Spatial Mobility and Returns to Education:
Some Evidence from a Sample of French Youth
Philippe Lemistre∗ and Nicolas Moreau+
∗
French Center for Research on Education, Training and Employment (Cereq) and Lirhe, University of Toulouse
1. Email: [email protected]. Mail: Lirhe, Université des Sciences Sociales de Toulouse, Bât. J, Place
Anatole-France, 31042 Toulouse Cedex, France.
+
Toulouse School of Economics, GREMAQ and IZA. Email: [email protected]. Mail: Gremaq,
Université des Sciences Sociales de Toulouse, Manufacture des Tabacs, Aile J-J Laffont, 21 Allée de Brienne,
31000 Toulouse, France.
We thank Rachid Boumahdi, Marie-Benoît Magrini, Ken Ritter and participants in the International Conference
"Economics of Education: Major Contributions and Future Directions" in Dijon, France, June 20-23, 2006 for
helpful comments and suggestions. The usual disclaimers apply.
1
Spatial Mobility and Returns to Education:
Some Evidence from a Sample of French Youth
Abstract
The purpose of this article is to reevaluate the returns to geographic mobility
and to the level of education for men and women, taking into account the
interaction between these two variables, and using the distance covered as a
proxy for migration costs. We have at our disposal an original French database
that permits precise calculation of the distance between the place of education
and the location of first employment. We thus capture mobility without a priori
regarding the geographical areas selected, and we use kilometric thresholds to
estimate the returns to spatial mobility. Our specification enables us to evaluate
the returns according to the level of studies and to the amplitude of the
migration. For men, our results suggest decreasing returns to spatial mobility
as the distance covered rises and increasing returns to mobility with higher
levels of education. The returns to migration often seem positive for low levels
of qualification. While we observe slightly higher returns to schooling for
women than for men, the marginal returns to spatial mobility for women are
lower across the board. Moreover, they are strongly dependent on the selected
distance thresholds and clearly positive only for women with diplomas of
higher education.
Key words: spatial mobility, returns to schooling, earnings function
JEL classification: J31, J61, I21
2
Spatial Mobility and Returns to Education:
Some Evidence from a Sample of French Youth
1. INTRODUCTION
In this paper, we use an original French data set, the “Generation Survey 98” (Enquête
génération 98), to estimate the returns to education and to geographic mobility for a sample of
young men and women who left the French educational system in 1998. This mobility
concerns the acquisition of a first employment. The mobility in question is between the town
in which the youths complete their education and the town in which they obtain their first
employment. We look at what factors influence their education level and their spatial mobility
and the returns to these decisions.
In the literature, the returns to spatial mobility are often estimated by considering
“education” as an exogenous variable even though its endogeneity has been firmly established
– see Card (2001) for instance. The goal of this article is to reevaluate the returns to spatial
mobility and the returns to education when these variables are assumed endogenous. We use a
flexible specification that allows the returns to spatial mobility to depend on the level of
education.
For this purpose, we have at our disposition a database that allows us to collect precise
information, both on the determinants of the level of education and on migration. However,
the originality of the data resides especially in calculating the distances covered for each
decision of mobility between towns. We can thus characterize mobility without a priori
concerning the geographic areas selected by estimating the returns to spatial mobility for
different kilometric thresholds.
3
This method contrasts with studies such as Raphael and Riker (1999) on US data,
using a dummy variable for spatial mobility, and Détang-Dessendre, Drapier and Jayet (2004)
on French data, considering work migrations between administrative departments. Taking into
account the distance covered allows the evaluation of returns to geographic mobility as a
function of the costs of migration, of which the distance is a proxy. Also, since we focus on
the first job experience of individuals who all leave the educational system in 1998, we avoid
the traditional cohort effect problems and pitfalls related to work experience.1
For men, the results of our research suggest returns to spatial mobility that increase
with the level of education and that depend on the distance covered. It seems that the positive
effect of spatial mobility on earnings decreases with distance. For numerous distances, the
returns to migration are positive for low qualification levels, in the spirit of Raphaël and Riker
(1999), but contrary to the results of other studies, notably Détang-Dessendre, Drapier and
Jayet (2004) for France and Yankow (2003) for the United States.2 For women, the returns to
migration prove to be positive only for high levels of education. Moreover, the results are
sensitive to the kilometric thresholds employed.
Section 2 exposes the theoretical background. In Section 3, we present the empirical
model. The data are described in Section 4. In section 5, we present the results. We
summarize the empirical findings and discuss some of their implications. Section 6 concludes
the text.
2 ECONOMIC BACKGROUND: MOBILITY AND WAGES
2.1 A Job Search Framework
Following Détang-Dessendre and Molho (1999), we consider migration decisions as a
part of the job search process. Our presentation draws heavily upon Détang-Dessendre,
1
See Card and Lemieux (2001) for a clear account of cohort effects. See Knight and Sabot (1981) and Abraham
and Medof (1981) for problems related to job experience in earnings functions.
2
For Yankow (2003), returns to migration only appear beyond several years for high levels of education.
4
Drapier and Jayet (2004). Each agent is described by a vector X of individual characteristics,
including the attributes of the locality of origin, and looks for job opportunities characterized
by the expected value of the stream of wages over all future periods. In this setting, the worker
is engaged in a unique search process. Regardless of the location of the job offer, it has a
unique reservation value and can be compared to all job opportunities by their global value,
net of migration costs m. The global value of a job v is given by v = V(w, X), where w is the
wage. At home, the agent faces a search cost ch(X). If the agent extends his or her search to
another area, he or she entails an additional cost ce(X), so that the total search cost is cg=ch+ce.
The job searcher receives a home distribution of offers Fh(v, X) and also receives
additional offers extracted from the external distribution Fe(v, X). The global distribution of
offers is designated Fg, with Fg(v)= Fh(v)Fe(v + m). We assume Fh and Fe to be independent,
and we note that the external offers must compensate for the cost of migration. The searcher
has a global reservation value, Vg*(X), which is determined from the standard equation:
Vg*
+∞
Vg* = − cg ( X ) + β ∫ Vg*dFg ( v ) +β ∫ * vdFg ( v )
0
Vg
(1)
He or she accepts the first offer with a value, net of migration costs, that exceeds the
global reservation value:
vg ( X ) − m( X ) > Vg* ( X )
(2)
A migration occurs when the accepted offer is in an external market. Following
Nakosteen and Zimmer (1980), we let migration costs depend on X.
Under the assumption that the global value v is strictly increasing with w, the
individual will never move if the migration costs are not compensated by a higher wage. In
other words, if all individuals were identical, mobility would always be associated with higher
wages. In reality, individuals have different characteristics. They do not face the same
migration and search costs and do not receive the same internal and external distribution of
job offers. For instance, if two individuals face different internal and/or external job offer
5
distributions, ceteris paribus, one of them will gain a wage surplus from migration (better
external distribution), whereas the other will prefer to stay at home (better home distribution).3
Therefore, the aggregated distribution of migrants’ salaries does not necessarily dominate the
distribution of non-migrants’ salaries, even if contracted migration is associated with a wage
surplus on the individual level.
The problem is that neither search and migration costs nor the global reservation
values and distributions of wages offered are observable by the applied researcher, who only
imperfectly discerns the individual characteristics. Indeed, the decision to move may possibly
reflect contradictory effects of search and mobility costs that are difficult to disentangle
empirically. This problem could explain differences in results. For instance, Falaris (1988), on
a sample of young workers two years after exit from the education system, find negative
returns for two regions and non-significant returns for two others; whereas, many other
studies report non-significant returns. On the other hand, Gabriel and Schmitz (1995), using
the current population survey, displaced workers files for 1986, 1988 and 1990, and Raphaël
and Riker (1999), using data on young male workers between 1985 and 1991 from the SMSA,
show positive returns to migration on US data. Another possible explanation for the
differences in results may be the geographical areas retained in determining a migration.
2.2 Defining mobility empirically
Indeed, what is a “migration”? Plainly, the question does not have a clear empirical
answer. In this domain, the areas of mobility selected are either the result of administrative
divisions, whose logical structure is not necessarily justified from an economic standpoint, or
stem from efforts to define homogenous areas according to diverse criteria (economic
homogeneity, infrastructure endowment). Examples of the first case would be “regions” and
“departments”.
3
See Détang-Dessendre, Drapier and Jayet (2004) for a comprehensive discussion.
6
An example of the second, for France, would be “Employment zones”.4 Migration is
then defined as mobility from one of theses geographic areas to another of the same type.
Beyond problems of homogeneity within each spatial division, such conventions pose two
types of well-known problems: migrations close to the border and amplitude of the migrations
(Hunt and Mueller, 2004). To begin with, individuals situated close to the border of a
geographic area are counted as migrants even if they simply cross the borderline, perhaps
representing just a short trip from home to work for some of them. In addition, geographic
areas with the same nomenclature do not always have the same size and are evidently not all
equidistant. Hence, mobility between two small areas can be counted as a migration while
mobility within a single large geographic area will not, even though the distance covered in
the second case may well be much greater than in the first.
To solve this problem, at least partly, we make no hypothesis about the homogeneity
of geographic areas and compensate for the inconveniences linked to the amplitude of
mobility by using distances covered. Importantly, this gives a better approximation of
migration costs. In fact, two types of mobility costs, direct costs of mobility (transportation,
lodging, child care)5 and opportunity costs of mobility (earnings forgone while travelling,
searching for job, “psychic costs”, amenities)6 affect mobility choice. Important parts of these
costs will be function of the distance of migration (Sjaastad, 1962). In this paper, we
characterize mobility without a priori concerning the geographic areas selected by estimating
the returns to spatial mobility for different kilometric thresholds.
3 EMPIRICAL SPECIFICATION AND ECONOMETRIC ISSUES
4
5
6
It is similar to the “Travel to Work Area” in the United Kingdom or to the “Local Labour System” for Italy.
Each zone is defined so that the majority of the resident population has a job within its borders and that the
firms located there hire the majority of their workforce from within the same area.
The cost of housing is important in the case under study here because we are considering youth who obtain
their first job and who may therefore have to leave the lodgings of their parents.
Since people are often genuinely reluctant to leave familiar surroundings, family and friends, migration
involves a psychic cost (Sjaastad, 1962).
7
In the spirit of Détang-Dessendre, Drapier and Jayet (2004), we make the assumption
that the global value of a job is strictly increasing with the wage offered. We then specify an
empirical model that relates spatial mobility to the log of the individual’s earnings. The
distance covered is used as a proxy for mobility costs. We control for human capital
characteristics and allow returns to migration to differ according to the level of education.
Indeed, there is evidence in the literature that control over information is correlated with the
level of education (Hägerstrand, 1965) and that migration costs decrease with educational
level (Da Vanzo, 1983; Schwartz 1973, 1976). For these various reasons, we take into account
the interactions between the level of education and migration.
Also, since mobility near the border and short trips exist, we test several thresholds to
account for “relevant” spatial migration: 0 (a change between towns = a mobility), 20
kilometers,7 50 kilometers and 100 kilometers. For each threshold, mobility has a meaning
only beyond that threshold. For example, if the threshold is 20 kilometers, a youth who travels
55 km carries out a mobility of 35 km. The decision to migrate in this case is based on a
change of towns 20 km apart. For a 50 km threshold, the distance is 5 km; and for a threshold
of 100 km, the youth is considered non-mobile. Here we come back to the differences that can
exist between partition by region versus division by “Employment zones” for example. The
complete model is:
log(wage) = b0 + b1educ + b2dist + b12(educ×dist) + b22(dist)2 + b3Paris + X1b + u
educ = eWδ1 + e1
dist = Wδ2 + e2 if dist > 0 and P(dist = 0 | W) = 1 − Φ (Wβ2) otherwise
(1)
(2)
(3)
educ×dist = Wδ12 + e12 if educ×dist > 0, P(educ×dist =0|W) = 1 − Φ(Wβ12) otherwise (4)
(dist)2 = Wδ22 + e22 if y22 > 0 and P((dist)2 = 0 | W) = 1 − Φ (Wβ22) otherwise
(5)
Paris = 1I (Wδ3 + e3 > 0)
(6)
selec = 1I (Wδ4 + e4 > 0).
(7)
8
The first equation of interest is the structural equation where educ is the education
level and dist the distance covered between the student’s place of residence at the end of his
studies and the site of his first employment. The distance in the question is the distance
beyond the threshold of mobility (which is equal to 0, 20, 50 or 100 km, depending on the
estimation). We call this measure “spatial mobility” in what follows. Additionally, Paris is a
dummy variable for living in the Paris region. All else being equal, the Paris/province
partition clearly determines the most significant differences of salary in France. Finally, X1 is
a vector that includes socioeconomic variables.
This specification permits flexible responses of wage to spatial mobility. In addition,
the interaction term (educ×dist) accounts for different returns from migration according to the
level of education. Ideally, we should also include a quadratic term in years of schooling to
account for a possible non-monotonic relationship between education and wages. However,
education and its square are nearly collinear in our data.
As discussed in the literature, earnings equations suffer from heterogeneity and
selection bias. Heterogeneity is usually associated with individual ability and motivation that
are likely correlated with education and spatial migration. Endogeneity of living in the Paris
region is another potential problem. Finally, we observe the wage rate only for those who
choose to work, which may induce a selectivity bias. Therefore, we have chosen to instrument
the education level (equation (2)), the spatial migration (equation (3)), its interaction with
education level (equation (4)), its square (equation (5)) and the Paris region (equation (6)).
Equation (7) is the selection equation. We allow arbitrary correlation among u, e1, e2, e12, e22,
e3 and e4. The matrix W includes the exogenous regressors X1 as well as the matrix X2 of
excluded instruments. This matrix is described below.
7
It corresponds to the high end of the bracket for distances from home to work.
9
We estimate years of schooling with a Poisson regression model (Poisson QMLE)
since they only take on non-negative integer values.8 We use binary response models for the
Paris region and the selection equation.
Spatial mobility, its square and its interaction with education are estimated with a twostage model. It is partly continuous with many observations at 0. In addition, unlike the
standard Tobit model, this specification allows the initial decision of dist > 0 versus dist = 0
to be separate from the decision of how much dist given that dist > 0.
To estimate our model, we use a two-step procedure. First, the six reduced forms are
estimated separately; then the wage equation is estimated using OLS on:
log(wage) = b0 + b1educ + b2dist + b12(educ×dist) + b22(dist)2 + b3Paris + X1b +
(8)
γ1 vˆ educ + γ2 vˆ dist + γ12 vˆ educ×dist + γ22 vˆ dist² + γ3 vˆ Paris + γ4 vˆ selec + ε,
where ε is a random term that represents the unobserved heterogeneity. The vˆ are the
residuals from reduced forms to control for the potential endogeneity of years of schooling
( vˆ educ), distance to work ( vˆ dist, a generalized residual from a two-part model), the interaction
between distance to work and education ( vˆ educ×dist, another generalized residual from a twopart model), the square of distance to work ( vˆ dist², a last generalized residual from a two-part
model), the Paris region ( vˆ Paris, an inverse Mill’s ratio) and selection ( vˆ selec, another inverse
Mill’s ratio). It also provides a direct test of exogeneity by means of the t−statistics of the
estimates of γ; see Blundell, Duncan and Meghir (1998) for an application of this control
function approach.9
The asymptotic covariance matrix of the second step estimator is computed using the
results of Newey (1984) and Newey and McFadden (1994) to take into account that we are
8
9
We also estimate the level of education with a negative binomial model (QMLE). The results are similar.
They are available upon request.
In our context, this approach has the advantage over the more traditional IV method to account for a proper
treatment of mass points at zero, discrete variables and dummies. As mentioned in Browning (1992), the
usual practice of treating this type of variables as unbounded and continuous in the auxiliary equation may
10
conditioning on generated regressors (that is, vˆ educ, vˆ dist, vˆ educ×dist, vˆ dist², vˆ Paris and vˆ selec). This
estimator is robust to heteroskedasticity of unknown form.
At this stage, the selection of the instruments that appear in the auxiliary regressions
(2) to (7) requires some discussion. In the literature, differences of education levels are often
attributable to borrowing constraints. Two types of schooling costs affect the borrowing
constraints: opportunity costs of schooling (the value of earnings forgone while in school) and
direct costs of schooling (monetary cost of tuition, books, transportation, and board and room
if necessary)10. Carneiro and Heckman (2002) distinguish between two types of borrowing
constraints, one in the short term and one in the long term. The first concerns financing
capacity at a given moment (that is, at the date of the survey) and plays a limited role. The
second, the long-term constraint that weighs on the entire duration of schooling (and perhaps
at the entry into the labor market), reflects notably family background. In this perspective, the
differences in education levels for youth with identical, ex-ante aptitudes are primarily
attributed to differences in family background and family income. Thus for many studies,
family background and family income are likely to influence the level of education.
Mobility is also not independent of family background and income, which influences
the costs of migration. In addition, migration behaviour remains highly associated with
territorial characteristics. For example, information circulates more easily and in greater
quantity in urban markets than in rural markets, which reduces the cost of information
research. Spatial mobility therefore remains particularly constrained by the characteristics of
the territory. Another reason for taking into account the characteristics of territory is the
existence of amenities that are likely to compensate for unfavourable salary differences by
increasing the opportunity cost of migration (Hunt, 1993; Gabriel and Schmitz, 1995). Lastly,
cause problems. We may be predicting values outside the appropriate interval that may in turn affect the
estimates of the parameter of interest.
11
the economic situation of the territory and the characteristics of the local labor market; that is,
push and pull factors, are as many potential determinants of special mobility (Greenwood
1997).
Along these lines, with respect to family background and income variables, we include
as instruments a set of variables that represent the youths’ social origin by indicators of their
fathers’ professions and their parents’ nationalities. More innovative is that we account for the
parents’ occupational status at the time when the youth leaves the education system in 1998
(father or mother unemployed, father retired, mother in public sector). With respect to
territorial characteristics, amenities, and push and pull factors, we use as supplementary
instruments a set of variables that characterize the Employment zone of completion of
education. It includes the employment density, the unemployment rate, residence in a rural
area at the end of schooling, the log of the population, the share of the industrial sector in total
employment, the percentages of youth (population from 20 to 39 years old / population over
40), the share of qualified jobs, the share of students (students / total population), as well as
the relative net migratory balance (net migratory balance / total population). These variables
are used to measure the attractiveness, economic dynamism and competitiveness of hiring in
the home zone. Adamson, Clark and Partridge (2004), Audas and Dolton (1998) and
Greenwood, Hunt, Rickman and Treyz (1991), among others, use similar indicators.
However, some of these potential instruments are probably important determinants of
wages. All variables that are thought to influence both the educational decision, the distance
covered and wage outcomes must be included as regressors. As a matter of fact, some family
background factors (father’s profession: farmer, technician, parents’ nationalities) and some
regional characteristics (employment density, unemployment rate, share of the industrial
sector in total employment, share of qualified jobs, student population percentage, youth
10
See Card (1999, 2001) for a survey on effects of family background. Card (1995), then Cameron and Taber
(2004) capture the borrowing constraints, also by a proxy for the proximity of the learning establishment. We
12
percentage) appear to be significant determinants of male wages in our estimation. Therefore,
we decided to control for these variables in the male wage equation. All in all, there are 11
instruments excluded from the male wage equation – residence in a rural area at the end of
schooling, relative net migratory balance, log of population, parents’ occupational status (4
dummies) and father’s profession (4 dummies).
For women, the relative net migratory balance, the share of the industrial sector in total
employment, the percentage of students, the percentage of youth and the unemployment rate
appear to be significant determinants of wages. We therefore include these variables in the
female wage equation. As a consequence, there are 17 instruments excluded from the female
wage equation – employment density, log of population, share of qualified jobs, parents’
occupational status (4 dummies), father’s profession (6 dummies) and parents’ nationalities (4
dummies).
4 DATA
4.1 Sample selection
We have exploited data in the Céreq’s “Generation 98” survey in which 55,000 youths
who left the French educational system with an initial education in 1998 are observed over a
three-year period. They are representative of the whole generation of those leaving (700,000).
The sample retained in this study contains only young men and women who obtained a first
full-time job, which is 43,943 youths (24,587 men, 19,356 women). The restriction of the
scope of investigation to workers with a full-time job is due to the availability of monthly
salary data alone, without variables sufficiently detailed to take into account part-time work.
The fact that we have at our disposal a single generation of youth leaving the
educational system allows us to avoid several problems encountered in the estimation of the
gains functions. The problems linked to the inverse correlation between the duration of
schooling and experience in the labor market do not exist here: all individuals in the sample
do not have this variable in our data base.
13
enter the labor market at the same moment (1998).11 Furthermore, cohort effects are limited, a
single generation of labor-market entrants being present,12 and these cohort effects are not
neutral with respect to the returns to education (Harmon and Walker, 1995, Card and
Lemieux, 2001).
Lastly, our choice of restricting analysis to starting salaries when first hired is linked to
our intention of not making the analysis more complex than need be. If we were to take into
account the subsequent job history, the variables that characterize it would probably also be
endogenous, which would require a complex, dynamic model.
4.2 Mobility and education level variables
The distances covered between the towns are calculated “as the crow flies” between
the centers of the towns of departure and the towns of arrival. In a pair (x,y) representing the
geographic coordinates of a set of points, the distance between point A (departure) and B
(arrival) is given by the equation d ( A, B) = ( x B − x A )² + ( y B − y A )² . Departure is the place
of completion of education (1998) and arrival the place of the first job.
Now, many students attend school in one region (say 150 km from their birthplace),
but return “home” after they leave school. We argue that these people have merely returned to
their true place of origin and must not be considered as geographically mobile.13 More
precisely, we consider that every individual who finds a job within a radius of x km around
the town where he or she attended junior high school should be considered as non-mobile. For
the sake of consistency, the value of x corresponds to the threshold used to account for spatial
migration. For instance, an individual who works 40 km away from his or her place of origin
11
Even if they do not find a job immediately, they have still entered the labor market.
Youth who left the educational system in 1998 were evidently not all born in the same year. However,
dispersion in terms of age rarely exceeds 5 years, and birth cohorts usually concern a minimum of 5
consecutive years.
13
We thank an anonymous referee who pointed this out.
12
14
and 150 km away from his or her place of study will be consider as mobile for the thresholds
0 and 20, and non-mobile for the thresholds 50 and 100.14
The variable that characterizes the level of education is the highest level attained and
not the number of years of study. In fact, the latter reflects rather poorly the amount of
accumulated human capital since it often corresponds to a period of study superior to the
length time necessary to obtain the degree. For instance, for 15 years of study (counted
beyond the age of 6), which in theory corresponds to the French Baccalauréat + 2 years, only
50% of youth have reached this level, while 26% have the level of the Baccalauréat, (“Bac”
hereafter). The level of the remaining students is still lower. Moreover, less than 30% of youth
reach the level of the Bac after 13 years of study. The remaining students have a lower level,
even no qualification at all, after retaking numerous academic years.
The variable we construct takes on a limited number of values, corresponding to
different levels of schooling: without qualification, the first level of professional training,15
the Bac, Bac+1, Bac+2, Bac+3, Bac+4, Bac+5, >Bac+5.
Table (1) exhibits some descriptive statistics of the sample. One woman (or man) out
of four has completed at most 13 (12) years of study (P25). Among women (or men), 10%
earn at least 1599 (1829) euros per month (P90). About 71% (72%) of women (or men) work
in a town different from the one in which they finished their studies. Among them, for 50% of
women (or men), the distance between the two towns remains limited – less than 23 (20)
kilometers (P50), and the place of work is close to the place of origin. Among the 38% (37%)
of women (or men) not working in the place of their studies at the 20 km threshold, 71%
(74%) are considered to be actually mobile and the others to be returning to work in the
proximity of their place of origin. Let us now turn to the results.
TABLE (1) HERE
14
To check the robustness of our results, we have also estimated the model with different constant radiuses
around the place of origin. The results are very similar.
15
5 RESULTS
First, for all samples and all thresholds, the exogeneity of living in the Paris region
was not rejected by the data. The coefficient γ3 in equation (8) never appeared to be
statistically different from 0 at conventional levels. Therefore, we only present estimates of
the complete model with living in the Paris region assumed to be exogenous. Before we
present any further results, we report in Table (2) statistics that test the validity of the
excluded instruments in all the auxiliary regressions for all thresholds. Under the null
hypothesis, none of them is significant. The corresponding p-values are close to zero,
indicating that it is clearly rejected. These results provide evidence that the instruments are
significant for all the endogenous variables.
TABLE (2) HERE
The estimates of the auxiliary regressions are shown in Appendix A. To save space,
we only report the results at the 20 km threshold, for the education level (equation 2) and the
distance covered (equation 3).16 The variables generally have the expected sign. Concerning
the level of education, for example, the parents’ professions have the effects observed in
numerous studies: the more qualified the father or the mother, the higher the children’s
educational attainment. Also, having a foreign father or mother lowers the level of studies
attained. If either the father or the mother is unemployed at the time the child leaves the
educational system, the latter will tend to do shorter studies. It is clear that the short-term
borrowing constraints weigh heavily here since the child cannot compensate for his parents’
reduced income by a loan.
Concerning the migration, the fact of residing in a rural area while studying leads to
more migration as compared to urban residence, but leads to migrations of lesser amplitude.
Inversely, the greater the employment density, the less probable is the migration.
15
16
CAP-BEP: Certificat d’aptitude professionnel and Brevet d’études professionnelles.
Results of the other auxiliary regressions are available upon request.
16
Nevertheless, once a migration has been undertaken, its distance covered is greater. The effect
of the population size is similar. The effects of variables characterizing the parents’
professions while studying are similar to those found for the level of studies. Next, we discuss
the results of the wage equation.
5.2 Wage equation
The results for the estimation of the wage equation at the different thresholds of
mobility are presented in Table (3), where for brevity, only the key results are reported.
TABLE (3) HERE
For both men and women, the coefficient of the residual of the level of studies ( vˆ educ)
is significantly different from 0. The exogeneity of the level of studies is therefore rejected.
Also, there is evidence of selection bias. It is more pronounced for women. On the other hand,
the residuals corresponding to the distance variables do not have an impact at conventional
levels. This comes from the strong correlation between these residuals, leading to large
estimated standard errors. However, a joint test leads us to reject the null hypothesis H0: γ 2= γ
12=
γ 22=0 at conventional levels, whatever the estimation. This result provides evidence that
the distance covered is endogenous to the earnings function.
We see that living in the Paris region has a positive effect on wages. It mainly reflects
the higher standard of living in the Paris area. Note, however that this effect is stronger for
men than for women. Also, living as a couple has a positive effect on wages for both men and
women. This is consistent with the results reported in Gabriel and Schmitz (1995) and many
other studies. Nevertheless, the nationality variables do not have the generally observed effect.
Ceteris paribus, for men, to have parents of African origin increases salary. A similar effect
has been observed for France by Boumahdi and Giret (2005).17
We observe that the level of studies, which does not take into account the number of
non-validated years of study, has high returns. However, we do not have at our disposal
17
variables characterizing intrinsic individual aptitudes. Several studies have shown that taking
into account such variables limits the augmentation of returns to schooling when the
education level is instrumented (Eckstein and Wolpin, 1999, Heckman and Carneiro, 2002). It
is therefore probable that the returns to schooling are overestimated here.
Be that as it may, we observe slightly higher returns to schooling for women. A similar
finding appears in Havet (2006) for a different sample of French youths. Note too that for
men, the returns to schooling increase with the distance covered – the coefficient of
(educ×dist) being positive. For women, the effect of the distance covered on the returns to
schooling is less clear-cut. The coefficients are lower in magnitude than for men and are only
significant at the 20 km, 50 km and 100 km thresholds.
As for the effect of migration on the salary, we see that, for men, the coefficients are
positive for the interaction term (educ×dist) between the level of studies and the distance
covered on the one hand, and negative for both the distance term (dist) and the quadratic term
(dist2) on the other. Consequently, the returns to spatial mobility increase with the level of
studies but diminish with the distance covered. For women, the distance term is negative,
though not always significant. The quadratic term is positive and significant at the 0 km
threshold, and not significant otherwise. Overall, the estimated coefficients related to spatial
migration in the female earnings equation are sensitive to the thresholds used to account for
mobility.
TABLES (4) AND (5) HERE
Tables (4) and (5) present marginal returns to spatial migration, for men and women
respectively. We calculate the marginal returns for each of four levels of education: the lowest
(without qualification), then the Bac, Bac+2, and Bac+5, and four distances: 20, 100, 200 and
400 kilometers. As an example, at the 20 km threshold, for a distance covered of 100 km and
17
Recall that the nationality variables have no significant effect on salaries of women.
18
for a male (or female) youth with an education level of Bac+5, the marginal returns to spatial
mobility are 0.062% (0.024%).
For men without qualification, returns are generally not statistically different from 0,
excepting long distances for which the returns are negative. Again, for men without
qualification, returns are positive for short distances at the 0 and 100 km thresholds. At higher
levels of schooling, spatial returns to migration are positive excepting long distances for
which the returns are sometimes non-significant. For women, marginal returns to spatial
migration are negative at all education levels at the 0 km threshold, except for the 400 km
distance for which the returns are not significant. At other thresholds, the returns only appear
to be positive for women with diplomas of higher education. Even when they are positive, the
returns to migration appear to be smaller for women than for men. One possible explanation is
that women are captive movers to a greater extent than men, the argument being that men
move for labor-market reasons, whereas their wives follow. As an example, in France, family
arbitration seems to be unfavorable for women with low qualifications during a migration
(Courgeau and Meron, 1995). This hypothesis is partly corroborated by our data. Indeed, the
difference in returns between men and women diminishes when we estimate the model only
on unmarried women, without vanishing for as much.18 Moreover, returns to migration remain
negative at the 0 km threshold for unmarried women and remain positive at other thresholds
only for unmarried women with diplomas of higher education.
While the marginal returns reported in Tables (4) and (5) are weak, it is important to
keep in mind that they apply to an increase in distance of only one kilometer. The returns to
the whole distance are of course greater. They are presented in Tables (6) and (7) for three
distances and four education levels. Other covariates are at the sample median. For a male
youth with the Bac, at the mobility threshold of 20 km for example, migrating 20 km
increases the salary by 0.6% in comparison to the salary of a youth who does not migrate. For
19
a distance covered of 100 km, the salary increase is 1.5%. It is 3.1% for a migration distance
of 200 km. For a male youth with the Bac+5 level, these increases are respectively 1.4%,
7.2% and 15%.
TABLES (6) AND (7) HERE
A threshold effect manifestly exists. It is obvious for women since total returns are
negative at the 0 km threshold and positive for the other thresholds, but it is also true for men.
Let us consider a male youth who moves by 120 km for example. At the 20 km threshold, the
total return from the kilometers covered (100 km beyond the limit) is 7.2% for a Bac+5
education level. At mobility threshold of 100 km, this return is only 1.9% for the 20 km
distance covered beyond the threshold. For women, these values are respectively 2.3%
(threshold of 20 km, distance of 100 km) and 1.5% (threshold of 100 km, distance of 20 km).
Our results suggest that choosing a geographic area of great size may lead to an
underestimation of returns to the total distance of the migration.
Whatever the level of education, the total returns to distance covered appear positive
and statistically significant for men. In other words, taking into account the distances leads us
to nuance the results obtained on French data by Détang-Dessendre, Drapier and Jayet (2004),
who concluded that positive returns to geographic mobility for men existed exclusively at the
highest levels of education.
6 CONCLUSION
Taking into account distances covered and interactions between the level of education
and the first geographical migration leads to several original conclusions.
We have captured spatial mobility without a priori regarding geographical areas
selected and used different kilometric thresholds to characterize this migration. It appears that
returns to spatial mobility depend on the threshold selected. The total returns are in general
weaker at high thresholds.
18
Results are available upon request.
20
Also, taking into account the distance covered between the dwelling place at the end of
studies and the site of the first employment widens the range of what returns to spatial
mobility can be. Most earlier research in fact makes use of a simple indicator to measure
returns from migration. For men, our results suggest decreasing marginal returns to spatial
mobility with distance covered that increases with the level of education. This result is not in
contradiction with the theory of employment search, or with the hypothesis of a link between
the level of human capital and the propensity to migrate. For women, marginal returns to
spatial mobility seem to be positive only for those with diplomas of higher education.
This study is limited to the first migration on departure from the educational system.
Further investigations could complement this analysis, focusing in particular on migration
during careers on the labor market.
References
Abraham K.G. and J.L. Medoff, 1981. Are Those Paid More Really More Productive? The
Case of Experience Journal of Human Resources 16, 186-216.
Adamson, Dwight W.; Clark, David E.; Partridge, Mark D., 2004. Do Urban Agglomeration
Effects and Household Amenities Have a Skill Bias? Journal of Regional Science May 2004,
vol. 44 (2), 201-223.
Audas R. and Dolton P., 1998. The Effect of Local Labor Market Conditions on the Decisions
to Migrate among the UK Graduates, European Regional Science Association Conference
Vienne.
Blundell R., A. Duncan and C. Meghir, 1998. Estimating Labor Supply Responses Using Tax
Reforms Econometrica, vol. 66, 827-61.
Boumahdi, R. and J.-F. Giret. 2005. Une analyse économétrique des disparités d’accès à
l’emploi et de rémunérations entre jeunes d’origine française et jeunes issus de l’immigration
Revue Economique, May 2005, v. 56, v. 3, 625-36.
Browning M., 1992. Children and Household Economic Behavior Journal of Economic
Literature 30, v.3, 1434-1475.
21
Cameron S. and C. Taber, 2004. Estimation of Educational Borrowing Constraints Using
Returns to Schooling Journal of Political Economy, n° 112, 132-182.
Card D., 2001. Estimating the Return to Schooling: Progress on Some Persistent Econometric
Problems Econometrica 69, 1127-1160.
Card D., 1999. The causal Effect of Education on Earnings. In: O. Ashenfelter and D. Card
(Eds), Handbook of Labor Economics, vol. 3a, North-Holland, Amsterdam, 1801-1864.
Card, D., 1995. Using geographic variation in college proximity to estimate the return to
schooling, Aspects of labour market behaviour: essays in honour of John Vanderkamp,
University of Toronto Press, Toronto, 201-222.
Card D. and T. Lemieux, 2001. Education, Earnings, and the Canadian G. I. Bill Canadian
Journal of Economics 34, 313-344.
Carneiro P. and J.-J Heckman, 2002. The Evidence on Credit Constraints in Post-secondary
Schooling Economic Journal 112, 705-734.
Courgeau D. and M. Meron, 1995. Mobilité résidentielle, activité et vie familiale des couples
Economie et statistique 290: 17-31.
Da vanzo J., 1983, Repeat Migration in the United States: Who Moves Back and Who Moves
on? Review of Economics and Statistics, vol. 65 (4), 552-559.
Détang-Dessendre C., C. Drapier and H. Jayet, 2004. The Impact of Migration on Wages:
Empirical Evidence from French Youth Journal of Regional Science, vol. 44, 661-691.
Détang-Dessendre C., and I. Molho, 1999, Migration and Changing Employment Status: a
Hazard Function Analysis Journal of Regional Science, vol. 39, 103-123.
Eckstein Z. and K. Wolpin, 1999. Why youth drop out of high school: the impact of
preferences, opportunities, and abilities Econometrica 67, 1295-1340.
Falaris M., 1988. Migration and wage of young men Journal of Human Resources 23, 514534.
Gabriel, P.E. and S. Schmitz, 1995. Favorable Self-Selection and the Internal Migration of
Young White Males in the United States Journal of Human Resources 30, 460-471.
Greenwood M.J., 1997. Internal Migration in Developed Countries. In: M. Rosenzweig and
O. Stark (Eds), Handbook of Population and Family Economics, v. 1B, Elsevier Science, 647720.
Greenwood M.J., G. Hunt, D. Rickman and G. Treyz, 1991. Migration, Regional Equilibrium,
and the Estimation of Compensating Differentials American Economic Review, vol. 81,
pp.1382-1390.
22
Hagerstrand T., 1965. Aspects of the spatial structure of social communication and the
diffusion of information Papers of the Regional Science Association, vol. 16, 27-42.
Harmon C. and I. Walker, 1995. Estimates of the economic return to schooling for the United
Kingdom American Economic Review 85, 1278-1286.
Havet N., 2006. Youth employement and public training schemes: male/female differences in
transitions Annales d'Economie et de Statistique, January-March 2006, vol. 81, 225-51.
Hunt G. and R.E. Mueller, 2004. North American Migration: Returns to Skill, Border Effects,
and Mobility Costs Review of Economics and Statistics 86, 988-1007.
Hunt G., 1993. Equilibrium and Disequilibrium in Migration Modelling Regional Studies 27,
341-349.
Knight J. B. and R.H. Sabot, 1981. The Returns to Education: Increasing with Experience or
Decreasing with Expansion? Oxford Bulletin of Economics and Statistics 43, 51-71.
McFadden D. and W. K. Newey, 1994. Large Sample Estimation and Hypothesis Testing. In:
R. F. Engle and D.L. McFadden (Eds.), Handbook of econometrics, vol. 4, Elsevier Science
B.V., 2111-2245.
Nakosteen R.A. and M.A. Zimmer, 1980. Migration and Income: The Question of SelfSelection Southern Economic Journal, vol. 46, 840-851.
Newey W.K., 1984. A Method of Moments Interpretation of Sequential Estimators
Economics Letters 14, 201-206.
Raphaël S. and D.A Riker, 1999. Geographic Mobility, Race, and Wage Differentials Journal
of Urban Economics 45, 17-46.
Sjaastad A. L., 1962. The Costs and Returns of Human Migration Journal of Political
Economy 70, 80-93.
Scharwz A., 1973, Interpreting the Effect of Distance on Migration The Journal of Political
Economy, vol. 81 (5), 1153-1169.
Scharwz A., 1976, Migration, Age and Education The Journal of Political Economy, vol. 84
(4), 701-719.
Yankow J., 2003. Migration, Job change, and wage growth: a new perspective on the
pecuniary return to geographic mobility Journal of Regional Science, vol. 43 (3), 486-516.
23
Table 1: Descriptive Statistics of the Sample
Men
Live in Paris region
Live in Couple
Migrants
Spatial Migration when > 0 (in kms)
Distance from the place of origin
Education Level
Age
Monthly Wage (in euros)
Women
Live in Paris region
Live in Couple
Migrants
Spatial Migration when > 0 (in kms)
Distance from the place of origin
Education Level
Age
Monthly Wage (in euros)
17.5%
26.2%
Threshold= 0
72% - 74%
Threshold= 20
37% - 74%
Threshold= 50
23% - 71%
Threshold= 100
17% -74%
P10
5
0
12
18
809
P25
9
4
12
19
915
P50
20
14
14
21
1067
P75
87
64
15
23
1296
17.8%
49%
Threshold= 0
71% - 76%
P10
5
0
12
19
809
Threshold= 20
38% - 71%
P25
9
3
13
20
884
Threshold= 50
35% - 65%
P50
23
15
15
22
1047
Threshold= 100
17% - 67%
P75
P90
92
344
65
320
17
18
24
26
1296
1599
P90
359
338
18
26
1829
Table 2: Reduced form equations - Validity of the instruments
Threshold=0
Threshold=20
Threshold=50
Threshold=100
Test
p-value
test
p-value
test
p-value
test
p-value
test
p-value
H0: δ2i=0
Men:
educ
2418
0
dist
711
0
436
0
298
0
265
0
902
0
520
0
332
0
294
0
educ×dist
968
0
492
0
309
0
268
0
(dist)2
selec
93
0
Women
Women: educ
2800
0
dist
875
0
625
0
430
0
312
0
875
0
675
0
458
0
340
0
educ×dist
831
0
676
0
457
0
320
0
(dist)2
selec
165
0
Notes: δ2i is the vector of parameters corresponding to the excluded instruments X2 in equation i, for i=2,3,4,5
and 7. LR tests are used for spatial migration and its related variables, and selection. A score statistic is used for
years of schooling.
24
Table 3: Log wage equation
Threshold of mobility
Gender
Intercept
Education level: educ
Distance: dist
educ×dist
dist²×1000
Paris region
Living in couple
Parents’ origin : Africa
(ref. France)
vˆ educ
vˆ dist
vˆ educ×dist
vˆ dist²
vˆ selec
Adjusted R2
0
Men
5.499***
(0.044)
0.093***
(0.003)
-0.027**
(0.014)
0.049***
(0.010)
-0.041**
(0.018)
0.050***
(0.009)
0.019***
(0.007)
0.046***
(0.010)
-0.024***
(0.003)
0.060
(0.120)
-0.098
(0.095)
0.068
(0.173)
-0.094*
(0.055)
0.367
20
Women
5.221***
(0.074)
0.111***
(0.004)
-0.044**
(0.019)
0.013
(0.013)
0.037***
(0.013)
0.090***
(0.009)
0.016***
(0.004)
--0.042***
(0.004)
0.176
(0.163)
0.212*
(0.113)
-0.631***
(0.135)
0.088**
(0.039)
0.305
Men
5.517***
(0.044)
0.092***
(0.003)
-0.070***
(0.020)
0.078***
(0.014)
-0.042
(0.030)
0.047***
(0.009)
0.019***
(0.007)
0.046***
(0.010)
-0.022***
(0.003)
0.551**
(0.216)
-0.440***
(0.142)
0.060
(0.308)
-0.106*
(0.055)
0.368
50
Women
5.222***
(0.075)
0.109***
(0.004)
-0.071**
(0.031)
0.052***
(0.020)
0.006
(0.018)
0.082***
(0.009)
0.016***
(0.004)
--0.040***
(0.004)
0.495
(0.328)
-0.260
(0.211)
-0.269
(0.186)
0.110***
(0.039)
0.304
Men
5.509***
(0.043)
0.093***
(0.003)
-0.067***
(0.021)
0.085***
(0.012)
-0.077***
(0.029)
0.047***
(0.009)
0.018***
(0.007)
0.046***
(0.010)
-0.023***
(0.003)
0.549**
(0.242)
-0.580***
(0.140)
0.471
(0.309)
-0.109**
(0.055)
0.368
100
Women
5.200***
(0.074)
0.110***
(0.004)
-0.033
(0.030)
0.034*
(0.019)
-0.010
(0.025)
0.083***
(0.009)
0.016***
(0.004)
--0.041***
(0.004)
-0.075
(0.331)
-0.042
(0.212)
0.050
(0.264)
0.114***
(0.039)
0.304
Men
5.493***
(0.043)
0.094***
(0.003)
-0.053**
(0.024)
0.083***
(0.013)
-0.099***
(0.035)
0.049***
(0.009)
0.018***
(0.007)
0.046***
(0.010)
-0.023***
(0.003)
0.222
(0.285)
-0.550***
(0.157)
0.865**
(0.373)
-0.107*
(0.055)
0.367
Women
5.201***
(0.074)
0.110***
(0.004)
-0.031
(0.033)
0.041**
(0.021)
-0.042
(0.032)
0.084***
(0.009)
0.016***
(0.004)
--0.041***
(0.004)
-0.181
(0.381)
-0.110
(0.227)
0.485
(0.358)
0.111***
(0.039)
0.304
Notes: Asymptotic standard errors in parentheses. Significance levels of 10, 5 and 1% are noted *, ** and *** respectively.
Other regressors are father’s profession, parents’ nationalities, employment density, unemployment rate, share of the industrial sector in total employment, share of qualified jobs, student
population percentage and youth percentage in the male wage equation, and relative net migratory balance, share of the industrial sector in total employment, percentage of students, percentage of
youth and unemployment rate in the female wage equation.
25
Table 4: Men’s marginal returns to spatial mobility
Threshold of mobility
Distance
20
Marginal returns (%)
without qualification 0.020***
(0.007)
Bac
0.035***
(0.007)
Bac+2
0.045***
(0.008)
Bac+5
0.059***
(0.010)
Threshold of mobility
Distance
20
Marginal returns (%)
without qualification
0.015
(0.012)
0.041***
Bac
(0.011)
0.058***
Bac+2
(0.011)
0.083***
Bac+5
(0.012)
0
100
20
200
400
20
100
200
400
0.014**
0.005
-0.011
0.006
-0.001
-0.009
-0.026*
(0.006)
(0.005)
(0.010)
(0.011)
(0.008)
(0.007)
(0.015)
0.028*** 0.020***
0.004
0.029*** 0.023*** 0.014*** -0.002
(0.005)
(0.003)
(0.007)
(0.011)
(0.007)
(0.004)
(0.013)
0.038*** 0.030*** 0.013** 0.045*** 0.038*** 0.030***
0.013
(0.005)
(0.003)
(0.006)
(0.012)
(0.007)
(0.003)
(0.012)
0.053*** 0.045*** 0.028*** 0.069*** 0.062*** 0.053*** 0.037***
(0.007)
(0.005)
(0.006)
(0.014)
(0.009)
(0.005)
(0.011)
50
100
100
200
400
20
100
200
400
0.003
-0.013* -0.044*** 0.026*
0.010
-0.010 -0.049***
(0.009)
(0.007)
(0.014)
(0.014)
(0.009)
(0.007)
(0.016)
0.028*** 0.013*** -0.018 0.051*** 0.035*** 0.015*** -0.024
(0.007)
(0.003)
(0.012)
(0.013)
(0.008)
(0.003)
(0.015)
0.045*** 0.030*** -0.001 0.068*** 0.052*** 0.032*** -0.008
(0.007)
(0.002)
(0.012)
(0.013)
(0.007)
(0.002)
(0.015)
0.071*** 0.056*** 0.025** 0.093*** 0.077*** 0.057***
0.018
(0.008)
(0.004)
(0.012)
(0.013)
(0.008)
(0.005)
(0.015)
Notes : Asymptotic standard errors in parentheses are computed with the Delta method. Significance levels of 10, 5 and
1% are noted *, ** and *** respectively.
Table 5: Women’s marginal returns to spatial mobility
Threshold of mobility
0
20
Distance
20
100
200
400
20
100
200
400
Marginal returns (%)
without qualification -0.029*** -0.023*** -0.016*** -0.001 -0.019
-0.018
-0.016
-0.014
(0.008)
(0.007)
(0.006) (0.008) (0.012)
(0.011)
(0.012)
(0.015)
Bac
-0.025*** -0.020*** -0.012*** 0.003
-0.003
-0.002
-0.001
0.002
(0.007)
(0.005)
(0.004) (0.005) (0.008)
(0.007)
(0.006)
(0.010)
Bac+2
-0.023*** -0.017*** -0.009*** 0.005
0.007
0.008 0.010*** 0.012*
(0.007)
(0.005)
(0.004) (0.005) (0.008)
(0.005)
(0.003)
(0.007)
Bac+5
-0.019**
-0.013*
-0.005
0.009 0.023** 0.024*** 0.025*** 0.028***
(0.009)
(0.007)
(0.006) (0.007) (0.010)
(0.008)
(0.006)
(0.007)
Threshold of mobility
50
100
Distance
20
100
200
400
20
100
200
400
Marginal returns (%)
without qualification
0.001
-0.001
-0.003
-0.007
0.008
0.002
-0.007
-0.024
(0.013)
(0.011)
(0.011) (0.016) (0.016)
(0.013)
(0.011)
(0.018)
0.011
0.009
0.007
0.004
0.020
0.014
0.005
-0.012
Bac
(0.010)
(0.007)
(0.005) (0.011) (0.013)
(0.009)
(0.006)
(0.014)
0.018*
0.016*** 0.014*** 0.010
0.028** 0.022*** 0.013*** -0.004
Bac+2
(0.010)
(0.006)
(0.003) (0.010) (0.012)
(0.007)
(0.003)
(0.013)
0.028**
0.026*** 0.024*** 0.020** 0.040*** 0.034*** 0.025*** 0.009
Bac+5
(0.012)
(0.009)
(0.006) (0.010) (0.014)
(0.009)
(0.006)
(0.014)
Notes : Asymptotic standard errors in parentheses are computed with the Delta method. Significance levels of 10, 5 and
1% are noted *, ** and *** respectively.
26
Table 6: Men’s total returns to spatial mobility
Threshold of mobility
Distance
20
Total returns (%)
without qualification 0.422***
(0.001)
Bac
0.718***
(0.001)
Bac+2
0.916***
(0.001)
Bac+5
1.214***
(0.001)
Threshold of mobility
Distance
20
Total returns (%)
without qualification 0.335***
(0.003)
0.850***
Bac
(0.003)
1.194***
Bac+2
(0.003)
1.713***
Bac+5
(0.003)
0
100
20
100
200
20
0.995***
(0.004)
1.742***
(0.004)
2.242***
(0.004)
5.867***
(0.004)
50
100
1.791***
(0.007)
3.301***
(0.007)
4.320***
(0.007)
11.158***
(0.007)
0.154***
(0.002)
0.624***
(0.002)
0.939***
(0.002)
1.414***
(0.002)
200
0.723***
(0.006)
2.019***
(0.006)
2.893***
(0.006)
8.195***
(0.006)
1.060***
(0.011)
3.679***
(0.011)
5.462***
(0.011)
15.269***
(0.011)
200
0.771***
(0.012)
3.161***
(0.012)
4.786***
(0.012)
15.071***
(0.012)
20
0.385***
(0.006)
1.568***
(0.006)
2.365***
(0.006)
7.271***
(0.006)
100
100
0.561***
(0.003)
1.065***
(0.003)
1.403***
(0.003)
1.911***
(0.003)
1.258***
(0.007)
2.532***
(0.007)
3.390***
(0.007)
9.062***
(0.007)
2.026***
(0.012)
4.610***
(0.012)
6.369***
(0.012)
16.617***
(0.012)
200
Notes: other covariates are at the sample median. Asymptotic standard errors in parentheses. Significance
levels of 10, 5 and 1% are noted *, ** and *** respectively.
Table 7: Women’s total returns to spatial mobility
Threshold of mobility
Distance
20
Total returns (%)
without qualification -0.861***
(0.004)
Bac
-0.861***
(0.004)
Bac+2
-0.861***
(0.004)
Bac+5
-0.861***
(0.004)
Threshold of mobility
Distance
20
Total returns (%)
without qualification 0.675***
(0.004)
0.878***
Bac
(0.004)
1.014***
Bac+2
(0.004)
1.218***
Bac+5
(0.004)
0
100
200
20
-2.085***
(0.009)
-2.085***
(0.009)
-2.085***
(0.009)
-3.948***
(0.009)
50
100
-3.948***
(0.018)
-3.948***
(0.018)
-3.948***
(0.018)
-7.052***
(0.018)
-0.378***
(0.002)
-0.068***
(0.002)
0.140***
(0.002)
0.452***
(0.002)
200
1.695***
(0.010)
2.210***
(0.010)
2.554***
(0.010)
6.239***
(0.010)
3.420***
(0.020)
4.468***
(0.020)
5.173***
(0.020)
12.868***
(0.020)
20
100
200
-1.878***
(0.012)
-0.338***
(0.012)
0.702***
(0.012)
4.616***
(0.012)
20
-0.943***
(0.006)
-0.169***
(0.006)
0.350***
(0.006)
2.282***
(0.006)
100
100
0.807***
(0.004)
1.051***
(0.004)
1.213***
(0.004)
1.458***
(0.004)
2.030***
(0.010)
2.647***
(0.010)
3.061***
(0.010)
7.504***
(0.010)
4.102***
(0.021)
5.365***
(0.021)
6.215***
(0.021)
15.571***
(0.021)
200
Notes: other covariates are at the sample median. Asymptotic standard errors in parentheses. Significance
levels of 10, 5 and 1% are noted *, ** and *** respectively.
27
Appendix A
Table A1: Reduced form equation − Men
Model
Intercept
Living in couple
Parents’ origin
France
Europe
Africa
Asia
Other countries
Paris region
Unemployment rate×100
Rural area end of schooling
Employment density×10000
Rel. net migratory balance
Log of the population
Share of qualified jobs
Share of industrial sector
Share of students
Percentage of youth
Fathers profession
farmer
corporate manager
professionals
technicians and associate
professionals
clerks
workers and elementary
occupations
unknown
Father is unemployed
Mother is unemployed
Father is retired
Mother in public sector
Education level
Poisson (QMLE)
2.583***
(0.018)
0.078***
(0.002)
ref.
-0.015***
-0.067***
-0.009
-0.013
0.050***
0.075**
-0.023***
0.051***
0.141***
0.005***
0.006
0.003
0.522***
-0.069***
(0.004)
(0.003)
(0.011)
(0.011)
(0.003)
(0.031)
(0.002)
(0.005)
(0.024)
(0.002)
(0.033)
(0.018)
(0.025)
(0.013)
-0.055***
-0.075***
ref.
-0.054***
(0.004)
(0.003)
Mobility (Threshold of mobility = 20 kms)
Probit
Distance > 0 (2-part model)
0.285
(0.178)
-239540***
(12088)
0.158**
(0.019)
2058
(3164)
ref.
-0.198**
-0.192**
-0.153
-0.143
0.596**
0.709***
0.207**
-0.076***
1.437
-0.055**
-2.728
1.074
2.978
-0.810
(0.039)
(0.036)
(0.104)
(0.107)
(0.027)
(0.310)
(0.024)
(0.049)
(0.233)
(0.015)
(0.312)
(0.181)
(0.241)
(0.130)
ref.
-46298***
-26743***
-43684***
3069***
44601***
299034***
-18149***
21611***
222937***
15089***
-95030***
-51545***
145898***
-119157***
(0.043)
(0.032)
(0.003)
-0.138**
-0.210**
ref.
-0.133**
(0.033)
-30891***
-14393**
ref.
-24491***
-0.095***
-0.123***
(0.003)
(0.003)
-0.237**
-0.300**
(0.026)
(0.028)
-23218***
-44317***
-0.084***
-0.016***
-0.017***
0.081***
0.015***
(0.004)
(0.005)
(0.005)
(0.003)
(0.002)
-0.256**
-0.041*
-0.059*
0.160**
0.059**
(0.034)
(0.051)
(0.054)
(0.029)
(0.020)
-8218
-1065
-4836
18596***
7490**
(11323)
(9492)
(2218)
(277)
(5480)
(59506)
(5848)
(7106)
(11528)
(543)
(5182)
(2332)
(7686)
(5451)
(9238)
(5725)
(6286)
(5004)
(7627)
(5780)
(10649)
(12131)
(4677)
(3433)
Notes: Asymptotic standard errors in parentheses. The coefficients of the two-part cannot be interpreted in terms of distance.
Only the signs count. The coefficients do not reflect marginal returns. Significance levels of 10, 5 and 1% are denoted *, **
and *** respectively.
28
Table A2: Reduced form equation − Women
Model
Intercept
Living in couple
Parents’ origin
France
Europe
Africa
Asia
Other countries
Paris region
Unemployment rate×100
Rural area end of schooling
Employment density×10000
Rel. net migratory balance
Log of the population
Share of qualified jobs
Share of industrial sector
Share of students
Fathers profession
farmer
corporate manager
professionals
technicians and associate
professionals
clerks
workers and elementary
occupations
unknown
Father is unemployed
Mother is unemployed
Father is retired
Mother in public sector
Education level
Poisson (QMLE)
2.660***
(0.018)
0.035***
(0.002)
ref.
-0.003
(0.004)
-0.075***
(0.004)
-0.022*
(0.013)
0.017
(0.010)
0.027***
(0.003)
0.014
(0.030)
-0.022***
(0.002)
0.039***
(0.004)
0.069***
(0.023)
0.005***
(0.001)
0.035
(0.032)
-0.038**
(0.018)
0.453***
(0.022)
-0.090***
(0.013)
Mobility (Threshold of mobility = 20 kms)
Probit
Distance > 0 (2-part model)
0.082
(0.187)
-111685***
(12647)
0.125**
(0.017)
-3957***
(1369)
ref.
ref.
-0.178**
(0.042)
-1353
(3235)
-0.320**
(0.041)
1592
(3315)
-0.06
(0.124)
-20282*
(10388)
0
(0.107)
-25316**
(10946)
0.618**
(0.029)
17106***
(2939)
-0.657*** (0.324)
138404***
(30350)
0.232**
(0.026)
-6238**
(2475)
-0.029*** (0.052)
9050***
(3073)
1.541
(0.242)
87496***
(10690)
-0.036**
(0.016)
6673***
(985)
-3.381
(0.334)
-39367***
(3753)
1.057
(0.190)
5035
(6702)
3.187
(0.237)
47581***
(6083)
-0.759
(0.133)
-43162***
(10625)
-0.056***
-0.058***
ref.
-0.044***
(0.004)
(0.003)
(0.043)
(0.033)
(0.003)
-0.018**
-0.097**
ref.
-0.091**
(0.036)
-12900***
-4229*
ref.
-7634***
-0.095***
-0.113***
(0.002)
(0.003)
-0.192**
-0.254**
(0.026)
(0.029)
-3865**
-7284***
-0.087***
-0.022***
-0.013***
0.061***
0.011***
(0.003)
(0.005)
(0.005)
(0.002)
(0.002)
-0.232**
-0.179*
-0.129*
0.076**
0.043**
(0.034)
(0.056)
(0.055)
(0.029)
(0.021)
-2891
4417
-7221
-252
-455
(4119)
(2299)
(2784)
(1810)
(2479)
(2337)
(3951)
(5165)
(1892)
(1352)
Notes: Asymptotic standard errors in parentheses. The coefficients of the two-part cannot be interpreted in terms of distance.
Only the signs count. The coefficients do not reflect marginal returns. Significance levels of 10, 5 and 1% are denoted *, **
and *** respectively.
29