We consider an overlapping generations economy populated by two-period lived
agents (adult and children). Following De la Croix and Doepke (2003, 2004), Galor
and Mountford (2006), Moav (2005) or Mountford and Rapoport (2007), adults'
utility function has two arguments, the amount of consumption and the total expected
income of children. The second component of the utility function re ects parental
altruism but it could also be compatible with the fact that parents care about old-age
security if children transfer money to their parents when the latter retire. We have
e
Ut = log(ct ) + log(wt+1 ht+1 nt ) (1)
s
where ct denotes parent' consumption, nt is the number of children (fertility), ht+1
e
is the human capital of each child and wt+1 is the expected wage per e¢ ciency unit of
labor of children. Uncertainty about future children wages arises from the fact that
children may stay in their origin country or emigrate to a richer country.
Adults are endowed with one unit of time that they can spend in supplying labor,
raising children or investing in their own education. Raising each child requires
units of time. Given their inherited level of human capital ht (resulting from their own
parents'decisions), adults may spend a fraction Et of their time in higher education
to increase their human capital. The training technology is given by
Ht = (Et ; ht ) (2)
0 0
such that E ; h 0. In the next sub-sections, we will consider variants where (:)
has a Cobb-Douglas analytical form and variants disregarding parents human capital
decisions, (Et ; ht ) = ht .
Parents can also invest in the human capital of their o¤spring. Investing et dollars
s
in children' basic education increases their human capital. We assume that
ht+1 = (et ) (3)
0 0
where e 0 and e 0. In the next sub-sections, we will consider variants with
ht+1 = et with 2 [0; 1] and variants with exogenous education choices, ht+1 = h.
The adult budget constraint is given by:
ct = (1 Et nt )wt Ht nt et + rt (4)
where rt stands for non labor income (including remittances received in adulthood)
s
and wt denotes adult' wage.
Assuming that adult education arises before employment, adults are uncertain
about their future place of work. If they stay in the South (with probability pt ), the
8
h
wage rate is given by wt = wt . If they move to the North (with probability 1 pt ), the
f
wage rate becomes wt = wt > wt . The production functions in the South and in the
North are linear in labor (in e¢ ciency unit). It implies that the local and foreign wage
h f h
rates wt and wt are time invariant. Without loss of generality, wt can be normalized
to unity and we can write ! = wf 1.
Adults are also uncertain about the place of work of their children. Children will
become adult at time t+1 and will be able to emigrate with a probability pt+1 . The
expected wage for each child in (1) is given by
wt+1 = pt+1 wf + (1
e pt+1 )wh = 1 + pt+1 !: (5)
The migration probability depends on country characteristics (such as geograph-
ical position, colonial links, linguistic proximity, etc.) and individual characteristics.
In particular, it can be reasonably assumed that the probability increases in human
capital. We have:
pt = p0 : (Ht ) (6)
0 00
where p0 captures country characteristics and (Ht ); such that 0 and 0,
re ects the fact that educated agents have a higher probability to emigrate.
Let us now solve particular variants of this general model, based on particular
analytical speci...cations for our technological functions (:), (:) and (:):
3.1 Fertility, migration and adults'higher education
We ...rst focus on the relationship between migration prospects and human capital
formation, as stated in the new brain drain literature (Mountford, 1997, Beine et
al., 2001 and 2008, or Docquier et al., 2008). To, address this issue, let us consider
a simpli...ed model in which children' human capital h is exogenous. Think about
s
a mandatory education system totally subsidized by the government. The cost of
education can therefore be removed from the budget constraint (et = 0). We also
disregard remittances (rt = 0).
Parents can invest Et in higher education to increase their productivity and their
own probability to emigrate. After education, they will work abroad and earn a wage
wf with a probability pt . They will work at home and earn a wage equal to one with
a probability 1 pt .
The timing is the following. First, parents decide whether or not to invest. Second,
they emigrate or stay in their home country. Third, they work, have children and
consume. Parents thus take tow decisions, Et and nt . The choice of Et is made under
uncertainty about the place of work.
Parents care about the expected income of their o¤spring. For mathematical
tractability, we assume that the probability that a child will live abroad do not de-
e
pend on parents'location. This implies that wt+1 is given in (1). Considering that
children born abroad have a much higher probability to stay would induce parents to
9
invest more in human capital. This would simply reinforce our mechanism. As ht+1
is also exogenous, the second component of the utility function (1) only depends on
the number of children, nt .
The following speci...cations are used:
Parents'probability to emigrate in (6) has a logarithmic form: (:) = log(Ht ).
Parents'productivity is endogenous and (2) has a Cobb Douglas form: (:) =
1
AEt h .
s
Children' human capital in (3) is ...xed: (:) = h.
remittances are nil: rt = 0.
In Appendix 7.1, we solve the model in two steps and proceed backwards. First,
for a given location, parents choose their optimal number of children. Second, after
substituting this number in the utility function, parents decide how much to invest in
education taking into account the endogenous probability to emigrate. The optimal
fertility rates of migrants and stayers are identical5 and amount to
(1 E)
nt = ; (7)
(1 + )
Agents then maximize the expected utility function, given the endogenous prob-
ability to emigrate. The optimal investment in higher education is given by
1 + p0 log(wf )
Et = h 1
i (8)
1+ + 1 + p0 log(Ah wf )
To summarizer, parents'investments in higher education increase with the prob-
ability to emigrate (@Et =@p0 > 0). Hence, given (7), openness induces human capital
and reduces fertility at origin since @nt =@Et < 0. The mechanism is simple. As
argued in the new brain drain literature, migration prospects to richer countries
stimulates human capital formation. This reduces the maximal amount of time that
parents can devote to children education and labor. In empirical regressions, this ...rst
e¤ect of migration on fertility can be easily accounted for by controlling for parents'
human capital.
3.2 s
Fertility, migration and children' education
s
Let us now focus on the links between children' human capital and their probabil-
ity to emigrate. In the second variant, we assume that parents have no possibility
5
We could easily extend the model to account for the fact that fertility is lower in rich countries.
10
to invest in human capital: equation (2) is such that Et = 0 and Ht = ht is pre-
determined. They do not receive remittances. For simplicity, we assume that the
probability that a child will emigrate is linearly increasing in human capital. The
following speci...cations are used:
Parents'probability to emigrate in (6) has a linear form: (:) = Ht .
Parents'productivity in (2) is predetermined: (:) = ht .
s
Children' human capital in (3) is endogenous: (:) = et .
Remittances are nil: rt = 0.
The optimization problem for remaining adults can thus be written as following:
fnt ; et g = arg max flog [(1 nt )Ht nt et ] + log [nt et (1 + !p0 et )]g
As shown in Appendix 7.2, an explicit analytical solution to this problem can be
obtained for = 1 :6 The optimal fertility rate and investment in children education
2
are characterized by the following system
Ht
nt = (9)
(1 + )( Ht + et )
h p i2
et = Ht !p0 + ( Ht !p0 )2+ H (10)
t
First, for a given parental income, we have a negative relationship between fertility
s
and investment in children' education. Clearly, in the absence of migration (p0 = 0),
we have et = Ht and nt = (1+ )2 . The fertility rate is independent on parental
income. With migration prospects, the optimal investment in education increases
in p0 . Hence, for a given wage rate, fertility decreases with migration. This result
contracts with Chen (2006) who shows that when the probability to emigrate is
exogenous, it does not a¤ect the optimal education of children and fertility. In our
framework with endogenous probability of migration, it comes out that p0 > 0 implies
the optimal fertility rate decreases with parental income. In empirical regressions, this
second e¤ect of migration on fertility can be accounted in two ways. First, average
rate of migration of the sending country, as a proxy for p0 ; can be introduced as a
direct determinant of the home country fertility. Second, it is desirable to control for
parents'human capital, which might be measured by the residents'education level.
6 1
Numerical experiments reveal that similar qualitative results would be obtained with 6= 2 .
11
3.3 Fertility and remittances
Migration also impacts on fertility through remittances sent by previous generation
of migrants and /or members of the community. Indeed, We can reasonably consider
that the amount of remittances positively depends on the stock of contemporaneous
compatriots living abroad. In the third variant, we assume that parents have no
possibility to invest in human capital (Et = 0 and Ht is predetermined) and that
children face an exogenous probability to emigrate. However, we now introduce non
labor income, which can be here interpreted as the amount of remittances. The
following speci...cations are used:
Parents'probability to emigrate in (6) has a linear form: (:) = 1.
Parents'productivity in (2) is predetermined: (:) = ht .
s
Children' human capital in (3) is endogenous: (:) = et .
Remittances are positive and exogenous: rt > 0.
The optimization problem of remaining adults can thus be written as the following
fnt ; et g = arg max flog([(1 nt )Ht nt et + rt ] + log [nt et (1 + !p0 )]g
s
As shown in Appendix 7.3, the optimal fertility rate and investment in children'
education are given by
rt
(1 ) 1+ Ht
nt = (11)
(1 + )
Ht
et = (12)
1
The optimal fertility rates increases with the amount of remittances (linked to the
number of migrants abroad).
The latter result is closely linked to the choice of the utility function and the
timing of remittances. Assume that the second component of the utility function
(1) is not due to parental altruism but to the fact that parents care about old-age
security. Assuming that working-aged children transfer a fraction of their income
to their parents and parents also receive other transfers when old, the utility function
would become:
o
e
Ut = log(ct ) + log( wt+1 ht+1 nt + rt+1 ) (13)
o
where rt+1 includes remittances sent by extra-family members to old parents.
Adults'optimization problem can thus be written as the following
o
fnt ; et g = arg max log([(1 nt )Ht nt et ] + log nt et (1 + !p0 ) + rt+1
12
s
The optimal fertility rate and investment in children' education become
o
(1 ) rt+1
nt = (14)
1+ (1 + ) et (1 + !p0 )
Ht
et = (15)
1
Under the old-age security hypothesis, the optimal fertility rates decreases with
the expected amount of remittances received when old. In sum, the e¤ect of extra-
family remittances is thus ambiguous. It can be positive of the income e¤ect dominates
or negative if the old-age security e¤ect dominates.
3.4 Transfers of norms
As argued in Fargues (2007), one could also argue that migrants transfer fertility
norms to those left behind. To model this hypothesis, let us consider the altruistic
variant of our model and introduce alternative preferences regarding fertility. The
novelty is that, in deciding on the number of children, parents internalize the gain of
utility from conformity to the norm for fertility. Katav-Herz (2003) applied this idea
to the choice of fertility, child education and child labor.
It is well documented that migrants abroad progressively assimilate in terms of
fertility choices. In particular, the average fertility rate of ...rst-generation immigrants
from developing countries is lower than the fertility rate at origin, although higher
than the average fertility rate of natives at destination. Just as migrants facilitate
transfers of knowledge and ideas, they are also likely to transfer fertility norms to
e
those left behind. We formalize this idea by introducing a reference level nt of fertility
(or norm) in the utility function and assume that adults derive utility from nt nt e
(instead of obtaining utility from nt , adults derive utility from having generally more
children than the reference number of children).
In this variant, we consider that parents cannot invest in education (Ht = ht is
predetermined) and the probability of migration is exogenous (for parent and chil-
dren). The following speci...cations are used:
Parents'probability to emigrate in (6) has a linear form: (:) = 1.
Parents'productivity in (2) is predetermined: (:) = ht .
s
Children' human capital in (3) is endogenous: (:) = et .
Remittances are nil: rt = 0.
Introducing the norm in the utility function (1), the optimization problem of
non-migrant adults becomes
13
fnt ; et g = arg max flog([(1 nt )Ht nt et ] + log [(nt e
nt ) et (1 + !p0 )]g
As shown in Appendix 7.4, the optimal fertility rate and investment in children's
education are given by
q
e
nt + (1 e
) + [ nt + (1 )]2 + 4 (1 + )
nt = (16)
2 (1 + )
e
Ht (nt nt )
et = (17)
n(1 e
)+n
(1
When n = 0, we have nt = (1+ ) as in the usual model. When n is positive, it is
e )
e
e
obvious that the optimal fertility is an increasing function of n (and is independent on
parental income). Hence, if a transfer of norms reduces the reference level of fertility
in the origin society, it impacts negatively on the optimal fertility rate.
The di¤usion technology can be a complex function of the geographical distribu-
tion of the population and of fertility rate di¤erentials across countries. We disregard
here the link between the fertility norm and the lagged fertility rate of the domes-
tic country to concentrate on the part of the norm a¤ected by emigration. If d;t
is thePproportion of the emigrant population living in country d, we can de...ne by
d
n = d d nd the average fertility rate at destination (d = 1; :::; D are foreign destina-
tions). Since p0 denotes the average emigration rate, a reasonable di¤usion technology
can be written as
n = N (p0 ; nd )
e (18)
0 0
with N1 0 and N2 0.
(p0 ) 0
For example, if N (:) = nd , we have ln (e) = (p0 ) : ln nd with
n 0.
0
It is a priori unclear whether is high or low. On the one hand, it can reasonably
be argued that a transfer of norms can only be detected if the size of the diaspora
abroad is large enough, a hypothesis which is perfectly compatible with the fact that
the average fertility at destination weights foreign fertility rates by the proportions
of emigrants located in these destinations. On the other hand, as the size of the
diaspora becomes larger, the marginal impact of p0 could decrease or even become
negative: large diasporas are less likely to socialize and assimilate abroad, and may
0
have less contacts with those left behind. If is low or not signi...cant, it means that
the di¤usion of norms is relatively independent of the intensity of migration, and
0
emigrants-based norms can be seen as a "public good". If is high, the di¤usion of
norms depends on the intensity of migration.
4 Empirical analysis
The model presented in section 3 enables us to identify the various channels through
which migration can a¤ect fertility in source countries. It also emphasizes the role of
14
education for explaining the prevailing fertility behavior. The identi...ed channels are
embedded in the regression model speci...ed in the following sub-section.
4.1 Econometric speci...cation
Our dependent variable is the log of the fertility rate in source countries, log (nt ).
The main explanatory variables of interest are the following.
The "norms-di¤usion" model predicts that the fertility rate should be increasing
in the average fertility rate at destination. We assume a linear form for (p0 ) ;i.e.
(p0 ) = a1 + a2 :p0 . From (18), the log of the norm has two additively separable
terms and can be written as log (e) = a1 log(nd ) + a2 p0 log(nd ). We expect
n
a non-negative sign for the estimates of a1 and a2 in the empirical equation
(19) below. A similar technology is used by Spilimbergo (2008) who empirically
showed that foreign-educated individuals promote democracy in their home
country, but only if the foreign education is acquired in democratic countries.
He constructs an index of average democracy in host countries, which is de...ned
as the weighted average of democracy indices in host countries where a country' s
weight is the share of students going to that country over all foreign students
from the origin country. Here, we transpose the same conceptual di¤usion model
to fertility behavior.
s
By altruism, the fertility rate should be decreasing with a country' average
emigration rate and with quality-selective immigration policies at destination.
This suggests the use of two relevant explanatory variables. The ...rst one is the
average emigration rate (p0 ). It is measured by the total emigration rate. The
second one aims at measuring selection in migration ows and is proxied by the
ratio of migrants to residents for skilled relative to unskilled labor (S). These
variables are taken in logs and we expect a negative sign for the estimates of a3
and a4 in the empirical equation (19) below.
Fertility is ambiguously a¤ected by the amount of remittances sent by extra-
family compatriots abroad. The income e¤ect predicts that the fertility rate
should be increasing in remittances received when adult. However, the old-
age security model predicts that fertility should be decreasing in remittances
received before retirement since part can be saved for retirement. The expected
sign for a5 is therefore ambiguous. Controlling for remittances R in empirical
regressions will reveal the sign of the global impact of this variable.
Theory predicts that migration prospects can stimulate the education of adults.
Since educated parents have a higher opportunity cost of time, one expects the
fertility rate to decrease in adults' human capital. In our regression, we will
use the proportion of adults aged 25+ with secondary and/or post-secondary
education (denoted by H) and expect a negative sign for the estimate of a6 .
15
Finally, we also control for a set of K explanatory variables Xk (k = 1; :::; K)
which are not necessarily linked to international migration but potentially have
an impact on fertility decision. We include the log of GDP per capita, the
urbanization rate, regional dummies as well variables capturing the type and
intensity of religious practice in source countries.
The benchmark empirical model can thus be written as:
log (n) = a0 + a1 : ln(nd ) + a2 :p0 ln(nd ) + a3 : ln (p0 )
X
+a4 : ln (S) + a5 : ln (R) + a6 :H + bk :Xk + " (19)
k
where a0 is a constant and "t is a iid error term.
Our main coe¢ cient of interest are a1 which determines the signi...cance and mag-
nitude of the di¤usion of fertility norms. The expected theoretical sign of a1 is
positive. We also interact the (log of) fertility rate at destination with the size of
the diaspora to assess the robustness of the impact to the intensity of migration. In
this set up, the emigration rate is also associated to the incentive channel (a4 ). We
assume that the incentive e¤ect follows a concave pattern and is measured by ln (p0 ) :
The expected theoretical sign of a4 is negative, as higher migration prospects increase
the incentive to invest in education and reduce fertility at home.
4.2 Data
Our regression involve cross section data because the migration stocks used to build
fertility at destination nd are only available for 2000: Data on fertility rates (nt )
are taken from the World Development Indicators. The fertility rate is the average
number of children that women have during their lives, from age 15 to age 50. To
compute data on average emigration rates (p0 ) and geographic shares of the emi-
grant population by destination ( d;t ), we use the data set developed by Parsons,
Skeldon, Walmsley and Winters (2007). They provide four versions of an interna-
tional bilateral migration stock database for 208 countries and territories of origin
and destination for the year 2000. We use the fourth comprehensive version which
uses a variety of techniques to estimate the missing data. The ...nal matrix, compris-
ing only the foreign-born reconcile all of the available information in order to provide
the researcher with a single and complete matrix of international bilateral migrant
stocks.
One striking picture coming out of the data is the importance of South-South
migration (to non OECD countries). Not surprisingly, North-South migration is neg-
ligible as all OECD countries send most of their migrants to other OECD countries.
To illustrate the importance South-South migration, it turns out that 47 percent
of developing countries have their main destination in a non-OECD countries. Out
of those countries, 81 percent send their migrants mostly to a neighboring country.
16
This is line with the well known stylized fact of prevailing liquidity constraints in
international migration (Lopez and Schi¤, 1998; Mayda, 2006).
The intensity of South-South migration is important for our analysis for two rea-
sons. First, it suggests that migrants are by no means concentrated in OECD coun-
tries. Therefore, the fertility norms that they will transmit from abroad are much
more heterogeneous that one would expect if most migrants were located in OECD
countries. Second, it suggests that the impact of fertility norms might work in both
directions. In fact, 83 countries have an average fertility at destination that is higher
than the one prevailing at home (i.e. 40 percent of the full sample of 208 countries).
If we consider only developing countries, this situation is observed in 44 cases (i.e. 28
percent of the sample of 155 developing countries)
Combining the fertility data with the bilateral migration matrix allows us to com-
pute weighted average fertility rate at destination, as in equation (18). As expected
from the observed patterns of migration and the importance of South-South migra-
tion as well as the range of fertility rates, the average fertility rates at destination nd
exhibit a high degree of variability, ranging from 1.40 to 5.58.
Table 1 describes the distribution of the fertility rates by country groups. The
countries are classi...ed along di¤erent criteria: income (high income vs developing
countries), data availability regarding important variables such as the remittances,
geographical location and religion. In 2000, the fertility rates vary signi...cantly across
location of the country, from 4.8 children per woman in Africa to 1.4 in Europe. At
the world level, the fertility rate is equal to 3.2 children per woman on average and
ranges from 0.9 in Macau to 7.95 in Niger. Table 1 also provides fertility rates at
destination, i.e. values for nd : The fertility rates at destination are signi...cantly higher
for Sub Saharan countries and to a lesser extent Islamic countries. This might be
explained by their emigration patterns. In Sub-Saharan Africa, a large share of the
migration ows is towards neighboring countries that also display high fertility levels.
In Islamic countries such as Egypt of Pakistan, a lot of workers migrate to Gulf
countries in which the fertility rates are relatively high.
Data on human capital (Ht ) on positive selection in emigration (St ), proxied by
the skilled-to-unskilled ratio of emigration rates to rich countries, are computed by
Docquier et al. (2007). Emigration rates (p0 ) are also computed from the Parsons et
al. (2007) database. Data on remittances are taken from the IMF database. In our
set of controls, we include the urbanization rate (available in the World Development
Indicators)7 , regional and religious dummies. Regional dummies are consistent with
the World Bank de...nition. Given the low number of countries (8) in South Asia,
we join those countries with the Europe and Central Asia (ECA) region. Religion
variables are measured in two ways. We ...rst use the proportion of Catholics and
Muslims living in the country. Alternatively, we introduce a dummy for countries
belonging to the organization of the Islamic Conference.
7
Sato (2007) and Sato and Yamamoto (2005) discuss the e¤ect of agglomeration and urbanization
on fertility rates.
17
Table 1. Descriptive statistics on fertility rates and norms
Nb. of Emig. Home-country Fertility Fertility at destination
obs. rate (%) Mean St. dev Min Max Mean St. dev Min Max
All countries 208 11.2 3.19 1.72 0.90 7.96 2.58 1.01 1.40 5.58
High-income 53 14.8 1.87 0.55 0.90 3.10 2.14 0.61 1.40 4.10
Developing 155 10.0 3.64 1.76 1.15 7.95 2.73 1.08 1.44 5.58
All countries (data on remit.) 155 10.3 3.10 1.66 0.98 7.96 2.55 1.04 1.44 5.58
Developing (data on remit.) 126 10.5 3.43 1.66 1.15 7.96 2.68 1.09 1.44 5.58
All countries (no data on remit.) 53 14.0 3.42 1.90 0.90 7.77 2.68 0.91 1.40 5.33
Muslims (% of pop) 56 7.6 4.24 1.73 1.84 7.96 3.17 1.19 1.46 5.58
Catholics (% of pop) 190 11.4 3.14 1.68 0.90 7.96 2.61 1.01 1.40 5.58
MENA countries 13 6.95 3.61 1.33 2.09 6.08 2.84 0.73 2.01 4.12
Latin Am. & Carib. 54 19.6 2.76 1.07 1.23 7.77 2.22 0.47 1.66 4.00
Sub Saharan Africa 68 12.1 4.58 1.77 1.23 7.96 3.41 1.16 1.47 5.58
18
East Asia & Paci...c 48 15.1 2.87 1.40 0.90 7.77 2.27 0.54 1.40 4.00
South Asia 28 18.1 3.20 1.63 1.23 7.77 2.50 0.64 1.66 4.00
Own computations based on fertility (World Development Indicators) and migration data (Parsons et al., 2007)
4.3 Results
Tables 2 to 5 report the results for the benchmark regressions on the full sample or the
sub-sample of developing countries. These are obtained using OLS (in Tables 2 and 4)
and IV (in Table 3 and 5) estimations accounting for potential endogeneity problems
related to an important control variable, the average emigration rate p0 . Each table
reports the estimation results from a set of variants in terms of speci...cations and
the used variables. Then, Table 6 and 7 deal with the potential endogeneity of
migrants'destination choices, a key variable used to construct the average fertility at
destination. Finally, Table 8 provides results obtained with a dynamic speci...cation.
OLS regressions. Let us start with OLS regressions in Table 2 using the
benchmark speci...cation (19). In this set up, the norm is supposed to be given
by ln (et ) = ln(nd ) + p0 ln(nd ). Following Spilimbergo (2008), the interaction term
n
p0 ln(nd ) tests whether the impact of fertility at destination (the fertility norm) de-
pends on the intensity of migration or not.
The data constraints tend to in uence signi...cantly the sample size. If we use
the full set of explanatory variables, data unavailability for a couple variables such
as remittances, GDP per head, share of Catholics in the countries and the selection
ratio reduces the sample size. The most important reduction comes from the use
of remittances, which are unavailable for 53 countries. Therefore, we estimate the
model with and without remittances as an explanatory variable. The estimation
results without (resp. with) are presented in columns (1) and (2) (resp. columns (3)
and (4)). The use of the other variables leads to a further reduction of the usable
sample, with 175 countries in the largest sample and 145 countries when remittances
are included. In all Tables, we estimate for each sample a full speci...cation (columns
1 and 3) as well as a parsimonious one (columns 1 and 4) to increase e¢ ciency in the
estimation.
Table 2 reports the estimates of model (19) for the sample of all countries. Let
us ...rst focus on parameters of interest. First, all estimations point to a positive
and signi...cant impact of fertility norms. For the two samples (with and without
remittances), we ...nd a signi...cant a1 coe¢ cient. The elasticity ranges between 0.27
and 0.34. In contrast, in all regressions, the interaction term p0 ln(nd ) turns out
to be insigni...cant, suggesting that the transfer of norm does not depend on the
intensity of migration. It is worth noticing that in most regressions, Spilimbergo
(2008) reached the same conclusion in his paper on democracy. As stated above, a
plausible explanation is that larger diasporas socialize and assimilate less abroad, or
have less contacts with those left behind. Another explanation could be that transfers
of norms can only operate between young individuals in age of fertility. Hence, the
variable p0 (based on global stocks of emigrants, including old guest workers arrived
after World War 2) does not perfectly capture the intensity of migration of young
individuals. Hypothesizing that cross-country di¤erences in destination choices (in
d
d;t ) are more stable than in migration intensity (in p0 ), the e¤ect of ln(n ) can
then be statistically signi...cant whereas the interaction term is not. In subsequent
19
regressions, this interaction term is dropped in order to increase e¢ ciency in the
parameter. Our estimates therefore suggest that on average, a decrease of 1% of
the fertility at destination leads to a decrease of about 0.30% in the home country
fertility.
The estimation results also point to a (weakly) signi...cant incentive impact, as
re ected by the negative and signi...cant parameter a3 ; as re ected by the results of
the parsimonious regressions (columns 2 and 4). The results support the idea that
higher migration prospects tend to slightly reduce fertility at home, possibly through
a higher investment in education. The coe¢ cient associated to the selection ratio (a4 )
is however non signi...cant and is also dropped in the subsequent parsimonious regres-
sions. The negative estimate of the adults' education level (a6 ) is in line with the
incentive impact of migration through investment in education. The weakly signi...-
cant coe¢ cient might be explained by the high collinearity with the level of GDP per
head, especially given the cross sectional dimension data. 8 It is therefore dropped in
parsimonious speci...cations. As for the impact of remittances (a5 ), we ...nd moderate
support for a positive impact on fertility at home in the parsimonious speci...cation.
This could suggest that the income e¤ect slightly dominates the negative impact
associated to the old-age security. Nevertheless, one might expect that old-age secu-
s
rity e¤ect of the migrant' transfers play a higher role in developing countries than
in developed countries. We check this point below when restricting the sample to
developing countries only.
As for the control variables included in regression (19), our results are mostly
in line with the expected impact. Fertility rates are found to decrease with income
per capita. They are found to be higher in Islamic countries and increase with the
proportion of Catholics in the country. Compared to the ECA-SA region, fertility
rates are higher Sub Saharan Africa, Latin America and East Asia and Paci...c.
8
The correlation between the adults'human capital and incoime per head amounts to 0.67 (resp.
0.57) for the full sample (resp. sample of developing countries).
20
Table 2. OLS regressions (dep = log of fertility rate) - All countries
(1) (2) (3) (4)
Constant 1.200 1.037 1.459 1.432
(5.51)*** (6.23)*** (5.11)*** (5.92)***
Log of fertility at dest 0.343 0.383 0.273 0.291
(3.35)*** (4.37)*** (2.40)** (2.76)***
p0 :Log of fertility at dest -0.202 -0.405
(0.77) (1.54)
log of p0 -0.025 -0.040 -0.018 -0.053
(0.78) (1.90)* (0.59) (2.44)**
Selection ratio (sec+tert) 0.001 0.001
(0.04) (0.02)
Log of remittances 0.018 0.032
(1.16) (2.18)**
Urbanization -0.005 -0.005 -0.004 -0.003
(2.62)*** (2.88)*** (2.54)** (2.24)**
GDP per capita -0.076 -0.079 -0.100 -0.117
(2.46)** (3.32)*** (2.80)*** (3.86)***
s
Adult' education -0.256 -0.190
(1.70)* (1.09)
East Asia & Paci...c 0.272 0.277 0.308 0.272
(2.85)*** (2.99)*** (3.29)*** (2.90)***
Sub-Saharan Africa 0.427 0.420 0.537 0.521
(4.67)*** (4.72)*** (4.75)*** (5.40)***
Latin Am. & Carib 0.350 0.323 0.451 0.486
(5.13)*** (5.31)*** (6.49)*** (8.03)***
Mena 0.115 0.159
(1.21) (1.26)
High-income 0.089 0.229 0.239
(0.96) (2.22)** (2.70)***
Muslims (% of pop) 0.004 0.004 0.003 0.003
(4.40)*** (5.59)*** (2.96)*** (4.55)***
Catholic (% of pop) 0.001 0.002 0.001
(1.81)* (2.68)*** (1.43)
Observations 175 175 145 145
R-squared 0.78 0.77 0.83 0.81
Robust t statistics in parentheses
* signi...cant at 10%; ** signi...cant at 5%; *** signi...cant at 1%
21
IV regressions. The OLS estimation of model (19) rests on the assumption
that all covariates are independent of "t . Nevertheless, it might be argued that some
variables might depend on fertility, invalidating this assumption. In particular, higher
fertility rates should increase labor supply and depress wages in domestic countries,
a¤ecting international migration. In other words, ln(p0 ) that captures the incentive
channel might depend on the level of the home country fertility rate ln (nt ) : In this
case, reverse causality might a¤ect directly the quality of the estimation of a3 but also
the ones of other coe¢ cients such as a1 9 . If such an e¤ect is quantitatively important,
it might be desirable to carry out instrumental variable estimation. Note that since
we are using migration shares across destination countries ( d ) rather than stocks of
migrants to build the norm variable; we can rule out any reverse causality running
from ln (nt ) to ln nd .
t
Table 3 reports the IV estimates of model (19) for the whole sample of countries.
We consider the following instruments of the (log of) emigration rate: a dummy vari-
able for islands, the (log of the) size of the country measured by its surface (in squared
kilometers) and (the log of the) distance to main destination of the migrants.10 It
is worth emphasizing that the two necessary conditions for instrumentation are ful-
...lled in our regressions. First, the ...rst stage estimation results indicate that we have
strong instruments. The F statistics of the ...rst stage regressions are most of the
time above 10. The partial correlations of the ...rst stage regression show that the 3
instruments explain a signi...cant part of the variability of emigration rates. Second,
as suggested by the p-value of the Hansen overidenti...cation test, those instruments
are found to be independent of the fertility rates. The test fails to reject the null
hypothesis of independence between the instrument set and the error term.
The main ...ndings of the IV estimations are broadly similar to those of the OLS
estimations. We ...nd evidence of a shifting norm e¤ect (a1 ) and reject the existence of
an interaction e¤ect with the emigration intensity (a2 ). The elasticity of the transfer
of norm also remains quite in line with the one estimated by the OLS regressions. It
is however slightly lower for the sample of countries for which data of remittances are
available. We ...nd moderate evidence of an incentive e¤ect of migration on fertility,
as suggested by the estimate of a3 in the parsimonious speci...cation including the
remittances. In this set up, the slightly positive impact of remittances on fertility
rates is also con...rmed. On the whole, the impact of the other variables are in general
in line with the one found in OLS regressions.
9
Theoretically speaking, the existence of a reverse causality between migration and fertility im-
plies that the interaction term (emigration rate times the fertility norm) should also be instrumented.
Nevertheless, since this term is insigni...cant in OLS regressions, we focus on the instrumentation of
the emigration rate only.
10
The ...rst stage regressions of the IV estimations yield estimates that are in line with intuition.
In particular, size is negatively associated to emigration rates. Islands are found to display higher
migration rates and higher distance to main destination negatively a¤ects emigration rates. The
estimation results are not reported here to save space but can be obtained upon request to the ...rst
author.
22
Table 3. IV regressions (dep = log of fertility rate) - All countries
(1) (2) (3) (4)
Constant 1.253 1.116 1.482 1.101
(5.99)*** (5.90)*** (6.30)*** (5.86)***
Log of fertility at dest 0.385 0.390 0.236 0.273
(3.44)*** (4.18)*** (1.91)* (2.57)**
p0 :Log of fertility at dest -0.223 -0.090
(0.41) (0.23)
log of p0 -0.020 -0.019 -0.060 -0.075
(0.28) (0.52) (1.12) (2.45)**
Selection ratio (sec+tert) -0.005 -0.017
(0.18) (0.58)
Log of remittances 0.019 0.032
(1.29) (2.17)**
Urbanization -0.005 -0.005 -0.005 -0.005
(2.79)*** (2.69)*** (2.97)*** (2.88)***
GDP per capita -0.075 -0.072 -0.107 -0.063
(2.17)** (2.81)*** (3.18)*** (2.74)***
s
Adult' education -0.282 -0.204
(1.87)* (1.24)
East Asia & Paci...c 0.238 0.245 0.268 0.220
(2.51)** (2.63)*** (2.70)*** (2.38)**
Sub-Saharan Africa 0.372 0.434 0.499 0.495
(4.35)*** (5.15)*** (4.69)*** (5.22)***
Latin Am. & Carib 0.321 0.371 0.435 0.412
(4.57)*** (6.27)*** (5.92)*** (7.18)***
Mena 0.076 0.238
(0.73) (2.39)**
High-income 0.169 0.191 0.205 0.134
(1.81)* (2.26)** (1.98)** (1.62)
Muslims (% of pop) 0.235 0.220 0.227 0.210
(4.17)*** (4.02)*** (3.88)*** (3.68)***
Catholic (% of pop) 0.001 0.001
(1.17) (1.21)
Partial Corr First Stage 0.174 0.342 0.212 0.400
F-stat First Stage 12.39 20.27 11.26 36.36
Hansen J Test (p-value) 0.130 0.319 0.63 0.67
R-squared 0.774 0.757 0.824 0.803
Observations 174 175 144 144
Robust t statistics in parentheses
Instruments for ln(p0): island, ln(size), ln(distance to main destination)
* signi...cant at 10%; ** signi...cant at 5%; *** signi...cant at 1%
23
Focusing on developing countries. The results of the benchmark regressions
tend to support the existence of a transfer of norms between countries in terms
of fertility. In this section, we assess the sensitivity of the choice of the included
countries in the sample. Here, we restrict our attention to the determinants of fertility
rates in the developing countries only. This robustness check stems at least for two
reasons. First, although we have no direct evidence for that, it might be expected
that norms might in the ...rst instance be from developed to developing countries.
Second, other channels through which migration a¤ect fertility might di¤er between
developed and developing countries. One obvious example concerns the impact of
remittances. One might expect that the e¤ect of transfers associated to old-age
security concerns is stronger in developing countries in which legal pension systems
are much less developed. In this case, we should expect the impact of remittances on
fertility rates to be less positive or even to be negative as the income e¤ect will be
more o¤set by the old-age security e¤ect.
Table 4 and Table 5 report the results with the sample of developing countries
only. A country is considered developed if it is classi...ed as a high income country
in the World Bank classi...cation. The results might summarized as follows. We ...nd
con...rmation of strongly signi...cant e¤ect of a transfer of norms in all regressions.
Nevertheless, we do not ...nd any signi...cant di¤erence in the elasticity between the
full sample and the one including the developing countries only. We also ...nd con...r-
mation that this impact does not depend directly on the intensity of migration, as the
in uence of the interaction term p0 ln(nd ) is always strongly rejected in all regressions.
We ...nd moderate evidence of an incentive e¤ect of migration through investment in
education (see Table 5 and in particular IV estimate of a3 in a parsimonious speci...-
cation with the remittances included as a control variable). Interestingly, we ...nd less
positive impact of remittances on the fertility rates for developing countries, although
our regressions clearly rejects the hypothesis that old-age security e¤ect dominates
the income one.
24
Table 4. OLS regressions (dep = log of fertility rate)
Developing countries
(1) (2) (3) (4)
Constant 1.270 1.267 1.587 1.525
(5.40)*** (6.20)*** (5.17)*** (6.29)***
Log of fertility at dest 0.337 0.338 0.267 0.261
(3.28)*** (3.65)*** (2.26)** (2.34)**
p0 :Log of fertility at dest -0.181 -0.368
(0.60) (1.21)
log of p0 -0.019 -0.028 -0.009 -0.034
(0.54) (1.17) (0.28) (1.49)
Selection ratio (sec+tert) -0.001 -0.013
(0.02) (0.41)
Log of remittances 0.008 0.016
(0.52) (1.14)
Urbanization -0.004 -0.003 -0.005 -0.004
(1.97)* (1.84)* (2.87)*** (2.38)**
GDP per capita -0.088 -0.094 -0.112 -0.116
(2.67)*** (3.25)*** (2.85)*** (3.81)***
s
Adult' education -0.375 -0.422 -0.350 -0.383
(2.30)** (2.67)*** (1.91)* (2.15)**
East Asia & Paci...c 0.406 0.387 0.407 0.363
(3.50)*** (3.40)*** (3.98)*** (3.66)***
Sub-Saharan Africa 0.428 0.387 0.519 0.439
(4.55)*** (4.34)*** (4.52)*** (4.40)***
Latin Am. & Carib 0.335 0.298 0.414 0.344
(4.81)*** (4.38)*** (6.15)*** (5.23)***
Mena 0.115 0.192
(1.11) (1.40)
Muslims (% of pop) 0.004 0.005 0.004 0.004
(4.51)*** (5.44)*** (3.11)*** (4.93)***
Catholic (% of pop) 0.002 0.002 0.002 0.003
(2.62)*** (2.86)*** (3.03)*** (3.30)***
Observations 143 143 119 119
R-squared 0.77 0.76 0.84 0.83
Robust t statistics in parentheses
* signi...cant at 10%; ** signi...cant at 5%; *** signi...cant at 1%
25
Table 5. IV regressions (dep = log of fertility rate)
Developing countries
(1) (2) (3) (4)
Constant 1.319 1.388 1.529 1.422
(5.35)*** (6.02)*** (5.62)*** (6.42)***
Log of fertility at dest 0.375 0.360 0.219 0.229
(3.45)*** (3.70)*** (1.91)* (2.12)**
p0 :Log of fertility at dest -0.695 0.041
(1.14) (0.10)
log of p0 0.048 0.005 -0.070 -0.069
(0.62) (0.12) (1.24) (2.22)**
Selection ratio (sec+tert) 0.016 -0.024
(0.44) (0.71)
Log of remittances 0.015 0.024
(1.07) (1.67)*
Urbanization -0.004 -0.003 -0.005 -0.004
(2.06)** (1.41) (2.95)*** (2.62)***
GDP per capita -0.070 -0.099 -0.118 -0.107
(1.81)* (3.29)*** (3.28)*** (3.69)***
s
Adult' education -0.413 -0.482 -0.318 -0.321
(2.55)** (3.01)*** (1.83)* (1.82)*
East Asia & Paci...c 0.425 0.396 0.364 0.334
(3.93)*** (3.72)*** (3.33)*** (3.26)***
Sub-Saharan Africa 0.445 0.378 0.502 0.451
(4.43)*** (4.08)*** (4.99)*** (4.97)***
Latin Am. & Carib 0.353 0.276 0.390 0.361
(4.78)*** (3.96)*** (5.62)*** (5.62)***
Mena 0.098 0.179
(0.88) (1.47)
Muslims (% of pop) 0.004 0.004 0.004 0.004
(4.31)*** (5.20)*** (3.51)*** (5.61)***
Catholic (% of pop) 0.002 0.002 0.003 0.003
(2.22)** (2.87)*** (3.38)*** (3.35)***
Partial Corr First Stage 0.150 0.336 0.183 0.367
F-stat First Stage 9.46 19.00 8.03 25.40
Hansen J Test (p-value) 0.298 0.420 0.803 0.623
R-squared 0.755 0.758 0.829 0.824
Observations 142 142 118 118
Robust t statistics in parentheses.
Instruments for ln(p0): island, ln(size), ln(distance to main destination)
* signi...cant at 10%; ** signi...cant at 5%; *** signi...cant at 1%
26
Endogeneity of fertility norm. It could be argued that the correlation between
home fertility and fertility at destination is driven by reverse causality or unobserved
heterogeneity. In particular, if migration ows tend to respond to variables directly
related to fertility rates, the previous results might be a¤ected by some endogeneity
bias. For instance, if migrants choose their destination because of similarity in fertility
behavior between the origin and the destination country, our key explanatory variable
P
ln(nd ) = ln( d d nd ) might be considered endogenous with respect to the fertility rate
of the origin country through the endogeneity of d ; i.e. the proportion of migrants
living in country d. Spilimbergo (2009) faces a similar problem when analyzing the
relationship between democracy and foreign education since this latter variable is
built using bilateral migration ows of students. If foreign students were choosing
the location of their education abroad on the basis of the democratic distance between
origin and destination, then the relevance of the OLS results would be questioned.
Of course, it is neither obvious that migration decisions are made on the basis
of similarity in fertility behaviors between origin and destination countries, nor that
there is a missing variable such as cultural similarity related to fertility in uencing
the decision to migrate. Focusing on South-North migration, one sees that a lot of
developing countries with colonial links with a Northern country exhibit very high
fertility rates and thus very di¤erent fertility behaviors between origin and destination
countries. Colonial links are known to favour international migration (Mayda, 2005
for instance). Countries such as Congo, Chad and Mali exhibit the highest fertility
rates (well above 6) while their colonizer (Belgium and France) is subject to low
demographic growth. Nevertheless, since our migration data also include South-
South migration, one cannot exclude that there might be some degree of endogeneity
of the migration rates with respect to fertility.
Given the fact that the possible endogeneity of ln(nd ) is related to the endogeneity
of d ; there is no standard econometric procedure such as instrumental variable to
deal with this issue. One intuitive econometric strategy is to use the predicted values
of d using a gravity model relating bilateral migration stocks to exogenous variables
or variables not related to fertility. A similar procedure was used by Spilimbergo
(2009) who uses predicted stocks of foreign students. The econometric procedure
therefore involves two steps. In the ...rst step, we predict the migration weights on
the basis of a set of bilateral exogenous variables. In turn, these predicted weights
allow us to build an alternative measure of the fertility rate at destination. In the
second step, we run the previous regressions using this alternative measure instead
of the observed fertility rate at destination.
We start from the bilateral stocks of migrants from each origin country i to desti-
nation country d observed in 2000, Mid . We regress ln(Mid ) on a set of three types of
variables11 : geographic distance in kilometers (DISid ); existence of a colonial relation-
ship after 1945 (COLid ); since we also investigate South-South migration, a dummy
11
The data come from the CEPII database dist_cepii including a large set of bilateral distance
measures.
27
variable equal to one if the two countries shared a common colonizer (CCOid ); and
linguistic proximity (LINid ). In a ...rst speci...cation, we use a dummy capturing
whether the two countries have the same o¢ cial language. In a second speci...cation,
we use a dummy equal to one if 9 percent of the population in the two countries
speak the same language. The regression model includes also country ...xed e¤ects at
origin and at destination to improve the ...t of the model and to reduce the possibility
of mispeci...cation. The ...rst-stage model is:
ln(Mid ) = + + d + 1 ln(DISid ) + 2 ln(COLid )
i (20)
+ 3 ln(CCOid ) + 4 ln(LINid ) + id
Table 6 reports the estimation results. Not surprisingly, our explanatory variables
are strong determinants of migration stocks. Distance, as a proxy of migration costs
is negatively related to those stocks while colonial links and linguistic proximity favor
migration. For both regressions, the R2 amount to 78 percent, which ensures that
the stocks are well predicted. Both speci...cations yield very similar results, both in
the ...rst step and in the second step. In the subsequent analysis, we use the predicted
weights based on speci...cation (2) but results are strikingly similar using speci...cation
c
(1). Using speci...cation (2), we can predict the log of bilateral stocks ln(Mid ). In
turn, this allows to build an alternative measure of the fertility rate at destination
P ^ ^ X
given by nd =
b c
id nid where id = Mid =
c
Mik :12
i d k
12
Actually, due to unvailaibility of some bialteral variables used in speci...cation (20), the estima-
^
tions involve only 200 countries instead of 208. Nevertheless, the predictions of d are made for all
countries, allowing to keep the full initial sample in step 2.
28
Table 6 : Gravity regressions - All countries
(dep = log of bilateral migration stocks)
(1) (2)
Constant 14.123 15.399
(71.06)*** (78.39)***
Log(distance) -1.067 -1.070
(70.57)*** (71.46)***
Colonial Link 2.061 2.086
(14.00)*** (14.16)***
Common colonizer 0.328 0.373
(9.49)*** (10.98)***
Common o¢ cial language 0.499
(17.15)***
Linguistic Proximity 0.512
(16.75)***
Origin dummies Yes Yes
Destination dummies Yes Yes
Observations 39800 39800
R-squared 0.78 0.78
Robust t statistics in parentheses
* signi...cant at 10%; ** signi...cant at 5%;
*** signi...cant at 1%
In the second step, we use ln(bd ) instead of ln(nd ) to reestimate model (19). We
n
start from the parcimonious speci...cations. Columns (1) and (2) of Table 7 report the
results for the whole sample of countries while columns (3) and (4) concern developing
countries only. Columns (1) and (3) report OLS restimates while columns (2) and (4)
use instrumental variable estimation accounting for the possible endogeneity of the
global emigration rate. It should therefore be obvious that those IV estimates account
simultaneously for two sourves of reverse causality : one due to the endogeneity of
bilateral migration rates used to build the fertility variable at destination and one due
to the endogeneity of the global emigration rate of country i with respect to its fertility
rate. Results of the OLS estimations in Table 7 are fairly similar to results of Tables
2 and 4. The elasticity of the norm amounts to about 0.35 and 0.43, well in line with
the one estimated before. The IV results exhibit a decrease in the signi...cance of the
ln(bd ) variable. This decrease is due to an increase in the standard error rather than
n
the size of the estimated coe¢ cient and is related to the instrumenting procedure of
the global migration rate. A similar results had also been observed before (see Tables
3 and 5). To sum up, we ...nd that the impact of fertilty norms brought by migrants
still survives concerns about the endogeneity of migration weights.
29
Table 7 : Using predicted bilateral stocks - All countries
(Dep=Log of Fertility rate)
OLS IV OLS IV
Constant 0.890 0.977 1.138 0.849
(4.16)*** (3.70)*** (4.72)*** (3.88)***
Log of fertility at dest 0.352 0.406 0.398 0.445
(2.19)** (1.91)* (2.39)** (1.71)*
Log of p0 -0.050 -0.024 -0.032 -0.019
(2.82)*** (0.61) (1.76)* (0.46)
Urbanization -0.006 -0.006 -0.004 -0.005
(3.75)*** (3.23)*** (2.47)** (3.15)***
GDP per capita -0.059 -0.053 -0.081
(2.38)** (1.92)* (2.61)**
s
Adult' education -0.449 -0.723
(2.40)** (4.25)***
East Asia & Paci...c 0.275 0.228 0.391 0.271
(3.71)*** (2.32)** (4.51)*** (2.45)**
Sub-Saharan Africa 0.546 0.546 0.487 0.457
(6.77)*** (6.96)*** (5.58)*** (5.24)***
Latin America & Carib 0.364 0.400 0.350 0.297
(5.04)*** (6.60)*** (3.98)*** (3.96)***
Mena 0.230 0.150 0.185
(2.48)** (1.48) (1.78)*
Muslims 0.005 0.241 0.004 0.213
(6.16)*** (4.35)*** (4.55)*** (4.01)***
Catholics 0.002 0.002
(2.70)*** (2.32)**
Partial Corr First Stage 0.353 0.323
F-stat First Stage 18.14 14.51
Hansen J Test (p-value) 0.932 0.667
R-squared 0.75 0.740 0.76 0.731
Observations 174 174 142 146
Robust z statistics in parentheses
signi...cant at 10%; ** signi...cant at 5%; *** signi...cant at 1%
Col (1)-(3): OLS; C (2)-(4): IV
Instruments for ln(p0): island, ln(size), ln(distance to main destination)
Col (1)-(2) : all countries; Col (3)-(4) : developing countries only
30
Dynamics. We supplement our cross sectional evidence by running a dynamic
model of fertility. The unavailability of migration data for periods prior to 2000
prevents us to estimate a panel regression model. Nevertheless, it might be interesting
to estimate a model linking the change in fertility rates with the distance between the
prevailing fertility rate and the fertility rate at destination. Introducing inertia in a
dynamic model makes the implications of our results stronger, for instance in terms
of convergence. Since the fertility norm validated by our cross sectional analysis is
log(nd ), we run the following regression:
ln (nt+1 ) ln (nt ) = a0 + a1 : ln (nt ) ln(nt d ) + a2 : ln (p0;t ) (21)
+a3 : ln (St ) + a4 : ln (Rt ) + "
The key coe¢ cient is a1 : Our model of transfer of norms implies a1 < 0 : countries
with fertility rates higher (resp. lower) than their fertility at destination are expected
to see a decrease (resp. increase) in their fertility rate. We use the change in fertility
rates between 2000 and 2005 as the dependent variable. The model is estimated with
OLS and involves di¤erent samples. Table 8 and 9 summarize the main ...ndings.
Table 8 use the observed migration stocks to build nt d : Columns 1 to 4 use OLS
while columns 5 and 6 use IV to deal with the prossible endogeneity of ln (p0;t ) :
Table 9 provides exactly the same estimation results, but using predicted bilateral
migration stocks instead of observed ones, as done in the previous section. This
con...rms the existence of a -convergence process. The average value of a1 across the
four regressions is about -1/8. Focusing on the terms in nt and nt d , equation (21) can
be rewritten as:
7 1
ln (nt+1 ) ln (nt ) = (1 + a1 ): ln (nt ) a1 ln(nt d ) + ::: = : ln (nt ) + ln(nt d ) + ::: (22)
8 8
Equation (22) indicates that an equal proportionate increase in the 2000 fertility
rate and the 2000 fertility norm raises the 2005 fertility rate by the same proportionate
amount, with 7/8 of that increase due to the increase in the 2000 fertility rate and 1/8
due to the increase in the fertility norm. The migration rate has a positive impact on
the 2000-to-2005 change in fertility rate in three of the four fertility change regressions
(signi...cant at the 1% level in two of the four regressions and at the 10% level in a third
one), which seems to make more sense than the negative and signi...cant coe¢ cients
obtained in four of the sixteen level regressions.
31
Table 8: Dynamic speci...cation (dep = ln (nt+1 ) ln (nt ))
OLS OLS OLS OLS IV IV
Constant -0.028 -0.050 0.008 -0.014 -0.055 -0.055
(1.11) (1.79)* (0.37) (0.57) (1.54) (1.52)
ln (nt ) ln(nt d ) -0.131 -0.115 -0.122 -0.121 -0.118 -0.117
(5.34)*** (4.28)*** (5.43)*** (4.73)*** (5.25)*** (4.68)***
ln (p0;t ) 0.021 0.029 0.015 0.019 -0.005 0.008
(3.05)*** (4.16)*** (1.54) (1.67)* (0.38) (0.51)
ln (St ) -0.002 0.009 -0.014 -0.002 -0.012 0.003
(0.41) (1.53) (1.44) (0.19) (1.71)* (0.36)
ln (Rt ) -0.008 -0.009
(1.89)* (2.04)**
Partial Corr First Stage 0.326 0.334
F-stat First Stage 18.23 15.84
Hansen J Test (p-value) 0.179 0.169
Observations 146 119 192 153 184 149
R-squared 0.33 0.30 0.24 0.18 0.168 0.187
Robust t statistics in parentheses
signi...cant at 10%; ** signi...cant at 5%; *** signi...cant at 1%
Instruments in IV estimations for ln(p0): island, ln(size), ln(distance to main destination)
Table 9: Dynamic speci...cation with predicted bilateral migration stocks
OLS OLS OLS OLS IV IV
Constant -0.045 -0.068 -0.019 -0.041 -0.089 -0.090
(1.70)* (2.43)** (0.85) (1.65) (2.17)** (2.16)**
ln (nt ) ln(bd )
nt -0.118 -0.116 -0.124 -0.128 -0.124 -0.127
(4.98)*** (4.65)*** (6.00)*** (5.38)*** (5.32)*** (4.91)***
ln (p0;t ) 0.017 0.026 0.012 0.017 -0.012 0.002
(2.35)** (3.55)*** (1.22) (1.53) (0.82) (0.16)
ln (St ) 0.004 0.018 -0.004 0.009 -0.004 0.013
(0.50) (2.56)** (0.44) (0.77) (0.57) (1.62)
ln (Rt ) -0.007 -0.008
(1.58) (1.74)*
Partial Corr First Stage 0.314 0.320
F-stat First Stage 16.83 14.74
Hansen J Test (p-value) 0.203 0.287
Observations 145 118 188 151 182 147
R-squared 0.30 0.30 0.23 0.18 0.13 0.17
Robust t statistics in parentheses
signi...cant at 10%; ** signi...cant at 5%; *** signi...cant at 1%
Instruments in IV estimations for ln(p0): island, ln(size), ln(distance to main destination)
32
Policy implications. The ...ndings presented here have policy implications for
both source countries in the South and host countries in the North.
Developing countries experiencing rapid population growth have typically looked
at migration as one of the means of reducing population pressure and thus of reducing
any social, economic and political problems associated with it. This paper has shown
that South-North migration can also lead to a reduction in fertility rates, and thus
to a more permanent reduction in population pressure in the South, by serving as a
channel for the transfer of low-fertility norms and by raising the incentive to acquire
education.
We can infer from this that an inter-temporal substitution (or tradeo¤) between
present and future migration exists, with an increase in current immigration resulting
in a decrease in future population pressure in the South and thus in a decrease in
future immigration pressure in the North. Developed host countries would bene...t by
taking the inter-temporal substitution in migration into account in the design of their
immigration policy. Doing so should result in a more relaxed immigration policy.
Second, developing emigration countries could achieve a greater reduction in pop-
ulation pressure by ...nding ways of directing emigrants towards the OECD countries
with the lowest fertility rates. This endeavor should be made easier by the fact that
such countries would be likely to be more open to immigration than countries with
higher fertility rates.
5 Conclusion
Though numerous studies have examined the impact of migration on the fertility
of migrants and their household, this paper is the ...rst one to provide a systematic
analysis of the impact of migration on fertility in migrants'home countries. Its main
s
objective was to identify migration' impact on the transfer of destination countries'
norms to migrants'home countries and hence its impact on home countries'fertility
rates.
The paper ...rst provided a theoretical analysis of the various channels through
which international migration might impact fertility in migrants'home country. The
model shows that migration raises adults'incentive to invest in their and their chil-
s
dren' education and thus reduces fertility, that it raises remittance levels and that
these have an ambiguous impact on fertility, and that the transfer of norms from low-
(high-) fertility destination countries reduces (raises) fertility in migrants'countries
of origin.
s
The model' predictions are supported by the empirical results. Regarding the
transfer of norms, we found that a one percent decrease in the fertility norm to
which migrants are exposed reduces home country fertility by about 0.3 percent for
developing countries as well as for all countries. Thus, migration from high-fertility
home countries to low-fertility destination countries reduces fertility in the former
ones.
33
The ...ndings presented here have a number of policy implications. Developing
countries'authorities that have experienced rapid population growth continue to be
greatly concerned with the potential social, economic and political problems associ-
ated with it. These countries have typically looked at migration as one of the (static)
ways of reducing population pressure. This paper has shown that South-North mi-
gration can lead to a reduction in fertility rates and thus contribute to a reduction
in home country population pressure by serving as a channel for the transfer of low-
fertility norms and by raising the incentive to acquire education.
The tradeo¤ implicit in the impact of migration on fertility should be taken into
account by developed host countries since accepting more migrants in the short run
may reduce the migration pressure in the long run. Moreover, emigration countries
could achieve a greater reduction in population pressure by ...nding ways of directing
emigrants towards the OECD countries with the lowest fertility rates. These countries
are also likely to be more open to migration than those with higher fertility.
Finally, we should note that further research of various aspects of this issue is
on our research agenda including possible di¤erences in the home country fertility
impact of a transfer of host country norms by men and by women, and di¤erences in
that impact for fertility norms that are higher or lower than the home country fertility
s
, and we hope that this paper will trigger other people' interest in contributing to
the research e¤ort in this area.
6 References
Beine, M., F. Docquier and H. Rapoport (2001), "Brain drain and economic
growth: theory and evidence", Journal of Development Economics, 64(1), 275-
289
Beine, M., F. Docquier and H. Rapoport (2008), "Brain Drain and Human
Capital Formation in Developing Countries: Winners and Losers", Economic
Journal, 118(528), 631-652.
Ben-Porath, Y. (1973), "Economic Analysis of Fertility in Israel: Point and
Counterpoint", Journal of Political Economy 81(2): 202-233.
Brockero¤, M. (1995), "Fertility and Family-planning in African Cities: the
Impact of Female Migration."Journal of Biosocial Science 27(3): 347-358.
Chen, Hung-Ju (2006), "International migration and economic growth: a source
country perspective", Journal of Population Economics, 19: 725-748.
De la Croix, D. and M. Doepke (2003), "Inequality and growth: why di¤erential
,
fertility matters" American Economic Review, 93(4): 1091-1113.
34
De la Croix, D. and M. Doepke (2004), " Private versus public education when
,
di¤erential fertility matters" Journal of Development Economics, 73(2): 607-
629.
Docquier, F., O. Faye, P. Pestieau (2008), "Is migration a good substitute for
education subsidies?", Journal of Development Economics 86, 263-276.
Docquier, F., B. L. Lowell and A. Marfouk (2007), "A Gendered Assessment of
the Brain Drain," IZA Discussion Papers 3235, Institute for the Study of Labor
(IZA).
Ebanks, E., P.M. George, C. E. Nobbe (1975), "Emigration and Fertility Decline:
The Case of Barbados."Demography 12(3): 431-445.
Farber, S.C. and B.S. Lee (1984), "Fertility Adaptation of Rural-to-Urban Mi-
grant Women: A Method of Estimation Applied to Korean Women." Demog-
raphy 21(3): 339-345.
Fargues, Ph. (2007), "The demographic bene...t of international migration: a
hypothesis and its application to Middle Eastern and North African countries",
in: Ozden, C. and M. Schi¤ (eds), International migration, economic develop-
ment and policy, World Bank and Palgrave Macmillan: Washington DC.
Freedman, R. and D. P. Slesinger (1961), "Fertility Di¤erentials for the Indige-
nous Non-Farm Population of the United States." Population Studies 15(2):
161-173.
Galor, O. and A. Mountford (2006), "Trade and the great divergence: the family
.
connection" American Economic Review, 96(2): 299-303.
Goldberg, D. (1959), "The Fertility of Two-Generation Urbanites."Population
Studies 12(3): 214-222.
Goldberg, D. (1960), "Another Look at the Indianapolis Fertility Data." Mil-
bank Memorial Fund Quarterly 38(1):23-36.
Goldstein, S. (1978) "Migration and Fertility in Thailand, 1960-1970." Cana-
dian Studies in Population 5: 167-180.
Goldstein, S. and A. Goldstein (1981), "The Impact of Migration on Fertility:
An `Own Children'Analysis for Thailand."Population Studies 35(2): 265-284.
Gould, D.M. (1994), "Immigrant Links to the Home Country: Empirical Im-
plications for U.S. Bilateral Trade Flows,"Review of Economics and Statistics,
76: 302-16.
35
Hervitz, M.H. (1985), "Selectivity, Adaptation, or Disruption? A Comparison
of Alternative Hypotheses on the E¤ects of Migration on Fertility: The Case of
Brazil."International Migration Review 19(2): 292-317.
Hiday, V.A. (1978), "Migration, Urbanization, and Fertility in the Philippines."
International Migration Review 12(3): 370-385.
Javorcik, B., C. Özden, M. Spatareanu and C. Neagu (2006) "Migrant Net-
works and Foreign Direct Investment." World Bank Working Paper no. 4046.
Washington DC.
Katav-Herz, Shirit (2003), "A Model of Parental Demand for Child Labor with
High Fertility Norms", Review of Economics of the Household, 1(3): 219-233.
Kugler, M. and H. Rapoport (2006), "Skilled Emigration, Business Networks
and Foreign Direct Investment."Economic Letters.
Kulu, H (2003), "Post-war Immigration to Estonia: A Comparative Perspec-
tive," in R. Ohliger, K. Schönwälder and T. Triada...lopoulus (eds.) "Euro-
pean Encounters, 1945-2000: Migrants, Migration and European Societies since
1945."Aldershot: Ashgate, pp. 38-52.
Lee, B.S. and L.G. Pol (1993), "The In uence of Rural-Urban Migration on
Migrants Fertility in Korea, Mexico and Cameroon."Population Research and
Policy Review 12(1): 3-26.
Lee, B.S. and S.C. Farber (1985), "The In uence of Rapid Rural-Urban Migra-
tion on Korean National Fertility Levels." Journal of Development Economics
17:47-71.
Lindstrom, D.P. and S. Giorguli Saucedo (2002), "The Short- and Long-Term
s
E¤ects of U.S. Migration Experience on Mexican Women' Fertility." Social
Forces 80(4): 1341-1368.
Lindstrom, D.P. and Munoz-Franco (2005), On contraceptives
Lopez, R. an M. Schi¤ (1998), Migration and the Skill Composition of the
Labor Force: the Impact of Trade Liberalization in LDCs, Canadian Journal of
Economics, 31(2), 318-336.
Martine, G. (1975), "Migrant Fertility Adjustment and Urban Growth in Latin
America."International Migration Review 9(2): 179-191.
Mayda, A.M. (2007), International Migration: a Panel data Analysis of the
Determinants of Bilateral Flows, Georgetown University Working paper.
36
Moav, O. (2005), " Cheap children and the persistence of poverty," Economic
Journal, 115: 88-110.
Mountford, A (1997), "Can a Brain Drain Be Good for Growth in the Source
Economy", Journal of Development Economics 53 (2): 287-303.
Mountford, A. and H. Rapoport (2007), "The brain drain and the world distri-
bution of income and population", Discussion paper n. 04/07, CReAM, Uni-
versity College London.
Myers, G.C. and E. W. Morris (1966), "Migration and Fertility in Puerto Rico."
Population Studies 20(1): 85-96.
Park, J.Y. and I.H. Park (1976), "Migration and Female Labor Force Impact on
Korean Fertility." In "Dynamics of Migration: Internal Migration and Fertility."
Occasional Monograph Series Vol. 1 No. 5 Interdisciplinary Communications
Program, Smithsonian Institution.
Parsons, C.R., R. Skeldon, T.L. Walmsley and L.A. Winters (2007), "Quanti-
fying International Migration: A Database of Bilateral Migrant Stocks", World
Bank Policy Research Working Paper No. 4165, World Bank.
Rauch, James (2001), "Business and Social Networks in International Trade,"
Journal of Economic Literature 39: 1177-1203.
Rauch, James and Vitor Trinidade (2002), "Ethnic Chinese Networks In Inter-
national Trade,"Review of Economics and Statistics 84(1): 116-130.
Rosenzweig, M.R and T.P. Schultz (1985), "The Demand for and Supply of
Births: Fertility and its Life Cycle Consequences", American Economic Review
75(5): 92-105.
Sato, Y. and K. Yamamoto (2005), "Population Concentration, Urbanization,
,
and Demographic Transition" Journal of Urban Economics 58(1): 45-61.
,
Sato, Y. (2007) "Economic Geography, Fertility and Migration" Journal of
Urban Economics 61(2): 372-387.
Spilimbergo, A. (2009), Democracy and foreign education, American Economic
Review, 528-543, 99(1).
Stephen, E.H. and F.D. Bean (1992), "Assimilation, Disruption and the Fertility
of Mexican-origin Women in the United States."International Migration Review
26(1): 67-88.
37
Umezaki, M. and R. Ohtsuka (1998), "Impact of Rural-Urban Migration on
Fertility: A Population Ecology Analysis in the Kombio, Papua New Guinea."
Journal of Biosocial Science 30(3): 411-422.
White, M.L., L. Moreno, and S. Guo (1995), "The Interrelation of Fertility
and Geographic Mobility in Peru: a Hazards Model Analysis." International
Migration Review. 29(2): 492-514.
38
7 Appendix
7.1 Analytics of section 3.1
We solve the model in two steps and proceed backwards. First, for a given location,
parents choose their optimal number of children. Second, after substituting this
number in the utility function, parents decide how much to invest in education taking
into account the endogenous probability of emigrating.
conditional'utility function is given by
In the case of migration, the '
h 1
i
Utf = log (1 Et nt )AEt h wf + log [nt ] + C:
where the constant term C stands for the given levels of human capital and expected
wage of their children.
The optimal fertility rate is equal to
(1 E)
nt = ;
(1 + )
and is clearly decreasing with the time spent by adults in higher education (before
having children). Substituting the optimal fertility rate in the utility function gives
the quasi-indirect utility function which depends on parents'education choice:
Vtf (Et ) = (1 + ) log(1 Et ) + log(Et ) + log(wf ) +
h i
where log (1+ ) log(1 + ) + log(A) + (1 ) log(h) + C is a constant.
In the case of staying, their conditional utility function is given by
h 1
i
Uth = log (1 Et nt )AEt h + log [nt ] + C
The optimal fertility rate is identical to the one of migrants and the quasi-indirect
utility function becomes
Vth (Et ) = (1 + ) log(1 Et ) + log(Et ) +
Agents then maximize the expected utility function, (1 pt )Vth +pt Vtf . The choice
of higher education solves the following optimization problem
h 1
i
f
fEt g = arg max(1 + ) log(1 Et ) + 1 + p0 log(Ah w ) log(Et )
The optimal investment in higher education is given by
1 + p0 log(wf )
Et = h 1
i
1+ + 1 + p0 log(Ah wf )
39
7.2 Analytics of section 3.2
The optimization problem for remaining adults can thus be written as following:
fnt ; et g = arg max flog [(1 nt )Ht nt et ] + log [nt et (1 + !p0 et )]g
The ...rst order conditions (with respect to nt and et ) can be expressed as
Ht + et
=
(1 nt )Ht nt et nt
nt !p0 et 1
= +
(1 nt )Ht nt et et 1 + !p0 et
The ...rst condition is standard and implies that the total cost of children (raising
s
cost + education) is proportional to the parent' maximal wage at the equilibrium
nt ( Ht + et ) = Ht ;
1+
This implies
Ht
nt = ;
(1 + )( Ht + et )
s
i.e. fertility decreases with children' education for a given parental income.
Combining the conditions yields an implicit polynomial solution for the optimal
investment in education
(1 2 )!p0 et +1 + (1 )et 2 Ht !p0 et Ht = 0
1
Assuming = 2 , the implicit function above becomes quadratic in et and gives
rise to an explicit solution. The optimal investment in children education becomes
h p i2
et = Ht !p0 + ( Ht !p0 ) 2+ H
t
7.3 Analytics of section 3.3
In the ...rst sub-case, we assume that remittances rt are received by young parents.
The optimization problem of remaining adults can thus be written as the following
fnt ; et g = arg max flog([(1 nt )Ht nt et + rt ] + log [nt et (1 + !p0 )]g
The ...rst order conditions (with respect to nt and et ) can be expressed as
Ht + et
=
(1 nt )Ht nt et + rt nt
nt
=
(1 nt )Ht nt et + rt et
40
s
As usual, the optimal cost of children is proportional to the parent' maximal
income
nt ( Ht + et ) = (Ht + rt ):
1+
Combining the ...rst order conditions yields the following explicit solution for hu-
man capital investments
Ht
et = ;
1
and for the fertility rate
r
(1 ) 1 + Htt
nt =
(1 + )
In the second sub-case, we assume that working-aged children transfer a fraction
o
of their income to their parents. Parents also receive other remittances rt+1 from
extra-family members when old, the utility function would become:
o
e
Ut = log(ct ) + log( wt+1 ht+1 nt + rt+1 )
Adults'optimization problem can thus be written as the following
o
fnt ; et g = arg max log([(1 nt )Ht nt et ] + log nt et (1 + !p0 ) + rt+1
The ...rst order conditions (with respect to nt and et ) can be expressed as
Ht + et et (1 + !p0 )
= o
(1 nt )Ht nt et rt+1 + nt et (1 + !p0 )
nt nt et 1 (1 + !p0 )
= o
(1 nt )Ht nt et rt+1 + nt et (1 + !p0 )
Combining the ...rst order conditions yields the following explicit solution
Ht
et =
1
and for the fertility rate,
o
(1 ) rt+1
nt =
1+ (1 + ) et (1 + !p0 )
7.4 Analytics of section 3.4
The optimization problem of non-migrant adults becomes
fnt ; et g = arg max flog([(1 nt )Ht nt et ] + log [(nt e
nt ) et (1 + !p0 )]g
41
The ...rst order conditions (with respect to nt and et ) become
Ht + et
=
(1 nt )Ht nt et nt e
nt
nt
=
(1 nt )Ht nt et et
From the second condition, we can easily derive the optimal investment in children
education as a function of the fertility rate
e
Ht (nt nt )
et =
n(1 e
)+n
Substituting this equation in the ...rst conditions gives, after straightforward ma-
nipulations, a quadratic implicit function in nt :
(1 + )n2
t e
[ nt + (1 )] nt e
nt = 0
The single positive root of this equation is the optimal fertility rate:
q
e
nt + (1 e
) + [ nt + (1 )]2 + 4 (1 + )
nt =
2 (1 + )
42