Human capital background and the educational attainment
of second-generation immigrants in France #
Manon Domingues Dos Santos *
François-Charles Wolff **
First revision, Economics of Education Review
November 2010
Abstract: In this paper, we study the impact of parental human capital background on ethnic
educational gaps between second-generation immigrants using a large data set conducted in
France in 2003. Estimates from censored random effect ordered Probit regressions show that
the skills of immigrants explain in the most part, the ethnic educational gap between their
children. Fluency in French and the length of their parents’ stay in France also matter. The
impact of the immigrants’ education on the educational attainment of their children further
depends on their country of origin, their place of schooling, and their proficiency in French.
Keywords: Immigrants, second generation, educational attainment, France
JEL Classification: I21, J15, J24, J61
#
We would like to thank three anonymous referees and the Associate Editor, Mikael Lindahl, for their very
helpful comments and insightful suggestions on a previous draft. We are also indebted to Pierre Cahuc, Arnaud
Chevalier, Francis Kramarz and participants at the TEMA seminar (University Paris 1), the Crest seminar
(INSEE), the Journées de Microéconomie Appliquée (Fribourg), the Annual Conference of the European Society
for Population Economics (Chicago) and the Annual Conference of the European Association of Labour
Economics (Oslo) for useful comments. The usual disclaimer applies.
*
OEP, University Paris-Est Marne-la-Vallée and CREST, INSEE, France.
Email : manondds@ensae.fr http://www.crest.fr/pageperso/manondds/manondds.htm
**
Corresponding author. LEMNA, Université de Nantes ; CNAV and INED, France.
Email: francois.wolff@univ-nantes.fr http://www.sc-eco.univ-nantes.fr/~fcwolff
1
1. Introduction
In France, 42% of the young men and 27% of the young women whose parents are
Northwest Africans leave school without any diploma. Among the children whose parents are
natives or Southern Europeans, this proportion is about two times lower amongst men and
nearly three times lower among women (Lainé and Okba, 2005). In many countries,
significant differences are also observed in the educational achievements of children from
different origins (OECD, 2006). What can explain such inequalities? As emphasized in the
Chiswick’s seminal paper (1988), three main hypotheses may explain why different ethnic
groups achieve different levels of educational attainment.
Firstly, some communities are likely to have a greater preference for schooling, which
may be due to cultural, religious or historical factors. This particular taste for schooling can
lead members of these communities to invest more in the human capital of their descendants.
Secondly, ethnic differences in educational attainment may arise from discrimination. During
their studies, children from some communities may be discriminated against with regard to
access to schooling, quality of schooling, grade retention or tracking decisions (Losen and
Orfield, 2002). During their working life, they can face less favorable job conditions. As the
returns of their investment in human capital are lower, they would then be less motivated to
invest in skills (Coate and Loury, 1993). Thirdly, some ethnic communities may be overrepresented in the most disadvantaged socio-economic groups. Following the human capital
theory, immigrants of these communities would then invest less in the education of their
children.
Do we have to promote special educational programs in favour of some secondgeneration youths? Do we have to assist immigrants themselves in their educative mission?
Are anti-discriminatory policies which encourage anonymous applications able to narrow the
educational gaps? To evaluate the relative efficiency of such policies, it is necessary to
rigorously identify the determinants of the educational attainment gaps. In this paper, we
investigate how differences in human capital background among immigrant communities
explain the educational attainments gaps among their descendants using a data set on
immigrants living in France in 2003. We extend the growing literature which intends to
understand the role of the family socio-economic background on the educational attainment of
the second-generation immigrants in the following ways.
On the one hand, our contribution is the first study to focus on this issue using French
data. Previous empirical studies have dealt with Anglo-Saxon countries, Germany (Gang and
Zimmermann, 2000) and the Netherlands (Van Ours and Veenman, 2003). It is not obvious
2
that these results can be readily extrapolated to the situation of immigrants living in other
countries. Institutional differences in immigration policies and education systems between
Anglo-Saxon and European countries could affect the educational achievement of children of
immigrants. On the other hand, the quality of the data allows us to better take the migratory
history of the parents of second-generation immigrants into account. For instance, our
regressions control for their fluency in the host country’s language, their length of stay in
France or their place of study.
We show that skills of immigrants explain, in the most part, the ethnic educational gap
between their descendants in France. French fluency and the length of stay in France of
parents also matter. The impact of the immigrants’ education on the educational attainment of
their children further depends on their country of origin, their place of schooling and their
proficiency in French. The remainder of our paper is organized as follows. In Section 2, we
briefly review the literature on human capital background and educational attainment of
second-generation immigrants. We describe the French data used for our empirical analysis in
Section 3 and provide some descriptive statistics in Section 4. We present our econometric
strategy in Section 5 and discuss our results in Section 6. Finally, Section 7 concludes.
2. Background on the educational attainment of second-generation immigrants
According to the Programme for International Students Assessment (OECD, 2006), in
many OECD countries the second-generation students defined as those who are born in the
host country and those who have arrived as children before school age perform at a lower
level with respect to mathematical literacy, reading literacy and scientific literacy than their
native peers. Understanding the determinants of this achievement gap is a key challenge.
According to previous studies, the educational attainment of parents, language spoken at
home and the age at immigration of children significantly influence the educational
achievement of the second generation.
As pointed out in Haveman and Wolfe (1995), the main determinant of the educational
attainment of children is the educational attainment of their parents. The PISA surveys show
that in OECD countries the parents of first-generation and second-generation students have,
on average, completed fewer years of schooling than the parents of native students. A simple
explanation of the educational gap between second-generation immigrants and their nativeparentage peers is hence that parents of the former group are on average less skilled than
those of the latter.
3
Following Chiswick’s (1988) seminal contribution, Borjas (1995) finds a positive
correlation between parental skills and the skills of children in the US, but this correlation is
not sufficiently high to remove ethnic skill differentials. This result is also confirmed by Card
et al. (2000). Nevertheless, both studies are based on data that do not directly link the skills of
a given child with the skills of its own parents. They only exhibit correlations between the
average skills of a cohort of immigrants and the average skills of a cohort of immigrants’
children with respect to groups of different national origins and do not distinguish between
second-generation immigrants by country of origin.
Focusing on the largest groups of immigrants living in Germany (Turkish, Yugoslav,
Greek, Italian, Spanish), Gang and Zimmermann (2000) showed that Turkish and Yugoslav
pupils obtain less favorable results than German pupils. Parental schooling plays no role in
the educational attainment of foreign born children, whereas it has a major role in the
educational attainment of the native parentage children. In the Netherlands, Van Ours and
Veenman (2003) found that differences in parental education explain differences in
educational attainment between ethnic groups. Turkish, Moroccan and Antilleanpupils
perform worse than Dutch pupils, but there is no difference between the school attendance of
second-generation immigrants (whatever their ethnic origin) and natives once the educational
level of parents is taken into account.
So, the intergenerational transferability of skills may differ not only between natives
and immigrants, but also between various ethnic communities. Several studies have further
stressed that the returns to foreign experience and education were lower than those obtained
domestically (Chiswick and Miller, 1985, Kossoudji, 1989, Schaafsma and Sweetman, 2001).
As long as immigrants are unable to completely transfer the human capital accumulated in
their home country to the labor market of the host country, it is also plausible that the impact
of their skills on the educational attainment of their children depends on the place where these
skills have been acquired. This suggests that the effect of the human capital of immigrants on
the educational achievement of their children should further be affected by the immigrants’
place of schooling as well as by their length of stay in the host country.
Language spoken at home should also affect the educational achievement of children
of foreign origin. Taking into account parents’ educational and occupational status, several
contributions have shown that educational attainment was lower among young people who
did not speak the language of the host country at home (Jones, 1987, Schaafsma and
Sweetman, 2001). Educational learning will be less easy for children not fluent in the host
country’s language, as they will face more difficulties revising their lessons and doing their
4
schoolwork. Language spoken at home is not only an instrument for transmitting
intergenerational effects, it will also indicate the degree of assimilation of parents1. At the
same time, the effect of language is a little bit more complex in France since there are
immigrants from former French colonies, where French is still a dominant language.
Place of birth and age at arrival affect the academic success of immigrants’ children,
as shown by Gonzales (2003), Chiswick and DebBurman (2004), Van Ours and Veenman
(2006) and Böhlmark (2008). Children who arrive in their teenage years achieve a much
lower educational attainment than those who arrive at a very young age. It is easier for
younger children to assimilate the culture of the host country. Age at immigration also
influences the ability to acquire destination language skills because of the greater exposure to
the destination language at school and the greater ability of children to learn a new language
(Bleakley and Chin, 2004, Newport, 2002). As there are large cultural differences among
immigrants depending on their country of origin, age at immigration and place of birth are
likely to affect children from different origins differently.
To summarize, the educational achievement of children belonging to different ethnic
groups should depend not only on their parents’ level of education, but also on the migration
history itself through the place of parents’ schooling, the length of stay in the host country and
host language proficiency. In what follows, we consider a unique data set containing detailed
characteristics of both immigrant parents and their children to study the educational
attainment of the second-generation immigrants in France.
3. The Passage to Retirement of Immigrants survey
For our purpose, we consider a French data set entitled Passage to Retirement of
Immigrants survey (PRI hereafter). This survey was conducted by the Caisse Nationale
d’Assurance Vieillesse and the Institut National de la Statistique et des Etudes Economiques
from November 2002 to February 2003. The sample comprises foreign respondents born in a
foreign country between 1932 and 1957 and living in France at the date of the survey. The
survey is representative of the different nationalities of the first-generation of immigrants
living in France in 2002-2003 and whose age is between 45 and 70 (for a detailed description,
see Attias-Donfut et al., 2006).
Given the focus of the PRI survey on older migrants, first-generation immigrants
between 35 and 44 years of age are by definition excluded from our empirical analysis. These
1
Chiswick et al. (2004) find a large positive correlation in the unmeasured determinants of proficiency between
parents and children after taking into account age, family status and years of schooling.
5
younger immigrants represent approximately 20% of the migrant population living in France,
meaning that a significant proportion of immigrants is not covered by our survey2. However,
the problem is perhaps less severe than it seems. Indeed, younger first-generation immigrants
are expected to have, on average, young children themselves. Among those who are old
enough to attend school, many of them are likely to be enrolled and to attend either primary or
secondary schools. There is unfortunately little to learn about educational attainment when
considering these young children: we just know that they will complete additional schooling.
Among other topics, the PRI survey contains basic information on demographic and
socioeconomic characteristics like gender, age, family status, education, financial status as
well as work trajectories. It also includes original information on the migration history of the
respondent, like economic status in the country of origin, date of immigration and level of
proficiency in French. A unique feature of this survey is that it provides detailed information
on the respondent’s extended family.
For each child, we have information on gender, year of birth, country of birth,
citizenship, current place and, when relevant, year of arrival in France. The survey includes
several additional questions only when the children are at least 16 years old3. In particular, we
know whether each child is still enrolled in school or is a student as well as their highest level
of diploma according to the following categories: ‘no education’, ‘primary or secondary
schooling’, ‘vocational school’, ‘high school’, ‘undergraduate studies’, ‘graduate studies’, and
‘postgraduate studies’.
The core sample of the survey includes 6211 respondents. From this ‘parent’ sample,
we constructed a ‘child’ sample where each child is counted as one observation. This leads to
a sample of N=19,285 parent-child pairs. Among them, 83.3% are at least 16 years old. We
deleted the younger children (N=3,234) since we have no information on their schooling in
the survey. The only thing we know is that all these children should be enrolled since
education is mandatory until the age of 16 in France (since January 1959). However, leaving
out the younger children does not lead to selection bias as the selection is based on age, which
is an exogenous covariate (see the discussion in Ejrnaes and Pörtner, 2004).
As older children may have experienced very different educational conditions, we
dropped a small number of children aged above 50 (N=29) from the sample. A shortcoming
of the survey is that we do not know where child’s schooling was completed. For that reason,
2
Immigrants more than 45 years old represent about 50% of the immigrants living in France.
As this is a recollection date from the parents, then the data about the children may suffer from measurement
errors. Controlling for unobserved heterogeneity using family specific effects will reduce the underlying bias.
3
6
we choose to exclude all the children who are not living in France at the date of the survey as
they may have completed their studies in a foreign country (N=1,735). Since the conventional
definition of second generation refers to children born in the host country or immigrated
before school age, we deleted those who arrived in France after six years (N=2,718). Finally,
we deleted children with missing information on the educational attainment of both the father
and the mother (N=1,634) along with the few children with missing educational outcomes
(N=31).
These different selections leave us with a full sample of N=9,904 children belonging
to N=4,118 families. About 83% of the respondents have at most three children. The
proportion is 28% for parents with one child satisfying the selection criteria, 33.8% for
parents with two children, and 21% for parents with three children. Since there are multiple
observations per family in many cases, we will be able to control for unobserved
heterogeneity at the family level through the use of family specific effect models. Actually,
many unobserved factors associated with the child’s educational attainment (like parental
altruism or parental ability) are presumably highly correlated within siblings4.
We present the distribution of education among the second generation in Figure 1. The
proportion of children having achieved more than high school is equal to 25.7%. We note that
there are large gender differences in education, the former proportion being respectively
30.2% among girls instead of 21.6% among boys. A significant proportion of children were
enrolled at the date of the survey (29.7%)5. These are censored observations, since by
definition we do not observe their final level of education. Given their importance, all the
children (either enrolled or not) will be included in our regressions when explaining the
educational attainment of the immigrants’ children.
Insert Figure 1 here
4. Descriptive statistics
We present some descriptive statistics in Table 1, both for the parents and the secondgeneration children. Concerning the former, the distribution with respect to their country of
origin reflects the French immigration pattern. The Southern European community
(essentially Portuguese, Spanish and Italian respondents) and the Northern African
community (Algerian, Moroccan and Tunisian migrants) represent more than 75% of the
4
In a recent contribution, Böhlmark (2008) also exploits within-family variation to study the role of age at
immigration for the school performance gap between native and immigrant pupils in Sweden.
5
Again, the proportion of children is much higher among girls (32.4%) than among boys (27.1%).
7
immigrant population living in France. Migrants from Asia and Turkey are much less
numerous, although their fraction increases over time.
Insert Table 1 here
The sample shows important differences between communities with respect to skills.
Female immigrants are on average less educated than male immigrants. However, differences
in skill acquisition between the women of two communities are rather similar to the ones
between the men of the same two communities. Immigrants coming from Northern Europe
and America are more educated, many of them having graduated. On the opposite, very few
North Africans have graduated and more than one-third of this population has never been
enrolled. The proportion of parents with no education is 10% among the Middle Eastern
community, but only 5% among the Southern European population. Finally, the Asian
community is more heterogeneous, 40% of Asians having graduated and one quarter of them
having no diploma.
With respect to proficiency in French, nearly one-fifth of Northern Europeans and
Americans have difficulties in speaking or writing French, whereas nearly half of North
Africans are concerned. Comparing Asians to Southern Europeans, we note the particularly
high fraction of Asians facing difficulties in French in spite of an education pattern close to
the education pattern of Southern Europeans. The younger mean age at arrival of Southern
Europeans and their longer length of stay in France could contribute to explaining their better
proficiency in French.
Regarding the socioeconomic status of immigrants in their home country, Southern
European and North African immigrants have a higher propensity to have grown up in a small
town or a village as well as in a poor or very poor financial context than other immigrants.
These findings are of course related to differences in economic development among the
various countries of origin. Finally, we observe significant differences in the number of
children. North Africans have on average three children above 16, while Northern and
Southern Europeans, as well as Americans, have less than two children.
Concerning the children, significant differences appear between communities with
respect to their educational attainment. About 40% of children originating from Northern and
Eastern Europe, America and Asia have completed more than high school education, while
this proportion reaches 30% among Southern Europeans, 19% among Northern Africans and
18% among Middle Eastern children. Conversely, the proportion of children with no
education or primary school is much higher in these ethnic groups. This proportion is
8
respectively 35% and 43% when parents are from North Africa and from other African
countries, while the average rate for the whole sample of children is 29%.
There are more boys than girls in our sample. The proportion of girls is 47% and 46%
when the parents originate from Southern Europe and Middle East respectively. As we only
focus on children currently living in France, this difference in the gender composition is due
to the fact that girls are more likely to live in the country of origin6. On average, children from
non-European countries are younger. Part of this age gap stems from differences in fertility
rates, as parents from Africa, the Middle East and to a lesser extent from Asia have more
children than other parents. The age pattern explains why children from Africa are more likely
to be enrolled. Nevertheless, differences in enrolment rates are also due to the increased
propensity of some groups to invest in the human capital of their children, which may explain
the high rate observed among Asians.
Finally, when considering the place of birth, almost all children originating from
Southern Europe are born in France, while about 30% of Asian and Middle Eastern children
are foreign-born. Migratory legislations concerning the free movement area and the
conditions under which immigrants have the right to have their family with them could partly
explain these disparities.
5. Econometric strategy
Since a large proportion of children were still in education at the time of the survey
(about 30% according to Table 1), we considered both children who were no longer enrolled
and children who had not yet completed schooling at the time of the survey in our
econometric analysis. The three following features of the PRI data have to be taken into
account when attempting to explain educational attainment.
First, the survey provides
ordered categorical information for the level of schooling. Specifically, the diploma pattern
denoted by e is defined in the following way: e  0 for ‘no education’, e  1 for ‘primary or
secondary level’, e  2 for ‘vocational studies’, e  3 for ‘high school’, e  4 for
‘undergraduate studies’, e  5 for ‘graduate studies’ and e  6 for ‘postgraduate studies’. We
assume that a continuous latent variable e* associated to the educational outcome exists,
which we express as a linear function of a set of family characteristics X , a vector of
coefficients  and a residual  :
6
When considering the whole sample of children (no restriction on age), 10.2% of girls are living in the country
of origin while the proportion is 9.1% for boys. As Dustmann (2003) points out, the welfare of the offspring
perceived by the parent may vary depending on the location of the child. Concerns about preserving traditions
may be more influential for female offspring than for male offspring.
9
e*  X  
(1)
Given the various categories, we assume that e*  1 when e  0 , 1  e*   2 when
e  1,  2  e*  3 when e  2 , .., and  6  e* when e  6 , 1 being normalized to 0 .
Assuming that the random perturbation  is normally distributed, the corresponding model is
a standard ordered Probit regression. The different parameters  j are a set of threshold levels
which have to be estimated jointly with the vector of coefficients  .
The second feature is that many children are still enrolled. Consider first the case of a
child who has completed schooling. From the definition of the ordered Probit model, the
probability Pr(e  j ) that a child has the diploma j is (with j  1,...,6 ):
Pr(e  j )  (  j 1  X )  (  j  X )
(2)
The case of children who are still enrolled is a little bit different. From the data, we know that
these children will end their schooling with at least the same diploma as they currently have
(this would occur in case of educational failure), and presumably with a higher diploma. For
these censored observations, the probability Pr( e  j ) may hence be expressed as:
Pr(e  j )  1  (  j  X )
(3)
In the general case, we thus get the following expression for a given level of education:
Pr(e  j )  c * (  j 1  X )  (  j  X )   (1  c) * 1  (  j  X ) 
(4)
with c a dummy variable equal to one when education is completed (uncensored observation)
and to zero otherwise (censored observation).
The last concern with the data is that we have information on several children for
many families. While siblings may be treated as independent observations, this assumption is
clearly unlikely to hold. Indeed, since children from a given family have the same parents,
their different educational levels are likely to be strongly correlated. Formally, this means that
the model we seek to estimate should have the following form:
e*fi  X fi    f   fi
(5)
where f and i refer respectively to the family and to the child. In (5),  f is an unobserved
family heterogeneity term. These family specific perturbations are supposed to be normally
distributed, with mean 0 and variance  2 . We assume that the disturbances  fi follow a
normal distribution, with mean 0 and variance  2 , and that X ,  and  are independent.
The likelihood function for the above model involves multivariate normal integrals, but one
can rely on quadrature techniques to estimate the corresponding random effect ordered Probit
10
model with censoring (see Frechette, 2001, Picard and Wolff, 2010). The contribution to the
final likelihood function is:
c *  (  j 1  X fi  )   (  j  X fi  ) 
Pr(ei1 ,..., eiN )   
  ( f ) d f
  (1  c ) * 1   (   X  ) 


j
fi
(3)
where  ( f ) is the density of N (0, 2f ) .
Since we estimate a random effect model, we suppose that the unobserved family
effects are uncorrelated with the covariates introduced into the regression. While an analysis
with fixed family effects (that may be correlated with family covariates) could be a
possibility, this would unfortunately impose to drop the most crucial covariates of the study as
parental characteristics would be picked up by the family fixed effects.
6. Econometric results
6.1. Unconditional returns to parental education
We first assume that the returns to the explanatory variables on the educational
attainment of the second-generation immigrants are independent of the parental migration
history. In Table 2, we report results from five random effect ordered Probit regressions. In
column 1, we control for the basic demographic characteristics of the children: gender, age,
birth cohorts, birth cohorts interacted with age, number of siblings, number of sisters, birth
order, place of birth and a dummy variables equal to one when the child was raised by both
parents till 12. We also add into the regression a set of dummy variables related to the
parental country of origin (with six groups), Northern Europe, Eastern Europe and America
being the reference.
Insert Table 2 here
As expected, the probability of having a higher level of education is greater for girls
than for boys. Following Picard and Wolff (2010), our specification means that we consider a
piecewise linear function for the child’s age. On the one hand, the small negative coefficient
associated to age indicates that education has slightly improved on average over time. On the
other hand, the birth cohort dummy for those born after 1985 shows that the youngest children
are characterized by much lower levels of education. This pattern is due to the fact that many
of them are still enrolled at the date of the survey. These children are expected to obtain much
higher diplomas once they will have completed their schooling.
Concerning family composition, both the number of siblings and birth order have a
negative impact on educational attainment, whereas the gender composition of the sibship is
11
not very significant. Having more sisters (for a given number of siblings) tends to improve
education. As shown in Table 2, family context matters since having been raised by both
parents during youth significantly improves schooling. We also find that educational
attainment is affected by the child’s migration history.
According to the PRI survey, being foreign-born has a strong negative impact on the
educational attainment of children (at the one percent level). Very similar results have been
found for the second generation living in the Netherlands by Van Ours and Veenman (2003).
This regression also shows significant differences between communities of origin with respect
to the child’s educational attainment. Southern European and North African children achieve
a significantly lower level of education than Northern European, Eastern European and
American children. The best performers originate from Asia, whereas the worst ones originate
from the Middle East.
As these community dummies are likely to pick up permanent differences in parental
characteristics, we then introduce the level of education of both the father and the mother in
the list of the covariates. The corresponding estimates lead to the following conclusions
(column 2, Table 2). Firstly, both the educational levels of the father and the mother have a
large positive impact on the child’s educational attainment. The child’s schooling is much
higher when either the father or the mother have completed more than high school studies.
However, it should be noted that there are no significant differences in the returns to paternal
and maternal education7. Secondly, once the educational attainment of the parents is
introduced in the regression, only two community effects differ from the others.
The negative impact of originating from the Middle East declines, but it remains
significant (at the one percent level), whereas the positive impact of originating from Asia is
now a little bit higher. Hence, except for the Middle Eastern and Asian populations, parents’
educational attainment seems to capture the main roots of the educational gap of children
between the migrants’ communities. We further add the current level of household income in
column 3. According to the data, we get a positive correlation between educational attainment
and parental income, but this additional covariate does not affect the magnitude of our
community coefficients8.
In column 4, we introduce additional variables dealing with parental migration history.
The length of stay in France when the child was 10 has a positive impact on the educational
7
We find an insignificant statistic when implementing a Wald test to assess the relevance of the hypothesis of
equal returns between paternal and maternal education.
8
Controlling for parental schooling and household income has also little effect on the influence of the child’s
characteristics.
12
attainment of the offspring, which is to be expected from the assimilation point of view. We
also observe a positive effect of parental age at migration. A possible explanation is that
migrants coming to France late are more experienced and devote more resources to invest in
the human capital of their children. Parental proficiency in speaking French has a large and
positive impact on the educational achievement of children. In our context, fluency in French
can stem from two main sources: assimilation if French has been learned in the post-migration
period or the country of origin (French speaking versus non-French speaking)9. Having poor
parents during childhood reduces educational attainment.
This last specification clearly shows that parental migration history influences the
educational potential of children in the context of the host country. At the same time, the
family human capital background explains, in the most part, educational attainment gaps
between communities in France. In columns 5 and 6 of Table 2, we report estimates from
separate models for boys and girls. An interesting result concerns the origin effect related to
the Middle East, which is negative and significant only among girls.
6.2. Conditional returns to parental human capital
To date, the effect of parental skills on the educational attainment of children was
supposed to be independent of parental migration history10. However, these returns are likely
to depend on the place where parents have been educated themselves. Returns to education in
the host country should be lower for immigrants coming from countries with less efficient
educational systems. The transferability of skills between countries may also depend on the
proximity between their education systems and on the ability of immigrants to transfer their
skills to their child in the context of host country. This ability could in particular depend on
host language fluency and length of stay in the host country.
To investigate these issues, we first convert the highest qualification of the parent into
years of schooling. Note that we restrict our attention to the education of the household head,
as the PRI questionnaire does not allow us to make a distinction between years of education
completed in the country of origin and years of education completed in France for both the
head and his/her spouse. In column 1 of Table 3, we present the results of the basic
specification with the parental years of schooling instead of the ordered levels of parental
9
This means that proficiency in French is not necessarily an indication of assimilation.
Chiswick and Miller (1985), Kossoudji (1989), Schaafsma and Sweetman (2001) point notably that the return
to foreign experience and education on the labor market is valued less than that obtained domestically.
10
13
education. We again find evidence of a significant positive effect of parental education, the
other estimates being rather similar to those described in column 3 of Table 211.
Insert Table 3 here
We then estimate a model where the number of years of schooling is interacted with
the country of origin of the parent (column 2, Table 3). We reach the following conclusions.
Firstly, the returns to the parental education are significantly lower when the parent originates
from North Africa. Secondly, taking into account the potential interaction between the returns
to education and the country of origin of the parent lowers the residual impact of the country
of origin. We can then argue that the origin of immigrants affects the impact of their
education on the educational attainment of their child in the context of the host country.
We also study whether the proficiency in French of immigrant affects their ability to
transfer their skills to their children in the host country. For that purpose, we introduce an
interaction term between years of schooling and proficiency in French in column 3. We find
that the returns to parental education on the educational attainment of the child are lower
when the parent faces difficulties in speaking French. Then, host language proficiency
facilitates the transfer of skill from immigrants to their children. Also, the crossed term
significantly reduces the returns to parental education for Middle Eastern immigrants,
whereas it increases the returns to parental education for Asian immigrants. Hence, Asian
immigrants have a comparative advantage in transferring their skills to their children whereas
Middle Eastern immigrants have a comparative disadvantage.
Finally, we take into account the age at immigration of the parent and break years of
schooling down into years of schooling in the country of origin and years of schooling in
France. Column 4 of Table 3 shows that the returns to years of schooling in the country of
origin are slightly lower than the returns to years of schooling in France. However, the
difference between the two coefficients is not significant at a conventional level. We also
introduce crossed effects between the place of schooling and the proficiency in French level.
These unreported additional results indicate that difficulties in speaking French have a
particularly detrimental effect on the return to foreign education for North Africans.
7. Conclusion
11
With respect to our third specification in Table 2 which includes four dummy variables for both paternal and
maternal education, we note some differences concerning the origin country dummies. The coefficient is now
negative when parents originate from Southern Europe (at the 5% level) and from North Africa (at the 10%
level), Northern and Eastern Europe being the reference category. Nevertheless, the largest estimates remain the
negative coefficient of Middle Eastern origin and the positive coefficient of Asian origin.
14
In this paper, we have investigated the determinants of the educational attainment of
the second-generation of immigrants living in France. Our regressions control for unobserved
heterogeneity at the family level and account for censoring of enrolled children. Our main
finding is that the skill of immigrants explain, for the most part, the ethnic educational gaps
between their children. Immigrants’ fluency in French and length of stay in the host country
also play a significant role, whilst the impact of immigrants’ educational attainment on the
educational attainment of their children depends on their place of origin, their place of
schooling as well as their proficiency in French.
When comparing the different ethnic groups, we find that educational attainment is
lower among children of Middle Eastern origin. Also, the returns to parental education are
lower when parents are from North Africa. From a public policy viewpoint, this suggests that
it could be useful to promote special educational programs in favor of these disadvantaged
groups. At the same time, it would be of interest to further understand why some parents tend
to under-invest in the human capital of their children. If for instance parents choose to invest
less in the education of their children because they expect their children to be discriminated
against on the labor market, then anti-discriminatory policies encouraging anonymous
applications would be effective in narrowing the educational gaps among children of different
origins.
A final comment is that we have only focused on the educational attainment of
second-generation immigrants in this paper. This is due to the fact that the PRI survey only
includes respondents of foreign origin. It would therefore be useful to further compare not
only educational outcomes, but also employment and earnings of both second-generation
immigrants and natives in France. At the same time, our paper clearly shows the necessity of
having large data sets on immigrants in developed countries to be able to study the magnitude
of potential differences in the behaviors of various ethnic groups. The focus often placed on
behavioral differences between natives and migrants should definitely not mask heterogeneity
within the immigrant community.
15
References
Attias-Donfut C., Davaut P., Gallou R., Rozenkier A., Wolff F.C., (2006), L’enracinement.
Enquête sur le vieillissement des immigrés en France, Armand Colin, Paris.
Bleakley H., Chin A., 2004. “Language skills and earnings: Evidence from childhood
immigrants”, Review of Economics and Statistics, 86 (481-496).
Böhlmark A., 2008. “Age at immigration and school performance: A siblings analysis using
Swedish register data”, Labour Economics, 15 (1366-1387).
Borjas G., 1995. “Ethnicity, neighbourhoods, and human capital externalities”, American
Economic Review, 85 (365-390).
Card D., DiNardo J., Estes E., 1998. “The more things change: Immigrants and children of
immigrants in the 1940s, the 1970s, and the 1990s”, NBER Working Paper, 6519.
Chiswick B., 1988. “Differences in education and earnings across racial and ethnic groups:
tastes, discrimination, and investment in child quality”, Quarterly Journal of
Economics, 103 (571-597).
Chiswick B., DebBurman N., 2004. “Educational attainment: Analysis by immigrant
generation”, Economics of Education Review, 23 (361-379).
Chiswick B., Miller P., 1985. “Immigrant generation and income in Australia”, Economic
Record, 61 (540-553).
Coate S., Loury G., 1993. “Will affirmative-action policies eliminate negative stereotypes ?”,
American Economic Review, 83 (1220-1240).
Dustmann C., 2003. “Children and return migration”, Journal of Population Economics, 16
(815-830).
Ejrnaes M, Pörtner C, 2004. “Birth order and the intrahousehold allocation of time and
education”, Review of Economics and Statistics, 86 (1008–1019).
Frechette G.R., 2001. “Random effects ordered Probit”, Stata Technical Bulletin, 159 (23–
27).
Gang I., Zimmermann K., 2000. “Is child like parent ? Educational attainment and ethnic
origin”, Journal of Human Resources, 35 (550-569).
Gonzalez A., 2003. “The education and wage of immigrant children: The impact of age at
arrival”, Economics of Education Review, 22 (203-212).
Haveman R., Wolfe B., 1995. “The determinants of children’s attainments: A review of
methods and findings”, Journal of Economic Literature, 4 (1829-1878).
Jones F., 1987. “Age at immigration and education: Further implications”, International
Migration Review, 21 (70-85).
Kossoudji S., 1989. “Immigrant workers assimilation: Is it a labour market phenomenon?”,
Journal of Human Resources, 24 (494-527).
Lainé F., Okba M., 2005. “L’insertion des jeunes issus de l’immigration: De l’école au
métier”, Collection Net-Doc du CEREQ, 15.
Losen D., Orfield G., 2002. Minority Issues in Special Education, The civil rights project at
Harvard University and Harvard Education Press, Cambridge.
16
Newport E., 2002. “Critical periods in language development”, in Nadel L., ed, Encyclopedia
of Cognitive Science, Mac-Millan Publishing, London.
OECD, 2006. Where immigrant students succeed. A comparative review of performance and
engagement in PISA 2003.
Picard N., Wolff F.C., 2010. ”Measuring educational inequalities: Evidence from Albania”,
Journal of Population Economics, 23 (989-1023).
Schaafsma J., Sweetman A., 2001. “Immigrants earnings: Age at migration matters”,
Canadian Journal of Economics, 34 (1066-1099).
Van Ours J., Veenman J., 2003. “The educational attainment of second generation immigrants
in Netherlands”, Journal of Population Economics, 35 (550-569).
Van Ours J., Veenman J., 2006. “Age at immigration and educational attainment of young
immigrants”, Economics Letters, 90 (310-316).
17
Figure 1. Distribution of educational attainment of the second generation
35
Frequency (in %)
Boys
Girls
30
All
25
20
15
10
5
0
No education
Primary/secondary
Vocational school
High school
Source: survey PRI 2003.
18
Undergraduate
Graduate
Postgraduate
Table 1. Descriptive statistics of the full sample
Country of origin of the parent
Variables
North & East
Europe
Characteristics of the parents
Father’s education No education
Primary
Secondary
High school
More than high school
Mother’s education No education
Primary
Secondary
High school
More than high school
Household’s income (log)
Age at migration
Language
Difficulty in speaking French
Difficulty in writing French
In a large town before migration
Financial status
Very poor
when 16
Poor
Fair
Good
Number of children above 16
Characteristics of the children
Education
No education
Primary/secondary
Vocational
High school
Undergraduate
Graduate
Postgraduate
Still enrolled in school
Female
Age
Number of siblings
Number of sisters
Rank within the siblings
Raised by both head and spouse till 12
Foreign born
Number of children
Number of parents
Source : survey PRI 2003.
South Europe
North Africa
Other Africa
America
Middle
Asia
All
0.02
0.15
0.26
0.17
0.41
0.02
0.18
0.27
0.19
0.33
10.19
22.64
0.18
0.21
0.47
0.11
0.23
0.36
0.30
1.95
0.05
0.55
0.29
0.06
0.05
0.05
0.54
0.29
0.07
0.05
9.87
18.30
0.30
0.36
0.23
0.26
0.32
0.34
0.08
2.04
0.34
0.33
0.22
0.06
0.05
0.42
0.29
0.19
0.06
0.04
9.62
23.47
0.48
0.50
0.32
0.24
0.25
0.38
0.13
3.05
0.19
0.17
0.18
0.11
0.35
0.21
0.18
0.24
0.17
0.20
9.93
27.08
0.28
0.31
0.65
0.16
0.20
0.37
0.28
2.23
0.04
0.15
0.21
0.06
0.54
0.02
0.17
0.15
0.17
0.48
9.96
28.13
0.21
0.17
0.58
0.04
0.19
0.33
0.44
1.69
0.10
0.50
0.20
0.05
0.16
0.30
0.37
0.13
0.08
0.11
9.73
26.70
0.73
0.69
0.41
0.19
0.17
0.48
0.17
2.27
0.09
0.15
0.21
0.15
0.40
0.12
0.19
0.26
0.17
0.26
9.95
28.18
0.60
0.47
0.64
0.16
0.14
0.42
0.28
2.13
0.16
0.39
0.25
0.07
0.12
0.20
0.37
0.24
0.09
0.09
9.81
21.86
0.38
0.41
0.34
0.22
0.27
0.37
0.14
2.41
0.05
0.09
0.21
0.23
0.15
0.14
0.12
0.27
0.48
28.20
1.99
0.97
2.03
0.96
0.11
670
344
0.08
0.12
0.30
0.19
0.17
0.08
0.05
0.16
0.47
28.47
1.92
0.92
2.02
0.98
0.03
3,490
1,710
0.17
0.18
0.25
0.21
0.11
0.06
0.02
0.35
0.49
24.82
4.51
2.26
3.21
0.97
0.15
4,379
1,434
0.16
0.27
0.13
0.23
0.10
0.07
0.05
0.59
0.49
21.42
4.21
2.23
2.54
0.94
0.23
486
218
0.02
0.18
0.18
0.25
0.17
0.10
0.09
0.56
0.50
23.47
1.86
0.90
1.91
1.00
0.19
88
52
0.21
0.18
0.27
0.16
0.10
0.04
0.04
0.31
0.46
22.88
3.48
1.74
2.90
0.97
0.27
384
169
0.08
0.20
0.10
0.22
0.17
0.14
0.09
0.52
0.49
23.10
3.00
1.48
2.47
0.96
0.30
407
191
0.13
0.16
0.25
0.20
0.14
0.08
0.04
0.30
0.48
26.01
3.28
1.64
2.62
0.97
0.12
9,904
4,118
19
Table 2. Estimates of the child’s educational attainment
(1)
(2)
(3)
All
All
All
Variables
Characteristics of the child
Female
Age
Birth cohort:
(Ref: <1970)
1970-1974
1975-1979
1980-1984
1985
Number of siblings
Number of sisters
Rank within the sibship
Raised by both head and spouse till 12
Foreign born
Characteristics of the parent
Education of father
(Ref : No education)
0.361***
(12.33)
-0.025**
(2.38)
0.684
(0.92)
0.127
(0.21)
-0.373
(0.66)
-7.178***
(6.78)
-0.169***
(9.90)
0.041*
(1.91)
-0.051***
(4.22)
0.433***
(4.64)
-0.215***
(4.15)
Primary
Secondary
High school
More than high school
Education of mother
(Ref : No education)
Primary level
Secondary
High school
More than high school
0.348***
(12.27)
-0.015
(1.45)
0.594
(0.82)
0.013
(0.02)
-0.540
(0.97)
-7.507***
(7.16)
-0.089***
(5.46)
0.033*
(1.67)
-0.059***
(4.38)
0.525***
(5.77)
-0.152***
(2.87)
0.004
(0.28)
0.642
(0.61)
2.090**
(2.44)
0.079
(0.10)
-4.924***
(3.38)
-0.075***
(3.53)
0.044*
(1.69)
-0.070***
(3.78)
0.408***
(3.51)
-0.119*
(1.71)
-0.030**
(2.00)
0.199
(0.18)
-1.111
(1.20)
-0.633
(0.74)
-11.171***
(6.74)
-0.118***
(5.27)
0.023
(0.82)
-0.039**
(2.02)
0.840***
(6.07)
-0.202**
(2.47)
0.163***
(2.72)
0.318***
(4.83)
0.565***
(5.95)
1.262***
(12.20)
0.025
(0.44)
0.132**
(2.01)
0.463***
(5.07)
0.811***
(7.23)
0.151**
(2.53)
0.299***
(4.53)
0.540***
(5.69)
1.213***
(11.67)
0.022
(0.38)
0.116*
(1.77)
0.444***
(4.88)
0.762***
(6.76)
0.091***
(3.72)
0.045
(0.59)
0.078
(0.95)
-0.040
(0.34)
-0.360***
(3.05)
0.465***
(3.87)
9,904
-13,439.2
0.041
(0.54)
0.089
(1.09)
-0.032
(0.28)
-0.349***
(2.95)
0.476***
(3.96)
9,904
-13,432.2
0.123**
(2.05)
0.226***
(3.35)
0.449***
(4.63)
1.074***
(10.07)
-0.033
(0.56)
0.029
(0.43)
0.323***
(3.47)
0.615***
(5.38)
0.090***
(3.68)
0.020***
(4.88)
0.019***
(4.72)
-0.224***
(4.55)
-0.063
(1.39)
0.041
(0.95)
0.171***
(3.26)
0.209***
(4.20)
0.228***
(3.34)
0.061
(0.80)
0.058
(0.71)
-0.097
(0.84)
-0.240**
(2.02)
0.560***
(4.63)
9,904
-13,394,5
0.156**
(2.14)
0.266***
(3.23)
0.477***
(3.90)
1.126***
(8.30)
-0.047
(0.66)
-0.050
(0.60)
0.268**
(2.33)
0.696***
(4.70)
0.065**
(2.14)
0.020***
(3.92)
0.023***
(4.37)
-0.224***
(3.68)
-0.112**
(2.00)
0.076
(1.43)
0.156**
(2.40)
0.208***
(3.42)
0.168**
(1.97)
0.111
(1.12)
0.022
(0.21)
-0.161
(1.10)
-0.150
(1.01)
0.598***
(3.86)
5,147
-7,128.3
0.050
(0.63)
0.157*
(1.74)
0.408***
(3.16)
1.062***
(7.26)
-0.001
(0.01)
0.097
(1.08)
0.357***
(2.82)
0.509***
(3.30)
0.129***
(3.85)
0.016***
(2.83)
0.011**
(2.02)
-0.205***
(3.15)
-0.031
(0.51)
-0.008
(0.14)
0.203***
(2.91)
0.187***
(2.79)
0.282***
(3.04)
-0.042
(0.40)
0.093
(0.84)
0.002
(0.02)
-0.378**
(2.31)
0.607***
(3.66)
4,757
-6,367.1
Years in France when the child was 10
Difficulty in speaking
Difficulty in writing
In a large town before migration
Poor
Fair
Good
Country of origin
Southern Europe
(Ref : Northern & Eastern Europe, America)
Northern Africa
Other Africa
Middle
Asia
-0.620***
(8.38)
-0.572***
(7.19)
-0.253**
(2.14)
-0.895***
(7.44)
0.325***
(2.65)
9,904
-13,697.2
(6)
Girls
0.347***
(12.21)
-0.020*
(1.94)
0.523
(0.72)
0.077
(0.13)
-0.457
(0.82)
-7.449***
(7.11)
-0.109***
(6.64)
0.032
(1.58)
-0.029**
(2.43)
0.533***
(5.85)
-0.188***
(3.70)
Age at migration
Financial status when 16
(Ref: very poor)
(5)
Boys
0.346***
(12.18)
-0.020*
(1.90)
0.600
(0.82)
0.127
(0.21)
-0.442
(0.79)
-7.464***
(7.11)
-0.108***
(6.61)
0.031
(1.54)
-0.030***
(2.58)
0.535***
(5.86)
-0.189***
(3.71)
Household’s income (log)
Proficiency in French
(4)
All
Number of observations
Log likelihood
Source: Survey PRI 2003.
Note: Estimates from random effect ordered Probit with censoring. Absolute values of t-statistics are in parentheses and levels of significance are respectively
1% (***), 5% (**) and 10% (*). Each regression also includes a set of variables multiplying age by birth cohort.
20
Variables
Characteristics of the child
Female
Table 3. Estimates of the child’s educational attainment, with crossed effects
(1)
(2)
Age
Birth cohort:
(Ref: <1970)
1970-1974
1975-1979
1980-1984
1985
Number of siblings
Number of sisters
Rank within the sibship
Raised by both head and spouse till 12
Foreign born
Characteristics of the parent
Years of education
(3)
(4)
0.356***
(12.42)
-0.015
(1.47)
0.592
(0.81)
-0.076
(0.12)
-0.529
(0.94)
-7.408***
(7.04)
-0.097***
(5.78)
0.040**
(1.96)
-0.063***
(4.62)
0.474***
(5.18)
-0.161***
(3.02)
0.356***
(12.42)
-0.015
(1.44)
0.608
(0.83)
-0.079
(0.13)
-0.514
(0.92)
-7.421***
(7.05)
-0.097***
(5.77)
0.040**
(1.96)
-0.063***
(4.62)
0.472***
(5.16)
-0.155***
(2.91)
0.358***
(12.52)
-0.015
(1.48)
0.619
(0.85)
-0.033
(0.05)
-0.500
(0.89)
-7.338***
(6.98)
-0.102***
(6.09)
0.044**
(2.16)
-0.063***
(4.61)
0.491***
(5.37)
-0.164***
(3.07)
0.354***
(12.36)
-0.015
(1.48)
0.633
(0.87)
-0.029
(0.05)
-0.516
(0.92)
-7.451***
(7.09)
-0.097***
(5.78)
0.038*
(1.88)
-0.062***
(4.56)
0.486***
(5.32)
-0.155***
(2.90)
0.054***
(9.52)
0.093***
(5.86)
0.054***
(9.42)
Years of education in country of origin
Years of education in France
Household’s income (log)
0.146***
(5.91)
0.026***
(6.26)
0.022***
(5.23)
-0.243***
(4.88)
-0.092**
(1.96)
-0.201***
(2.66)
-0.146*
(1.79)
-0.173
(1.48)
-0.433***
(3.63)
0.497***
(4.08)
Age at migration
Years in France when the child was 10
Proficiency in French
Difficulty in speaking
Difficulty in writing
Country of origin
Southern Europe
(Ref : Northern & Eastern Europe, America)
Northern Africa
Other Africa
Middle
Asia
Country of origin * education
Southern Europe * years of education
Northern Africa * years of education
Other Africa * years of education
Middle * years of education
Asia * years of education
Country of origin * difficulty
In speaking
Southern Europe * dif. in speaking
0.139***
(5.62)
0.024***
(5.82)
0.020***
(4.79)
-0.264***
(5.28)
-0.097**
(2.09)
0.161
(0.84)
0.429**
(2.27)
0.047
(0.20)
-0.174
(0.71)
0.491*
(1.83)
-0.031*
(1.71)
-0.071***
(4.14)
-0.012
(0.55)
-0.011
(0.43)
0.009
(0.38)
0.141***
(5.71)
0.025***
(6.09)
0.021***
(4.95)
-0.396**
(2.35)
-0.083*
(1.78)
-0.225***
(2.67)
-0.261***
(2.89)
-0.143
(1.06)
0.233
(1.04)
0.745***
(4.02)
0.119
(0.67)
0.305*
(1.71)
-0.040
(0.16)
-0.739**
(2.55)
-0.285
(1.06)
9,904
-13,480.6
Northern Africa * dif. in speaking
Other Africa * dif. in speaking
Middle * dif. in speaking
Asia * dif. in speaking
0.051***
(8.83)
0.068***
(6.78)
0.145***
(5.85)
0.028***
(6.48)
0.021***
(5.12)
-0.245***
(4.91)
-0.091*
(1.95)
-0.212***
(2.80)
-0.158*
(1.94)
-0.181
(1.56)
-0.449***
(3.75)
0.482***
(3.95)
Number of observations
9,904
9,904
9,904
Log likelihood
-13,494.1
-13,474,6
-13,492.6
Source: survey PRI 2003.
Note: Estimates from random effect ordered Probit with censoring. Absolute values of t-statistics are in parentheses and levels of
significance are respectively 1% (***), 5% (**) and 10% (*). Each regression also includes a set of variables multiplying age by birth cohort,
whether parents were living in a large town before migration, and parental financial status when 16.
21