Migration and Fertility: Competing
Hypotheses Re-examined
HILL KULU
Max Planck Institute for Demographic Research, Konrad-Zuse-Straße 1, 18057 Rostock,
Germany (e-mail: kulu@demogr.mpg.de).
Received 13 May 2004; accepted in final form 12 August 2004
Kulu, H., 2005, Migration and fertility: competing hypotheses re-examined, European Journal
of Population, 21: 51­87.
Abstract. Competing views exist concerning the impact of geographical mobility on childbearing
patterns. Early research shows that internal migrants largely exhibit fertility levels
dominant in their childhood environment, while later studies find migrants' fertility to
resemble more closely that of natives at destination. Some authors attribute the latter to
adaptation, others claim the selection of migrants by fertility preferences. Moreover, shortterm
fertility-lowering-effects of residential relocation have also been proposed and challenged
in the literature. This paper contributes to the existing discussion by providing an analysis of
the effect of internal migration on fertility of post-war Estonian female cohorts. We use
retrospective event-history data and apply intensity regression for both single and simultaneous
equations. Our analysis shows that first, the risk of birth for native residents decreases
with increasing settlement size and the decrease is larger for higher-order parities. Second, it
shows that migrants, whatever their origin, exhibit fertility levels similar to those of nonmigrants
at destination. We also observe elevated fertility levels after residential relocations
arising from union formation. Our further analysis supports the adaptation hypothesis. We
find no evidence on (strong) selectivity of migrants by fertility preferences.
Key words: Estonia, fertility, intensity regression, internal migration, simultaneous equations
Kulu, H., 2005, Migration et fé condité : un ré examen ďhypothe` ses concurrentes, Revue
Europé enne de Dé mographie, 21: 51­87.
Ré sumé . Les diffé rentes é tudes de ĺinfluence de la mobilité gé ographique sur la constitution de
la famille aboutissent a` des ré sultats contradictoires. Les premie` res recherches ont montré que
les migrants internes pré sentaient des niveaux de fé condité comparables a` ceux qui pré valaient
dans ĺenvironnement de leur enfance alors que des é tudes posté rieures font é tat ďune
fé condité des migrants ressemblant davantage a` celle de la population ďaccueil. Certains
auteurs attribuent cette observation a` un phé nome` ne ďadaptation tandis que ďautres mettent
ĺaccent sur un effet de sé lection des migrants en fonction de leurs choix de fé condité . De plus,
certains travaux insistent sur le phé nome` ne de baisse provisoire de la fé condité au moment du
dé placement. Cet article apporte une contribution au dé bat en analysant ĺeffet des migrations
internes sur la fé condité des gé né rations fé minines né es en Estonie apre` s la guerre. Nous
utilisons des donné es biographiques ré trospectives et appliquons une ré gression sur ĺintensité
pour des é quations simples et simultané es. Notre analyse montre ďune part que la probabilité
European Journal of Population (2005) 21: 51­87 Ó Springer 2005
DOI 10.1007/s10680-005-3581-8
de naissance pour les personnes né es sur place est ďautant plus faible que la taille du lieu de
ré sidence est importante et que cette relation est plus importante aux parité s é levé es. Ďautre
part elle montre que les migrants, quelle que soit leur origine, pré sentent des niveaux de
fé condité comparables a` ceux des non-migrants du lieu ďarrivé e. On observe é galement une
fé condité plus é levé e apre` s un dé placement lié a` la formation ďune union. Enfin notre analyse
renforce ĺhypothe` se ďadaptation. Aucune preuve ďune (forte) sé lection des migrants en
fonction de leurs pré fé rences de fé condité n'a é té retrouvé e.
Mots clé s: fé condité , migration interne, ré gression sur ĺintensité , é quations simultané es,
Estonie
1. Introduction
During the past century, many nations witnessed increasing population
spatial mobility and its concentration in the major centres of the country
(Woods, 2003). While the peak of rural­urban migration in (economically)
more developed countries passed a long time ago, and the level of urbanisation
has reached close to its maximum today, the overall spatial mobility
does not seem to exhibit any signs of decreasing. Considering Zelinsky's
(1971) `theory of the mobility transition', the different forms of spatial
mobility simply replace urbanisation when a country proceeds from the
`transitionaĺ to the `advanced stage' in its demographic development.
Increasing inter-urban movement, trends of sub-urbanisation and counterurbanisation
experienced by many European and North American countries
in the past (Geyer and Kontuly, 1996), give support for Zelinsky's general
argument, despite the fact that the theory had not been able to foresee all
these trends (Cadwallader, 1993).
Needless to say, moving from one place to another is an important life
event, accompanied by both short and long-term changes in an individuaĺs
life. While the effect of migration on different life domains of an individual
seems rather self-evident, different views exist concerning the impact of a new
social environment on childbearing preferences and behaviour of migrants.
Earlier research has proposed four partly complementary, partly contradictory
hypotheses about how the patterns of fertility might appear following
migration (Hervitz, 1985; Rundquist and Brown, 1989; Lee, 1992; Singley
and Landale, 1998).
The socialisation hypothesis relies on the premise that fertility behaviour of
migrants reflects the fertility preferences dominant in their childhood environment.
Therefore, migrants exhibit similar fertility levels to stayers at
origin and the convergence towards fertility levels of population at destination
occurs only in the next generation (given that the differences exist). The
adaptation hypothesis, in contrast, premises on an individuaĺs re-socialisation
possibility, and suggests that the fertility behaviour of migrants, sooner or
later, comes to resemble the dominant behaviour at the destination
HILL KULU52
environment. The selection hypothesis, in turn, argues that changing behaviour
is not a question, yet rather the fact that migrants are a specific group of
people whose fertility preferences are more similar to those of people at
destination than at origin. Finally, the disruption hypothesis suggests that
immediately following migration, migrants show particularly low levels of
fertility due to the disruptive factors associated with the migration process.
This paper contributes to existing discussion, providing an analysis of
fertility of internal migrants in Estonia. The objectives are as follows: first, to
examine the fertility differences between people who move and those who
stay at various origin and destination environments, second, to look at the
role of various factors proposed in the literature in accounting for observed
fertility patterns of migrants. The study uses retrospective event-history data,
and applies intensity regression for both single and simultaneous equations,
with the aim to arrive at a more comprehensive insight into the causes of
fertility behaviour of migrants. The paper proceeds as follows: first, we will
give an overview of previous research, and specify arguments of different
views. Then, we will briefly describe the study context, and introduce the
data, methods and variables. Thereafter, we will present the results of a
multivariate analysis. This leads us finally to the discussion on the role of
migration in shaping an individuaĺs fertility behaviour.
2. Views on the impact of migration on fertility
The rise of the socialisation hypothesis in internal migration-fertility literature
is largely associated with Goldberg's (1959, 1960) two studies. Goldberg's
main interest was to examine the socioeconomic differences in fertility in
urban areas, which many previous studies had found. While research had
established an inverse relation of fertility to socioeconomic status ­ whitecollar
families were smaller than blue-collar families ­, Goldberg hypothesised
that in reality this relationship might not be so simple as it appeared.
Namely, the larger fertility of blue-collar workers might result from occupational
selectivity of rural migrants whose fertility was expected to be higher
than that of native urban residents because of different childhood socialisation.
To test the hypothesis, Goldberg examined fertility of populations of
Detroit and Indianapolis. Both studies showed that people with a rural
background exhibited significantly higher levels of fertility than native (twogeneration)
urbanites, and the overall socioeconomic differences in fertility
could be attributed primarily to the fertility behaviour of rural migrants and
their concentration in lower social and economic positions in the city.
Several other papers studying the socioeconomic differences of fertility in
the U.S. gave indirect support for the socialisation hypothesis. Inspired by
Goldberg's research, Freedman and Slesinger (1961) analysed the data on the
MIGRATION AND FERTILITY: COMPETING HYPOTHESES RE-EXAMINED 53
U.S. total population and found that a traditionally observed negative correlation,
between either income or education with fertility, disappears when
we consider only a native urban population. Thus, overall socioeconomic
differences in fertility within an urban population did result from the differences
among rural­urban migrants and their over-representation in lower
income and educational groups. Duncan's (1965) research confirmed previously
observed patterns and led him to conclude that a `modern fertility
pattern' could be reached either by non-rural rearing or by prolonged contact
with the educational system. Later, however, McGirr and Hirschman (1979)
showed that significant socioeconomic differences did exist among earlier
cohorts of rural­urban migrants, but not among more recent generations.
Their research also indicated that despite decreasing socioeconomic differences,
rural­urban migrants had slightly higher fertility in most educational
groups.
Surprisingly, however, from later literature on fertility of internal migrants,
we can find very few studies dealing with the socialisation hypothesis.
One notable exception is a study on fertility of inter­regional migrants in
Brazil by Hervitz (1985), where he found only limited support for the
hypothesis. Meanwhile, many studies dealing with fertility of immigrants
have supported the main arguments of the socialisation hypothesis, although
using different rhetoric (the assimilation hypothesis). Rosenwaite's (1973)
study showed that first­generation Italian­Americans maintained their specific
fertility behaviour, while the second generation exhibited similar
behaviour to native Americans. More recently, Stephen and Bean (1992)
found similar inter-generational differences for Mexican­Americans, and
Kahn's (1994) research showed no evidence of changes in fertility behaviour
for most immigrant groups in the U.S.
While the socialisation hypothesis received support mainly in early
internal migration-fertility literature, the adaptation hypothesis seems to be
widespread and popular later on as well. Examples of early studies supporting
the adaptation hypothesis are research by Myers and Morris (1966),
and Goldstein (1973). The former examined fertility of internal migrants in
Puerto Rico using the census data on current residence and place of birth.
As opposed to dominant research at that time, their study showed that
migrants from rural to urban areas exhibited the same levels of fertility as
the native urban population. Goldstein (1973) arrived at largely similar
results when examining fertility of rural­urban migrants in Thailand. He
found that the fertility levels of migrants, especially in the capital city of
Bangkok, were well below those of the non-migrants in the rural areas from
which most of the migrants came. Later, Hiday (1978) showed that previous
findings also applied to fertility of internal migrants in the Philippines.
While all of these studies found the fertility levels of migrants to be more
similar to those of the population at destination than at origin, the authors
HILL KULU54
were still careful to give their full support for the adaptation hypothesis, as
it remained unclear whether migration did operate as a cause or an effect of
low fertility.
More recently, the adaptation hypothesis has been tested and supported
by several authors. Farber and Lee (1984) examined the effect of rural­urban
migration on fertility in Korea. To control the possible preference selectivity,
a model was constructed where they compared fertility paths of individuals
who had already migrated (post­migrants), and people who had not yet
migrated, but were known to migrate later (pre­migrants). The authors
found that two paths differed significantly, and concluded that rural­urban
migration slowed down the fertility rate of Korean women. Later Lee and
Pol (1993) showed a significant rural­urban adaptation in Mexico, but not in
Cameroon, which they attributed to the specific context of African fertility
transition (the fertility increasing effect of urban residency due to reduced
infertility).
Brockeroff and Yang (1994), however, found support for the adaptation
hypothesis in the African context as well. Their comparative study on fertility
of rural­urban migrants in six countries indicated that migrants' risk of
conception declined dramatically in all countries around the time of migration,
and remained low in the long run among most migrant groups. Further
analysis showed that the decline in migrants' fertility could be largely
attributed to a pronounced improvement in the standard of living after
migration and the increasing use of modern contraceptive methods. Brockeroff's
(1995) subsequent study on fertility of rural­urban migrants in 13
African countries confirmed previous findings. Clear evidence supporting the
adaptation hypothesis can likewise be found in Hervitz's (1985) research on
fertility of inter-regional migrants in Brazil, and in a recent paper by Umezaki
and Ohtsuka (1998) on fertility of rural­urban migrants among the
Anjangmui dialect group in Papua New Guinea.
The selection hypothesis has been discussed in many papers, but examined
in only a few studies. Myers and Morris (1966), and Goldstein (1973), raised
the question of migrant selectivity in the final sections of their papers. Still,
some studies at that time also addressed the issue. Macisco et al. (1970)
compared fertility of migrants and non-migrants in San Juan, Puerto Rico.
While both groups exhibited significantly lower levels of fertility than the
rural population, the fertility of migrants was even lower than that of urban
natives. The analysis showed that a higher activity rate and education level of
migrants explained some of the differences, but not all. This led the authors
to conclude that the rural­urban migrants in Puerto Rico were more oriented
toward achievement and innovation, as were the stayers. Early marriage and
children in rural areas might be viewed as obstacles to upward mobility, and
the response was to delay marriage and fertility, favouring higher education
and migration to the capital city (Zarate and Zarate, 1975, p. 125).
MIGRATION AND FERTILITY: COMPETING HYPOTHESES RE-EXAMINED 55
Hendershoťs (1971) similar analysis showed lower migrant fertility at older
ages, but higher fertility among younger migrants to Manila in the Philippines.
The author attributed the differences largely to changing migration
streams: While in early stages of urbanisation, rural­urban migration was
difficult and therefore selective, whereas in later stages it was less difficult and
therefore also less selective (Zarate and Zarate, 1975, p. 137).
Many subsequent papers discussed the issue of migrant selectivity
(Goldstein and Goldstein, 1981; Murphy and Sullivan, 1985). Yet, the study
by Courgeau (1989) on fertility of rural­urban and urban­rural migrants in
France also provided clear evidence supporting the hypothesis. The multivariate
analysis of longitudinal data showed that migration to the city significantly
reduced a woman's fertility, whereas migration to rural settlements
increased it. However, further analysis revealed that migration to rural areas
attracted women whose fertility before the move was similar to that of other
women in the urban area, while migration to urban areas attracted women
whose fertility was similar to that prevailing in the urban areas. Thus, urban­
rural migrants adapted to the behaviour dominant in the rural areas, while
rural­urban migrants were a selective group, according to their fertility
preferences. Recently, White et al. (1995) found evidence supporting the
selection hypothesis when analysing fertility of internal migrants in Peru.
More specifically, the new residents in larger locations in general, and in the
capital city in particular, were more likely to arrive with lower lifetime fertility
preferences.
The disruption hypothesis, assuming the short­term fertility­lowering­effect
of migration, has found direct or indirect support in many studies.
Goldstein's (1973) early analysis on migrant fertility in Thailand showed that
fertility levels of lifetime migrants were not very different from those of nonmigrants
at destination, while the fertility of recent migrants (those who have
been living in a new destination less than five years) was considerably lower.
He attributed this phenomenon to possible cohort changes or disruption
effect, resulting from spousal separation. In his later study, however, he and
his co-author tended to favour the disruption hypothesis, although they
alternatively proposed that the phenomenon might also be related to a low
overall fertility of migrants that later `caught up' to the corresponding levels
of urban population (Goldstein and Goldstein, 1981). A few years later,
Hervitz (1985) brought clear evidence supporting the disruption hypothesis in
his study on migrant fertility in Brazil. More recently, Brockerhoff (1995)
demonstrated a very low fertility of urban­rural migrants during their first
few years in cities in several African countries, which he attributed to the
unmarried status of migrants and to high levels of spousal separation among
married migrants. White et al. (1995) analysed migrant fertility in Peru using
longitudinal data, and showed that residential relocation lengthened the birth
interval of migrants.
HILL KULU56
Recently, however, several authors studying immigrant fertility have
implicitly or explicitly challenged the disruption hypothesis. Singley and
Landale (1998) compared the risk of the first birth of several groups of
Puerto Rican women (born in Puerto Rico, but residing in the U.S., residing
in Puerto Rico and the U.S.-born Puerto Ricans) using longitudinal data.
Their analysis revealed that single women migrating to the U.S. were much
more likely than their non-migrant counterparts in Puerto Rico to form
unions and experience a conception, either in unions or outside. The authors
concluded that migration to the U.S. should be seen as a part of the family
building process for many Puerto Rican women. Andersson (2004) arrived at
very similar conclusions when examining immigrant fertility in Sweden. The
analysis of risk of the first birth showed elevated levels of childbearing during
the first couple of years after immigration to Sweden. Moreover, the author
found migration to trigger, rather than disrupt the process of childbearing,
also for higher birth orders. The study by Mulder and Wagner (2001) on
family formation and home ownership in West Germany and the Netherlands,
in turn, demonstrated increasing rates of first childbirth shortly after a
couple had moved to their own house.
To sum up, different hypotheses have been proposed to predict and explain
fertility patterns of migrants. Each of these hypotheses has received
support in the literature, and has also been challenged. Each perspective
draws upon some theoretical view, assuming a variety of factors to be more
important than others in shaping migranťs childbearing behaviour. The
socialisation hypothesis emphasises the critical role of the childhood environment.
The norms and values dominant in a migranťs childhood environment
guide her/his later actions in other places as well. To the contrary,
the adaptation hypothesis assumes that what matters most in shaping the
migranťs fertility behaviour is her/his current socio-cultural and economic
environment. The selection hypothesis also seems to emphasise the importance
of the childhood environment. However, norms and values differ across
population sub-groups, and the `minority' later moves to locations where
values similar to theirs dominate. Finally, the disruption hypothesis argues
that economic and psychological costs of residential relocation cannot be
underestimated when studying fertility patterns of migrants.
Contradictory conclusions of studies often arise from different time,
context and types of migrations investigated. Various methodologies used
may also account for some differences. In this context, some critical remarks
on dominant research methodologies are inevitable. First, most studies use
cross-sectional (usually census) data while studying the effect of migration on
fertility. Longitudinal data have found only limited use, despite their dominant
position in many areas of population research. Clearly, the lack of
information on the precise timing of migration, fertility and other factors
restricts any causal inferences to be made about the migration-fertility
MIGRATION AND FERTILITY: COMPETING HYPOTHESES RE-EXAMINED 57
relationships. Second, most recent studies have successfully controlled for
selection of migrants by various socioeconomic factors, while assessing the
impact of migration on fertility. However, the possible unobserved selectivity
of migrants by fertility preferences has been addressed only in a few studies
(e.g. Courgeau, 1989; Montgomery, 1992; Michielin, 2002). This fact has
further intensified the difficulties when drawing conclusions about migrationfertility
relationships. Therefore, using retrospective event-history data, and
controlling for unobserved selection of migrants when examining the effect of
various factors on migrant fertility, should be seen as major contributions of
this paper. Before we introduce the data and methods, however, we will
briefly outline the context of the research.
3. Fertility and migration trends in estonia
The beginning of the fertility transition in Estonia can be traced back to the
mid-19th
century. The 1850­1860s signified a period when fertility levels
began to decrease and gradually approached levels characteristic of a
`modern fertility regime'. By the late 1920s, period fertility had already
reached below replacement level in Estonia (Katus, 2000, pp. 215­216). With
its relatively early fertility transition, Estonia (and neighbouring Latvia) has
been associated with the group of `pioneering nations' of the demographic
transition, along with France, Switzerland, Sweden and Norway (Katus
et al., 2002, p. 143). While earlier fertility developments in Estonia followed
patterns common in Western and Northern Europe, the post World War II
trends differed. Estonia did not experience a post-war baby boom, and period
fertility remained below replacement level until the mid-1960s. Thereafter it
slightly increased among its native population, and stayed above replacement
level until the late 1980s, when a rapid fertility decrease, characteristic of the
post-socialist transition period, began. Cohort fertility, based on the Estonian
Family and Fertility Survey, shows relatively stable and low levels
among cohorts of native people born from the 1910s to the 1940s, and a
slight increase among the cohorts born in the late 1940s and the 1950s (Katus
et al., 2002, p. 145).
While fertility levels remained rather stable in post­war Estonia, the
trends in population migration were far more dynamic. Besides the intensive
immigration from other parts of the Soviet Union (Kulu, 2003a), intensive
urbanisation also characterised the post-war period. In 1939, 33% of the
Estonian population lived in urban areas. The share of urban population
increased to 56% in 1959 and to 71% in 1989 (Tammaru, 2001, p. 580). The
cities grew both as a result of (internal) rural­urban migration and immigration
that was overwhelmingly directed to urban areas. While external
migration fed the urban growth until the late 1980s, trends in internal
HILL KULU58
migration were different. Since the late 1970s, urban­rural migration increased,
and in the following decade, rural areas in Estonia witnessed a
positive net migration for the first time in the country's history. The migration
turnaround has been attributed to increasing investments by the state in
agricultural production, and also to the changing preferences of people (cf.
Marksoo, 1992, p. 134). During the 1990s, the concentration in major centres
again became the dominant trend among the working age population, and
was later accompanied by increasing sub-urbanisation. The share of the urban
population, however, decreased from 71% in 1989 to 67% in 2000,
mostly as a result of emigration of Russians and other ethnic minorities
(Tammaru, 2001).
The fertility of internal migrants in Estonia has not been studied, nor is
much known about regional variation in fertility levels and urban­rural
differences. In this context, a follow-up of the recent census (2000) data on
the Estonian native population is highly informative. We see significant
differences that exist in fertility levels of population across settlement hierarchy
(Figure 1). Fertility in rural areas is clearly above replacement levels in
all birth cohorts who have already passed, or are about to reach the end of
their childbearing ages. Fertility in the urban population, however, remains
below replacement level, comprising about 80% and 65% of rural levels in
towns and the capital city of Tallinn, respectively (a very similar picture also
applies for the immigrant population). It is also striking that fertility differences
are rather stable, and do not change much across the cohorts, as one
might assume.
0.5
1.0
1.5
2.0
2.5
3.0
1910-14 1915-19 1920-24 1925-29 1930-34 1935-39 1940-44 1945-49 1950-54 1955-59 1960-64
Villages Towns Tallinn Total
Figure 1. Cohort fertility of native population by type of current settlement in 2000.
Source: Estonian Population Census 2000.
MIGRATION AND FERTILITY: COMPETING HYPOTHESES RE-EXAMINED 59
4. Data and research population
The data for our study come from the Estonian Family and Fertility Survey
(FFS). The Estonian FFS was carried out in 1994 among 5,021 women born
between 1924 and 1973 (see Katus et al., 1995), using the 1989 census as a
sample frame. The share of those surveyed was 81% (5,021 from 6,212), or
86%, leaving aside over­coverage (those who had died or had left Estonia in
1989-1994). A comparison between women who were surveyed with those
who were not, and the total female population by major socio­demographic
variables, showed that there were no significant differences (Katus et al.,
1995, pp. 18­21). As part of the Europe-wide FFS program, the survey was
based on the collection of event histories. All major demographic events that
had taken place in the life of the respondent were identified (to the accuracy
of a month), including births, co­residential unions and residential changes
since age 14. In the FFS program, the collection of migration histories was
optional, and only a few countries implemented this module. The high response
rate, the multiple retrospective histories, and the high quality of the
collected retrospective information in the Estonian FFS provide a good basis
for studying the effect of migration on fertility in more detail (see Katus
et al., 2000 on the quality of the Estonian FFS).
Our research population consists of the native female population born
from 1944 to 1973. We focus on only native people because we wish to have a
homogeneous population when testing competing hypotheses. In total, there
are 1,918 native women in the data set. However, we exclude 43 of them who
did not indicate the date of their union dissolution for some or another
reason. Therefore, our final research sample consists of 1,875 native females
born from 1944 to 1973. We study the impact of migration on their first,
second and third conceptions (leading to births). There are 1,556, 1,005 and
358 such events, respectively. We define migration as a residential change that
crosses the border of an urban settlement or rural municipality, and lasts for
more than three months. We go beyond the traditional urban­rural-dichotomy
and distinguish three types of settlements of origin and destination of
migration: (1) rural areas (less than 2,000 inhabitants); (2) small and mediumsize
towns (2,000­100,000); and (3) a large city or the capital (the capital
being Tallinn, with more than 400,000). The distribution of the sample
population across settlement-type is as follows: 46% of the women lived at
age 14 in rural areas, 36% in small towns, and 18% in the capital city. The
corresponding figures at the time of interview were 37%, 38% and 25%.
We split the data-set by conception episode, following the general logic of
event­history data set up. Individuals are at risk from age 14 onwards (for
the first conception) or previous birth (for the second and third conceptions).
The final censoring takes place at interview (actually, 9 months before) or at
age 45. Residential episodes outside Estonia are excluded from our analysis.
HILL KULU60
In addition, episodes outside union are excluded from the analysis of the
second and third births. If conceptions occur simultaneously (in the same
month) with migration and/or union formation, we use the sequence of
events as follows: migration, union formation and conception. Thus, we
assign simultaneous conceptions to the destination environment, which, as
we will see later, turns out to be a reasonable solution. We also build a multiepisode
data-set for migration, which we need for later simultaneous analyses.
The risk of migration starts at age 14, or at the previous migration. In
total, there are 3,063 migration events in our data-set: 1,183 of which were
moves to rural areas, 1,255 to small towns and 625 to the capital city.
5. Methods and analytic strategy
We use intensity regression or (multivariate) indirect standardisation (Hoem,
1993) as a research method. We estimate several models in order to further
examine various hypotheses proposed in the literature. We begin with a
simple model where we look at the effect of migration on conception, controlling
for only age and time since previous birth (for the second and third
conceptions). The results outline possible differences between migrants and
non-migrants at various places of origin and destination, and therefore give
us preliminary evidence about how and whether migration shapes fertility
behaviour. Thereafter, we also include in our analysis background variables
of individuals to control for demographic and socioeconomic selectivity of
migrants when assessing the impact of migration on fertility. Our basic model
is formulated as follows:
ln lit  yt 
X
k
zkuik  t 
X
j
ajxij 
X
l
blwilt; 1
where li(t) denotes the intensity of conception (first, second or third) for
individual i, y(t) denotes a piecewise linear spline that captures the impact of
baseline duration on the intensity. The zk(uik + t) denotes the spline representation
of the effect of a time­varying variable that is a continuous function
of t with origin uik. The xij represents the values of a time-constant variable
and wil(t) represents a time-varying variable whose values can change only at
discrete times.
After having outlined the basic differences between people who moved and
those who stayed at various origin and destination environments, we then
look at the possible role of unobserved selectivity accounting for differences
between migrants and non-migrants (which we expect to find). We have built
a simultaneous­equations model for that purpose, which jointly estimates
three equations for fertility, and three equations for migration (according to
destination of residential change). Both processes have their (person-specific)
heterogeneity terms, and allowing correlation between the residuals we
MIGRATION AND FERTILITY: COMPETING HYPOTHESES RE-EXAMINED 61
identify possible endogeneity of migration in the fertility process and control
for the unobserved selectivity when analysing the effect of migration on
fertility (Lillard, 1993). The model can be formalised as follows:
ln lC1
i t  yC1
t 
X
k
zC1
k uik  t 
X
j
aC1
j xij 
X
l
bC1
l wilt  eC
i ;
ln lC2
i t  yC2
t 
X
k
zC2
k uik  t 
X
j
aC2
j xij 
X
l
bC2
l wilt  eC
i ;
ln lC3
i t  yC3
t 
X
k
zC3
k uik  t 
X
j
aC3
j xij 
X
l
bC3
l wilt  eC
i ;
ln lR
imt  yR
t 
X
k
zR
k uimk  t 
X
j
aR
j ximj 
X
l
bR
l wimlt  eM
i ;
ln lS
imt  yS
t 
X
k
zS
kuimk  t 
X
j
aS
j ximj 
X
l
bS
l wimlt  eM
i ;
ln lL
imt  yL
t 
X
k
zL
k uimk  t 
X
j
aL
j ximj 
X
l
bL
l wimlt  eM
i ;
2
where li
C1
(t), li
C2
(t), li
C3
(t) denote the intensities of the first, second and
third conceptions, respectively, and lim
R
(t), lim
S
(t), lim
L
(t) represent the risks
of mth migration to rural, small urban and large urban destinations in the
competing risk framework.1
ei
C
and ei
M
are person­specific heterogeneity
terms for fertility and migration processes, respectively, and are assumed to
have a joint bivariate normal distribution. The identification of our model is
attained through within­person replication: many people have had several
births, and some people have also made several moves (see Lillard et al.,
1995, p. 446).
However, there is reason to believe that the nature of selectivity may
depend on the destination of migration. As the reviewed literature showed,
larger cities may attract people who prefer smaller families for some or another
reason, while migrants to rural settlements may desire many children.
Therefore, we have to extend our simultaneous­equations model allowing
separate (person-specific) heterogeneity terms for each migration equation:
ln lC1
i t  yC1
t 
X
k
zC1
k uik  t 
X
j
aC1
j xij 
X
l
bC1
l wilt  eC
i ;
ln lC2
i t  yC2
t 
X
k
zC2
k uik  t 
X
j
aC2
j xij 
X
l
bC2
l wilt  eC
i ;
ln lC3
i t  yC3
t 
X
k
zC3
k uik  t 
X
j
aC3
j xij 
X
l
bC3
l wilt  eC
i ;
ln lR
imt  yR
t 
X
k
zR
k uimk  t 
X
j
aR
j ximj 
X
l
bR
l wimlt  eR
i ;
ln lS
imt  yS
t 
X
k
zS
kuimk  t 
X
j
aS
j ximj 
X
l
bS
l wimlt  eS
i ;
ln lL
imt  yL
t 
X
k
zL
k uimk  t 
X
j
aL
j ximj 
X
l
bL
l wimlt  eL
i ;
3
HILL KULU62
where ei
R
, ei
S
and ei
L
are heterogeneity terms for migration to rural, small
urban and large urban areas, respectively. Again, allowing correlation between
(person-specific) residuals of fertility and migration equations, we test
endogeneity of migration in the fertility process, and thus eliminate a possible
bias while estimating the effect of residential change on fertility. To identify
the model, we have no need of instruments at this time, either, as some people
have made several moves towards the same destination.
So far our modelling strategy has focussed mostly on testing three basic
hypotheses in general, and the question of adaptation versus selection in
particular. However, the disruption hypothesis also needs examination. In
order to clarify the effect of migration on fertility, we have to make our static
models more dynamic. Therefore, in our final models, we allow the intensity
of conception at destination to vary over time since arrival in the settlement,
instead of assuming a constant risk. Technically, this is achieved by using the
linear spline representation to capture the effect of time at destination on
fertility.
6. Explanatory variables and hypotheses
We do have equations for both fertility and migration processes. In fertility
equations, variables reflecting an individuaĺs residential history hold a central
position. In the analysis we include a (time­varying) variable showing an
individuaĺs residence, and a variable indicating residence at age 14 (for migrants).
(We also include in the analysis a variable showing the number of
migrations.) Although the childhood settlement may have some role in later
fertility behaviour, we draw upon recent internal migration­fertility literature
and hypothesise that migrants display fertility levels similar to those of the
population at destination (Courgeau, 1989; Brockehoff, 1995; Lindstrom,
2003). If so, then finding an answer to the question of adaptation versus
(unobserved) selection becomes a major task of our further analysis. We
certainly expect to find some selection effects as residential relocation is often
related to union formation, and may thus be followed by elevated intensities
of (first) conception (Singley and Landale, 1998; Andersson, 2004). However,
this selectivity (by reason for migration) can be easily controlled for when
including the partnership status and union duration in the analysis. Finally,
we also expect to find evidence supporting the disruption hypothesis (White
et al., 1995).
We control for several demographic and socioeconomic variables when
testing different hypotheses concerning the impact of migration on fertility.
Our duration variables are age and time since previous birth (for the second
and third conceptions). We expect the probability of the first conception to
be the highest in the early twenties as previous studies have shown relatively
early childbearing of post-war Estonian women (Vikat, 1994; Katus et al.,
MIGRATION AND FERTILITY: COMPETING HYPOTHESES RE-EXAMINED 63
2002, pp. 154­155). The intensities of the second and third conceptions rise
rapidly in the first six months after a birth, and remain at high levels during
the subsequent year or two, before they begin to decrease (cf. Lillard, 1993,
p. 675; Katus et al., 2002, p. 158). Second, we include in our analysis, an
individuaĺs partnership status and union duration for those in union. We
expect union formation and marriage to significantly raise the probability of
conception, and the risk to gradually decrease with duration of union (Baizan
et al., 2003, p. 157). The next variable represents (calendar) time to capture
the impact of changing context. We hypothesise slightly increasing fertility
since the late 1960s, but a sharp decrease in fertility levels in the 1990s (Katus,
2000; Philipov, 2002, p. 7).
We include educational enrolment, employment status, and level of education
to control the effect of an individuaĺs socioeconomic characteristics.
We assume the probability of conception to be higher when an individual has
completed her/his studies (Singley and Landale, 1998, p. 1459; Baizan et al.,
2003, p. 157). We expect employed women to have a higher intensity of
conception (for the first conception, at least) compared to those who are
inactive in the context of full (and compulsory) employment. The possible
fertility patterns across educational groups are more difficult to predict.
Earlier studies showed an inverse relationship between education and fertility,
while more recent research has demonstrated a relatively high risk for
second and third births for educated women. However, that seems to disappear
when controlling for `relative' age of educated women, their partners'
characteristics and other (usually) unobserved factors (Hoem et al. 2001b;
Kravdal, 2001; Kreyenfeld, 2002). We hypothesise an inverse relationship
between fertility and education regarding first conception, and a relatively
high risk for second conception for educated women (Katus et al. 2002,
p. 177). However, that may vanish in the course of simultaneous analyses.
Finally, we assume the number of siblings to be positively related to an
individuaĺs fertility (Hoem et al., 2001a, p. 46; Baizan et al. 2002, p. 39), and
women belonging to the Russian (historical) minority in Estonia to exhibit
slightly lower levels of fertility (Sakkeus, 2000, p. 278).
While our paper focuses mostly on the impact of migration on fertility, the
analysis also allows us to examine the determinants of migration in post
World War II Estonia. All variables in the fertility equation may enter into
the migration equation (if needed), as our models are identified through
within­person replication. Our baseline variables are age and time since
previous residential change (for the second and higher migrations). In
keeping with the literature, we assume the intensity of migration to reach its
peak at late adolescence, when the majority of cohorts complete their (secondary)
education and continue their studies (often elsewhere), or enter into
the labour market (see also Ma and Liaw, 1997, p. 237; Katus et al., 1999,
p. 16). Concerning the effect of time since previous migration, we expect
HILL KULU64
increasing risks during the first few years, and decreasing intensities thereafter
(Gordon and Molho, 1995). Second, we hypothesise singles and divorced
people to be more mobile than those already in a union (cf. Mulder
and Wagner, 1993, p. 73; Newbold and Liaw, 1995, p. 125). Third, we assume
the presence of children in a family to significantly decrease the probability of
urban-ward migration, yet possibly to increase the propensity to move to
rural destinations (Courgeau, 1989, p. 140).
The next variable represents (calendar) time, and we hypothesise an
increasing migration risk towards rural areas in the 1980s, and decreasing
overall intensities in the early 1990s as a result of economic hardship arising
from post­socialist transition (Marksoo, 1992; Kulu and Billari, 2004).
Fifth, we assume increasing migration intensities after studies have been
completed, and also a higher risk when an individual is out of the labour
market (cf. Fischer and Malmberg, 2001, p. 265). Sixth, we hypothesise
increasing mobility as it correlates to an individuaĺs rising level of education.
This is a result of increasing options due to education and a larger dispersion
of jobs of more educated individuals (Courgeau, 1985, p. 159; Newbold, 1999,
p. 266). Next, there is also reason to believe that the presence of siblings raises
the probability of the first migration (home-leaving), at least (Courgeau, 1989,
p. 136). We also assume that ethnic Russians move less than Estonians do, to
rural areas in particular (Kulu and Billari, 2004). Finally, we hypothesise
decreasing migration intensity with increasing settlement size for non-migrants.
The pattern for migrants is likewise expected to depend on residence at age 14
and on the number of previous moves (Kulu, 2002).
7. Results
7.1. IMPACT OF MIGRATION ON FERTILITY
We began our modelling by running a set of models to examine the effect of
various destination environments on fertility of migrants with different origins
(e.g. a woman with a rural background in a small town, with a large city
origin in a rural settlement etc.). As we found no significant variation among
migrants with different origin (residence at age 14 or previous residence)
when living in the same (destination) environment, we decided to collapse the
categories of origin and leave only the destination of migration in our main
analysis, with one exception: the residence at age 14 is included in the models
for the third birth (as its impact turned out to be significant). Thus, there are
six residential categories in most cases: non-migrational and migrational
episodes in rural, small urban and large urban areas, respectively. The episode
is non­migrational if an individual has not moved since age 14.
Migrational episodes are defined according to destination of migration,
whatever the origin of migrants.
MIGRATION AND FERTILITY: COMPETING HYPOTHESES RE-EXAMINED 65
Let us now present the results of our main analysis. We begin with the
patterns of the first birth across two basic models, and then move to the
second and third births. In the first model, we look at the effect of migration
on conception, controlling only for baseline duration. We see that, first, the
intensity of the first conception of non-migrants in rural settlements and
small towns does not differ significantly (Tables 1 and 2, model 1). The major
dividing line runs between this pool of people and natives in a large city. The
latter have an about 1/5 lower risk of the first conception than the native
rural and small urban population. Second, migrants in rural areas exhibits a
58% higher risk of the first conception compared to non-migrants there,
while migrants in urban areas display intensities similar to those of native
residents. In the next model, we control for demographic and socioeconomic
selectivity of migrants. The fertility levels of migrants in rural areas decrease
significantly, and they now exhibit fertility levels similar to non-migrants at
destination (Tables 1 and 2, model 2). Our further analysis reveals that the
inclusion of partnership status and union duration mostly lead to the
decreasing levels. Clearly, the elevated fertility pattern for migrants in rural
areas we originally observed did result from a significant portion of residential
changes driven by family formation among rural­bound migrations.
The results of the analysis on the second and third conceptions are also
interesting, although the intensities vary much less across the models. As the
changes are small, we only report the results of the model where we already
control for the demographic and socioeconomic characteristics of population.
For the second conception, a major division exists between residents of
rural and urban areas. Non-migrants in small towns and large cities have a
41% lower risk of giving birth to a second child than the native rural population
(Tables 1 and 2, model 2). Migrants (whatever their origin) exhibit
the levels of risk similar to the non-migrants at destination. The impact of
settlement hierarchy is clearly present for the third conception ­ the larger the
settlement the lower the risk of having a third child. Natives in small towns
have a 26% and those in large cities have a 49% lower risk of the third
conception as compared to the non-migrant population in rural areas. At
first it seems that migrants have an even lower risk of the third conception
than non-migrants at destination. However, further analysis shows that the
differences are not significant. Residence at age 14 also shapes the patterns of
the third birth. Surprisingly, however, migrants who lived in small towns at
age 14 show the lowest intensity levels, whatever their later residence, while
the risk for those socialised in a large city seems to be the highest.
Thus, our analysis shows that the differences between native residents
of various settlements grow with increasing parity, and migrants (whatever
their origin) exhibit fertility levels similar to those of non­migrants at
destination. We also observe elevated first conception intensities for people
who move because of family formation (and there are many such people
HILL KULU66
Table 1. Intensity of conception leading to a birth (parameter estimates)
Model 1a
Model 2b
Model 3c
Model 4d
First conception
Constant (baseline) )7.82*** )7.40*** )7.58*** )7.58***
Age (baseline)
14­17 years (slope) 1.33*** 0.93*** 0.93*** 0.93***
18­21 years (slope) 0.24*** 0.04 0.08*** 0.08***
22­25 years (slope) )0.15 )0.03 )0.01 )0.01
26+ years (slope) )0.21*** )0.15*** )0.15*** )0.15***
Cohabitation (ref = single)
Enter cohabitation (constant) 2.90*** 2.92*** 2.92***
0­2 years (slope) )0.60*** )0.50*** )0.50***
2+ years (slope) )0.14*** )0.13*** )0.13***
Marriage (ref = cohabitant)
Enter marriage (constant) 0.30*** 0.30*** 0.30***
0+ years (slope) )0.03 )0.02 )0.02
Year
­1969 (slope) 0.04 0.05* 0.06*
1970­1979 (slope) 0.00 0.01 0.01
1980­1989 (slope) 0.02* 0.02 0.02
1990­1994 (slope) )0.12** )0.13** )0.13**
Enrolled in education )0.22*** )0.31*** )0.32***
Educational level (ref = secondary)
Basic 0.10 0.14* 0.14*
Higher 0.13 0.08 0.06
Employed 0.25*** 0.27*** 0.28***
Number of siblings (ref = 0­1)
2­3 0.08 0.11* 0.12*
4+ 0.25*** 0.32*** 0.32***
Non-Estonian ethnicity 0.13 0.13 0.12
Residential status
(ref = non-migrants in
rural areas)
Migrants in rural areas 0.46*** 0.01 )0.11 )0.16
Non-migrants in small towns 0.04 )0.03 )0.08 )0.08
Migrants in small towns 0.10 0.03 )0.09 )0.11
Non-migrants in large city )0.21** )0.36*** )0.49*** )0.49***
Migrants in large city )0.20* )0.22** )0.35*** )0.29**
2+ migrations (ref = 1 migration) 0.14** 0.04 0.02
MIGRATION AND FERTILITY: COMPETING HYPOTHESES RE-EXAMINED 67
Table 1. Continued
Model 1a
Model 2b
Model 3c
Model 4d
Second conception
Constant (baseline) )2.80*** )2.66*** )3.00*** )3.02***
Time since first birth (baseline)
0­0.5 years (slope) 4.28*** 4.85*** 4.81*** 4.81***
0.5­1 years (slope) )0.58** )0.54** )0.51* )0.52*
1­4 years (slope) )0.07* 0.03 0.02 0.02
4+ years (slope) )0.23*** )0.11*** )0.13*** )0.13***
Age
16­22 years (slope) )0.06 )0.02 )0.02
23­25 years (slope) 0.06 0.10** 0.10**
26+ years (slope) )0.11*** )0.09*** )0.09***
Cohabitation
0­0.5 years (slope) )2.54*** )2.57*** )2.56***
0.5+ years (slope) )0.08*** )0.05 )0.05
Marriage (ref = cohabitant)
Enter marriage (constant) 0.94*** 0.91*** 0.90***
0­1 years (slope) )0.56** )0.53** )0.53**
1+ years (slope) )0.05 )0.05 )0.05
Year
­1969 (slope) 0.05 0.05 0.05
1970­1979 (slope) 0.04*** 0.06*** 0.06***
1980­1989 (slope) 0.02* 0.02* 0.02*
1990­1994 (slope) )0.19*** )0.20*** )0.20***
Enrolled in education )0.33*** )0.38*** )0.39***
Educational level (ref = secondary)
Basic 0.19** 0.22** 0.21**
Higher 0.27*** 0.21* 0.20*
Employed 0.13 0.13 0.13
Number of siblings (ref = 0­1)
2­3 0.12 0.16* 0.17**
4+ 0.25*** 0.32*** 0.33***
Non-Estonian ethnicity )0.42*** )0.46*** )0.47***
Residential status
(ref = non­migrants in rural
areas)
Migrants in rural areas 0.01 )0.16 )0.24 )0.28*
Non­migrants in small towns )0.60*** )0.52*** )0.58*** )0.59***
Migrants in small towns )0.32** )0.40*** )0.50*** )0.52***
Non-migrants in large city )0.68*** )0.52*** )0.68*** )0.68***
Migrants in large city )0.45*** )0.51*** )0.61*** )0.54***
2+ migrations (ref = 1 migration) 0.18** 0.06 0.04
HILL KULU68
Table 1. Continued
Model 1 a
Model 2 b
Model 3 c
Model 4 d
Third conception
Constant (baseline) )3.12*** )0.50 )1.13 )1.13
Time since second birth (baseline)
0­0.5 years (slope) 2.92*** 3.41*** 3.48*** 3.48***
0.5­1 years (slope) )0.80* )0.36 )0.41 )0.40
1+ years (slope) )0.13*** )0.01 )0.03 )0.03
Age
18­21 years (slope) )0.26 )0.21 )0.22
22­33 years (slope) 0.02 0.05** 0.05**
34+ years (slope) )0.26*** )0.25*** )0.25***
Cohabitation
0­2 years (slope) )0.69*** )0.72*** )0.72***
2+ years (slope) )0.10** )0.09** )0.08**
Marriage (ref = cohabitant)
Enter marriage (constant) 0.59** 0.56* 0.55*
0­3 years (slope) )0.23** )0.22** )0.22**
3+ years (slope) )0.02 )0.03 )0.03
Year
­1969 (slope) )0.11 )0.10 )0.09
1970­1979 (slope) 0.00 0.02 0.02
1980­1989 (slope) 0.03 0.04* 0.04*
1990­1994 (slope) )0.06 )0.07 )0.07
Enrolled in education )0.07 )0.05 )0.06
Educational level (ref = secondary)
Basic 0.25* 0.32** 0.32**
Higher 0.11 0.09 0.07
Employed )0.04 )0.03 )0.03
Number of siblings (ref = 0­1)
2­3 0.07 0.14 0.15
4+ 0.13 0.22 0.23
Non-Estonian ethnicity )0.14 )0.12 )0.13
Residential status
(ref = non-migrants in
rural areas)
Migrants in rural areas )0.09 )0.16 )0.30 )0.33
Non­migrants in small towns )0.40 )0.30 )0.45 )0.46*
Migrants in small towns )0.51** )0.41 )0.57** )0.58**
Non­migrants in large city )0.97*** )0.68** )0.95*** )0.96***
Migrants in large city )1.19*** )1.01*** )1.18*** )1.11***
Residence at age 14 for migrants
(ref = rural area)
Small town )0.19 )0.27* )0.29* )0.29**
Large city 0.34 0.24 0.17 0.12
2+ migrations (ref = 1
migration)
0.15 0.03 0.00
MIGRATION AND FERTILITY: COMPETING HYPOTHESES RE-EXAMINED 69
among migrants to rural settlements). While our results give limited support
for the socialisation hypothesis (regarding the third birth), it is now
clear that our major task is to clarify why the fertility of migrants is
similar to that of natives at destination. Does this result from migrants'
adaptation, or rather (further) selectivity of migrants that is unobserved in
this case?
In the next stage, we include therefore person-specific residuals into fertility
and migration equations, allowing for correlation between heterogeneity
terms. The model fit improves significantly (the value of likelihood ratio
test statistic (LR) is 191.4 with 3 degrees of freedom, p­value is <0.01). The
standard deviation of residuals is significantly different from zero in both
cases (Table 1, model 3). Moreover, the correlation coefficient is positive
(0.39) and significant. Thus, migrants (some of them, at least) have unobserved
characteristics that increase their probability of childbearing. Controlling
for this unobserved selectivity, however, does not change our
previous results substantially. Only the difference between rural natives and
Table 1. Continued
Model 1 a
Model 2 b
Model 3 c
Model 4 d
reC 0.47*** 0.47***
reM 0.70***
qeC eM 0.39***
reR 1.12***
reS 0.77***
reL 0.82***
qeC eR 0.42***
qeC eS 0.48***
qeC eL 0.09
qeR eS 0.72***
qeR eL )0.04
qeS eL )0.07
Log-likelihood )15739.5 )35323.4 )35227.7 )35173.9
a
We only control for baseline duration.
b
We also control for (other) demographic and socioeconomic variables (see Equation 1). The
log­likelihood of the model is the sum of log-likelihoods of (three) fertility and migration
equations (Table 3).
c
We estimate jointly fertility and migration equations with two person-specific residuals and
correlation between them (Equation 2).
d
We estimate jointly fertility and migration equations with four person-specific residuals and
correlations between them (Equation 3).
r denotes a standard deviation of the person­specific error term, q denotes a correlation
between the person­specific error terms (the definition of subscripts is given in the Equations 2
and 3).
* p < 0.10; ** p < 0.05; *** p < 0.01.
HILL KULU70
other groups (including rural migrants) increases slightly. Thus, what we
have established is the fact that some (more) migrations are directly related to
the childbearing process, and/or that strong positive selection of migrants by
fertility preferences operates towards some destination. As a result, the
overall figures also follow this pattern. We should continue allowing possible
selection to vary across destination of migration.
To further examine the issue of unobserved selectivity, we include in our
analysis a person-specific residual for fertility equations and separate heterogeneity
terms for each migration equation, allowing for correlation between
Table 2. Effect of residential status on conception (relative risks)a
Model 1 Model 2 Model 3 Model 4
First conception
Non-migrants in rural areas 1 1 1 1
Migrants in rural areas 1.58*** 1.01 0.89 0.85
Non-migrants in small towns 1.04 0.97 0.93 0.92
Migrants in small towns 1.11 1.03 0.91 0.90
Non-migrants in large city 0.81** 0.70*** 0.61*** 0.61***
Migrants in large city 0.82* 0.81** 0.70*** 0.75**
Second conception
Non-migrants in rural areas 1 1 1 1
Migrants in rural areas 1.01 0.85 0.79 0.76*
Non-migrants in small towns 0.55*** 0.59*** 0.56*** 0.56***
Migrants in small towns 0.73** 0.67*** 0.61*** 0.60***
Non-migrants in large city 0.51*** 0.59*** 0.51*** 0.51***
Migrants in large city 0.64*** 0.60*** 0.54*** 0.58***
Third conception
Non-migrants in rural areas 1 1 1 1
Migrants in rural areas 0.92 0.85 0.74 0.72
Non-migrants in small towns 0.67 0.74 0.64 0.63*
Migrants in small towns 0.60** 0.66 0.57** 0.56**
Non-migrants in large city 0.38*** 0.51** 0.39*** 0.38***
Migrants in large city 0.30*** 0.36*** 0.31*** 0.33***
Residence at age 14 for migrants
Rural area 1 1 1 1
Small town 0.83 0.77* 0.75* 0.75**
Large city 1.41 1.28 1.19 1.13
a
The relative risks are obtained by performing exponentiation operations of the parameter
estimates presented in Table 1.
* p < 0.10; ** p < 0.05; *** p < 0.01.
MIGRATION AND FERTILITY: COMPETING HYPOTHESES RE-EXAMINED 71
the residuals. Again, the model fit improves significantly, as compared to the
previous one (LR = 107.6 with 7 degrees of freedom, p­value is <0.01). The
standard deviations of all four heterogeneity terms are significantly different
from zero (Table 1, model 4). More interestingly, while correlations between
the (person­specific) residual of fertility equations and that of migration to
rural and small towns are positive, the correlation with residual of equation for
the large urban destination is not different from zero. Therefore, the unobserved
selectivity of migrants, whatever its meaning, operates towards rural
and small urban destinations. However, the coefficients of our main interest do
not change as significantly­migrants exhibit fertility levels rather similar to the
native population at destination. Still, comparing the current results with that
of separate modelling (model 2), we notice that the coefficients for migrants to
rural and small urban areas are upward biased, although slightly, in the singleequation
model. Interestingly, the same holds regarding first, second and third
conceptions, and applies likewise to non-migrant groups. To sum up, the
simultaneous analysis largely supports our previous findings. Furthermore, it
shows that unobserved selection of migrants is a reality (in some cases), but its
impact to refute the adaptation hypothesis is not very strong.2
So far, we have assumed a constant fertility risk for migrants at destination.
Next, we extend our two basic models, allowing the intensity of conception
to vary over time since arrival in the settlement. This strategy enables
us not only to examine the disruption hypothesis in more detail, but also to
gain further insight into the selectivity issue. We focus only on the risk of the
first conception, as possible changes in time are expected to be most `dramatic'
here (and the number of events sufficient for more detailed analysis).
We present our results in the graph in order to assist in interpretation of the
results. Allowing the intensity of conception to vary over time improves the
model (1) fit significantly (LR  58.1 with 9 degrees of freedom, p-value is
<0.01). We notice different time patterns for migrants to rural and small
urban destinations, on the one hand, and for migrants to large urban destinations,
on the other hand (Figure 2). In the former, the risk after migration
is very high (migrants to rural areas have an about 3.7 times higher risk
than rural non­migrants, and those to small towns have a 2.9 times higher
risk than natives in small towns), which then quickly decreases. In the latter,
the intensity is low immediately after migration (38% lower than that of
urban natives) and then increases.
The second graph shows the fertility patterns after controlling for
demographic and socioeconomic selectivity of migrants. Again, the model
with time­varying intensity is better than where the risk is constant
(LR  14.6 with 9 degrees of freedom, p-value is  0.10). As expected, the
elevated risk of the conception for migrants in rural and small urban areas
observed previously decreases significantly (Figure 2). Once more, we receive
support to the fact that there are many people who move to rural settlements,
HILL KULU72
but also to small towns because of family formation. Interestingly, however,
a somewhat higher risk of conception remains right after migration for
people who move to rural and small areas even after controlling for their
demographic and socioeconomic selectivity, especially the over-representation
of union-formers among them. How to interpret this pattern? Further
analysis reveals that elevated fertility (that largely accounts for correlations
observed in the simultaneous analysis) is mostly related to people of two
groups: first, to those who first move and then become pregnant (and only
thereafter form a union); second, to those who first form a union and only
then move to their partner and conceive a child. Obviously, they can also be
considered as people moving because of family formation: the former become
pregnant right before union formation for one reason or another (of course,
this group may also include people for whom the pregnancy was actually not
(yet) planned), the latter have reported the start of the union to take place
before the actual co-residence.3
Let us now take a look at two other interesting patterns. First, a relatively
low risk right after migration for people who move to a large city seems to
give support to the disruption effect: getting settled in a large city takes some
time, and in most cases childbearing is postponed, although only for a period
of a few months (and may be also elsewhere if we exclude migrations related
to union formation). Second, and most importantly, the longer­term fertility
patterns of movers do not change very much ­ migrants in different
destinations still exhibit rather similar fertility levels to the non-migrants
there. (These patterns remain also in the course of simultaneous analysis
(results not shown).) Thus, while some migrations are directly related to
childbearing (and family formation), we find no evidence to conclude that
certain areas attract people with the fertility behaviour dominant there.
Rather, migrants tend to adapt to fertility levels prevalent at destination, and
sometimes postpone childbearing for a period of time in order to overcome
economic and psychological costs arising from a residential relocation.
7.2. IMPACT OF OTHER VARIABLES
Let us now discuss the effect of other variables on fertility. The results are
mainly as expected. The risk of the first conception is the highest in the early
twenties, confirming the relatively early start of childbearing of post­war
Estonian women (Vikat, 1994; Katus et al., 2002, pp. 154­155) (Table 1).
The baseline intensities for the second and third conceptions largely follow
patterns shown in other studies: they rise rapidly during the first six months
after the previous birth and then decrease (cf. Lillard and Waite, 1993, p. 666;
Hoem et al., 2001a, p. 46). As expected, both union formation (in case of the
first birth) and marriage significantly contribute to the increase of the
probability of conception. However, the rise is an extreme upward surge and
MIGRATION AND FERTILITY: COMPETING HYPOTHESES RE-EXAMINED 73
the subsequent decrease rather steep, which points to a concentration of
many conceptions (first two, at least) in the beginning of the union, contrary
to patterns found in other countries (Baizan et al., 2002, p. 39). Regarding
changes over time, the risk of the second birth (at least) rose in the 1970s and
the 1980s, and the intensities of all parities decreased in the 1990s as expected
(Katus, 2000; Philipov, 2002, p. 7). The lower risk of conception during
educational enrolment also corresponds to what is expected. Likewise, it is
not a surprise that there is a higher risk of the first conception during
employment, in the context of a planned economy where inactivity before
childbearing might indicate possible health problems.
Model 1 (extended)
0
0.5
1
1.5
2
2.5
3
3.5
4
0.0 0.5 1.0 1.5 2.0
Years since migration
Relativerisk
Model 2 (extended)
0
0.5
1
1.5
2
2.5
3
3.5
4
0.0 0.5 1.0 1.5 2.0
Years since migration
Relativerisk
Non-migrants in rural areas Migrants in rural areas Non-migrants in small towns
Migrants in small towns Non-migrants in large city Migrants in large city
Figure 2. Effect of migration on first conception.
Source: Own calculations based on the Estonian FFS.
HILL KULU74
Our analysis supports previous findings on the relatively high risk of a
second birth to highly educated women, showing them as `carriers' of the
`two­child norm' in post­war Estonia (Katus et al., 2002, p. 177). Also, a
higher intensity of the third conception among less educated women corresponds
to previous findings. The analysis, however, does not confirm the
inverse relationship between education and the first birth found in most
studies.4
The analysis also supports the role of siblings and ethnic origin in
shaping fertility patterns. As expected, the larger the number of siblings, the
higher the fertility. Belonging to the Russian minority, in turn, decreases the
probability of second and (obviously also) third conceptions. Finally, we also
tested the effect of variables showing parental divorce, whether a previous
child was conceived with the same partner or not, and whether a previous
child was born in the current residence. None of these variables had a significant
impact on fertility, and we therefore excluded them from our main
analysis. However, very religious women exhibited high levels of the third
birth intensities, as was expected (Hoem et al., 2001a, p. 46), but we also
excluded this variable as there were too few cases and events in the most
interesting group.
7.3. DETERMINANTS OF MIGRATION
Next, we briefly discuss determinants of internal migration of post-war
Estonian female cohorts. The analysis largely supports findings of previous
studies, although some differences can also be outlined. As expected, the
intensities of migration to all destinations are the highest at late adolescence,
and thereafter gradually decrease (Table 3) (Katus et al., 1999, p. 16). Our
analysis also shows increasing mobility of migrants during the first three to
four years after residential relocation, and subsequently a decrease. The fact
that people in a union have a lower risk of migration to cities than singles, in
addition to the higher mobility of separated people, corresponds to expectations
(cf. Mulder and Wagner, 1993, p. 73). We also find the presence of
children to significantly decrease the probability of moving to urban areas in
general, and to the capital city in particular, but we have not found this to
affect the mobility towards rural settlements, as shown in some other studies
(Courgeau, 1989, p. 140). Still, further analysis indicates that children
(especially the second or subsequent child) increase the probability of moving
to rural areas from smaller towns (but not from a large city). Decreasing
migration intensities towards urban areas in the 1980s, and also to rural areas
in the 1990s, are consistent with the socioeconomic changes in Estonia during
the late socialist and early transitional periods (Marksoo, 1992; Kulu and
Billari, 2004).
MIGRATION AND FERTILITY: COMPETING HYPOTHESES RE-EXAMINED 75
Lower intensity of migration during education and employment likewise
corresponds to expectations (Fischer and Malmberg, 2001, p. 265).
Increasing mobility as it correlates with an individuaĺs rising education is not
surprising either, as well as a significantly higher risk of educated people
moving to the capital city. The latter points to both better opportunities for
the highly educated to achieve their ends, and the location of their (major)
job-market (Newbold, 1999, p. 266), but obviously also to a specific character
of planned economies. Namely, some university graduates had been directed
to work in smaller towns and rural settlements which they left (alone or with
families) when their first (semi-compulsory) job-contract ended (cf. Rybakovskiy
and Tarasova, 1991). Concerning the effect of siblings on migration
intensities we see, as expected, a rising mobility (to rural and small urban
areas) of people having two or more siblings (Courgeau, 1989, p. 136).
However, further analysis shows that the impact is significant only on the risk
of the first migration, which actually is not surprising.
Finally, current residence, residence at the age of 14 years, and the number
of previous migrations are also important determinants of mobility. Moreover,
the effect of these characteristics appears to be sensitive to different
specifications. The results of a separate analysis show that native urban
residents have a significantly lower risk of moving to rural settlements than
migrants in cities or native rural populations (Table 3, model 2). This is not
surprising, and neither is the fact that migrants with an urban background
have significantly lower migration intensities as compared to those with a
rural origin (whatever their current residence). Most residential groups exhibit
a significantly lower risk of moving to small towns than native rural
residents. Again, the probability is lowest for natives in the capital city, which
confirms their modest desire to leave the large city (Marksoo, 1990). Residence
at age 14 matters: migrants with a small town background have the
highest and those with a large city origin exhibit the lowest intensities of
moving towards small towns. Again, migrants with two and more residential
changes have a lower risk of migration than those with only one move, which
may result from the fact that two migrations are often unavoidable for an
individual who has no intention to leave her/his childhood settlement permanently,
but decides to continue her/his studies.
Our analysis of migrations towards large cities shows that native rural and
small urban residents have a higher risk of migration than migrants in rural
environments. Women with a large city background have the highest probability
among migrants, pointing to a returning to their childhood environment.
Interestingly, however, some of the results described above, change in the
course of simultaneous analysis. More specifically, the differences in risk of
migrating to rural and small urban destinations increase between native residents
and migrants in rural and small urban areas (Table 3, model 4). Further
analysis shows that an upward­biased risk of moving to rural and small urban
HILL KULU76
Table 3. Intensity of migration by destination (parameter estimates)
Model 2a
Model 3b
Model 4c
Rural destination
Constant (baseline) )2.16*** )2.32*** )2.85***
Age (baseline for first migration)d
14­17 years (slope) )0.09** )0.05 )0.04
18­19 years (slope) )0.36*** )0.23* )0.20
20­22 years (slope) )0.19*** )0.14*** )0.13***
23­33 years (slope) )0.13*** )0.13*** )0.14***
34+ years (slope) )0.02 )0.03 )0.05
Time since previous migration
(baseline for second migration)
0.25­0.75 years (slope) 2.52*** 2.41*** 2.46***
0.75­3.5 years (slope) 0.14*** 0.18*** 0.22***
3.5­9 years (slope) )0.20*** )0.18*** )0.17***
9+ years (slope) )0.01 )0.01 0.01
Partnership status (ref = single)
Cohabiting )0.20* )0.17 )0.13
Separated 0.63*** 0.68*** 0.78***
Marriage (ref = cohabitant) )0.26** )0.30*** )0.31***
Parity 1 (ref = 0)e
)0.04 )0.11 )0.10
Parity 2 + (ref = 1) 0.03 )0.10 )0.15
Year
­1969 (slope) 0.04 0.04 0.05*
1970­79 (slope) 0.02* 0.03** 0.04***
1980­89 (slope) 0.01 0.01 0.02
1990­94 (slope) )0.16*** )0.16*** )0.16***
Enrolled in education )2.55*** )2.74*** )2.80***
Educational level (ref = secondary)
Basic )0.05 )0.02 )0.03
Higher 0.38*** 0.43*** 0.39***
Employed )0.56*** )0.52*** )0.47***
2+ siblings (ref = 0­1) 0.21*** 0.30*** 0.40***
Non-Estonian ethnicity )0.77*** )0.95*** )1.10***
Residential status
(ref = non-migrants in rural areas)
Migrants in rural areas 0.00 )0.27* )0.70***
Non-migrants in small towns )0.19** )0.25*** )0.32***
Migrants in small towns )0.04 )0.29* )0.26
Non-migrants in large city )0.86*** )1.18*** )1.23***
Migrants in large city )0.10 )0.34** 0.10
Residence at age 14 for migrants
(ref = rural area)
Small town )0.57*** )0.64*** )0.81***
Large city )0.86*** )1.16*** )1.66***
2+ migrations (ref = 1 migration) )0.40*** )0.94*** )1.05***
MIGRATION AND FERTILITY: COMPETING HYPOTHESES RE-EXAMINED 77
Table 3. Continued
Model 2a
Model 3b
Model 4c
Small urban destination
Constant (baseline) )1.12*** )1.26*** )1.37***
Age (baseline for first migration)
14­17 years (slope) )0.07* )0.04 )0.06
18­19 years (slope) )0.60*** )0.49*** )0.51***
20­22 years (slope) )0.11** )0.07 )0.07
23­33 years (slope) )0.11*** )0.10*** )0.11***
34+ years (slope) )0.05 )0.06 )0.07
Time since previous migration
(baseline for second migration)
0.25­0.75 years (slope) 1.78*** 1.67*** 1.60***
0.75­3.5 years (slope) 0.05 0.09* 0.10**
3.5­9 years (slope) )0.25*** )0.23*** )0.23***
9+ years (slope) 0.00 0.00 0.01
Partnership status (ref = single)
Cohabiting )0.70*** )0.66*** )0.63***
Separated 0.05 0.11 0.15
Marriage (ref = cohabitant) )0.31** )0.34** )0.33**
Parity 1 (ref = 0) )0.17 )0.21* )0.23*
Parity 2+ (ref = 1) )0.38*** )0.50*** )0.55***
Year
­1969 (slope) 0.06*** 0.06*** 0.07***
1970­79 (slope) 0.00 0.01 0.01
1980­89 (slope) )0.04*** )0.04*** )0.04***
1990­94 (slope) )0.05 )0.05 )0.04
Enrolled in education )2.78*** )2.93*** )2.95***
Educational level (ref = secondary)
Basic )0.55*** )0.54*** )0.56***
Higher 0.44*** 0.47*** 0.37***
Employed )1.05*** )0.96*** )0.94***
2+ siblings (ref = 0­1) 0.13** 0.20*** 0.21***
Non-Estonian ethnicity )0.55*** )0.70*** )0.68***
Residential status
(ref = non-migrants in rural areas)
Migrants in rural areas )0.29** )0.58*** )0.57***
Non-migrants in small towns )0.13* )0.18** )0.18**
Migrants in small towns )0.27** )0.52*** )0.56***
Non-migrants in large city )0.98*** )1.27*** )1.17***
Migrants in large city 0.09 )0.14 0.27
Residence at age 14 for migrants
(ref = rural area)
Small town 0.53*** 0.48*** 0.50***
Large city )0.86*** )1.12*** )1.31***
2+ migrations (ref = 1 migration) )0.58*** )1.06*** )0.99***
HILL KULU78
Table 3. Continued
Model 2a
Model 3b
Model 4c
Large urban destination
Constant (baseline) 0.02 )0.14 )0.17
Age (baseline for first migration)
14­17 years (slope) )0.06 )0.03 )0.02
18­19 years (slope) )0.74*** )0.58*** )0.59***
20­22 years (slope) )0.01 0.03 )0.01
23­33 years (slope) )0.12*** )0.10*** )0.10***
34+ years (slope) )0.26 )0.29* )0.27*
Time since previous migration
(baseline for second migration)
0.25­0.75 years (slope) 0.44 0.34 0.55
0.75­3.5 years (slope) 0.07 0.09 0.07
3.5­9 years (slope) )0.05 )0.03 )0.03
9+ years (slope) )0.05 )0.03 )0.05
Partnership status (ref = single)
Cohabiting )0.51** )0.47** )0.53**
Separated 0.81*** 1.02*** 0.95***
Marriage (ref = cohabitant) )0.07 )0.07 )0.05
Parity 1 (ref = 0) )1.66*** )1.75*** )1.80***
Parity 2+ (ref = 1) )0.88*** )1.05*** )0.99***
Year
­1969 (slope) )0.01 )0.01 )0.02
1970­79 (slope) 0.02 0.03 0.02
1980­89 (slope) )0.05*** )0.06*** )0.06***
1990­94 (slope) )0.15 )0.14 )0.16
Enrolled in education )3.43*** )3.59*** )3.61***
Educational level (ref = secondary)
Basic )1.05*** )0.98*** )0.92***
Higher 1.33*** 1.32*** 1.50***
Employed )2.17*** )2.09*** )2.21***
2+ siblings (ref = 0­1) )0.08 )0.04 )0.08
Non-Estonian ethnicity )0.43* )0.50* )0.47*
Residential status
(ref = non-migrants in rural areas)
Migrants in rural areas )0.61*** )0.91*** )0.72***
Non-migrants in small towns 0.08 0.06 0.07
Migrants in small towns )0.11 )0.37*** )0.22
Residence at age 14 for migrants
(ref = rural area)
Small town )0.06 )0.12 )0.06
Large city 1.74*** 1.57*** 2.03***
MIGRATION AND FERTILITY: COMPETING HYPOTHESES RE-EXAMINED 79
settlements, as we (originally) observed, results largely from strong interrelations
of migrations between these areas. This is not surprising, and could also
be concluded from the results presented earlier: a significant positive correlation
of person­specific residuals of two migration processes (Table 1).
8. Summary and discussion
Let us now summarise the major results of this study and discuss the observed
fertility patterns and their significance. We began our analysis
examining differences in fertility behaviour of residents of various settlements,
both migrants and non­migrants. Our analysis showed that first, the
risk of conception for native residents decreased significantly with increasing
settlement size, and the decrease was larger for higher­order birth parities.
Natives of the capital city had lower intensities of the first conception, as
compared to the rural non-migrants, a much lower risk of the second conception,
and even lower intensities of the third. Second, it appeared that
migrants, whatever their origin, exhibited fertility levels rather similar to the
non-migrants at destination. We also observed elevated fertility for people
who moved because of union formation.
Next, we also identified and controlled for possible unobserved selectivity
of migrants. Our simultaneous analysis showed the presence of unobserved
selectivity for migrants to rural and small urban destinations. However, previous
results did not alter much ­ migrants still exhibited fertility levels rather
similar to the non-migrants at destination (although in previous models we
had slightly overestimated the risk of conception for migrants to rural and
small urban settlements). Finally, to better understand the nature of unobserved
selectivity, we allowed the risk of first conception at destination to
Table 3. Continued
Model 2a
Model 3b
Model 4c
2+ migrations (ref = 1 migration) )0.22 )0.63*** )0.45***
Log-likelihood )35323.4 )35227.7 )35173.9
a
We estimate single-equation model of migration with competing destinations. The loglikelihood
of the model is the sum of log-likelihoods of fertility (Table 1) and migration
equations.
b
We estimate jointly fertility and migration equations with two person-specific residuals and
correlation between them (equation 2).
c
We estimate jointly fertility and migration equations with four person-specific residuals and
correlations between them (equation 3).
d
The variable also captures the effect of age for second and higher migrations.
e
The values of parity change at the moment of conception.
* p < 0.10; ** p < 0.05; *** p < 0.01.
HILL KULU80
change over time. We found elevated fertility for migrants to rural, and also
small urban areas immediately after moving (even after controlling for the
demographic and socioeconomic selectivity of migrants), and a relatively low
fertility for movers to a large city, while long-term fertility patterns for
migrants remained similar to those of natives at destination. We concluded
that migrations related to family formation and childbearing, were mostly
responsible for the unobserved heterogeneity we observed, and we found no
evidence on (strong) selectivity of migrants by fertility preferences.
Thus, while some evidence can be brought to support each of the four
hypotheses proposed in the literature, the results of our analysis place the
adaptation theory in a central position. Briefly, migrants, whatever their origin,
adapt to fertility levels prevalent at the destination environment. But why do
they adapt? What are the factors pushing migrants to change their behaviour
that may have been originally different? At least two explanations can be offered.
The first emphasises the critical role of resources in general, and the
housing conditions in particular. In Estonia (as in many other European
countries), most people in rural settlements live in single­family houses, while
in urban areas, especially in the larger cities, flats in multi­storey dwellings
dominate (Estonian Statistical Office, 2000). More importantly, living space is
significantly larger for people living in family houses (Kulu, 2003b, p. 904).
Therefore, it is likely that migrants moving from rural settlements (single family
houses) to urban areas (flats) have less living space after migration, while migrants
from urban to rural areas usually move to more spacious housing. Both
adopt their family plans (the former more, the latter less) according to the new
conditions. Besides less favourable housing conditions, higher overall living
and opportunity costs can also be seen as responsible factors for lower fertility
of urban residents, both natives and migrants (Michielin, 2002).
The second explanation draws upon the cultural approach in fertility
studies, emphasizing the critical role of norms and values. It is well documented
in the literature that `modern fertility behaviour' spread first among
urban elites in Europe, and only later reached lower social classes and the
rural areas (Pollak and Watkins, 1993, p. 469). While values and norms
associated with the `modern fertility regime' are equally spread among rural
and urban populations in Estonia today, there is still evidence that life in
rural settlements has remained more `traditionaĺ and the notion of family is
stronger there (cf. Katus et al., 2002, p. 329). Furthermore, a rural population
can be considered one of the major `regional sub­cultures' in the
country, distinct from other(s), such as urban one(s) (cf. Lesthaeghe and
Neels, 2002). Moving from one sub­culture to another has an effect.
Everyday interaction with new friends and peers in a new environment
moulds an individuaĺs beliefs and desires, as well as their behaviour (Kohler,
2000). Migrants assume the fertility behaviour dominant at destination.
MIGRATION AND FERTILITY: COMPETING HYPOTHESES RE-EXAMINED 81
While we believe that most of the results of this study are also valid in
other contexts, some particularities arising from the post­war Estonian
context should be mentioned. First, elevated fertility observed for migrants to
rural and also small urban areas did result (partly, at least) from the fact that
the timing of union formation and first conception largely coincided in postwar
Estonian female cohorts (Vikat, 1994; Katus et al., 2002, p. 323).
Therefore, conceptions accompanied migration related to union formation,
which we also observed for migrants who moved to rural settlements and
small towns. Conceiving a child right after union formation has been seen as
a strategy of young couples to accelerate receiving the state housing under
central planning (Katus et al., 2002, p. 156). In countries where union formation
did not automatically lead to conception for one reason or another,
fertility patterns immediately after migration might thus be different, or if
similar, they might become evident in migrating couples who are intending to
have a child immediately thereafter.
Second, sub­urbanisation and counter­urbanisation, characteristic to
many Western countries during the post-war period, did not spread extensively
in Estonia. Many people left cities to sub-urban and rural areas in the
1980s, but most of them became employees of agricultural farms that offered
relatively good salaries and often provided labour with housing (Tammaru
et al., 2004, p. 215). Thus, our study does not deal with fertility patterns of
migrants to sub­urban areas or rural destinations when urban-type
employment continues and an `urban lifestyle' is maintained. The effect of
sub­ and counter­urbanisation on childbearing, however, needs detailed
research as `urban life combined with rural environmenť is becoming more
extensively spread among the populations in many developed countries,
including post­Soviet Estonia (Tammaru et al., 2004).
The study inspires research into two interrelated directions. First, we
should continue research based on the FFS data, including the analysis of
variables reflecting housing conditions of the population and extending
our research beyond one single country. The inclusion of data on housing
enables us to examine the validity of one explanation proposed above, at
least. Comparing the effect of migration on fertility in two or three
European countries with different institutional and socioeconomic development,
in turn, allows us to gain further insight into patterns and causes
of migrant fertility. Second, the register data from the Nordic countries
looks very attractive, no doubt. Rich longitudinal data on large samples
would enable us to examine the effects of various migrations and time at
destination on fertility of migrants more closely. Perhaps then the migrants
with specific (long-term) fertility preferences (if they do exist) will
also reveal themselves. `Migration makes a difference' is no doubt the
main message of this paper.
HILL KULU82
Acknowledgements
I express my sincerest gratitude to Andres Vikat for his detailed comments
and suggestions when preparing this study. I also thank Gunnar Andersson,
Jan M. Hoem, Francesca Michielin and the referees for their valuable
comments, and Yvonne Sandor for the English editing. I am grateful to the
Advisory Group of the FFS program for comparative research and the
Estonian Interuniversity Population Research Centre for their permission
(granted under identification number 75) to use the Estonian FFS data.
Notes
1
The reason behind considering only destination and not the origin of migration will become
clear in the course of the data analysis.
2
We decided to estimate one more model (5), where we included separate (person-specific)
heterogeneity terms for equation of the first birth, on the one hand, and for equations of the
second and third births, on the other hand. Our different specifications (with and without
further instruments) showed that the model was identified in its current form (without
instruments), despite the fact that there was only one birth episode per individual in the former
case. The results of analysis (not shown) were largely similar to that obtained in a previous
step (model 4).
3
In the Estonian FFS, the respondent was asked to report when she actually started to think
of her partner as being her partner. Thus, the beginning of the union did not necessarily
coincide with the start of the co-residence (Katus et al., 1995).
4
Our further analysis reveals that this pattern holds regarding younger birth cohorts, and
also when separate (person-specific) residuals are allowed for equations of the first and secondthird
births. The issue, however, requires more detailed treatment, which goes beyond the
scope of this study.
References
Andersson, G., 2004. `Childbearing after migration: fertility patterns of foreign-born women
in Sweden', International Migration Review 38(1): 364­392.
Baizan, P., Aassve, A. and Billari, F. C., 2002. Institutional Arrangements and Life Course
Outcomes: The Interrelations Between Cohabitation, Marriage and First Birth in Germany
and Sweden. Max Planck Institute for Demographic Research. Rostock MPIDR Working
Paper 2002-026.
Baizan, P., Aassve, A. and Billari, F. C., 2003. `Cohabitation, marriage, and first birth: the
interrelationship of family formation events in Spain'. European Journal of Population
19(2): 147­169.
Brockerhoff, M., 1995. `Fertility and family-planning in African cities: the impact of female
migration'. Journal of Biosocial Science 27(3): 347­358.
Brockerhoff, M. and Yang, X. S., 1994. `Impact of migration on fertility in Sub-Saharan
Africa'. Social Biology 41(1­2): 19­43.
Cadwallader, M., 1993. `Classics human geography revisited: commentary 2'. Progress in
Human Geography 17(2): 215­217.
MIGRATION AND FERTILITY: COMPETING HYPOTHESES RE-EXAMINED 83
Courgeau, D., 1985. `Interaction between spatial mobility, family and career life­cycle: a
French survey'. European Sociological Review 1(2): 139­162.
Courgeau, D., 1989. `Family formation and urbanization'. Population: An English Selection
44(1): 123­146.
Duncan, O. D., 1965. `Farm background and differential fertility'. Demography 2: 240­249.
Estonian Statistical Office, 2000. Public database of the census 2000. http://www.stat.ee.
Farber, S. C. and Lee, B. S., 1984. `Fertility adaptation of rural-to-urban migrant women: a
method of estimation applied to Korean women'. Demography 21(3): 339­345.
Fischer, P. A. and Malmberg, G., 2001. `Settled people don't move: on life-course and
(im)mobility in Sweden'. International Journal of Population Geography 7(5): 357­371.
Freedman, R. and Slesinger, D. P., 1961. `Fertility differentials for the indigenous non­farm
population of the United States'. Population Studies 15(2): 161­173.
Geyer, H. and Kontuly, T., (eds.) 1996. Differential Urbanization: Integrating Spatial Models.
Edward Arnold, London.
Goldberg, D., 1959. `The fertility of two­generation urbanites'. Population Studies 12(3): 214­
222.
Goldberg, D., 1960. `Another look at the Indianapolis fertility data'. Milbank Memorial Fund
Quarterly 38(1): 23­36.
Goldstein, S., 1973. `Interrelations between migration and fertility in Thailanď. Demography
10(2): 225­241.
Goldstein, S. and Goldstein, A., 1981. `The impact of migration on fertility: an `own children'
analysis for Thailanď. Population Studies 35(2): 265­281.
Gordon, I. R. and Molho, I., 1995. `Duration dependence in migration behaviour: cumulative
inertia versus stochastic change'. Environment and Planning A 27(12): 1961­1975.
Hendershot, G. F., 1971. Cityward migration and urban fertility in the Philippines. Paper
presented at the 1971 annual meeting of the Population Association of America.
Hervitz, H. M., 1985. `Selectivity, adaptation, or disruption? A comparison of alternative
hypotheses on the effects of migration on fertility: the case of Braziĺ. International
Migration Review 19(2): 293­317.
Hiday, V. A., 1978. `Migration, urbanization, and fertility in the Philippines'. International
Migration Review 12(3): 370­385.
Hoem, J. M., 1993. Classical Demographic Models of Analysis and Modern Event­history
Techniques. Stockholm Research Reports in Demography 75. Stockholm University,
Demography Unit, Stockholm.
Hoem, J. M., Neyer, G. and Prskawetz, A., 2001a. Autonomy or Conservative Adjustment? The
Effect of Public Policies and Educational Attainment on Third Births in Austria. MPIDR
Working Paper 2001­016. Max Planck Institute for Demographic Research, Rostock.
Hoem, J. M., Neyer, G. and Prskawetz, A., 2001b. `Autonomy or conservative adjustment?
The effect of public policies and educational attainment on third births in Austria'.
Population Studies 55(3): 249­261.
Kahn, J. R., 1994. `Immigrant and native fertility during the 1980s: adaptation and expectations
for the future'. International Migration Review 28(3): 501­519.
Katus, K., 2000. `General patterns of post-transitional fertility in Estonia'. Trames 4(3): 213­
230.
Katus, K., Puur, A., Sakkeus, L. and Silver, B., 1995. Estonian Family and Fertility Survey:
Methodological Report. Population Studies A 39. Estonian Interuniversity Population
Research Centre, Tallinn.
Katus, K., Kupiszewski, M., Rees, P., Sakkeus, L., Herm, A. and Powell, D., 1999. Internal
Migration and Regional Population Dynamics in Estonia. Population Studies B 37. Estonian
Interuniversity Population Research Centre, Tallinn.
HILL KULU84
Katus, K., Pungas, E., Puur, A. and Sakkeus, L., 2000. `Estonian family and fertility survey:
experience from the transforming statistical environmenť. Trames 4(3): 317­346.
Katus, K., Puur, A. and Pô ldma, A., 2002. Eesti pôlvkondlik rahvastikuareng [Population
Development in Estonia across Generations]. Rahvastiku-uuringud D 2. Eesti Kô rgkoolidevaheline
Demouuringute Keskus, Tallinn.
Kohler, H.-P., 2000. `Social interaction and fluctuations in birth rates'. Population Studies
54(2): 223­232.
Kravdal, ., 2001. `The high fertility of college educated women in Norway: an artefact of the
separate modelling of each parity transition'. Demographic Research 5(6): 185­216.
Kreyenfeld, M., 2002. `Time-squeeze, partner effect or self-selection? An investigation into the
positive effect of women's education on second birth risks in West Germany'. Demographic
Research 7(2): 13­48.
Kulu, H., 2002. `Socialization and residence: ethnic return migrants in Estonia'. Environment
and Planning A 34(2): 289­316.
Kulu, H., 2003a. `Post-war immigration to Estonia: a comparative perspective', in R. Ohliger,
K. Scho¨ nwä lder and T. Triadafilopoulos (eds.), European Encounters, 1945­2000: Migrants,
Migration and European Societies Since 1945. Ashgate, Aldershot, 38­52.
Kulu, H., 2003b. `Housing differences in the late Soviet city: the case of Tartu, Estonia'.
International Journal of Urban and Regional Research 27(4): 897­911.
Kulu, H. and Billari, F. C., 2004. `Multilevel analysis of internal migration in a transitional
country: the case of Estonia'. Regional Studies 38(6): 679­696.
Lee, B. S., 1992. `The influence of rural­urban migration on migranťs fertility behavior in
Cameroon'. International Migration Review 26(4): 1416­1447.
Lee, B. S. and Pol, L. G., 1993. `The influence of rural­urban migration on migrants fertility in
Korea, Mexico and Cameroon'. Population Research and Policy Review 12(1): 3­26.
Lesthaeghe, R. and Neels, K., 2002. `From the first to the second demographic transition: an
interpretation of the spatial continuity of demographic innovation in France, Belgium and
Switzerlanď. European Journal of Population 18(4): 325­360.
Lillard, L. A., 1993. `Simultaneous equations for hazards: marriage duration and fertility
timing'. Journal of Econometrics 56(1­2): 189­217.
Lillard, L. A. and Waite, L. J., 1993. `A joint model of marital childbearing and marital
disruption'. Demography 30(4): 653­681.
Lillard, L. A., Brien, M.J. and Waite, L. J., 1995. `Premarital cohabitation and subsequent
marital dissolution: a matter of self-selection'. Demography 32(3): 437­457.
Lindstrom, D. P., 2003. `Rural­urban migration and reproductive behaviour in Guatemala'.
Population Research and Policy Review 22(4): 351­372.
Ma, Z. and Liaw, K.-L., 1997. `Explaining hierarchical and interprovincial migrations of
Chinese young adults by personal factors and place attributes: a nested logit analysis'.
Mathematical Population Studies 6(3): 217­239.
Macisco, J. J., Bouvier, L. F. and Weller, R. H., 1970. `The effect of labor force participation
on the relation between migration status and fertility in San Juan, Puerto Rico'. Milbank
Memorial Fund Quarterly 48(1): 51­70.
Marksoo, A., 1990. `Tallinn Eesti rahvarä nde su¨ steemis [The place of Tallinn in the migration
system of Estonia]'. Eesti Geograafia Seltsi Aastaraamat 25: 53­66.
Marksoo A., 1992. `Dynamics of rural population in Estonia in the 1980s', in J.-M. Punning
(ed.), Estonia: Man and Nature. Estonian Geographical Society, Tallinn, 129­153.
McGirr, N. J. and Hirschman, C., 1979. `The two­generation urbanite hypothesis revisiteď.
Demography 16(1): 27­35.
MIGRATION AND FERTILITY: COMPETING HYPOTHESES RE-EXAMINED 85
Michielin, F., 2002. Lowest Low Fertility in an Urban Context: When Migration Plays a Key
Role. MPIDR Working Paper 2002-050. Max Planck Institute for Demographic Research,
Rostock.
Montgomery, M., 1992. `Household formation and home-ownership in France', in J. Trussell,
R. Hankinson and J. Tilton (eds.), Demographic Applications of Event History Analysis.
Clarendon Press, Oxford, 94­119.
Mulder, C. H. and Wagner, M., 1993. `Migration and marriage in the life course: a method for
studying synchronized events'. European Journal of Population 9(1): 55­76.
Mulder, C. H. and Wagner, M., 2001. `The connection between family formation and firsttime
home ownership in the context of West Germany and the Netherlands'. European
Journal of Population 17(2): 137­164.
Murphy, M. J. and Sullivan, O., 1985. `Housing tenure and family formation in contemporary
Britain'. European Sociological Review 1(3): 230­243.
Myers, G. C. and Morris, E. W., 1966. `Migration and fertility in Puerto Rico'. Population
Studies 20(1): 85­96.
Newbold, K. B., 1999. `Spatial distribution and redistribution of immigrants in the metropolitan
United States, 1980 and 1990'. Economic Geography 75(3): 254­271.
Newbold, K. B. and Liaw, K.-L., 1995. `Return and onward migrations in Canada, 1976­
1981: an explanation based on personal and ecological variables'. The Canadian Geographer
39(1): 16­30.
Philipov, D., 2002. Fertility in Times of Discontinuous Societal Change: The Case of Central
and Eastern Europe. MPIDR Working Paper 2002-024. Max Planck Institute for Demographic
Research, Rostock.
Pollak, R. A. and Watkins, S. C., 1993. `Cultural and economic approaches to fertility: proper
marriage or mé salliance?'. Population and Development Review 19(3): 467­496.
Rosenwaite, I., 1973. `Two generations of Italians in America: their fertility experience'.
International Migration Review 7(3): 271­280.
Rundquist, F.-M. and Brown, L. A., 1989. `Migrant fertility differentials in Ecuador'.
Geografiska Annaler B 71(2): 109­123.
Rybakovskiy, L. L. and Tarasova, N. V., 1991. `Contemporary problems of migration of the
population of the USSR'. Soviet Geography 32(7): 458­473.
Sakkeus, L., 2000. `Demographic behaviour patterns of immigrants and national minority of
the same ethnic background: case of Estonia'. Trames 4(3): 268­285.
Singley, S. G. and Landale, N. S., 1998. `Incorporating origin and process in migrationfertility
frameworks: the case of Puerto Rican women'. Social Forces 76(4): 1437­1464.
Stephen, E. H. and Bean, F. D., 1992. `Assimilation, disruption and the fertility of Mexicanorigin
women in the United States'. International Migration Review 26(1): 67­88.
Tammaru, T., 2001. `Urbanization in Estonia in the 1990s: Soviet legacy and the logic of
transition'. Post-Soviet Geography and Economics 42(7): 504­518.
Tammaru, T., Kulu, H. and Kask, I., 2004. `Urbanisation, suburbanisation and counterurbanisation
in post-Soviet Estonia'. Eurasian Geography and Economics 45(3): 212­229.
Umezaki, M. and Ohtsuka, R., 1998. `Impact of rural­urban migration on fertility: a population
ecology analysis in the Kombio, Papua New Guinea'. Journal of Biosocial Science
30(3): 411­422.
Vikat, A., 1994. Family Formation in Estonia. Publications of the Finnish Demographic
Society 15. Finnish Demographic Society, Helsinki.
White, M. L., Moreno, L. and Guo, S., 1995. `The interrelation of fertility and geographic
mobility in Peru: a hazards model analysis'. International Migration Review 29(2): 492­514.
Woods, R., 2003. `Urbanisation in Europe and China during second millennium: a review of
urbanism and demography'. International Journal of Population Geography 9(3): 215­227.
HILL KULU86
Zarate, A. and Zarate, A. U. de, 1975. `On the reconciliation of research findings of migrantnonmigrant
fertility differentials in urban areas'. International Migration Review 9(2): 115­
156.
Zelinsky, W., 1971. `The hypothesis of the mobility transition'. The Geographical Review 61(2):
219­249.
MIGRATION AND FERTILITY: COMPETING HYPOTHESES RE-EXAMINED 87