Access to this full-text is provided by Springer Nature.
Content available from Social Indicators Research
This content is subject to copyright. Terms and conditions apply.
Vol.:(0123456789)
Social Indicators Research (2025) 176:51–106
https://doi.org/10.1007/s11205-024-03437-1
ORIGINAL RESEARCH
Using Survey Data toEstimate Intergenerational Mobility
inIncome andEducation inPortugal
LuísClemente‑Casinhas1 · LuísFilipeMartins1,2· AlexandraFerreira‑Lopes1
Accepted: 17 September 2024 / Published online: 14 October 2024
© The Author(s) 2024
Abstract
Previous studies about intergenerational mobility for the Portuguese economy find that
education and income persistence is very high in comparison with other developed econo-
mies. We construct relative, absolute, global and local measures of mobility for Portugal,
comparing them with existing evidence for this and other countries. These are the inter-
generational income elasticity (computed using the two-sample two-stage least squares
method), income correlation, rank-rank slope, bottom to top income level probability, the
share of individuals earning more than their fathers and also the intergenerational educa-
tion correlation, the low to high education level probability, and the share of individuals
with a higher education level than their fathers. We consider the 1968–1988 cohorts and
the 1995 and 2019 waves of the European Community Household Panel and the Euro-
pean Union Statistics on Income and Living Conditions, respectively. Overall, based on the
point estimates, women seem to present more mobility in income. Upward income mobil-
ity is verified at the bottom while persistence exists at the top. Women present a greater
absolute educational mobility. More than 80% of individuals have a higher education than
their fathers and full upward education mobility exists for children of low-educated fathers.
Mobility in education is higher for the offspring of medium–high-income fathers. Individ-
uals with a high education level, in the medium–high income level or with occupations
requiring a higher education level show higher mobility.
Keywords Intergenerational mobility in income and education· Relative and absolute
intergenerational mobility indicators· Global and local intergenerational mobility
measures· Two sample two stage least squares· Ordered probit· Mincer equations
JEL Classification E24· C26· I24· J62· O15
* Luís Clemente-Casinhas
luis_miguel_casinhas@iscte-iul.pt
1 Instituto Universitário de Lisboa (ISCTE-IUL) andBusiness Research Unit (BRU-IUL), Avenida
das Forças Armadas, 1649-026Lisboa, Portugal
2 CIMS-University ofSurrey, Guildford, UK
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
52
L.Clemente-Casinhas et al.
1 Introduction
The lack of intergenerational mobility prevents an efficient allocation of resources. When
children of more educated parents are more likely to obtain more education and higher-
paying jobs regardless of their innate abilities, the role of individual talent is suppressed.
This pattern, highlighted in the works of Marrero and Rodriguez (2013) and Van de Gaer
etal. (2001), reveals a vicious cycle between intergenerational persistence and inequality:
high inequality promotes an unequal distribution of parental investments and opportunities,
which then harms mobility in the next generation, perpetuating further inequality. Moreo-
ver, low mobility shapes individuals’ perceptions regarding fairness and aspirations in a
negative way (Ray, 2006; Ross, 2019; Weintraub etal., 2015), with a lower tolerance for
inequality and policies to fight it, thereby discouraging growth and social stability.
Despite the positive changes that were made in Portugal through the years, the findings
of several studies for the Portuguese economy made by international and national organiza-
tions, stress some persistence regarding income and education mobility.
On one hand, the educational level of the Portuguese population has been improving
steadily since the Carnation Revolution (25th of April 1974) when a democratic regime
was instituted. In the decades that followed, a strong consolidation of the social and eco-
nomic development was verified, with the expansion of the welfare state being its main
pilar. The State’s increased spending and intervention in the economy was reflected in
higher investment in infrastructure, technological modernization, and education, as well as
in the expansion of social policies to overcome poverty and unemployment. A comparison
between the 1981 and 2021 CENSUS of the population shows that the illiterate popula-
tion with 15years or more, decreased from 37 to 6%, while the same population with a
higher education degree increased from 2 to 20%. Most of the population with 15 or more
years has at least the basic or secondary education completed. Compulsory education has
also increased: individuals are now required to stay in school until they are 18years old.
Regarding income, the National Statistics Office (INE) reports that around 70% of house-
hold income is derived from labour. The Gini index in the last 30years has been between
32 and 38%, showing persistence.
On the other hand, studies seem to support the view that low educational attainment is
likely to perpetuate, and a high education persistence should exist from one generation to
another (Bank of Portugal, 2022; Clements, 1999; OECD, 2019). Besides, according to the
OECD (2018), a five-generation time window is needed for someone who belongs to the
10% poorest population to reach the median income.
We aim to answer the general research question: “What is the current state of intergen-
erational mobility in income and education in Portugal?” This is followed by other three
minor questions: (i) “Are there gender differentials in mobility?” (ii) “How does the Por-
tuguese mobility compare to other countries?”; and (iii) “Are there differences in mobility
when considering different individual characteristics?” Therefore, our goal is to summarize
the different aspects of intergenerational mobility to allow for a more general overview of
the phenomenon. To reach this goal, we construct measures of intergenerational mobility
in income and education for Portugal using the 1968–1988 cohorts of the European Com-
munity Household Panel (ECHP) and the 1995 and 2019 waves of the European Union
Statistics on Income and Living Conditions (EU-SILC).
To the best of our knowledge, we are the first to compute global and local measures
of absolute and relative intergenerational income mobility for Portugal. These are the
intergenerational income elasticity (IGE), the intergenerational correlation coefficient, the
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
53
Using Survey Data toEstimate Intergenerational Mobility in…
rank-rank slope, the share of individuals earning more than their fathers, and the prob-
ability that a child born with a low-income father has of reaching the top income level in
his or her generation (we define it as the bottom to top income level probability), comple-
mented by an ordered logit transition matrix. Additionally, we compute global and local
intergenerational educational mobility measures in relative and absolute terms to comple-
ment income mobility measures. We calculate the intergenerational education correlation
and the probability that a child born with a low-educated father has of reaching the highest
education level (denoted by low to high education level probability), also complemented by
an ordered logit transition matrix. Also, the share of individuals with more education than
their fathers is computed. Each measure is computed for both genders separately as well as
together. Furthermore, we analyse different subsamples to uncover which characteristics
of fathers and children may be related to more or less mobility, including their education
levels, occupation categories, income levels, and status in employment.
This paper is organized as follows. In Sect.2, the state of the art on intergenerational
mobility in Portugal is revised. Section3 details the methodology. In Sect.4, we describe
the data and sample construction. Section5 provides a discussion of the results. Section6
concludes.
2 Literature Review
In this section, we present the state of the art regarding intergenerational mobility studies,
with a particular focus on Portugal. We also identify the shortcomings in existing works
and define our contribution to the literature.
The recent availability of proper databases made the empirical study of intergenera-
tional mobility in income and education possible. However, deep single-country research
is mainly focused on the USA and Canada (e.g., Chetty etal., 2014, 2016, 2017, 2020a,
2020b, 2020c; Chetty & Hendren, 2018a, 2018b, Hilger, 2016; Latif, 2017, 2018; Fletcher
& Han, 2019). For the European Continent, existing work is still focused on Scandinavian
countries, the UK, Germany and Austria (as Björklund & Jäntti, 1997; Dearden & Reed,
1997; Blanden etal., 2004; Bauer & Riphahn, 2006; Nicoletti & Ermisch, 2008; Heidrich,
2017; Neidhöfer and Stockhausen, 2018; Brandén, 2019; Eriksen & Munk, 2020; Kyzyma
& Groh-Samberg, 2020) while for Southern European countries the Italian and Spanish
cases were addressed in the works of (Acciari etal., 2022; Cervini-Plá, 2015; Mocetti,
2007; Piraino, 2007).
Studies on education mobility in Portugal are scarce. For income, which should have
a close relationship with human capital formation, they are almost non-existent. Carneiro
(2008) uses transition matrices to show that educational persistence is strong in Portugal:
more than 90% of children who do not complete high school, with fathers who did not
complete primary education is verified, while almost no children complete less than high
school if their fathers have a university degree. Evidence shows that parental generational
differences in educational attainment create differences in opportunities for their children
and these differences in educational attainment differ from generation to generation.
Pereira (2010) studies the transmission of higher educational attainment in Portugal
through the use of probabilistic regression with data for individuals aged between 18 and
64years old. The author concludes that parents’ education strongly matters for children’s
higher educational attainment. The likelihood of reaching a higher education degree is
greater for individuals born into families with higher education (about eight times higher
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
54
L.Clemente-Casinhas et al.
in comparison with offspring of parents with 9 or fewer years of education), meaning that
low education levels are likely to perpetuate over time. Men generally perform more poorly
than women, meaning that they have overall lower mobility: they show a mean increase in
the probability of higher education degree attainment around 10%.
Bago d’Uva and Fernandes (2017) use multinomial probabilistic models and lin-
ear regression analysis to study the educational mobility of individuals born from 1940
to 1985. Mobility presented by the 1940 cohort is low: more than 80% of children with
low educated fathers didn’t reach tertiary education and 75% of individuals attain the
same level as fathers considering tertiary education. Upward mobility is generally lower
in Portugal when compared to the European Union. In this country, individuals born in
the 1970s present higher upward mobility than the ones born in the 1940s (40.6% versus
15.4%, respectively). The increase in mobility was more pronounced for Portugal when
compared to the European Union, with the gap between the two being shrinking from the
1970s on, being around 5.8% points. The difference in upward mobility between Portugal
and the European Union for the youngest cohort is mainly due to men (33.4% for Portugal
and 42.8% for the EU), while the share of girls reaching a higher education level than their
parents is close to the European average (47.9% for Portugal and 49.8% for the EU).
Six works identified include Portugal along with other countries. Comi (2004) uses data
on current income for the 1994–1998 period considering 12 European countries. Portugal,
Ireland, and the Mediterranean countries are the most relatively persistent in income and
education. Relative persistence in income measured by the earnings elasticity is stronger
for the pair father-son when compared to the pair father-daughter (0.20 compared to 0.15,
respectively). The same occurs for the eigenvalues of transitional educational matrices
(around 0.30 for men and 0.17 for women). Another study reporting that Portugal is the
least mobile country of those belonging to the OECD is Causa and Johansson (2010), who
computed wages’ persistence (as a proxy of income) as the difference between wage pre-
mium and wage penalty, around 70% points when corrected for distributional differences.
Schneebaum etal. (2014) consider 20 European countries and find that for the intergen-
erational correlation in education, Portugal presents the highest mobility considering the
pair father-sons, equal to 0.24, while for the pair father-daughters, being equal to 0.26, it
is surpassed by France, the Nordic and Anglo-Saxon countries, Greece, Czech Republic,
and Poland. In Nybom’s (2018) analysis of intergenerational persistence in education from
a linear regression on educational outcomes and individuals born around 1980, there is
cross-country heterogeneity in high-income countries, with Portugal standing amongst the
most persistent, along with Hungary and Uruguay. Evidence from the OECD (2018) shows
that, through the use of earnings elasticities, men present higher mobility than women. The
Portuguese men’s persistence (income elasticity of 0.4) is above countries such as Canada
and below the USA, Italy, France and Brazil. Regarding women, more mobility is veri-
fied in the country in comparison with the UK and Australia, while lower than Spain. The
OECD (2018) also computes the regression coefficient between the average of parents and
child’s years of education, finding that persistence verified in Portugal (coefficient around
65%) is below Indonesia and India. For the intergenerational correlation, around 50%,
mobility is higher than in Spain, Belgium, Chile, Slovenia, Indonesia and India. Addition-
ally, when considering the share of sons in the top quartile of earnings when the father is
at the bottom, Portugal is the country presenting the highest mobility, after Chile and Den-
mark, with approximately 20% of individuals. The Global Database on Intergenerational
Mobility (GDIM, 2018), constructed by Narayan etal. (2018), also presents mobility esti-
mates for Portugal, for both income and education. These are given by the intergenerational
coefficient of the regressions between child and parental income or education, respectively.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
55
Using Survey Data toEstimate Intergenerational Mobility in…
It is found that mobility in income is equal to 0.28, regardless of considering all children,
sons or daughters, moms and dads. For education, is around 0.6–0.7. The authors consider
that income mobility is lower than what is expected for the level of education mobility, for
the 1960 and 1970 cohorts, as it occurs with Ethiopia, the former Yugoslav Republic of
Macedonia, Nepal, and Romania. For the intergenerational correlation in education, the
values range between 0.44–0.46, while the probability a child from the bottom half of the
distribution ends up in the highest quartile is around 10–25%. Summing up, there is mixed
evidence for Portugal, though most papers appear to confirm International Organizations’
concerns about the high persistent level in the country.
Our work is the most comprehensive when analysing mobility across generations, given
the limitations that may be found in existing literature which we briefly describe. Firstly,
most of the works do not analyse income and education simultaneously, which we consider
of extreme relevance, since previous theoretical research, using models, has shown that
education should determine income mobility. For example, in Becker and Tomes (1979)’s
work, the income of children is part of the parental utility, which is maximized with an
optimal investment in children’s both non-human and human capital. The authors show,
that, among others, the inheritability of endowments and the likelihood parents have to
invest in their children are responsible for the equilibrium levels towards which mobility in
income tends to. Solon (2004) transforms the previous model to rationalize the log-linear
regression for intergenerational income mobility and also shows that an effective invest-
ment in human capital and progressive public investment on human capital contribute to
the patterns shown by the income elasticity’s steady-state value. Additionally, if education
is important for income mobility, some connection may also be expected between both edu-
cation and income mobility, as shown in the theory developed by Becker etal. (2018), in
which the children’s human capital production function is increased by the parents’ human
capital. They show that, when parental human capital and investments in children have
complementarities in the production of children’s human capital, richer parents will invest
more in their children’s human capital, in comparison with poor parents: this is translated
in economic status persistence across generations. Though a relationship is expected, it can
be broken due to, for example, a labour market where education is not easy to monetize or
other characteristics that individuals can’t control. Therefore, it is important to study both
these dimensions in simultaneous, to make an integrated analysis, that will enable to prop-
erly address policy implications given the patterns observed.
Furthermore, the use of mobility measured in both absolute and relative terms is also of
interest in this context. Absolute mobility provides the extent to which individuals are bet-
ter off than their parents. According to Deutscher and Mazumder (2023), absolute upward
mobility captures broad rising levels of education and economic growth. This means that
upward mobility will result with no ambiguity in a welfare improvement, considering the
Pareto Principle. Absolute mobility measures such as the share of individuals earning more
than their father (or with a higher education level) and the bottom to top income (educa-
tion) level probability provide different information about absolute mobility. While the first
measure indicates the proportion of children that are better off (in terms of education and
income) than their parents, hence providing a current measure; the second measure gives
us a more dynamic perspective, by estimating the probability of a children whose parents
are in the bottom quintile of the parental (or educational) distribution to reach the top quin-
tile (when adults).
On the other hand, relative mobility regards the extent to which the relative position
of children is connected to the relative position of parents in their respective generations.
Deutscher and Mazumder (2023), state that changes in relative mobility may capture a
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
56
L.Clemente-Casinhas et al.
variety of income movements, with different directions. To capture this wide range of pos-
sibilities regarding the outcome of relative mobility measures we use the intergenerational
income elasticity, the intergenerational income (education) correlation and the rank-rank
slope. The elasticity provides information on how the child’s income will change (in per-
centage) if the parental income changes by one percentage point. The larger the coeffi-
cient (in absolute terms) the stronger the impact. The interpretation of the intergenerational
income (or education) correlation is similar to the elasticity but excludes changes in ine-
quality as an explanation for changes between parents and children. Finally, the rank-rank
slope gives us information about movements between positions in the income distribution,
typically between income percentiles, which is different from the information provided by
the elasticity.
In a study about social mobility, the OECD (2018) uncovers, that, as countries become
more developed, absolute mobility slows down, meaning that the focus on relative mobil-
ity becomes stronger. This is the same as saying that absolute mobility is a consequence
of development, therefore the debate on relative mobility gains more attention because it
allows to have a better assessment of how unequal a society is.
Finally, we should devote attention to both global and local mobility measures.
Deutscher and Mazumder (2023) define global mobility variables as the ones that sum-
marize the joint distribution of income or education, while local measures regard isolated
portions of the distributions. Analysing global measures allows policymakers to have an
overall understanding of the intergenerational transmission of socioeconomic status, while
with local measures they have insights into the mobility of particular groups, namely the
children of parents who are poor and can reach the other extreme point of the income dis-
tribution (the same applies for educational attainment). Local measures provide informa-
tion to design tailor-made policies for specific groups. Our local (absolute) measure is the
bottom to top income (education) level probability.
Examining both absolute and relative mobility with global and local dimensions pro-
vides a full understanding of the mobility landscape in the country, which can’t be captured
by a single measure, an argument raised by Deutscher and Mazumder (2023) and also by
Corak (2019). At the exception of the seminal paper of Chetty etal. (2014), we don’t have
knowledge of previous research that computed global and local measures for absolute and
relative mobility in the same work.
Other concerns appear in the income mobility literature, namely regarding income
life-cycle effects since the relationship between current and permanent income changes
throughout an individual’s life: during the early stages of a career, incomes may be low,
increasing with skills and experience acquirement and then stabilizing or even declining
during retirement. When not accounted for, mobility estimates may not be well measured.
We incorporate this issue in our analysis. To mitigate this problem, we restrict our sample
to the individuals’ ages where there is a stable relationship between current and permanent
income, i.e., between 30 and 50years old.
Connected with this are the several biases that may appear in the analysis. For example,
when there is no information about permanent income for both generations, current income
has to proxy for it, introducing a measurement error in mobility estimates. Another exam-
ple that may lead to sensitive measures is the inclusion of samples in the analysis that are
not representative of the total population, as in the case of co-residents or siblings. We are
the first developing a sensitive analysis to assess the robustness of our results to the meas-
urement or selection problems that should also exist in other works.
Finally, as pointed out by Chadwick and Solon (2002), “daughters’ own earnings often
comprise a minority share (…) of her family income”, i.e., their individual income, when
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
57
Using Survey Data toEstimate Intergenerational Mobility in…
married, may not be a true measure of their socioeconomic status. When considering both
genders and married women are included in the analysis of income mobility, different
authors may not have properly considered the role of marital status. Others simply discard
women from the analysis and consider men only to avoid mismeasurement of mobility.
Since we advocate that both genders should be analysed, we calculate a measure of the
couple’s average income for women when they are married and use it when computing the
intergenerational mobility measures in income.
We try to overcome these shortcomings identified in the literature, and estimate both
mobility in income and education, in relative and absolute terms, with global and local
measures, for both genders, while incorporating different types of biases in the analysis and
testing how results are sensitive to them. Our approach aims to capture various dimensions
of social mobility, offering a broad understanding of the phenomenon.
3 Methodology
In this section, we present the intergenerational mobility measures that are used in this
work both in relative and absolute terms. Absolute mobility regards the extent to which the
younger generation is better off than the older generation while relative mobility concerns
the extent to which the socioeconomic relative position of children is connected to the one
of parents, in their respective generations. Intergenerational mobility can also be character-
ized as global or local measures. The first ones summarize the joint income or education
distributions and the second ones comprehend only some parts of those distributions.
Each mobility measure,
IM
, will have a specific functional form,
g
, such that
IM
=g
(
S
c
,S
p)
where
c
stands for children,
p
stands for parents and
S
is a measure of indi-
vidual’s socioeconomic status, as income or education.1 Grounded on methodological fun-
damentals on mobility measurement, we describe how the functional form of each measure
should be constructed conditional on the type of data we have.
3.1 Intergenerational Mobility inIncome
We now present the income mobility measures considered in this work. Table1 summa-
rizes them according to the framework provided by Deutscher and Mazumder (2023). This
regards the relative versus absolute and global versus local criteria.
The larger the value of relative mobility measures the lower mobility is, while the oppo-
site occurs with absolute mobility measures.
3.1.1 Relative Mobility Measures
3.1.1.1 Intergenerational Income Elasticity (IGE) The coefficient
(𝛽1)
obtained by regress-
ing the log of child
i
’s permanent income
(yc
i)
on the log parental permanent income
(
y
p
i)
,
2which is the canonical measure used for relative mobility:
1 As a very simple exercise, we present in Appendix A3 a relationship between relative mobility in income
and education based on the well-known Mincer (1974) wage equations.
2 Permanent/lifetime income can be defined as the average income during an individual’s lifetime (Fried-
man, 1957)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
58
L.Clemente-Casinhas et al.
where
i∈[1;N]
stands for the pair child-parent, from a total of
N
pairs. It is an elastic-
ity and therefore interpreted as the child’s income percentage change resulting from a one
percentage point variation in the parental income. The larger the coefficient is in absolute
terms the stronger the impact that parental income has on child’s income and vice-versa. It
is a measure of global mobility as it will consider the entire income distribution.
The estimation of Eq.(1) is possible only when at least two generations’ lifetime income
is available. For this purpose, researchers would need long panels to link parents and chil-
dren during their entire lives. However, data are usually available in short panels where
individuals (parents and children) are observed for a few years only and, therefore, dif-
ferent authors use current income
(yit)
in period t as a proxy for permanent income
(yi)
and assume their relationship to be constant and equal to one. The standard least squares
estimator for (1) using the current income may have inconsistency problems. In light of the
classic errors-in-variables model, this procedure is associated with a measurement error,
𝜏it,
When parental permanent income, i.e., our explanatory variable, is proxied by current
income, IGE is subject to an attenuation bias, as pointed out by Solon (1992)3: the meas-
urement error leads to an underestimation of the true relationship between both genera-
tions’ socioeconomic status. Also, as recent non-classic measurement error research points
out, the relationship between permanent and current income changes during the life-cycle
of individuals (children and parents): current income usually starts low when entering the
labour market, increases in mid work life and declines when reaching retirement, fluctuat-
ing around permanent/lifetime income (more stable measure for the long-term). Therefore,
grounded on Nybom and Stuhler (2016), Eq.(2) should be generalized to account for the
changes in time of this relationship (
𝜆t
), as
meaning that besides the standard attenuation bias, an associated life-cycle bias should also
exist.4
Our work is no exception in the framework of intergenerational mobility estimates
because the survey we use for Portugal contains information only about children’s current
income. We cannot directly observe parental income as the data are not available, so we
use the two-samples two-stage least squares method (TSTSLS). Two samples are needed
for this purpose: one for children used in the second step and another for parents used in the
first step. In the first step, we predict parental current income (
y
p
it
) by proxying their lifetime
income with parental characteristics reported by children: we use parental education, occu-
pation and managerial position. In the second stage, we estimate intergenerational mobility
by regressing child’s observed income on parental predicted current income. Furthermore,
we must account for the uncertainty arising from the regressor used in the second stage
(1)
yc
i
=𝛽
0
+𝛽
1
y
p
i
+𝜔
i
(2)
yit =yi+𝜏it
(3)
yit =𝜆tyi+𝜏it
3 For the attenuation bias, we have that plim �
𝛽1=𝛽1
Var(y
p
i)
Var
(
yp
i)
+Var(𝜏p
it
)
<𝛽
1
and
plim
𝛽1
→
0
if
Var (
𝜏
it)
→
+∞
,
i.e., beta becomes attenuated (Nybom and Stuhler, 2016).
4 The life-cycle bias (if income profiles change throughout life for both generations) is reflected by
plim
𝛽1=𝛽1𝜆c
it𝜆p
it
Var(yp
i
)
𝜆p
it
2Var
(
yp
i)
+Var(𝜏p
it)
. Depending on
𝜆c
it
and
𝜆p
it
, different results may arise (Nybom and Stuhler,
2016).
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
59
Using Survey Data toEstimate Intergenerational Mobility in…
(parental income, which is predicted from the first stage,
y
p
it
). Pagan (1984) pointed out
that the final steps’ coefficients may be in general consistent but the standard errors not. As
suggested by Björlund and Jäntti (1997), Piraino (2015), and OECD (2018), we compute
second step standard errors by employing a bootstrapping methodology.
For the life-cycle bias, controlling for individuals’ age
(A)
and its square
(A2)
to account
for life-cycle effects is by itself not sufficient (Jenkins, 1987). One should therefore restrict
the sample to the age range in which there should be a stable relationship between current
and permanent income and
𝜆it
equals one (Haider & Solon, 2006).5 The authors found that
for the USA economy this should occur between the early thirties and mid-forties (there-
fore around 40years old), a result corroborated by Brenner (2010) for Germany, and by
Böhlmark and Lindquist (2006) for Sweden. Regarding the attenuation bias, the most com-
mon way to deal with it in the literature is to average parents’ current income over time
(Solon, 1992).6
Therefore, the IGE is computed through the following equation:
3.1.1.2 Intergenerational Income Correlation Assuming that
y
p
it
is orthogonal regarding
Ac
it
and A
c
it
2
, we have:
where
sd(yc
it)
and
sd
(
y
p
it)
are the standard deviations of the (logged) child’s current income
and predicted parental current income, respectively, and
𝜌
y
c
it
,y
p
it
is the partial correlation
between those two variables. This correlation is the second measure we compute because
since
sd(
y
c
it)
≠sd
(
y
p
it)
, we have an intergenerational income elasticity distinct from the
(4)
yc
it
=𝛽0+𝛽1y
p
it
+𝛾c
1
Ac
it
+𝛾c
2
Ac
it
2
+𝜔c
it
(5)
𝛽
1=𝜌yc
it,yp
it
sd(y
c
it)
sd(yp
it
)
⇒𝜌yc
it,yp
it
=𝛽1
sd(
y
p
it)
sd(yc
it
)
,
Table 1 Properties of Intergenerational Mobility in Income Measures
NA stands for Not Applicable
Global Measures Local Measures
Relative Mobility - Intergenerational Income Elasticity
- Intergenerational Income Correlation
- Rank-rank Slope
NA
Absolute Mobility - Share of Individuals Earning More than their
Parents - Bottom to Top
Income Level
Probability
5 If we consider
𝜆c
it
=𝜆
p
it
=
1
in plim
𝛽1=𝛽1𝜆c
it𝜆p
it
Var(y
p
i)
𝜆p
it
2Var(yp
i
)+Var(𝜏p
it)
, we only have to worry about the standard
attenuation bias.
6 If we adopt this approach to compute the average, we will have to drop a lot of observations to guarantee
parents remain in the benchmark sample for the periods considered. Besides, that does not guarantee that
the bias disappears, as found by Mazumder (2005): the authors shows that even with an average computed
for five years results in a 30% bias. Therefore, our study uses a single year to predict parental income, pre-
serving sample size and acknowledging that the true relative income mobility may be lower.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
60
L.Clemente-Casinhas et al.
intergenerational income correlation,
𝛽
1
≠𝜌
yc
it
,y
p
it
. In other words, we adjust the elasticity
which, as argued by Bukodi and Goldthorpe (2018), reflects earning’s association and also
changes in inequality across generations. It is also classified as a global mobility measure.
3.1.1.3 Rank‑Rank Slope Dahl and DeLeire (2008) suggest another measure of relative
intergenerational income persistence, which is the rank-rank slope, adopted also by Chetty
etal. (2014). It may be computed by first rank children and parents in their respective
permanent income percentiles’ distribution. Second, for each parental income percentile
rank
r
(y
p
i)
, obtain the average children’s income percentile ranks,
r(yc
i)
. Third, regressing it
against parental income percentile ranks, as follows:
The resulting coefficient (
𝜅1
) measures the relationship between the positions children
and parents have in their respective income distributions. As with the intergenerational
income elasticity, the greater is the coefficient the greater intergenerational persistence will
be, and vice-versa, in absolute terms.7 As pointed out by Deutscher and Mazumder (2023),
the rank-rank slope, also a global mobility measure, is preferable to the intergenerational
income elasticity if the interest lies in positional mobility rather than the regression to the
mean rate, i.e., if the focus is on the movements between positions is the distribution of
income and not on the incomes that are connected to them.
We rank the predicted values for parental income,
r(
y
p
it)
. 8Then, for each one, there is a
given number of corresponding children about which we observe their percentile income
ranks and compute the average,
r
(y
c
it)
. We should face the same constraints as before in
terms of income (mis)measurement. Therefore, we should consider the strategies explained
above to smooth the life-cycle associated bias, although we don’t know its full extent. Fol-
lowing Chetty etal. (2014), Eq.(6) will therefore be rewritten as
We estimate Eq.(7) through OLS. Percentile ranks for children and ranked bins for par-
ents will be based on the entire sample throughout our analysis.
(6)
r
(y
c
i
)=
𝜅
0+
𝜅
1r(y
p
i
)+
𝜓i.
(7)
r(
y
c
it)
=𝜗
0
+𝜗
1
r
(
y
p
it)
+𝜛
it,
7 Chetty etal. (2014) argue that the rank-rank slope and the intergenerational income correlation have a
close relationship, since they are scale invariant. This does not occur with the intergenerational income elas-
ticity, because inequality should be different across generations. When inequality is greater for the child’s
generation, an increase in parental income may have a greater effect on children’s income when compared
to a scenario where inequality is lower. In other words, the rank-rank slope and the intergenerational income
correlation are not affected by changes in inequality, while the intergenerational income elasticity is.
8 Since we predict parental income, our rank-rank slope will not be the same as the standard rank-rank
slope used in the literature. This is because parental income will be predicted grounded on the available but
limited set of predictors, reducing the number of distinct values it can assume. Therefore, it is not possible
to split them into percentile ranks, as we do for children. Though we don’t have percentile ranks, we have
ordered bins. For the sake of our analysis, we consider that only the interpretation of the slope changes,
though the reasoning that can be taken from conclusions is the same. Instead of reflecting what will be the
change in the average percentile rank if the parental percentile rank changes, the rank-rank slope will give
us the change in the average percentile rank if the parental bin increases to a higher order.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
61
Using Survey Data toEstimate Intergenerational Mobility in…
3.1.2 Absolute (Upward) Mobility Measures
Besides looking at relative mobility, one should be interested in measuring absolute upward
mobility as well.9 As Chetty etal. (2014) argue, while improvements in relative mobility
may occur at the expense of rich people’s income being harmed, improvements in absolute
mobility for a given level of income, ceteris paribus, should result in a welfare improve-
ment according to the Pareto Principle. This is the same as saying that, holding other things
constant, absolute upward mobility de facto reflects beneficial changes in income of indi-
viduals from a given background. We follow their work and compute two main measures of
absolute upward mobility.
3.1.2.1 Share ofIndividuals Earning More thantheir Parents The first measure of absolute
upward and also global mobility suggested by Chetty etal. (2014) is the share of individuals
whose income exceeds their parents’ income in real value.
3.1.2.2 Bottom toTop Income Level Probability Following Chetty etal. (2014), the other
measure one can use for upward absolute mobility is the bottom to top quintile probability,
which is the probability that children whose parents are in the bottom quintile of the parental
income distribution have of reaching the top quintile of the children’s income distribution
when adults. Since it covers specific sections of the income distribution, it is considered a
local measure of mobility. This would be the well-known “American Dream”. We measure
in this way the bottom to top income level probability because, as mentioned above, we are
unable to construct percentile ranks for parents. This also prevents us from transforming
data into quartiles or quintiles. Therefore, we consider a specific cell of the Ordered Logit
Transition Matrix, which we describe below.
Suppose that we assign each child’s income level
inclevc
i
in one specific category, i.e.,
we have
inclevc
i
∈{1,2, ..., H
}
where
H
denotes the number of possible income categories,
which will be defined later in this work: the same is considered for the parental income
level categories.
The ordered logit transition probability will be estimated by
with the cumulative distribution function of the logistic defined by
G
(ch−Ψinclevp
i)= ech−Ψinclev
p
i
1+e
ch−Ψinclevp
i
.
Ψ
is estimated using the maximum likelihood estimator.
The bottom to top income level probability is given by
Pr
(inclev
c
i
=H
|
inclev
p
i
=1
)
, i.e., it
corresponds to the probability that a child with parents classified as low income has of
becoming classified as a high-income level earner.
(8)
Pr
(inclevc
i=h�inclevp
i)=
⎧
⎪
⎨
⎪
⎩
G(c1−Ψinclevp
i),ifh =1
G(ch−Ψinclevp
i)−G(ch−1−Ψinclevp
i),if 1<h≤H−
1
1−G(cH−1−Ψinclevp
i),ifh =H
9 We acknowledge the possibility of downward movements, but the focus should be on the upward direc-
tion, as it is connected with higher income growth and shared prosperity (GDIM, 2018).
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
62
L.Clemente-Casinhas et al.
3.2 Intergenerational Mobility inEducation
The educational mobility measures considered in this work are now presented. As we
did for income, Table2 summarizes them according to Deutscher and Mazumder (2023),
namely regarding the relative versus absolute and global versus local criteria.
As it occurred with income, the larger the value of relative mobility measures the lower
mobility is, with the opposite occurring with measures of absolute mobility.
The preferred measure in the literature of relative intergenerational mobility in education
is analogous to the relative mobility measure used for income and consists of the coefficient
obtained by regressing the total years of educational attainment of children on the total years
of education of parents. However, our data characteristics do not allow us to compute it.10
3.2.1 Relative Mobility Measure
We rely on the Pearson correlation between parental and child’s education levels to meas-
ure relative mobility in education:
where
ec
i
is a variable for the ordered education levels of children,
ep
i
is a variable for the
ordered education levels of parents, and the respective average education levels in the sam-
ple are
ec
=
1
N∑N
i=
1e
c
i
and
ep
=
1
N∑N
i=
1e
p
i
.
The coefficient ranges between − 1 and 1. From its sign it is possible to infer if we have
positive or negative monotonic relationships between the education levels of parents and
children, with 0 meaning that no such type of correlation should exist. The closer the coef-
ficient is to the extremes, the stronger the relationships are, while the opposite occurs if it is
near zero. It is also a global mobility measure.
3.2.2 Absolute Mobility Measures
In order to measure mobility in education in absolute terms, two measures are considered.
The first is the share of individuals with a higher education level than their fathers. The
second is the probability of low to high education level, which corresponds to the prob-
ability children have of reaching the highest education level conditional on the father’s edu-
cation being the lowest one. Therefore, it is classified as a local measure of mobility. This
corresponds to a specific cell of the Ordered Logit Transition Matrix described below.
Similar to the case of income levels, we model the probability of children having
attained a specific observed category in terms of education,
ec
i
, conditional on the observed
educational category of their parents
ep
i
. Suppose that for the educational levels of children
we have
ec
i∈{1,2, ..., M}
where
M
denotes the number of educational categories we have
(9)
P
=
∑N
i=1(e
c
i−e
c
)(e
p
i−e
p
)
�∑
N
i=
1(ec
i
−ec)2
�∑
N
i=
1(ep
i
−ep)2
,
10 It would only be possible if we had information on both parents’ and children’s educational attainment,
expressed in completed years of education. However, that is not considered in the surveys we use. Instead,
educational attainment is reported in categories of completed education levels: the disaggregation is not
the same for both generations. Therefore, by making both categorizations comparable and attributing them
years of education, we could lose information in the end.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
63
Using Survey Data toEstimate Intergenerational Mobility in…
for our dependent variable: the same categories are considered for the case of parents. We
have an index model for parental educational attainment described as
where
e∗c
i
is an unobserved latent measure of the years of education of children and
ep
i
is a
variable for the ordered education levels of parents.
Θ
is the regression coefficient associ-
ated with the explanatory variable, estimated using maximum likelihood.
𝜉p
i
is the error
term, which follows a logistic distribution. Furthermore, the latent variable crosses specific
thresholds,
tm
, which are also unknown, such that:
For each value of the transition matrix, we will estimate
with the cumulative distribution function
G
(tm−Θep
i)= etm−Θe
p
i
1+e
tm−Θep
i
. The low to high edu-
cation probability is given by
Pr
(e
c
i
=M
|
e
p
i
=1
)
.
4 Data andSample Construction
In this section, we present the databases that are used not only to construct the mobility
measures but also to estimate the relationship between relative mobility in income and edu-
cation, through the use of Mincer (1974) equations. Besides, we describe how our sample
is constructed.
4.1 Data
To estimate our benchmark measures of mobility in income and education, we use two data-
bases. Both are provided by INE (Instituto Nacional de Estatística, the Portuguese National Sta-
tistics Authority) and are the Portuguese components of two main European Union surveys.
The first survey is the Painel dos Agregados Domésticos Privados da União Europeia, part
of the European Community Household Panel (ECHP), developed for 14 Member States. The
second is the Inquérito às Condições de Vida e Rendimento das Famílias, which is a part of
the European Union Statistics on Income and Living Conditions (EU-SILC) and was launched
in 2003, replacing the first survey. Individuals are between 16 and 80years old. Our sample of
children is restricted to the latest survey wave, in which there is retrospective data on their par-
ents. We use the 2019 wave of the EU-SILC as it contains a module aimed at providing infor-
mation on intergenerational transmission of poverty. Individuals considered are between 30 and
50years old. Here, personal information is used, in particular individuals were asked about their
(10)
e∗c
i
=Θe
p
i
+𝜉
p
i,
(11)
e
c
i=
⎧
⎪
⎨
⎪
⎩
1, ife∗c
i≤t1
m,iftm−1<e∗c
i≤tm
M,ife∗c
i>tM−1
.
(12)
Pr
(ec
i=m�ep
i)=
⎧
⎪
⎨
⎪
⎩
G(t1−Θep
i),ifm =1
G(tm−Θep
i)−G(tm−1−Θep
i),if 1<m≤M−
1
1−G(tM−1−Θep
i),ifm =M
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
64
L.Clemente-Casinhas et al.
parents’ characteristics when they were about 14years old. The pseudo-parents’ samples used
in our analysis concern the 1995–1999 waves of the ECHP, since they are the ones closer to
the periods in which the adults in our main sample are 14years old. In the EU-SILC survey, an
income reference period is defined as the period that income is related to. In most of the EU-
member States it corresponds to the previous calendar year (fixed 12-month period). Hence, the
outcomes’ periods for specific variables considering the 2019 wave is 2018. The same applies
to the 1995–1999 ECHP waves, where the reference period is 1994–1998.
Additionally, although research about intergenerational income mobility is mainly
focused on fathers and sons, in this work we consider both genders for children. The rea-
son, as stated above, is because Portugal has some very specifics characteristics regarding
the female labour market and educational attainment for women.
4.2 Sample Construction
The sample construction is now presented. We describe how we deal with unobserved
parental income, life-cycle effects in income measurement, differences between permanent
and current income, and income measurement conditional on gender. We also show how
we make information comparable across surveys and detail the definitions of income, edu-
cation, occupation and managerial position related variables.
4.2.1 Income
4.2.1.1 Predicting Father’s Income We follow the common methodology of a variety of
previous studies in which the datasets share the same characteristics as ours and father’s
income has to be predicted, namely, Björklund and Jäntti (1997), Leigh (2007), Lee and
Solon (2009), and Nuñez and Miranda (2010). Our strategy can be formalized as follows.
Consider that the log of parents’ current income (in t) can be defined as the sum of per-
manent income
yp
i
and time-varying characteristics, namely age
(A)
and its square
(A2)
to
control for life-cycle effects in income:
In the current wave of the survey (main sample) we cannot observe parental current
income, y
p
it
. We also cannot link parents and children across waves. Although this is the case,
we can observe in an earlier wave of the survey the current income of individuals, which are
assumed to be representative of the same population as the current one. We call it the auxil-
iary sample of pseudo-parents. Thus, let
Xp
ij
be a vector of dummies for each possible parental
characteristic (j
∈J
) which can proxy for lifetime income (again, not observed), such that:
(13)
yp
it
=yp
i
+𝛾p
1
Ap
it
+𝛾p
2
Ap
it
2
+𝜇p
it.
Table 2 Properties of Intergenerational Mobility in Education Measures
NA stands for Not Applicable
Global Measures Local Measures
Relative Mobility - Intergenerational Education (Pearson) Correlation NA
Absolute Mobility - Share of Individuals with a Higher Education Level
than their Parents - Low to High
Education Level
Probability
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
65
Using Survey Data toEstimate Intergenerational Mobility in…
Equation(13) becomes:
We estimate Eq.(15) through an OLS estimator in
t
(i.e., our results are computed for a
cross-section). The resulting coefficients are used to predict the current income of pseudo-
parents of children in the main sample,
yp
it
:
We consider as potential proxies of parental permanent income their individual charac-
teristics such as occupation, educational attainment and managerial position.
This approach has some issues attached to it that are worth mentioning. First, we use a
sample of pseudo-parents which is not the same as using parents, taken from the population
in our main sample. Second, the predicted income is not the same as the observed income.
Third, results may be biased due to the possible lack of validity of the instruments used. As
pointed out by Solon (1992), there is the possibility of these instruments not being exog-
enous and, in turn, having a relationship with children’s income that goes beyond the parental
income channel. Grounded in Nicoletti and Ermisch (2008) and supported by the evidence
presented by Björklund and Jäntti (1997), Cervini-Plá (2015) argues that these instruments
may positively influence the children’s income even after controlling for the parental income,
promoting an upward bias in the estimate of the elasticity. Thus, most authors that use this
method assume that the estimates are upper bounds of the true coefficient. We test how sensi-
tive our results are to the use of different combinations of characteristics that proxy for paren-
tal permanent income. Fourth, as parental income is predicted using a small number of differ-
ent instruments that proxy for their permanent income, we have a limited small set of distinct
values that these can assume and a lack of variability in parental income.11 In line with this,
there is the potential problem of missing variables, such as industries or sectors of activity
and years of experience. However, there is no more retrospective information about parents
reported by children available in the EU-SILC that could be used to predict their individual
income. All together these issues may influence the results and conclusions.
4.2.1.2 Life‑cycle andAttenuationBias To account for the life-cycle measurement error we
restrict our sample to individuals aged 30–50years old.12 Current income is used for both
generations. We predict parental income at 40years old, the age in the middle of the range
at which permanent income may be proxied,13.14 To address the standard attenuation bias,
existing evidence shows that a large time range would be needed to make it disappear. We
restrict the parents’ sample to fathers only: as we predict parental individual income and the
best option for women is to use family income/couple’s income (while for men, concerns
(14)
yp
it
=Φ
�p
j
X
p
ij
+𝜑
p
ij
(15)
y
p
it
=Φ
�p
j
Xp
ij
+𝛾p
1Ap
it
+𝛾p
2Ap
2
it
+𝜇p
it
+𝜑
p
ij
(16)
y
p
it
=Φ
�p
j
Xp
ij
+𝛾 p
1Ap
it
+𝛾 p
2Ap
2
it
11 This has implications for the rank-rank slope because we cannot rank predicted parental income in per-
centiles as it is done for children. Nevertheless, parental income is still ranked but in different bins.
12 Different authors used similar age ranges: e.g., 30–50 in Cervini-Plá (2015), 25–54 in Mendolia and
Siminski (2019), and 38–45 in Corak (2019).
13 We follow authors such as Leigh (2007) and Mendolia and Siminski (2019).
14 Results for the first stage are presented in Table16 in the Appendix.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
66
L.Clemente-Casinhas et al.
are not that clear in the literature), we do not have an intersection between both condi-
tions. Besides, we don’t know the parental marital status when income is measured. We
will therefore have estimates for the pairs father-children, father-son, and father-daughter. A
father is defined as the individual considered by the interviewed person as his or her father
when aged 14, having (or not) a biological relationship, even if the biological father was
known and alive. According to Mazumder (2005), a father’s income averaged for 5years
will still produce attenuated beta (IGE) estimates, which are 30% biased for the USA, and
even using a 25-year range period the bias would remain. As Cervini-Plá (2015) points out
for Spain (Spain’s data have the same characteristics as ours), when using instruments in the
TSTSLS approach to proxy for parental income and then predict parental current income,
one is already computing its average. By using a single year for parental income in our
benchmark sample, we assume that we are obtaining the most attenuated estimate of rela-
tive persistence in income, which means that relative mobility in income may be lower than
the one we obtain.15 Additionally, considering more than a single year implies guaranteeing
that individuals are in the cross-sectional samples for all periods, which reduces the number
of observations by a large amount. We perform a sensitivity exercise to assess how sensi-
tive our estimates are when using an average for parental income (i.e., using more than one
period to compute it).
4.2.1.3 Measurement Issues In our sample of children, individuals can be either single or
married. Chadwick and Solon (2002) show that in the case of married daughters, we should
use the couple’s income to better proxy for their economic status. Although this may justify
the use of couple’s income for women, it should not rule out the use of the couples’ income
as well for men. This is because in our sample women earn on average 45% of the couple’s
income. This makes us consider the couple’s income as well for men, when married.
Additionally, since we are studying intergenerational income mobility, we decide
to include only individuals with strictly positive income during the income reference
period.16 For singles we use individual income. For married individuals we use the com-
bined income of the couple, i.e., we add the total income of the couple and divide by two,
obtaining an average, following Chadwick and Solon (2002) and Raaum etal. (2008). Mar-
ried individuals who do not work but benefit from the income of his/her spouse are also
not considered, as what they earn is not a direct result from being active in the labour mar-
ket. In a later sensitivity exercise we include the partners with no individual income, but
with positive average couple’s income and test if results change. We also perform a sen-
sitivity analysis to evaluate the possible differences arising from using individual income
instead of average couple’s income when individuals are married. When using the ECHP,
we measure income as the wage and salary income for employees and self-employment
income for self-employed individuals. The corresponding variables available in the EU-
SILC are the net employee cash or near cash income and the net cash profits or losses from
self-employment. The reason why we use these “narrower” definitions of income is that the
15 For the standard attenuation bias, when we average the annual income of fathers from 1 to
T
and regress
yc
i
on
yp
i
=
1
T∑T
t=
1y
p
it
, we obtain that plim �
𝛽1=𝛽1
Var(y
p
i)
Var(yp
i
)+1
T
Var(𝜏p
it
)
<𝛽
1
. If
t→+∞
, beta becomes less attenu-
ated, which reflects more persistence (Björklund and Jäntti, 1997).
16 The reason is that the canonical measure considered in the literature and also in our paper is the inter-
generational earnings elasticity and we wanted to follow existing research. Income is logged, therefore it
has to be strictly positive.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
67
Using Survey Data toEstimate Intergenerational Mobility in…
characteristics reported about parents by children, in the EU-SILC database, mainly con-
sider labour-related income. To make both generations comparable, this is also the type of
income chosen for the children’s subsample.
In the second survey, the first variable is defined as the gross cash or near cash income,
deducted from tax at source and/or social insurance contributions. In turn, gross cash or
near cash income consists of the cash monetary component of employees’ compensation
paid by an employer, including the value of income taxes and social contributions that are
paid either by the employee or by the employer to tax authorities and/or social insurance
schemes (on behalf of the employee).17 The second variable can be defined as the net of
tax at source and/or social insurance contributions net operating profit or loss for owners/
partners that work in an unincorporated company, with interest on business loans deducted,
plus royalties (writing, inventions, among others) and rentals from equipment.18 To make
income comparable across surveys we use the Consumer Price Index (CPI with a base year
in 2010 to obtain income in real values).
We also define income levels for both children and parents. We ground our definition
for each income category on the OECD definition for low and high pay workers.19 We con-
sider the low-income level to be the one in which individuals earn less than two-thirds of
the median national income, while the high-income level comprehends individuals earning
one and a half the median income. Individuals classified as middle-level earners are those
between, and are split into two categories, middle-low and middle-high, according to the
intermediate value of the category’s possible values’ range. We again apply the CPI base
year 2010. For parents, the log income’ bounds separating classifications are 8.81, 9.29,
and 9.62. For children we have 8.96, 9.45, and 9.77.
4.2.2 Education
Data for educational attainment is taken from the ECHP and the EU-SILC. Education is
classified using the International Standard Classification of Education (ISCED) of the
United Nations Scientific and Cultural Organization (UNESCO). There exist two categori-
zations. The first one, ISCED 1997, considers 7 levels of education. Data in the 1995 wave
of the ECHP cover three valid groups, which have correspondence with the ISCED 1997
classification. The second categorization, ISCED 2011, was used in the 2019 wave, cover-
ing 9 levels. When asked about their parents’ education, children’s responses are divided
into low, medium, and high educational levels, which have correspondence with ISCED
2011 classification. This means that to estimate intergenerational mobility in income—in
17 As described in the EU-SILC variables definitions, it includes wages and salaries, payments for time
not worked as holidays, overtime rates, directors’ fees, piece rate payments, commissions, gratuities and
tips, supplementary payments as the thirteenth month, shared profits and bonuses, productivity payments,
allowances for remote working and transport, sickness, disability and maternity supplements. It excludes
reimbursements, severance and termination pay, purely work-related expenses, lump sum transfers at retire-
ment time and union strike pay.
18 According to the EU-SILC documentation, when computing the net operating profit, one should sum
market output, market value of goods and services consumed by the entrepreneur but bought for the unin-
corporated company, property income, subsidies, and subtract intermediate goods, compensation of
employees, taxes, interest, rents and fixed capital consumption. The documentation also states that income
from self-employment excludes directors’ fees earned by owners of incorporated enterprises (included in
the gross employee cash or near cash income), dividends paid by incorporated companies, profits from cap-
ital invested in other enterprise where the individual does not work, rent from land and rentals from dwell-
ings.
19 https:// data. oecd. org/ earnw age/ wage- levels. htm.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
68
L.Clemente-Casinhas et al.
which we predict parental income grounded on educational attainment—as well as in inter-
generational education mobility, we must match children’s own education levels in the
pseudo-fathers’ education categories, which is presented in Table3 below.
Information about education in the pseudo-parents’ sample is only used to proxy for their
permanent income and then predict their current income (which is not available) in a first
stage, which is the estimation of intergenerational mobility in income. For the estimation of
intergenerational mobility in education we only need to use the children’s samples where ret-
rospective information about education is directly available. As we aim not only to analyse
mobility in education, but also to identify patterns regarding its joint behaviour with mobility
in income, we should consider the same individuals in both analyses, which implies that the
age range we first chose is the same. We also have to ensure that individuals are not enrolled
in school. Therefore, we include in the analysis only individuals between 30 and 50years
old, which have finished school and are not enrolled in any type of education at the time of
the survey, following Urbina (2018). In 2018, 5% of the Portuguese individuals aged 30–34
where still enrolled in school, 4% for the age range of 34–39, and 2% for 40–64years old.20
4.2.3 Occupation
The ECHP and the EU-SILC are the sources for occupation related data. The International
Standard Classification of Occupations (ISCO) from the International Labour Organization
(ILO) is considered in our work. For the 1995 wave of the ECHP the ISCO-88 classifica-
tion is used, while for the 2019 wave of the EU-SILC the ISCO-08 classification is consid-
ered. The correspondence is in Table4.
4.2.4 Managerial Position
Another characteristic we use to proxy for father’s permanent income is his managerial
position. The parent can be either in a supervisory or non-supervisory position. We create a
dummy variable that is equal to 1 in the first case, if the individual has formal responsibil-
ity for an employees’ group, with direct supervision of the work, and 0 otherwise. Expect-
edly, for the same occupation category and education level, an individual in a superior
managerial position should have higher income than one in a lower managerial stage. Data
for managerial position is also taken from the ECHP and the EU-SILC.
Summary statistics are presented in Tables14 and 15 in the Appendix.
5 Empirical Results andDiscussion
In this section we present our benchmark results for the measures of intergenerational
mobility in income and education for the Portuguese economy.21 As pointed out by
Deutscher and Mazumder (2023), there are no correct or incorrect mobility measures.
Therefore, the information contained in the different mobility measures is distinct and
20 https:// stats. oecd. org/ Index. aspx? DataS etCode= EAG_ ENRL_ RATE_ AGE.
21 The surveys we use provide individual weights that are computed accounting for the sample design and
individuals’ characteristics. They reflect the structure of the population: the greater the weight the stronger
the representativeness an individual has on the population, which cannot be ignored. We therefore use pop-
ulation weights in our analysis.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
69
Using Survey Data toEstimate Intergenerational Mobility in…
serves the heterogeneous interests that policymakers may have. We also present possible
explanations for our evidence, which do not imply causality, but may be explored in future
research. Policy directions are addressed.
5.1 Intergenerational Mobility inIncome
Table 5 presents the benchmark results for the intergenerational mobility in income for all
children regardless of gender and also for male and female children separately.22
The evidence presented shows that the income elasticity, equal to 0.26, is higher rela-
tive to the intergenerational income correlation of 0.20. This result is expected because the
standard deviation of children’s income (0.56) is higher than the one of parents (0.44), i.e.,
inequality is expected to have increased throughout time. This resembles the evidence pre-
sented by the World Inequality Database on which inequality has been increasing through-
out time for this country.23 Results also suggest that, although 53.11% of the individuals
have experienced upward income mobility relative to their parents, the probability that
individuals have of reaching the top income level when raised in the low-income level is
still low given that only 7.15% managed to do so. In other words, although more than half
the population is better off than their parents, it is still not easy for someone born in the
worst income scenario to attain full prosperity.
Gender differences also evident. Women show more intergenerational income mobility
than men, with the exception of the bottom to top income level probability. Our evidence
is also verified in the literature regarding other countries. Borisov and Pissarides (2019)
show that mobility is higher in correlation ranks for females in Russia. For this measure
as well as for the intergenerational income elasticity, Helsø (2020) finds daughters to be
more mobile than sons for Denmark and USA, while ambiguous findings are reported by
Kyzyma and Groh-Samberg (2020) for Germany. Acciari etal. (2022) show that mobility
is higher for women when considering the rank-rank slope for Italy. Considering the work
of Comi (2004), for an older generation and with differences in variables’ definitions and
sample construction, the same finding is presented for Portugal regarding the intergenera-
tional income elasticity, according to which girls show more mobility. Regarding the evi-
dence of the OECD (2018), also following an instrumental variable approach and similar
sample restrictions, men present higher mobility than women, as opposed to our evidence.
When analysing intergenerational mobility measures, a main goal is to stress how high
or low mobility is. This is done through comparisons between countries. We must be care-
ful in the comparisons because estimates are sensitive to measures of income, estimation
methods, and sample selection, among others. This means that we try to choose works
that make choices to ours in terms of sample and methods.24 Most of these studies address
mainly the case for intergenerational income elasticity for a single gender (usually men).
By country and for sons, we have elasticities being around: 0.1–0.3 (Blanden etal.,
2004), 0.20–0.25 (Nicoletti & Ermisch, 2008), and 0.56–0.59 (Dearden etal., 1997) for
the UK25; 0.19–0.22 for Canada (Fortin & Lefebvre, 1998); 0.28 for Sweden (Björklund
& Jäntti, 1997); 0.2–0.3 (Leigh, 2007), 0.35 (Mendolia and Siminski, 2019) for Australia;
22 Results are the same when age is centred at 40years old for both generations, regarding income mobility.
23 https:// wid. world/ count ry/ portu gal/.
24 Slight differences between ours and the cited studies may lead to wrong conclusions (see Solon, 2002).
This is also true for education mobility estimates.
25 Large differences for the UK may be due to differences in the cohorts or the surveys used by the authors.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
70
L.Clemente-Casinhas et al.
0.4 for France (Lefranc & Trannoy, 2005); 0.42 for Spain in Cervini-Plá (2015); 0.45–0.53
(Solon, 1992), 0.34–0.49 (Lee & Solon, 2009), and 0.52 (Björklund & Jäntti, 1997) regard-
ing the USA; 0.5 for Italy (Mocetti, 2007; Piraino, 2007); 0.58 (Ferreira & Veloso, 2006)
and 0.69 (Dunn, 2007) for Brazil. Our estimated value for the elasticity of males, 0.3, is
similar to some of the estimates for the UK, Sweden, and Australia, but higher than the
estimates found for Canada, and lower than the ones for France, Spain, the USA, Italy, and
Brazil. In the OECD (2018), when comparing persistence across countries, Portugal, with
an elasticity of almost 0.4 (higher than ours) is also above Canada and below the USA,
Italy, France and Brazil. The UK and Australia surpass Portugal, as in some of the cases
presented, while Spain is below the country’s estimate.
For daughters we have elasticities ranging about: 0.05–0.46 in the USA (Lee & Solon,
2009); 0.1–0.3 (Blanden etal., 2004) and 0.63–0.70 (Dearden & Reed, 1997) in the UK;
0.3 in France (Lefranc & Trannoy, 2005). Our estimate of 0.22 fits in the interval of
some of the estimates made for the USA and the UK, but lower than the estimate made
for France. This is in line with results presented by the OECD (2018), though results for
France are not presented: Portugal presents an elasticity slightly above 0.4 (also higher
than ours).
The Global Database on Intergenerational Mobility (GDIM, 2018), constructed by
Narayan etal. (2018), also presents elasticity estimates for Portugal, finding that is equal to
0.28, for each gender separately as well as together.
Mendolia and Siminsky (2019) also compute the intergenerational income correlation,
which is around 0.233 for men in Australia, similar to our findings. We can only compare
our estimates for men and women with those in the literature. None of the authors instru-
menting and predicting parental income compute the other mobility measures. For sons,
Portugal may stand amongst the most relative mobile countries in income, being similar
to the UK and Australia. Regarding daughters, it fits in all the estimates for the countries
described.
Figure 1 presents the transition probabilities between father and children (sons and
daughters) income levels in the respective generations, which complements the previous
measures.
There is a strong degree of intergenerational mobility when the father is classified as
low-income earner: the majority of individuals are likely to arrive at higher income lev-
els when adults. This means that almost no child with low-income fathers keeps that
position and the majority is able to be better off when adults. At the same time, about
75% of individuals remain in the low- and medium–low income levels. Connected with
this, the majority of children with medium–low income level earners keep that position,
which reveals that for children born below the medium–high and high-income levels, per-
sistence is high. The probability of keeping a high income level is lower than the one of
reaching a lower income level (downward mobility is high for children of high income
fathers). Besides, the upward probabilities decrease the higher the fathers’ income levels,
as expected (the more fathers earn, the less room for being surpassed by children there is).
The chances of reaching a high-income level are lower for all fathers’ income levels. The
likelihood of departing from a low-income level and reaching the highest is lower than the
opposite movement. The chances of ending up in the medium–low income level are the
highest. These are higher for females with fathers in the medium high and high-income
levels (39.93% and 36.72% for women when compared to the 39.25% and 34.01% for men,
respectively), and higher for males with fathers in the low and medium low-income levels
(39.60% and 40.96% for women when compared to the 42.83% and 42.43% for men with
fathers, respectively).
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
71
Using Survey Data toEstimate Intergenerational Mobility in…
5.2 Intergenerational Mobility inEducation
We present the benchmark results for intergenerational mobility in education in Table6.
Results show that there is a positive association between children and father’ educa-
tional attainment according to which children will have, on average, higher education levels
when their father also have higher levels of education, as reflected by the intergenerational
education correlation of 0.26. We also see that there is a 44.35% chance of individuals
born into low-educated environments reaching the highest education level, with 84.47% of
individuals attaining improving their education in comparison with the oldest generation.
This means that less than 20% of individuals keep their father’s education level or reach a
lower one, which appears to be good news for a country where, according to International
Organizations, low education levels were likely to perpetuate, i.e., there was a high chance
of individuals keeping the same education level than their parents.
The last result may be connected with the decrease in school drop-out rates between
1968 and 2018: for the primary and secondary education levels, the percentage changes
were approximately equal to 99.6% and 96.35%, respectively. The reasoning is that oppor-
tunities are equalized across individuals from different educational backgrounds as pointed
outby Narayan etal. (2018). This is in line with the description made in Clements (1999)
from the IMF to highlight an education reform that took place in the nineties and that sub-
stantially decreased school dropout rates which were found to be responsible for the low
education levels by the OECD (1995). The measures adopted by the Portuguese Govern-
ment included the expansion of preschool on which participation increased by 30% points
(from 34 to 64% between 1988 and1997/1998; the Guaranteed Minimum Income Program
which requires children from recipients to be at school; the expansion of professional
schools at the secondary level, associated with an increase in enrolled individuals of about
26 p.p. (from 13 to 29%) between 1991/1992 and 1998/1999. Some of the measures are
still in place nowadays, and the OECD (2019) corroborates that progress in the last decade
continued to be made, as continued policy efforts to reduce out-of-school rates made the
share of individuals attaining upper-secondary or post-secondary non-tertiary education
level increased 13 p.p. between 2008 and 2018.
Men are more relatively mobile than women (0.24 compared to 0.29 for the intergenera-
tional correlation) while more persistent in absolute terms (36.93% compared to 49.98%
Table 3 Correspondence between ISCED Classifications across Surveys
Adapted from Eurostat online tables (correspondence between ISCED 2011 and 1997 levels). Source http://
www. uis. unesco. org/ Educa tion/ Pages/ inter natio nalst andard- class ifica tion- of- educa tion. aspx
ECHP 1995 (ISCED 1997) EU-SILC 2019 (ISCED 2011) Retrospective
question about
parents
Less than second stage of secondary education Primary Low level
Lower secondary
Second stage of secondary education Upper secondary Medium level
Recognized third level education Short cycle tertiary High level
Bachelor or equivalent
Master or equivalent
Doctorate or equivalent
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
72
L.Clemente-Casinhas et al.
for the low to high education level probability and 82.54% compared to 85.99% for the
share of individuals with more education than their fathers). Gender differences in relative
mobility may be also related with differences in school drop-out rates, resembling the find-
ings of Hilger (2016) for the USA regarding high school enrolment. In Portugal, women
appear to have higher dropout rates than men in more than 90% of times, considering the
primary education level between 1968 and 2018. For the lower secondary level, this share
is close to 89%.26 Besides, regarding the primary level of education, the reduction in school
dropouts in the same period was more pronounced for men than for women (99% for men
Table 4 Correspondence between ISCO Classifications across Surveys
Source International Labour Organization (https:// www. ilo. org/ public/ engli sh/ bureau/ stat/ isco/)
ECHP 1995 (ISCO-88) EU-SILC 2019 (ISCO-08)
Legislators, senior officials and managers Managers
Professionals Professionals
Technicians and associate professionals Technicians and associate professionals
Clerks Clerical support workers
Service workers and shop and market sales workers Services and sales workers
Skilled agricultural and fishery workers Skilled agricultural, forestry and fishery workers
Craft and related trades workers Craft and related trades workers
Plant and machine operators and assemblers Plant and machine operators and assemblers
Elementary occupations Elementary occupations
Table 5 Benchmark Results for Intergenerational Mobility in Income
Standard errors are presented in parentheses. *,**,*** stands for statistically significant at 10%, 5%, 1%
levels, respectively. Elasticity stands for the Intergenerational Income Elasticity (IGE), Corr. stands for the
Intergenerational Income Correlation. Rank-rank stands for the Rank-Rank Slope. Prob. stands for the Bot-
tom to Top Income Level Probability. Share stands for the Share of Individuals Earning More than their
Fathers. These are described in Sect. 3.1. Only fathers are considered. Probabilities obtained using an
ordered logit and the share of individuals earning more than their fathers are expressed in %. The share
of individuals earning more than their fathers does not have an associated significance level. n stands for
the number of observations in the sample and N for the total population represented by those observations
using survey weights
Elasticity Corr Rank-rank Prob Share
All individuals
n = 2549 | N = 980,083 0.26***
(0.04) 0.20***
(0.03) 0.45***
(0.01) 7.15***
(0.01) 53.11
Males
n = 1027 | N = 431,849 0.3***
(0.05) 0.24***
(0.04) 0.48***
(0.02) 8.21***
(0.01) 52.90
Females
n = 1522 | N = 548,234 0.22***
(0.06) 0.17***
(0.04) 0.42***
(0.02) 6.42***
(0.01) 53.28
26 Data on dropout rates for the primary education level for males and females are available in https://
data. world bank. org/ indic ator/ SE. PRM. UNER. MA. ZS and https:// data. world bank. org/ indic ator/ SE. PRM.
UNER. FE. ZS, respectively. For the lower secondary level, these are presented in https:// data. world bank.
org/ indic ator/ SE. SEC. UNER. LO. MA. ZS and https:// data. world bank. org/ indic ator/ SE. SEC. UNER. LO. FE.
ZS, respectively.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
73
Using Survey Data toEstimate Intergenerational Mobility in…
and 90% for women), while these were similar between genders for the lower secondary
education level. All together may have made women present more relative education per-
sistence than men.
Some authors compute an analogous measure to our probability measure. Lam and Liu
(2019) find that for primary and lower secondary educated fathers (both in our low level
of education), the chances children have of reaching the high education level are in the
26.63–33.07% range, for both-generation Hong Kong born individuals, 32.78–40.61% for
second-generation Mainland immigrants, and 16.11–20.35% for both-generation Mainland
immigrants. Schneebaum etal. (2016) show that for Austria, this likelihood is around 8%
for males and 7% for daughters. For Portugal, Bago d’Uva and Fernandes (2017) find that
this is around 20% considering male children, which is about 17% points below our 36.93%
estimate. Although cohorts used are similar to ours (1970–1985), differences should be
noted in the methodology. They use a multinomial logit and their calculations involve the
2005 and 2011 waves of the EU-SILC. All these are below our estimates and the differ-
ences between genders are the opposite to what we obtain. The share of individuals with
more education than their fathers is higher than 80%. This value is larger than the one
found in Lam and Liu (2019) for Hong Kong-born children with Hong Kong-born fathers
(78.06%), while lower than the one for Hong Kong-born children of Mainland immigrant
fathers (89.47%). Both generation Mainland immigrants fall in the middle (86.54%). Due
to the lack of comparability in the literature that, for our measures, is scarce, we cannot
infer if Portugal has high absolute mobility in education (or not) in the World.
Education correlations are the most studied measure in the literature and mainly use
years of education instead of education levels. Considering that there may be a strong
link between years of education and education level attained, we abstract from this last
issue. Urbina (2018) is the only investigator studying the pair father-children and finds a
correlation that is between 0.45 and 0.51 regarding Mexico. As before, reported studies
often confront the analysis for each gender separately. Schneebaum etal. (2014) consider
20 European countries. In general, mobility is lower for sons (0.33) when compared to
daughters (0.26). They include Portugal in their analysis, finding values similar to ours: the
intergenerational correlation for Portugal for the pair father-son is equal to 0.24 (the same
as we obtain), while the pair father-daughter is equal to 0.26 (lower than our estimate).
This country presents the highest mobility when considering men. Regarding daughters,
Portugal is surpassed in terms of mobility by France (0.24), all the Nordic countries (aver-
age correlation of 0.20), the Anglo-Saxon countries (average correlation of 0.23), Greece
(0.22), Czech Republic (0.20), and Poland (0.21). The highest persistence value is found
for Italy (0.40). Latif (2018) shows for Canada that boys are on average less mobile than
girls with the education correlation being equal to 0.33 for boys and 0.32 for girls. Schnee-
baum etal. (2016) found that persistence appears to be greater for girls, 0.43, than for boys,
0.41, for Austria. Azam and Bhatt (2015) find that the correlation between father and son’s
education is around 0.64 for India. To sum up, Portugal is the most relatively mobile coun-
try in education for sons when considering the intergenerational correlation in education
whereas for daughters it is in the middle of known World’s estimates.
The fact that Portugal is the one presenting a larger relative change in the government
expenditures as a share of GDP may be leading our evidence (an increase around 186%
between 1968 and 201827). According to Narayan etal. (2018), this may be the result of
27 https:// data. world bank. org/ indic ator/ SE. XPD. TOTL. GD. ZS
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
74
L.Clemente-Casinhas et al.
higher public spending on education is associated with larger relative mobility in educa-
tion in richer countries, by compensating the inequality in private investments in education
between poor and rich parents, in line with the findings of Ferreira and Gignoux (2014). It
also should be connected with the policies implemented to improve enrolment in schools
which can be again considered to explain differences between countries. Portugal is the
country with the highest decrease in the school drop-out rate for men, considering primary
education, while for women it lays in the middle of the group of countries’ estimates. How-
ever, women do not maintain the same position when compared to other countries in terms
of primary school dropout rates’ decrease as the one they have regarding relative education
mobility. This reinforces the argument of Clements (1999) that early education is one of
the main drivers of educational achievement in the Portuguese economy.
Figure 2 presents the transition probabilities for education levels considering both
generations.
An interesting result emerges when we analyse the transition probabilities for inter-
generational mobility in education. The probability of staying in the same low education
level as the father is equal to 0%, i.e., individuals present full absolute mobility when
raised in a low educated environment. This result appears to be stronger than the one
found by Bago d’Uva and Fernandes (2017): noting the same differences in methodol-
ogy mentioned above, sons with low educated fathers appear to have almost 50% chance
of reaching a higher education level. When the father is classified as medium educated,
children’s chance of surpassing that level is higher than the one they have of obtaining
the same level. The probability of remaining in the same education level of the father is
higher for men regarding the medium education level and for women regarding the high
education level (45.07% compared to 24.35% for the first case and 90.62% compared to
71.72% in the second case). Moreover, the chances of completing the highest education
level is always higher for females when compared to males for all the father’s educa-
tion levels. Finally, the likelihood that an individual has of reaching or remaining in the
high-education level is increasing on the father education level, which reflects a high
persistence at the top of the education classification: this finding is similar to that pre-
sented by Bago d’Uva and Fernandes (2017), regarding sons born from 1950 on.
As we stated before, we perform several sensitivity analysis, which we present in the
Appendix2. Overall, our analysis shows that results are robust to most of the sensitivity
exercises. The rank-rank slope may be upward or downward biased, as found when per-
forming the sensitivity analysis for different instruments for parental income. The other
measures are likely to be attenuated. We also show that it is a fair choice to consider
average total household income for married individuals instead of individual income,
because household structure persistence influences the transmission of socioeconomic
status.
5.3 Results byIndividual Characteristics
Literature reports that individual’s characteristics are associated with more or less mobility.
Examples include Causa and Johansson (2010) and Gallagher etal. (2019) that find that
there is a connection between parental education and income mobility, in OECD countries
and the USA, respectively. Emran etal. (2019) and Alesina etal., (2021, 2023) also find
that different occupations or sectors of activity present heterogeneous education mobility
patterns, respectively considering India and China, and African countries. Acciari et al.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
75
Using Survey Data toEstimate Intergenerational Mobility in…
(2022) analysed Italy and distinct status in employment also appeared to be associated with
differences in income mobility. We analyse intergenerational mobility in income and edu-
cation in different subsamples according to individual characteristics to assess these pre-
vious findings. Hence, the benchmark analysis is extended by children/father education,
occupation, income levels, and employment status. This analysis allows us to understand
which characteristics are associated with more or less mobility. We are not however cap-
turing the relative importance of any of those variables, but exploring the within-group
inequality that may exist for each characteristic.
Fig. 1 Intergenerational Transition Probabilities in Income Using an Ordered Logit. Probabilities
obtained using an ordered logit are expressed in % and are all statistically significant at 1%. Parental indi-
vidual income (in logs) is predicted at the age of 40years old, with results for the first stage presented in
Table16 in the Appendix and using father’s education, occupation, and managerial position as instruments
for permanent income. Children’s income (in logs) correspond to the average of the couple’s income when
married and to individual income when not married. Results can be found in Table17
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
76
L.Clemente-Casinhas et al.
5.3.1 Education
We present the disaggregation by own and father’s education levels, in Tables7 and 8,
respectively.
In the majority of the indicators, individuals with a high education level present the
highest relative and absolute mobility in income, pointing to the possibility that there is an
absolute advantage in income of completing the highest education level.
The share of individuals earning more than their fathers is the measure for which results
appear to have opposite findings, as reported in the work of Causa and Johansson (2010)
for the OECD. The authors show that highly educated households are associated with more
relative mobility, while we observe that children whose fathers have a low education level
are the ones with higher relative mobility. The opposite happens for indicators of absolute
mobility in comparison with the entire sample. Regarding education, when fathers have
a low education level, children have more absolute mobility in comparison to the entire
sample. Also, children of high educated fathers show more persistence in income than the
sample for which we consider all individuals.
5.3.2 Occupation
Results by own occupation and father occupation categories are in Table 9 and 10
respectively.
Mobility in income is always higher than in the benchmark sample when considering
the (significant) subsamples of individuals with occupations in the following categories:
legislators, senior officials, and managers, and professionals. The opposite occurs for
skilled agricultural and fishery workers, plant and machine operators and assemblers,
and elementary occupations. Subsamples where individuals have occupations classified
in the technicians and associate professionals and also service workers and shop and
market sales workers categories are the ones presenting more relative mobility but less
absolute mobility in income in comparison with the entire sample. Regarding educa-
tion, relative mobility is higher than in the benchmark sample, except when considering
the subsamples where individuals work as skilled agricultural and fishery workers or
have elementary occupations. Absolute mobility is also lower than in the sample with
Table 6 Results for
Intergenerational Mobility in
Education
Standard errors are presented in parentheses. *,**,*** stands for sta-
tistically significant at 10%, 5%, 1% levels, respectively. Probabilities
obtained using an ordered logit and the share of individuals with more
education than their fathers are expressed in %. The share of individu-
als with more education than their fathers does not have an associated
significance level. n stands for the number of observations used and N
for the total population represented by those observations using survey
weights
Correlation Prob Share
All individuals
n = 2549 | N = 980,083 0.26*** 44.35***
(0.01) 84.47
Males
n = 1027 | N = 431,849 0.24*** 36.93***
(0.02) 82.54
Females
n = 1522 | N = 548,234 0.29*** 49.98***
(0.02) 85.99
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
77
Using Survey Data toEstimate Intergenerational Mobility in…
all individuals for the technicians and associate professionals’ category and also for the
elementary occupations.
In comparison with the entire sample, when fathers’ occupations are classified in the
legislators, senior officials, and managers as well in the professional’s category, children
always have lower relative mobility in income. In turn, the absolute income mobility
is always higher than in the benchmark case when fathers belong to clerks and skilled
agricultural and fishery workers occupations. Regarding the rank-rank slope, most of
the professional categories show lower mobility, with exceptions being for technicians
and associate professionals. Relative mobility in education presented in each subsam-
ple is higher than in the benchmark case for legislators, senior officials, and manag-
ers, technicians, and associate professionals, and craft and related trades workers, but
lower for professionals. Absolute mobility in education is higher than in the benchmark
subsample when children have fathers working as clerks, but lower than in the entire
sample when considering the subsample of children whose fathers are classified as
professionals.
5.3.3 Income Level
Tables11 and 12 present the results by own and father income levels, respectively.
Mobility in income is in most cases higher than the one verified in the entire sample.
For the majority of measures of intergenerational mobility in education, relative mobility
is also higher in each income level partition when compared to the entire sample, while
absolute mobility is always higher than in the benchmark case when individuals’ income
belongs to the medium–high category.
Only when fathers are classified as medium–low income earners do children present
more (relative and absolute) mobility in income than the one in the entire sample. When
individuals have fathers in medium–high- and high-income levels, (relative and absolute)
mobility in income appears to be lower than in the benchmark case. Relative mobility in
education is higher than the entire sample when parents have a medium–low income level
and is lower than in the entire sample when parents have a high-income level. In terms
of absolute mobility in education, for children of parents with medium–high- and high-
income level, mobility is higher.
5.3.4 Status inEmployment
In this section we analyse children by their status in employment (either self-employed or
employees), with results in Table13.28
Overall, children who are self-employed present higher persistence in income and edu-
cation when compared to the entire sample, while the opposite occurs for the subsample of
children who are employees. For the first group the exception is the low to high education
level probability. For the last group, lower mobility is verified for the bottom to top income
level probability and the low to high education level probability.
28 This exercise cannot be performed by fathers’ employment status, because we cannot distinguish in the
sample of children the cases for which their fathers were employees only, employers only, or both.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
78
L.Clemente-Casinhas et al.
5.4 Policy Directions
Although Portugal exhibits a positive evolution in terms of intergenerational mobility, the
above results allow us to conclude that differences across groups in the Portuguese econ-
omy still exist. This research aims to present the current framework of intergenerational
mobility in income and education. Some of our findings are in line with the results of exist-
ing research that point to possible mechanisms for the evidence presented and therefore
are highlighted when applicable. This does not mean that we are defining causality for the
evidence presented, which should be a topic of further research. However, given that there
is a likelihood that these factors are possible explanations for the results we get, we address
aligned possible policy directions, that should be properly evaluated in what concerns effi-
ciency and applicability, given that these are context-dependent. These are targeted at clos-
ing the gap between advantaged and disadvantaged individuals by improving the outcomes
of the latter. All of these find support in the World Bank report of Narayan etal. (2018).
First, early childhood development should be part of policymakers’ agenda, as it is
supposed to make children succeed in school, developing skills that are rewarded later in
life through, for example, productivity. A positive empirical relationship with education
Fig. 2 Intergenerational Transition Probabilities Using an Ordered Logit. Probabilities obtained
using an ordered logit are expressed in % and are all statistically significant at 1%. Results can be found in
Table18
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
79
Using Survey Data toEstimate Intergenerational Mobility in…
mobility is also found in the works of Bauer and Riphahn (2006) for Switzerland and
Daude and Robano (2015) for Latin American countries. This also finds support in the
theory designed by Daruich (2018) on which the macroeconomic life-cycle general-equi-
librium model incorporates parental investments in their children’s skills through time
and money over several periods. Early childhood development is considered to improve
social mobility. It gains relevance since differences and changes in school dropouts may be
responsible for the results we present, particularly for the early stages of education. It rein-
forces the idea of Clements (1999) that with the expansion of preschool as a part of early
childhood development, higher educational attainment in a country where low education
levels were likely to perpetuate is now in place.
The work of Clements (1999) also shows that offering technology and vocational
courses through the regular education system may have led to a considerable increase
in the secondary school enrolment rate, which appears to be improved according to the
OECD (2019). This highlights the importance of our second policy direction which is the
access to education as well as its quality, as found by Chetty etal., (2014, 2020a, 2020b),
Chetty and Hendren (2018b) and Hilger (2016) for the USA, Acciari etal. (2022) for Italy,
and Nimubona and Vencatachellum (2007) for South Africa. All together appear to ground
our result on which more than 80% of individuals attain a higher education level than their
fathers and a chance of about 44% of being raised in a low educated environment and
reaching the highest education level.
The third policy implication is the basis of the other already described. Now we have
the relevance of an efficient investment of public resources in education, by equaliz-
ing opportunities across individuals from different backgrounds. First, there is evidence
that government expenditures on early education are positively connected with education
mobility in the works of Daude and Robano (2015) for Latin America, Urbina (2018) for
Mexico and Lee and Lee (2020). This supports the points raised for early childhood devel-
opment-related policies and their role in achieving better outcomes in the country. For gov-
ernment total spending on education, the same is presented by Chu and Lin (2020) for Tai-
wan regarding income mobility, Daude and Robano (2015) for Latin American countries,
Latif (2017) for Canada, and Urbina (2018) for Mexico regarding education mobility. In
the theoretical model of Solon (2004), it is also shown that a progressive public investment
Table 7 Results by Own Education Level
Standard errors are in parentheses. *,**,*** stands for statistically significant at 10%, 5%, 1% levels,
respectively. Probabilities obtained using an ordered logit and the share of individuals earning more/with
more education than their fathers are expressed in %. Correlations and the share of individuals earning
more/with more education than their fathers does not have an associated significance level. n stands for the
number of observations used and N for the total population represented by those observations using survey
weights
Intergenerational Mobility in Income Intergenerational
Mobility in Educa-
tion
Own Education Level Elasticity Corr Rank-rank Prob Share Corr Prob Share
Medium
n = 1291 | N = 477,439 0.19***
(0.05) 0.13***
(0.04) 0.33***
(0.04) 4.85***
(0.01) 49.44 – – 87.51
High
n = 1258 | N = 502,644 0.18***
(0.05) 0.20***
(0.06) 0.3***
(0.02) 11.51***
(0.02) 56.6 – – 87.00
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
80
L.Clemente-Casinhas et al.
in human capital will decrease the income elasticity’s steady state value, contributing to
promote mobility in income. Our findings suggest that more mobility is associated with a
higher education level, so investing in that stage of education is of utmost importance in
the Portuguese context.
Connected with higher mobility being verified in high education levels but also in
occupations requiring a high education level, easing the access to capital markets through
credit to finance not only education but also entrepreneurial undertakings may be another
strategy to implement. Becker etal. (2018) developed a stylized model of intergenera-
tional mobility, which is connected with cross-sectional inequality. The authors show that
even without credit constraints or innate ability differences, richer parents invest more in
their children’s education in comparison with poorer ones which reduces intergenerational
mobility. Therefore, as pointed in the work of Piketty (2000), it is expected that with credit-
constrained economies, as it is the common case, persistence becomes stronger, because
parental investments are constrained by the availability of resources, a restriction which
impacts to a greater extent the poor.
Policymakers can also target education and income mobility to promote a feedback
effect, having a long-term perspective. This is because if the current generation sees its
mobility in income improved, their ability to invest in their children is promoted, which
impacts education mobility and therefore income mobility again.
6 Conclusion
Published work on intergenerational mobility in Europe has been focused on Scandinavian
countries while research on Southern Europe is still limited. In this group, literature is scarce
for Portugal in terms of income mobility, although some developments have been made regard-
ing the study of educational mobility. Our work analyses intergenerational mobility in income
and education for this country by constructing several relative and absolute measures of inter-
generational mobility. For income mobility we compute the intergenerational income elasticity,
Table 8 Results by Father Education Level
Standard errors are in parentheses. *,**,*** stands for statistically significant at 10%, 5%, 1% levels,
respectively. Probabilities obtained using an ordered logit and the share of individuals earning more/with
more education than their fathers are expressed in %. Correlations and the share of individuals earning
more/with more education than their fathers does not have an associated significance level. n stands for the
number of observations used and N for the total population represented by those observations using survey
weights
Intergenerational Mobility in Income Intergenerational
Mobility in Education
Father Education Level Elasticity Corr Rank-rank Prob Share Corr Prob Share
Low
n = 2040 | N = 745,593 0.2***
(0.05) 0.12***
(0.03) 0.37***
(0.02) 6.74***
(0.01) 44.38 – – 100.00
Medium
n = 275 | N = 124,035 0.36**
(0.15) 0.19**
(0.08) 0.49***
(0.08) 9.84*
(0.03) 49.44 – – 66.34
High
n = 234 | N = 110,455 0.22
(0.35) 0.06
(0.09) 1.66***
(0.22) – 17.69 – – 0.00
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
81
Using Survey Data toEstimate Intergenerational Mobility in…
Table 9 Results by Own Occupation Category
Standard errors are in parentheses. *,**,*** stands for statistically significant at 10%, 5%, 1% levels, respectively. Probabilities obtained using an ordered logit and the share
of individuals earning more/with more education than their fathers are expressed in %. Correlations and the share of individuals earning more/with more education than their
fathers does not have an associated significance level. n stands for the number of observations used and N for the total population represented by those observations using sur-
vey weights
Intergenerational Mobility in Income Intergenerational Mobility in
Education
Own occupation Elasticity Corr Rank-rank Prob Share Corr Prob Share
Legislators, senior officials, and managers
n = 183 | N = 82,768 0.25(0.15) 0.18(0.10) 0.29***(0.08) 17.87***(0.06) 59.87 0.2*** 61.71***(0.05) 82.68
Professionalsn = 858 | N = 328,837 0.15**(0.07) 0.12**(0.06) 0.21***(0.03) 14.87***(0.02) 59.5 0.03 95.27***(0.01) 80.46
Technicians and associate professionals
n = 410 | N = 157,272 0.2**(0.09) 0.17**(0.08) 0.36***(0.07) 6.03***(0.02) 52.21 0.2*** 33.78***(0.04) 81.45
Clerks
n = 298 | N = 110,206 0.02(0.11) 0.02(0.09) 0.11(0.07) 4.51**(0.02) 53.34 0.24** 19.85***(0.03) 88.38
Service workers and shop and market sales workers
n = 474 | N = 163,753 0.09(0.08) 0.07(0.06) 0.18**(0.07) 2.62**(0.01) 41.91 0.04 14.67***(0.02) 89.81
Skilled agricultural and fishery workersn = 12 | N = 3318 0.13(0.59) 0.11(0.49) 1.7**(0.75) - 48.5 0.68* 3.25(0.04) 84.88
Craft and related trades workers
n = 103 | N = 44,842 0.31(0.2) 0.19(0.13) 0.32***(0.09) 3.21(0.03) 53.14 0.33 9.78**(0.04) 95.34
Plant and machine operators and assemblers
n = 94 | N = 48,608 0.13(0.09) 0.13(0.10) 0.22(0.19) 1.82(0.01) 50.56 0.17 10.01**(0.04) 86.27
Elementary occupations
n = 103 | N = 34,717 − 0.03(0.12) − 0.02(0.08) − 0.01(0.11) 1.2*(0.01) 36.21 0.38** 9.14**(0.05) 81.92
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
82
L.Clemente-Casinhas et al.
Table 10 Results by Father Occupation Category
Standard errors are presented between parentheses. *,**,*** stands for statistically significant at 10%, 5%, 1% levels, respectively. Probabilities obtained using an ordered
logit and the share of individuals earning more/with more education than their fathers are expressed in %. Correlations and the share of individuals earning more/with more
education than their fathers do not have an associated significance level. n stands for the number of observations used and N for the total population represented by those
observations using survey weights
Intergenerational Mobility in Income Intergenerational Mobility in
Education
Father occupation Elasticity Corr Rank-rank Prob Share Corr Prob Share
Legislators, senior officials, and managersn = 126 | N = 64,302 0.46**(0.22) 0.23**(0.11) 0.75***(0.06) 7.85*(0.05) 41.59 0.16* 71.86***(0.06) 67.36
Professionals
n = 213 | N = 98,992 0.06(0.27) 0.02(0.09) 0.69***(0.07) 4.42(0.14) 14.58 0.34*** 39.18***(0.11) 20.51
Technicians and associate professionalsn = 403 | N = 170,812 0.27*(0.15) 0.12*(0.07) 0.2***(0.06) 15.33*(0.08) 33.14 0.12* 54.01***(0.04) 82.66
Clerksn = 186 | N = 65,185 0.17(0.23) 0.06(0.08) − 0.47***(0.16) 11.92*(0.07) 45.07 0.11 54.65***(0.06) 94.21
Service workers and shop and market sales workersn = 359 |
N = 114,019 0.36**(0.15) 0.16**(0.07) 0.69***(0.13) 5.55***(0.02) 41.14 0.1 42.18***(0.04) 93.40
Skilled agricultural and fishery workersn = 122 | N = 26,053 − 0.17
(0.35) − 0.05
(0.10) − 1.39***
(0.25) 7.03
(0.05) 80.67 − 0.07 31.83***
(0.07) 97.24
Craft and related trades workersn = 593 | N = 246,746 0.12(0.17) 0.04(0.05) 0.13(0.1) 6.2***(0.02) 77.5 0.11* 43.24***(0.03) 96.17
Plant and machine operators and assemblersn = 362 |
N = 143,457 0.49**(0.23) 0.01**(0.003) 0.71***(0.04) 5.37**(0.02) 62.59 0.04 35.45***(0.03) 99.01
Elementary occupationsn = 185 | N = 50,517 − 0.61(0.6) − 0.12(0.12) − 1.32***(0.41) 5.54**(0.02) 87.96 0.18 - 100
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
83
Using Survey Data toEstimate Intergenerational Mobility in…
the intergenerational correlation coefficient, the rank-rank slope, the share of individuals earning
more than their fathers, and the bottom to top income level probability. For education mobility
we compute the intergenerational education (Pearson) correlation, the low to high education level
probability, and the share of individuals with more education than their fathers. Both income
and education mobility measures are complemented by ordered logit transitions matrices. We
uncover the patterns that exist and which individual characteristics present more or less mobility,
for individuals born in 1968–1988. Both genders are considered. Two Portuguese components of
European datasets are used: the European Community Household Panel and the European Union
Statistics on Income and Living Conditions.
Our benchmark results reveal gender differences, showing that women generally pre-
sent higher mobility in income than men, a finding also for Russia, Denmark, the USA,
Italy, and in other studies that included Portugal. When considering transition probabili-
ties between income levels, we observe that there is a strong degree of intergenerational
mobility when fathers are at the low-income level. At the same time, persistence is high for
children born below the medium–high and high income levels. Additionally, upward prob-
abilities decrease the higher the father’s income level. Our value estimates are according to
estimates previously done. As in the case of income, women have a higher probability of
passing from a low to a high education level than men, with previous studies for Portugal
reaching lower probabilities than ours. In Portugal the share of individuals with more edu-
cation than their fathers is higher than 80% and the probability of staying in a low educa-
tion level, if that is the case of the father, is 0%, a finding that improved relative to other
estimates for Portugal and is higher than previous estimates for Hong Kong and Austria.
The likelihood that an individual has of reaching or remaining in the high-education level
is increasing on the father’s education level, confirming published findings.
We analyse intergenerational mobility in different subsamples according to individual charac-
teristics to check which own and father’s characteristics are associated with more or less income
and education mobility when compared to the benchmark sample. We assess characteristics
such as education level, occupation, income level, and status in employment. Contrary to what is
found in the literature, individuals with a high education level show higher income and education
Table 11 Results by Own Income Level
Standard errors are in parentheses. *,**,*** stands for statistically significant at 10%, 5%, 1% levels,
respectively. Probabilities obtained using an ordered logit and the share of individuals earning more/with
more education than their fathers are expressed in %. Correlations and the share of individuals earning
more/with more education than their fathers does not have an associated significance level. n stands for the
number of observations used and N for the total population represented by those observations using survey
weights
Intergenerational Mobility in Income Intergenerational Mobility in
Education
Own income level Elasticity Corr Rank-rank Prob Share Corr Prob Share
Low
n = 609 | N = 230,374 -0.09
(0.1) -0.07
(0.08) -0.03***
(0.01) – 11.55 0.26*** 28.11***
(0.03) 87.15
Medium–low
n = 1007 | N = 390,927 0.03**
(0.01) 0.10**
(0.04) 0.07***
(0.02) – 55.00 0.22*** 37.31***
(0.02) 87.28
Medium–high
n = 602 | N = 228,948 0.03***
(0.01) 0.14***
(0.05) 0.06***
(0.01) – 71.68 0.22*** 64.62***
(0.03) 85.23
High
n = 331 | N = 129,834 0.14***
(0.05) 0.21***
(0.07) 0.06***
(0.01) – 88.43 0.19*** 69.05***
(0.04) 69.93
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
84
L.Clemente-Casinhas et al.
Table 12 Results by Father Income Level
Standard errors are in parentheses. *,**,*** stands for statistically significant at 10%, 5%, 1% levels, respectively. Probabilities obtained using an ordered logit and the share
of individuals earning more/with more education than their fathers are expressed in %. Correlations and the share of individuals earning more/with more education than their
fathers does not have an associated significance level. n stands for the number of observations used and N for the total population represented by those observations using sur-
vey weights
Intergenerational Mobility in Income Intergenerational Mobility in
Education
Father income
level Elasticity Corr Rank-rank Prob Share Corr Prob Share
Low
n = 275 |
N = 68,264
5.33
(4.48) 0.18
(0.15) 6.18*
(3.17) - 89.27 0.07 35.68***
(0.05) 98.95
Medium–low
n = 1262 |
N = 488,814
0.12
(0.12) 0.03
(0.03) 0.22***
(0.02) - 67.86 0.13*** 40.68***
(0.02) 97.84
Medium–high
n = 451 |
N = 182,071
0.41
(0.37) 0.08
(0.07) 0.82***
(0.17) - 43.8 0.04 58.67***
(0.04) 91.08
High
n = 561 |
N = 240,934
0.22
(0.17) 0.08
(0.06) 0.63***
(0.08) - 19.99 0.33*** 47.62***
(0.04) 48.25
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
85
Using Survey Data toEstimate Intergenerational Mobility in…
mobility. This is a further advantage of having more education and corroborates the findings for
occupations in the legislators, senior officials, and managers and professionals’ categories for
which mobility in income is higher: they require a higher education level than occupations as
skilled agricultural and fishery workers, and plant and machine operators and assemblers, which
show lower income mobility when compared to the benchmark sample. Also, individuals with
elementary occupations always present lower mobility and mobility in education is higher when
fathers work as clerks. Also, a finding against previous literature, children whose fathers have
a low education level are those presenting higher relative income mobility. Mobility in income
and education is higher for individuals in the medium–high income level and more mobility in
education occurs for these individuals when fathers also belong to the medium–high-income
level category. However, medium–low income fathers bring more mobility in income to their
offspring. Self-employed individuals present lower income mobility when compared to the entire
sample. Vis-à-vis these results and noting that any policy-making targeting mobility improve-
ments needs a strong study justifying it (meaning we are not addressing causality here), policies
such as the ones proposed in Narayan etal. (2018), which promote early childhood development,
provide access to quality education, aim to reduce segregation, strengthen institutions and infra-
structures, and an efficient public investment in education should help to close the gap between
advantaged and disadvantaged individuals and serve as basis for new research on the topic. The
same applies to the ease of access to capital markets and a robust economic growth which should
have a feedback effect of education and income mobility in future generations.
We have also performed sensitivity analyses to some of our initial methodological hypoth-
eses, namely in terms of income definition (average total income and individual income, parental
income, inclusion of individuals with no individual income, co-resident bias, presence of sib-
lings and attenuation bias) to determine if our benchmark results hold. There is some degree of
intergenerational persistence in household structure, as reported in previous literature. Further-
more, if we consider four years, instead of one year, to estimate parental income, the designated
attenuation bias, this changes the results. However, neither the inclusion of individuals with no
individual income or taking in consideration two generations living in the same home affect our
benchmark results.
Some shortcomings can be pointed to our work. First, the datasets we use do not provide
direct information on father’s income when children were around 14years old. Following the
literature, we predict parental income which is not observed, through education, occupation,
Table 13 Results by Status in Employment
Standard errors are in parentheses. *,**,*** stands for statistically significant at 10%, 5%, 1% levels,
respectively. Probabilities obtained using an ordered logit and the share of individuals earning more than
their fathers are expressed in %. Correlations have the associated significance level of the elasticities used
to compute them. The share of individuals earning more than their fathers does not have an associated sig-
nificance level. n stands for the number of observations used and N for the total population represented by
those observations using survey weights
Intergenerational Mobility in Income Intergenerational Mobility
in Education
Elasticity Corr Rank-rank Prob Share Corr Prob Share
Self-employed
n = 155 | N = 70,871 0.44**
(0.18) 0.23**
(0.10) 0.62***
(0.11) 6.76**
(0.03) 53.11 0.29*** 53.18***
(0.06) 80.64
Employee
n = 2260 | N = 851,447 0.23***
(0.04) 0.19***
(0.04) 0.41***
(0.02) 6.61***
(0.01) 53.89 0.25*** 42.66***
(0.02) 85.70
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
86
L.Clemente-Casinhas et al.
and managerial position information, i.e., by using a pseudo-parent’s sample, which has impli-
cations for results. Second, our datasets provide a set of retrospective questions about parental
characteristics, which allow us to predict their income. But the range of available characteristics
is insufficient: the higher the number of instruments to proxy for parental permanent income,
the more unique values parental income could assume, which increases the heterogeneity of the
pseudo-parents’ sample. Additionally, there is the potential problem of missing variables, such as
industries or sectors of activity and years of experience. However, there are no more retrospective
questions reported by children in the EU-SILC survey that could be used to predict their par-
ents’ individual income. Regional proxies are a simple example that would fill this need. Third,
it is possible to follow individuals in both generations over time, but for the children’s subsample
this is done at the expense of no retrospective questions about parents, and hence mobility could
not be computed. This implies that analyses that require addressing temporal behaviours for
the measures we compute to complement our cross-sectional framework cannot be performed.
Although we try to make the possible adaptations, the final analysis of the effect of education
mobility on income mobility is not performed using the same sample as in the other analysis,
which can influence the results. Overcoming these problems would improve our work, although
it is a difficult task, since the majority of problems are due to the nature of the data supplied by
the existing surveys. Finally, the biases we address are mainly studied in the literature for relative
mobility measures, but the restrictions to avoid them should influence absolute mobility meas-
ures as well. This topic needs further research. As before, we also recognize the need for future
research to investigate what drives mobility in Portugal to uncover which specific policy actions
should take place to improve mobility in this country. This can be done by studying the role of
different mediators of the relationship between parents and children’s incomes, as in the work of
Blanden etal. (2005). Finally, given our evidence and that Deutscher and Mazumder (2023) find
intergenerational persistence measures and inequality of opportunity to have a strong correlation,
we consider that there should be a further analysis with a more profound study about the role of
inequality of opportunity in shaping intergenerational mobility. Some research on inequality of
opportunity has covered Portugal (e.g., Carranza, 2022), although the relationship between both
has not been examined.
Appendix1: Descriptive Tables
See Tables14, 15, 16, 17 and 18
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
87
Using Survey Data toEstimate Intergenerational Mobility in…
Table 14 Summary (unweighted) statistics: pseudo-parents sample
Variables Obs Mean Std. Dev Min Max
Age 1025 39.23 5.85 30 50
Education: Low level 1025 0.84 0.36 0 1
Education: Medium level 1025 0.11 0.31 0 1
Education: High level 1025 0.05 0.21 0 1
Main occupation: Legislators, senior officials and managers 1025 0.02 0.13 0 1
Main occupation: Professionals 1025 0.04 0.20 0 1
Main occupation: Technicians and associate professionals 1025 0.09 0.29 0 1
Main occupation: Clerks 1025 0.10 0.30 0 1
Main occupation: Service workers and shop and market sales workers 1025 0.14 0.35 0 1
Main occupation: Skilled agricultural and fishery workers 1025 0.04 0.20 0 1
Main occupation: Craft and related trades workers 1025 0.29 0.45 0 1
Main occupation: Plant and machine operators and assemblers 1025 0.16 0.37 0 1
Main occupation: Elementary occupations 1025 0.12 0.32 0 1
Managerial position: Supervisory 1025 0.07 0.26 0 1
Managerial position: Non-supervisory 1025 0.93 0.26 0 1
Individual income (in logs) 1025 9.07 0.56 6.31 11.03
Table 15 Summary (unweighted) statistics
Variables Obs Mean Std. Dev Min Max
Children’s characteristics
Age 2549 40.70 5.93 30 50
Education: low level 2549 0 0 0 0
Education: medium level 2549 0.51 0.50 0 1
Education: high level 2549 0.49 0.50 0 1
Individual income (in logs) 2549 9.10 0.73 4.54 11.64
Father’s characteristics (recalled by children)
Education: low level 2549 0.80 0.40 0 1
Education: medium level 2549 0.11 0.31 0 1
Education: high level 2549 0.09 0.29 0 1
Main occupation: Legislators, senior officials and managers 2549 0.05 0.22 0 1
Main occupation: Professionals 2549 0.08 0.28 0 1
Main occupation: Technicians and associate professionals 2549 0.16 0.36 0 1
Main occupation: Clerks 2549 0.07 0.26 0 1
Main occupation: Service workers and shop and market sales work-
ers 2549 0.14 0.35 0 1
Main occupation: Skilled agricultural and fishery workers 2549 0.05 0.21 0 1
Main occupation: Craft and related trades workers 2549 0.23 0.42 0 1
Main occupation: Plant and machine operators and assemblers 2549 0.14 0.35 0 1
Main occupation: Elementary occupations 2549 0.07 0.26 0 1
Managerial position: Supervisory 2549 0.28 0.45 0 1
Managerial position: Non-supervisory 2549 0.72 0.45 0 1
Father’s predicted individual income (in logs) 2549 9.25 0.62 3.95 11.94
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
88
L.Clemente-Casinhas et al.
Table 16 First stage results: 1995 pseudo-parents’ sample
Variable Benchmark Education Occupation Managerial position Education
and occupa-
tion
Education and
managerial posi-
tion
Occupation and
managerial posi-
tion
Education: Low level (omitted)
Education: Medium level 0.10*
(0.06) 0.32***
(0.08) 0.12**
(0.06) 0.26***
(0.07)
Education: High level 0.46***
(0.13) 1.10***
(0.06) 0.53***
(0.15) 0.84***
(0.08)
Main occupation: Managers (omitted)
Main occupation: Professionals 0.33**
(0.17) 0.38*
(0.22) 0.18
(0.18) 0.52***
(0.20)
Main occupation: Technicians and associate professionals 0.19
(0.19) -0.20
(0.22) -0.02
(0.21) 0.05
(0.20)
Main occupation: Clerks -0.001
(0.19) -0.51**
(0.22) -0.29
(0.21) -0.17
(0.20)
Main occupation: Service workers and shop and market
sales workers 0.05
(0.20) -0.46**
(0.22) -0.21
(0.22) -0.15
(0.20)
Main occupation: Skilled agricultural and fishery workers -0.54**
(0.22) -1.09***
(0.24) -0.83***
(0.24) -0.74***
(0.23)
Main occupation: Craft and related trades workers -0.28
(0.19 -0.82***
(0.21) -0.56***
(0.21) -0.48**
(0.20)
Main occupation: Plant and machine operators and assem-
blers -0.20
(0.20) -0.73***
(0.21) -0.47**
(0.22) -0.40**
(0.20)
Main occupation: Elementary occupations -0.55***
0.21) -1.09***
(0.22) -0.84***
(0.23) -0.75***
(0.21)
Managerial Position: Supervisory 0.40***
(0.08) 0.86***
(0.07) 0.52***
(0.05) 0.53***
(0.08)
Managerial Position: Non-supervisory (omitted)
Age -0.002 0.01
(0.06) 0.01
(0.05) 0.01
(0.06) 0.002
(0.05) 0.005
(0.05) -0.0001
(0.05)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
89
Using Survey Data toEstimate Intergenerational Mobility in…
Table 16 (continued)
Variable Benchmark Education Occupation Managerial position Education
and occupa-
tion
Education and
managerial posi-
tion
Occupation and
managerial posi-
tion
Age20.00005
(0.001) -0.0001
(0.001) -0.0001
(0.001) -0.0001
(0.001) -6.43e-06
(0.001) 0.00001
(0.001) 0.00001
(0.001)
Intercept 9.20***
(0.95) 8.61***
(1.05) 9.60***
(0.98) 8.64***
(1.06) 9.37***
(0.98) 8.78***
(1.01) 9.38***
(0.95)
No. Of observations 1025 1025 1025 1025 1025 1025 1025
Total Population 774,800 774,800 774,800 774,800 774,800 774,800 774,800
R2
43.13% 26.25% 38.65% 20.28% 40.45% 31.75% 41.83%
Standard errors are in parentheses. *,**,*** stands for statistically significant at 10%, 5%, 1% levels, respectively. Parental individual income (in logs) is predicted at the age
of 40 years old
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
90
L.Clemente-Casinhas et al.
Table 17 Predicted Probabilities for Income Mobility using an Ordered Logit
Standard errors are in parentheses. *,**,*** stands for statistically significant at 10%, 5%, 1% levels,
respectively. Probabilities are expressed in %. n stands for the number of observations used and N for the
total population represented by those observations using survey weights
Father Income Levels Children’s Income Levels
Low Medium–low Medium–high High
All individuals Low
n = 275 | N = 68,264 35.88***
(0.02) 40.86***
(0.01) 16.11***
(0.01) 7.15***
(0.01)
Medium–low
n = 1262 | N = 488,814 27.27***
(0.01) 41.59***
(0.01) 20.83***
(0.01) 10.31***
(0.01)
edium–high
n = 451 | N = 182,071 20.08***
(0.01) 39.62***
(0.01) 25.65***
(0.01) 14.65***
(0.01)
High
n = 561 | N = 240,934 14.41***
(0.01) 35.41***
(0.02) 29.79***
(0.02) 20.39***
(0.02)
Males Low
n = 98 | N = 29,180 31.65***
(0.03) 42.83***
(0.02) 17.32***
(0.02) 8.20***
(0.01)
Medium–low
n = 484 | N = 203,553 23.47***
(0.02) 42.43***
(0.02) 22.21***
(0.02) 11.89***
(0.01)
Medium–high
n = 185 | N = 81,735 16.88***
(0.02) 39.25***
(0.02) 26.94***
(0.02) 16.93***
(0.02)
High
n = 260 | N = 117,382 11.85***
(0.02) 34.01***
(0.02) 30.59***
(0.02) 23.55***
(0.03)
Females Low
n = 177 | N = 39,084 38.56***
(0.03) 39.6***
(0.02) 15.42***
(0.02) 6.42***
(0.01)
Medium–low
n = 778 | N = 285,261 30.07***
(0.02) 40.96***
(0.02) 19.87***
(0.01) 9.10***
(0.01)
Medium–high
n = 266 | N = 100,336 22.76***
(0.02) 39.93***
(0.02) 24.56***
(0.02) 12.75***
(0.01)
High
n = 301 | N = 123,552 16.79***
(0.02) 36.72***
(0.02) 28.91***
(0.02) 17.58***
(0.02)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
91
Using Survey Data toEstimate Intergenerational Mobility in…
Appendix2: Sensitivity Analysis
In this section, we check how sensitive our benchmark estimates of intergenerational
mobility are to changes in variables definitions and sample construction.
Income Denitions
We start by considering different income definitions. We use individual income as opposed
to the benchmark estimation in which the average couples’ income is used. We are able
to do this exercise just for children, not only because the characteristics used to proxy for
father’s permanent income pertain to individual income, but also because the father’s mari-
tal status is not known. Results are presented in Table19.
The intergenerational income elasticity and the share of individuals earning more than
their fathers are the ones for which there is only a slight increase in persistence compared
to the benchmark (and therefore they may be considered as reasonably robust to the income
definition). On other hand, when analysing the intergenerational income correlation and the
rank-rank slope, one may conclude that there is a change of about 10% and 13%, respec-
tively, meaning that persistence is higher in the first scenario. The bottom to top income
level probability shows a 31% increase between the two cases (7.15% in the benchmark
compared to 9.36% when using an individual measure of income), which is the biggest
Table 18 Predicted Probabilities
for Education Mobility using an
Ordered Logit
Standard errors are in parentheses. *,**,*** stands for statistically
significant at 10%, 5%, 1% levels, respectively. Probabilities are
expressed in %. n stands for the number of observations used and N
for the total population represented by those observations using survey
weights
Father Education Levels Children’s Education
Levels
Medium High
All individuals Low
n = 2040 | N = 745,593 55.65***
(0.01) 44.35***
(0.01)
Medium
n = 275 | N = 124,035 34.41***
(0.02) 65.59***
(0.02)
High
n = 234 | N = 110,455 17.98***
(0.03) 82.02***
(0.03)
Males Low
n = 800 | N = 321,123 63.07***
(0.02) 36.93***
(0.02)
Medium
n = 131 | N = 60,726 45.07***
(0.03) 54.93***
(0.03)
High
n = 96 | N = 50,001 28.28***
(0.05) 71.72***
(0.05)
Females Low
n = 1240 | N = 424,470 50.02***
(0.02) 49.98***
(0.02)
Medium
n = 144 | N = 63,309 24.35***
(0.02) 75.65***
(0.02)
High
n = 138 | N = 60,454 9.38***
(0.02) 90.62***
(0.02)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
92
L.Clemente-Casinhas et al.
change. This is in line with relative persistence increases when using household related
measures in the work of Murray etal. (2018). Although the most obvious reason for the
correlation change is related with the increase in the variability in children’s income, when
considering individual income, it is likely that assortative mating had play its role for both
measures: this is the process according to which individuals select a partner with similar
backgrounds. Torche (2015) argues that if the characteristics of individuals with which one
shares a life are approximately the same, it is therefore expected that persistence will be
higher in those cases, when compared to the scenario for which this type of mating does
not occur. A simple exercise allows us to have a clue on the likelihood this has of occurring
in our estimation sample. About 32.19% of individuals who are married and have fathers
in the medium–high and high-income levels, have selected individuals with fathers in those
same levels. The scenario is more evident when considering married individuals with
parents in the low and medium–low income levels, with that share being approximately
43.74%.
When individual income is considered, men have the intergenerational income correla-
tion and the rank-rank slope decreasing more than women. Persistence increases when the
couples’ income is considered, with a higher percentage change for men. This may reflect
the fact that men are more likely to be married to individuals with similar backgrounds
than women. From the medium–low parental income level on, the shares of men in this
situation are approximately 51.37, 31.29, and 34.12% against 48.95, 29.14, and 33.43% for
women. The exception is the low level, where women surpass men by 2% points, with a
share equal to 4%. Interestingly, absolute persistence for men increases when the couples’
income is used in comparison with individual income (the opposite occurs for women, who
benefit in terms of mobility when average couples’ income is considered).
All in all, the exercise of using individual income instead of average couple’s
income provides different results from the benchmark analysis. This reinforces our
Table 19 Sensitivity of intergenerational mobility in income to alternative income definitions for the bench-
mark sample
Standard errors are in parentheses. *,**,*** stands for statistically significant at 10%, 5%, 1% levels,
respectively. Probabilities obtained using an ordered logit and the share of individuals earning more than
their fathers are expressed in %. Correlations have the associated significance level of the elasticities used
to compute them. The share of individuals earning more than their fathers does not have an associated sig-
nificance level. n stands for the number of observations used and N for the total population represented by
those observations using survey weights
Income definitions for
children Elasticity Correlation Rank-rank slope Prob Share
All individuals
n = 2549
N = 980,083
Average total family
income 0.26***
(0.04) 0.20***
(0.03) 0.45***
(0.01) 7.15***
(0.01) 53.11
Individual income only 0.27***
(0.05) 0.18***
(0.03) 0.39***
(0.01) 9.36***
(0.01) 52.30
Males
n = 1027
N = 431,849
Average total family
income 0.30***
(0.05) 0.24***
(0.04) 0.48***
(0.02) 8.21***
(0.01) 52.90
Individual income only 0.27***
(0.07) 0.18***
(0.05) 0.31***
(0.02) 16.1***
(0.02) 59.30
Females
n = 1522
N = 548,234
Average total family
income 0.22***
(0.06) 0.17***
(0.04) 0.42***
(0.02) 6.42***
(0.01) 53.28
Individual income only 0.23***
(0.06) 0.16***
(0.04) 0.40***
(0.02) 5.14***
(0.01) 46.49
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
93
Using Survey Data toEstimate Intergenerational Mobility in…
decision to consider average total income instead of individual income only, since the
marital status of individuals plays a role. In other words, the point made by Chad-
wick and Solon (2002) is clear: there is some degree of intergenerational persistence in
household structure which cannot be ignored.
Alternative Specications forParental Income
As mentioned above, it is likely that our estimators may suffer from an upward bias,
since the instruments used to proxy parental permanent income and then predict the cur-
rent income may be endogenous. Table20 presents our estimates using different combi-
nations of the available instruments (we exclude those using one instrument only).
Results are robust to this sensitivity exercise with the exception of the rank-rank
slope, which is unstable when different combinations of instruments are considered.
This makes us unable to guarantee the direction of the bias which is very likely to be
present and may be a consequence of not observing parental income. However, we also
recognize that the case with more instruments makes the rank-rank slope more efficient,
with lower standard errors.
Inclusion ofIndividuals withNo Individual Income
There may exist mobility mismeasurement in our benchmark analysis by including only
individuals that work. Hence, we tested the sensitivity of the results by including indi-
viduals with no individual income derived from work. Results are presented in Table21
Results are almost unchanged for all measures, both for all individuals and for each
gender. Percentage changes in the estimates are no higher than 5%.
Co‑residents Bias
Following Azam and Bhatt (2015), the co-resident bias may exist in our context. The
idea is that if parents are part of the same household as children, they can still influence
their offspring’s decisions about education. The authors point out that the use of sam-
ples with co-residents (children-parents) may lead to problems related to sample selec-
tion, as co-resident individuals may not represent the adult population. We consider
this to be true also for work-related decisions and therefore income, although published
research is mainly related to education. Our benchmark sample includes not only co-res-
ident fathers and children, but also individuals who do not live with their fathers. Now,
we compare the original estimates to a sample with no co-residents and see whether the
results change significantly. Results are presented in Table22 for income mobility and
Table23 for educational mobility.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
94
L.Clemente-Casinhas et al.
Table 20 Sensitivity of Intergenerational Mobility in Income to Alternative Instruments for Father Income for the Benchmark Sample
Standard errors are in parentheses. *,**,*** stands for statistically significant at 10%, 5%, 1% levels, respectively. Probabilities obtained using an ordered logit and the share
of individuals earning more than their fathers are expressed in %. Correlations have the associated significance level of the elasticities used to compute them. The share of
individuals earning more than their fathers does not have an associated significance level. n stands for the number of observations used and N for the total population repre-
sented by those observations using survey weights
Instruments Elasticity Correlation Rank-rank slope Prob Share
All individuals
n = 2549 | N = 980,083 Education, managerial position 0.26***
(0.04) 0.19***
(0.03) 3.75***
(0.06) 6.99***
(0.01) 55.66
Occupation, managerial position 0.25***
(0.04) 0.28***
(0.04) 1.12***
(0.02) 7.13***
(0.01) 50.85
Education, occupation 0.27***
(0.04) 0.20***
(0.03) 0.99***
(0.02) 6.95***
(0.01) 56.04
Education, occupation, managerial position 0.26***
(0.04) 0.20***
(0.03) 0.45***
(0.01) 7.15***
(0.01) 53.11
Males
n = 1027 | N = 431,849 Education, managerial position 0.31***
(0.06) 0.23***
(0.04) 3.98***
(0.12) 7.51***
(0.01) 59.07
Occupation, managerial position 0.28***
(0.05) 0.22***
(0.04) 1.16***
(0.05) 8.17***
(0.01) 51.08
Education, occupation 0.31***
(0.06) 0.23***
(0.05) 1***
(0.05) 8.05***
(0.13) 57.19
Education, occupation, managerial position 0.3***
(0.05) 0.24***
(0.04) 0.48***
(0.02) 8.21***
(0.01) 52.9
Females
n = 1522 | N = 548,234 Education, managerial position 0.22***
(0.07) 0.15***
(0.05) 3.45***
(0.07) 6.53***
(0.01) 52.98
Occupation, managerial position 0.21***
(0.06) 0.17***
(0.04) 1.06***
(0.03) 6.4***
(0.01) 50.67
Education, occupation 0.24***
(0.06) 0.17***
(0.04) 0.96***
(0.04) 6.15***
(0.01) 55.12
Education, occupation, managerial position 0.22***
(0.06) 0.17***
(0.04) 0.42***
(0.02) 6.42***
(0.01) 53.28
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
95
Using Survey Data toEstimate Intergenerational Mobility in…
Table 21 Sensitivity of Intergenerational Mobility in Income to the Inclusion of Individuals with no Individual Income
Standard errors are in parentheses. *,**,*** stands for statistically significant at 10%, 5%, 1% levels, respectively. Probabilities obtained using an ordered logit and the share
of individuals earning more than their fathers are expressed in %. Correlations have the associated significance level of the elasticities used to compute them. The share of
individuals earning more than their fathers does not have an associated significance level. n stands for the number of observations used and N for the total population repre-
sented by those observations using survey weights
Elasticity Correlation Rank-rank slope Prob Share
All individuals Benchmark sample
n = 2549 | N = 980,083 0.26***
(0.04) 0.20***
(0.03) 0.45***
(0.01) 7.15***
(0.01) 53.11
Considering individuals with no individual income
n = 2665 | N = 1,024,434 0.27***
(0.04) 0.20***
(0.03) 0.45***
(0.01) 6.93***
(0.01) 51.84
Males Benchmark sample
n = 1027 | N = 431,849 0.3***
(0.05) 0.24***
(0.04) 0.48***
(0.02) 8.21***
(0.01) 52.9
Considering individuals with no individual income
n = 1059 | N = 441,624 0.31***
(0.06) 0.24***
(0.04) 0.48***
(0.02) 8.11***
(0.01) 52.1
Females Benchmark sample
n = 1522 | N = 548,234 0.22***
(0.06) 0.17***
(0.04) 0.42***
(0.02) 6.42***
(0.01) 53.28
Considering individuals with no individual income
n = 1606 | N = 582,810 0.23***
(0.06) 0.17***
(0.04) 0.41***
(0.02) 6.14***
(0.01) 51.64
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
96
L.Clemente-Casinhas et al.
Overall, we can observe a slight change in the income mobility measures when compar-
ing the original (benchmark) sample and the one without co-residents, i.e., an increase in
persistence. However, although differences exist, the sizes of the potential biases may be
considered negligible, as mobility measures are around the same values with and without
co-residents in the sample. Previous literature on the topic, (e.g., Nicoletti & Francesconi,
2006) found a lower intergenerational income elasticity when using a sample of co-resi-
dents, only in comparison with a sample of parents and children who do not co-reside.
For education, marginal differences are also verified. The work of Muñoz and Siravegna
(2023), who use the first two measures, confirms this behaviour.
Summing up, both income and education mobility sensitivity analysis contain similar
results to the benchmark estimates. Almost unchanged results may result from the small
disparity regarding the sizes of the samples represented in the two scenarios analysed. This
happens because there is not a high degree of co-residency. Our evidence is consistent with
individuals leaving their parents’ home, on average, before their 30s. According to the
Eurostat,29 in 2018 the estimated age at which young people leave their parents’ home is
26.3years (27.2 for males and 25.2 for females) for the EU-27—below Portugal, for which
the age is around 28.2years old (29.9 for men and 28 for women). In turn, the influence
parents might exert on children is residual and this bias can also be ignored.
Table 22 Sensitivity of Intergenerational Mobility in Income to the Exclusion of Co-residents
Standard errors are in parentheses. *,**,*** stands for statistically significant at 10%, 5%, 1% levels,
respectively. Probabilities obtained using an ordered logit and the share of individuals earning more than
their fathers are expressed in %. Correlations have the associated significance level of the elasticities used
to compute them. The share of individuals earning more than their fathers does not have an associated sig-
nificance level. n stands for the number of observations used and N for the total population represented by
those observations using survey weights
Elasticity Correlation Rank-rank slope Prob
All individuals Benchmark sample
n = 2549 | N = 980,083 0.26***(0.04) 0.20***(0.03) 0.45***(0.01) 7.15***(0.01)
Without co-resident
fathers
n = 2279 | N = 901,644
0.27***(0.04) 0.21***(0.03) 0.47***(0.02) 7.42***(0.01)
Males Benchmark sample
n = 1027 | N = 431,849 0.3***(0.05) 0.24***(0.04) 0.48***(0.02) 8.21***(0.01)
Without co-resident
fathers
n = 902 | N = 395,392
0.31***(0.06) 0.25***(0.04) 0.5***(0.03) 8.41***(0.01)
Females Benchmark sample
n = 1522 | N = 548,234 0.22***(0.06) 0.17***(0.04) 0.42***(0.02) 6.42***(0.01)
Without co-resident
fathersn = 1377 |
N = 506,252
0.23***(0.07) 0.18***(0.05) 0.43***(0.02) 6.69***(0.01)
29 https:// ec. europa. eu/ euros tat/ datab rowser/ view/ ILC_ LVPS0 8$DV_ 1041/ defau lt/ table? lang= en
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
97
Using Survey Data toEstimate Intergenerational Mobility in…
Siblings
We also analyse the role of siblings in influencing our benchmark mobility measures,
because they share many characteristics that are specific to their family, namely parental
involvement, schools or neighbourhoods. Therefore, their socioeconomic status may be
strongly correlated (Fletcher etal., 2023), and may influence the mobility estimates made
for income and education. In the same line as the analysis for co-residents, the use of sam-
ples with over-representation of siblings may lead to problems related to sample selection,
since they do not represent the adult population. Our benchmark sample includes individu-
als together with their brothers and sisters. This way, we evaluate if results are influenced
by the existence of such relationships. Results are presented in Tables24 and 25 for income
and educational mobility, respectively.
Overall, most mobility values remain unchanged, when the benchmark sample and the one
with no siblings are compared. The differences are predominant in the bottom to top income
level probability and the share of individuals earning more than their fathers. Nevertheless,
these differences are negligible and therefore can be discarded along with any potential bias.
For education, again relative mobility does not appear to change, while for absolute
mobility the differences that take place are too small.
We can conclude that our benchmark evidence for income and education is robust to the
exclusion of siblings. As it occurred in the case of co-residency, the sample size of siblings
is not large, which makes these sensitivity results very similar to the benchmark ones.
AttenuationBias
For the parental income measure, we now compare our benchmark estimates using a
single year to an estimate obtained by using a 4year period average. Here the estimates
based on income measured using a single year will be different from the ones presented
in Table3. This is because we must ensure that the pseudo-parental sample remains
Table 23 Sensitivity of Intergenerational Mobility in Education to the Exclusion of Co-residents
Standard errors are in parentheses. *,**,*** stands for statistically significant at 10%, 5%, 1% levels,
respectively. Probabilities obtained using an ordered logit and the share of individuals with more education
than their fathers are expressed in %. The share of individuals with more education than their fathers does
not have an associated significance level. n stands for the number of observations used and N for the total
population represented by those observations using survey weights
Corr Prob Share
All individuals Benchmark sample
n = 2549 | N = 980,083 0.26*** 44.35***
(0.01) 84.47
Without co-resident fathers
n = 2279 | N = 901,644 0.26*** 44.69***
(0.02) 84.66
Males Benchmark sample
n = 1027 | N = 431,849 0.24*** 36.93***
(0.02) 82.54
Without co-resident fathers
n = 902 | N = 395,392 0.24*** 37.19***
(0.02) 82.80
Females Benchmark sample
n = 1522 | N = 548,234 0.29*** 49.98***
(0.02) 85.99
Without co-resident fathers
n = 1377 | N = 506,252 0.28*** 50.36***
(0.02) 86.11
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
98
L.Clemente-Casinhas et al.
constant from year 1 to year 4 for results to be comparable. In other words, we have to
guarantee that the same individuals remain in the different survey waves used to com-
pute the average incomes. This allows us to make some inference about what might hap-
pen to our main estimates if we were able to keep the entire initial pseudo-parents sam-
ple, which would guarantee that the differences are mainly due to the number of years
used to compute parental average income, instead of changes in the sample composition
(Murray etal., 2018). According to Solon (1992), the larger the number of periods used
to compute the parental average income, the more reduced the attenuation bias should
Table 24 Sensitivity of Intergenerational Mobility in Income to the Exclusion of Siblings
Standard errors are in parentheses. *,**,*** stands for statistically significant at 10%, 5%, 1% levels,
respectively. Probabilities obtained using an ordered logit and the share of individuals earning more than
their fathers are expressed in %. Correlations have the associated significance level of the elasticities used
to compute them. The share of individuals earning more than their fathers does not have an associated sig-
nificance level. n stands for the number of observations used and N for the total population represented by
those observations using survey weights
Elasticity Correlation Rank-rank slope Prob Share
All individuals Benchmark sample
n = 2549 | N = 980,083 0.26***
(0.04) 0.20***
(0.03) 0.45***
(0.01) 7.15***
(0.01) 53.11
Without siblings
n = 2519 | N = 973,655 0.26***
(0.04) 0.20***
(0.03) 0.45***
(0.01) 7.20***
(0.01) 53.10
Males Benchmark sample
n = 1027 | N = 431,849 0.3***
(0.05) 0.24***
(0.04) 0.48***
(0.02) 8.21***
(0.01) 52.9
Without siblings
n = 1010 | N = 428,503 0.3***
(0.06) 0.24***
(0.03) 0.48***
(0.02) 8.19***
(0.01) 52.8
Females Benchmark sample
n = 1522 | N = 548,234 0.22***
(0.06) 0.17***
(0.04) 0.42***
(0.02) 6.42***
(0.01) 53.28
Without siblings
n = 1509 | N = 545,152 0.22***
(0.06) 0.17***
(0.03) 0.41***
(0.02) 6.52***
(0.01) 53.32
Table 25 Sensitivity of Intergenerational Mobility in Education to the Exclusion of Siblings
Standard errors are in parentheses. *,**,*** stands for statistically significant at 10%, 5%, 1% levels,
respectively. Probabilities obtained using an ordered logit and the share of individuals with more education
than their fathers are expressed in %. The share of individuals with more education than their fathers does
not have an associated significance level. n stands for the number of observations used and N for the total
population represented by those observations using survey weights
Corr Prob Share
All individuals Benchmark sample
n = 2549 | N = 980,083 0.26*** 44.35***(0.01) 84.47
Without siblings
n = 2519 | N = 973,655 0.26*** 44.41***(0.01) 84.52
Males Benchmark sample
n = 1027 | N = 431,849 0.24*** 36.93***(0.02) 82.54
Without siblings
n = 1010 | N = 428,503 0.24*** 37.14***(0.02) 82.59
Females Benchmark sample
n = 1522 | N = 548,234 0.29*** 49.98***(0.02) 85.99
Without siblings
n = 1509 | N = 545,152 0.29*** 49.91***(0.02) 86.03
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
99
Using Survey Data toEstimate Intergenerational Mobility in…
be regarding the intergenerational income elasticity. The same can be considered for the
rank-rank slope as shown in Chetty etal. (2014). Results are presented in Table26.30
See Table26
Regarding the intergenerational income elasticity, when we increase the time on which
parental individual income is averaged to four years, we obtain an estimate that is 22%
higher (from about 0.23 to 0.28). Mazumder (2005) simulates by how much intergen-
erational elasticity is attenuated when considering several periods on which they aver-
age parental income. The author shows that when four years are used, the estimates may
be downward biased by about 31.3%. Our evidence suggests that attenuation bias plays a
considerable role in our estimates. If we perform the same exercise as Mazumder (2005)
on our benchmark estimates using the corresponding attenuation factor for a single year
(0.526), the intergenerational income elasticity in Table3 should be approximately 0.49
instead of 0.26. For men this would compute an elasticity of 0.57 instead of 0.3 and for
girls 0.42 instead of 0.22. This attenuation bias would change the benchmark correlations
as well: for children, males, and females, they would be equal to 0.38 instead of 0.20, 0.46
instead of 0.24, and 0.32 instead of 0.17, respectively.31
Nybom and Stuhler (2017) and Murray etal. (2018) find the rank-rank specification
to be more robust to attenuation bias than the log–log specification. Our estimates for the
elasticity and rank slope may have also increased more than in some works (23% from 0.52
to 0.64 in our case), which use family income as a measure for parental income. This is
the case of Chetty etal. (2014), who consider that in this context using individual meas-
ures of economic status, such as individual income, may lead to larger differences when
Table 26 Sensitivity of Intergenerational Mobility in Income to Attenuation Bias
Standard errors are in parentheses. *,**,*** stands for statistically significant at 10%, 5%, 1% levels,
respectively. Probabilities obtained using an ordered logit and the share of individuals earning more than
their fathers are expressed in %. Correlations have the associated significance level of the elasticities used
to compute them. The share of individuals earning more than their fathers does not have an associated sig-
nificance level. n stands for the number of observations used and N for the total population represented by
those observations using survey weights
Number of periods Elasticity Correlation Rank-rank slope Prob Share
All individuals
n = 2549
N = 980,083
1year 0.23***
(0.04) 0.17***
(0.03) 0.52***
(0.02) 8.13***
(0.01) 58.37
4years 0.28***
(0.04) 0.21***
(0.03) 0.64***
(0.02) 6.55***
(0.01) 48.73
Males
n = 1027
N = 431,849
1year 0.28***
(0.06) 0.22***
(0.05) 0.56***
(0.04) 8.69***
(0.02) 58.65
4years 0.33***
(0.06) 0.25***
(0.05) 0.67***
(0.03) 7.35***
(0.01) 49.41
Females
n = 1522
N = 548,234
1year 0.19***
(0.07) 0.13***
(0.05) 0.47***
(0.03) 7.70***
(0.01) 58.16
4years 0.24***
(0.06) 0.17***
(0.05) 0.59***
(0.02) 5.95***
(0.01) 48.19
30 For children, we could use longitudinal samples to also address the attenuation bias, but these do not
contain retrospective information on parents. Besides, we cannot link longitudinal to the cross-sectional
waves where that information would be available.
31 We cannot state that, in opposition to our previous finding, results for Portugal would now be part of the
most persistent countries in the literature, since the studies used for comparison can also suffer from attenu-
ation bias: if this is true, the relative positions of the countries should remain the same.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
100
L.Clemente-Casinhas et al.
comparing estimates for different period averages because individual income fluctuates
more across years. Income measured on a single year should also be noisy for the measures
not directly influenced by the attenuation bias: we have a pattern of mobility declining for
the remaining measures. Concerning the bottom to top income level probability, there is a
fall of 19% (from 8.13 to 6.55%), and in the case of the share of individuals earning more
than their fathers this measure falls 17% (from 58.37 to 48.73%). Gender patterns are simi-
lar to the findings for the benchmark sample.
Appendix3: The Relationship betweenRelative Mobility inIncome
andEducation
As pointed out by Narayan etal. (2018), mobility in education and mobility in income
should be related. The authors argue that this relationship is likely to be positive because
income persistence is verified due to the endowments that are inherited and to the invest-
ments parents make in children, namely their education. Restuccia and Urrutia (2004) and
Marrero and Rodriguez (2013) find that education is an important factor for economic per-
sistence. Hence, we can formalize their relationship as follows.
Theorem A1
Assuming that the logged current income of parents,
yp
it
, is orthogonal with respect to
the age of children,
Ac
it
, and its squared, Ac
it
2
, the responsiveness of intergenerational rela-
tive mobility in income (
𝛽1
) to marginal changes in intergenerational relative mobility in
education (
𝜕1
) is given by considering the model defined by
where the first two equations reflect, for children and for parents, respectively, the Mincer
(1974) wage equations, which measure the change in logged current income (
yit
) due to
an additional year of current maximum education attained (
Edit
), reflected by
𝜚
, after con-
trolling for other factors (
Wit
); the third regression corresponds to the standard equation
used to estimate the intergenerational income elasticity; the last expression estimates the
relationship between the maximum years of education attained of parents and children; and
uit,𝛼it
and
𝜋i
are the error terms. The proof of this theorem is presented below.
Proof of Theorem A1
We have that
𝛽
1=Cov(y
c
it,y
p
it
)
Var (yp
it
) , which, for the Mincer-type equation implies32:
d𝛽
1
d𝜕
1
=𝜚c𝜚pVar
(
Edp
it
)[
Var
(
yp
it
)]
−1≥
0
⎧
⎪
⎨
⎪
⎩
y
c
it =
𝜚c
Ed
c
it +
𝜒�c
W
c
it +u
c
it
yp
it =𝜚pEdp
it +𝜒�pWp
it +up
it
yc
it =𝛽0+𝛽1yp
it +𝛾c
1Ac
it +𝛾c
2Ac
it
2+𝛼
it
Edc
i
=𝜕
0
+𝜕
1
Edp
i
+𝜋
i
𝛽
1=
Cov
(
𝜚
c
Ed
c
i+𝜒�
c
W
c
i+u
c
i,𝜚
p
Ed
p
i+𝜒�
p
W
p
i+u
p
i
)
Var (𝜚pEdp
i+𝜒�pWp
i+up
i)
=
𝜚c𝜚pCov(Edc
i,Edp
i)+Cov(𝜒�cWc
i+uc
i,𝜚pEdp
i)+Cov(𝜚cEdc
i+𝜒�cWc
i+uc
i,𝜒�pWp
i+up
i)
Var
(
𝜚pEdp
i
+𝜒�pWp
i
+up
i)=
32 We will omit the subscript t for simplicity.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
101
Using Survey Data toEstimate Intergenerational Mobility in…
We are now able to uncover how mobility in income responds to changes in mobility in
education:
Acknowledgements We express gratitude to nine referees, the editor, and a co-editor for their valuable
help. We also thank João Moura and Ana Balcão Reis for helpful contributions. We extend our appreciation
to the participants of the Society for Computation in Economics 29th International Conference on Com-
puting in Economics and Finance and the 16th Annual Meeting of the Portuguese Economic Journal for
=
𝜚
c
𝜚
p
𝜕1Var
(
Ed
p
i
)
+Cov
(
𝜒
�c
W
c
i+u
c
i,𝜚
p
Ed
p
i
)
+Cov
(
𝜚
c
Ed
c
i+𝜒
�c
W
c
i+u
c
i,𝜒
�p
W
p
i+u
p
i
)
Var (𝜚pEdp
i+𝜒�pWp
i+up
i)
=
𝜚c𝜚p𝜕1Var (Edp
i)+Cov(𝜒�cWc
i+uc
i,𝜚pEdp
i)+Cov(𝜚cEdc
i+𝜒�cWc
i+uc
i,𝜒�pWp
i+up
i
)
(𝜚p)2Var (Edp
i)+Var (𝜒�pWp
i)+Var (up
i)+2𝜚pCov(Edp
i,𝜒�pWp
i)
=
𝜚c𝜚p𝜕1+Cov(𝜒�cWc
i+uc
i,𝜚pEdp
i)+Cov(𝜚cEdc
i+𝜒�cWc
i+uc
i,𝜒�pWp
i+up
i)
Var (Edp
i)
(𝜚p)2+Var (𝜒�pWp
i)
Var
(
Edp
i)
+Var (up
i)
Var
(
Edp
i)
+2𝜚pCov(Edp
i,𝜒�pWp
i)
Var
(
Edp
i)
d𝛽
1
d𝜕1
=
𝜚c𝜚p
(𝜚p)2+Var(𝜒�pWp
i)
Var(Ed p
i)+Var(up
i)
Var(Ed p
i)+2𝜚pCov(Edp
i,𝜒�pWp
i)
Var(Ed p
i)
=1
𝜚p
𝜚c+1
𝜚c𝜚p
Var(𝜒�pWp
i+up
i)
Var(Ed p
i)+2Cov(Edp
i,𝜒�pWp
i)
Var(Ed p
i)
=1
𝜚p
𝜚c+1
𝜚c𝜚p
Var(𝜒�pWp
i+up
i)
Var(Ed p
i)+2
𝜚c
Cov(Edp
i,𝜒�pWp
i)
Var(Ed p
i)
=1
𝜚p
𝜚c+1
𝜚c𝜚p
Var(yp
i−𝜚pEdp
i)
Var(Ed p
i)+2
𝜚c
Cov(Edp
i,𝜒�pWp
i)
Var(Ed p
i)
=1
𝜚p
𝜚c+1
𝜚c𝜚p
Var(yp
i)+(𝜚p)2Var(Ed p
i)−2𝜚pCov(yp
it,Edp
i)
Var(Ed p
i)+2
𝜚c
Cov(Edp
i,𝜒�pWp
i)
Var(Ed p
i)
=1
2𝜚p
𝜚c+1
𝜚c𝜚p
Var(yp
i)
Var(Ed p
i)+2
𝜚c[Cov(Edp
i,𝜒�pWp
i)−Cov(yp
i,Edp
i)
Var(Ed p
i)]
=1
2𝜚p
𝜚c+1
𝜚c𝜚p
Var(yp
i)
Var(Ed p
i)+2
𝜚c[Cov(Edp
i,𝜒�pWp
i−yp
i)
Var(Ed p
i)]
=1
2𝜚p
𝜚c+1
𝜚c𝜚p
Var(yp
i)
Var(Ed p
i)+2
𝜚c[Cov(Edp
i,−𝜚pEdp
i−up
i)
Var(Ed p
i)]=1
2𝜚p
𝜚c+1
𝜚c𝜚p
Var(yp
i)
Var(Ed p
i)+2
𝜚c[Cov(Edp
i,−𝜚pEdp
i)
Var(Ed p
i)
]
=1
2𝜚p
𝜚c+1
𝜚c𝜚p
Var(yp
i)
Var(Ed p
i)−2𝜚p
𝜚c[Cov(Edp
i,Edp
i)
Var(Ed p
i)]
=1
2𝜚p
𝜚c+1
𝜚c𝜚p
Var(yp
i)
Var(Ed p
i)−2𝜚p
𝜚c[Var(Ed p
i)
Var(Ed p
i)]=1
2𝜚p
𝜚c+1
𝜚c𝜚p
Var(yp
i)
Var(Ed p
i)−2𝜚p
𝜚c
=1
1
𝜚c𝜚p
Var(yp
i)
Var
(
Edp
i)
=𝜚c𝜚pVar (Edp
i)
Var (yp
i).
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
102
L.Clemente-Casinhas et al.
their comments. We acknowledge financial support from FCT—Fundação para a Ciência e a Tecnologia
(National Science and Technology Foundation) through grants 2020.04449.BD (https:// doi. org/ 10. 54499/
2020. 04449. BD) and UIDB/00315/2020 (https:// doi. org/ 10. 54499/ UIDB/ 00315/ 2020).
Author Contributions All authors contributed to the study conception and design, including literature
review, data collection and analysis. The final version of the manuscript was also written and approved by
all authors.
Funding Open access funding provided by FCT|FCCN (b-on). Luís Clemente-Casinhas has received finan-
cial support from FCT—Fundação para a Ciência e a Tecnologia (National Science and Technology Foun-
dation) through grant 2020. 04449.BD (https://doi.org/10.54499/2020.04449.BD). All the authors acknowl-
edge financial support from FCT—Fundação para a Ciência e a Tecnologia through grant UIDB/00315/2020
(https://doi.org/10.54499/UIDB/00315/2020).
Data Availability Data is available upon request.
Declarations
Conflict of interest The authors declare that they do not have competing interests that may influence the work
reported in this paper.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License,
which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long
as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Com-
mons licence, and indicate if changes were made. The images or other third party material in this article
are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the
material. If material is not included in the article’s Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly
from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
References
Acciari, P., Polo, A., & Violante, G. (2022). “And yet it moves”: intergenerational mobility in Italy. Ameri-
can Economic Journal: Applied Economics, 14(3), 118–163. https:// doi. org/ 10. 1257/ app. 20210 151
Alesina, A., Hohmann, S., Michalopoulos, S., & Papaioannou, E. (2021). Intergenerational mobility in
Africa. Econometrica, 89(1), 1–35. https:// doi. org/ 10. 3982/ ECTA1 7018
Alesina, A., Hohmann, S., Michalopoulos, S., & Papaioannou, E. (2023). Religion and educational mobility
in Africa. Nature. https:// doi. org/ 10. 1038/ s41586- 023- 06051-2
Azam, M., & Bhatt, V. (2015). Like father, like son? Intergenerational Educational Mobility in India.
Demography, 52(6), 1929–1959. https:// doi. org/ 10. 1007/ s13524- 015- 0428-8
Bago d’Uva, T. & Fernandes, M. (2017). Mobilidade Social em Portugal. Estudos da Fundação Francisco
Manuel dos Santos, Lisboa
Bank of Portugal (2022). Economic Bulletin: The Portuguese Economy in 2021. Banco de Portugal, Lisboa
Bauer, P., & Riphahn, R. (2006). Timing of school tracking as a determinant of intergenerational transmis-
sion of education. Economics Letters, 91(1), 90–97. https:// doi. org/ 10. 1016/j. econl et. 2005. 11. 003
Becker, G., & Tomes, N. (1979). An equilibrium theory of the distribution of income and intergenera-
tional mobility. Journal of Political Economy, 87(6), 1153–1189.
Becker, G., Kominers, S., Murphy, K., & Spenkuch, J. (2018). A theory of intergenerational mobility.
Journal of Political Economy, 126(S1), S7–S25. https:// doi. org/ 10. 1086/ 698759
Björklund, A., & Jäntti, M. (1997). Intergenerational income mobility in Sweden compared to the United
States. American Economic Review, 87(5), 1009–1018.
Blanden, J., Goodman, P., & Machin, S. (2004). Changes in generational mobility in Britain. In M.
Corak (Ed.), Generational income mobility in north America and Europe (pp. 122–146). Cam-
bridge University Press.
Blanden, J., Gregg, P. & Machin, S. (2005). Intergenerational Mobility in Europe and North America.
Centre for Economic Performance
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
103
Using Survey Data toEstimate Intergenerational Mobility in…
Böhlmark, A., & Lindquist, M. (2006). Life-cycle variations in the association between current and life-
time income: replication and extension for Sweden. Journal of Labor Economics, 24(4), 879–896.
https:// doi. org/ 10. 1086/ 506489
Borisov, G., & Pissarides, C. (2019). Intergenerational earnings mobility in post-soviet Russia. Eco-
nomica, 87(345), 1–27. https:// doi. org/ 10. 1111/ ecca. 12308
Brandén, G. (2019). Does Inequality Reduce Mobility? The Great Gatsby Curve and its Mecha-
nisms. IFAU—Institute for Evaluation of Labour Market and Education Policy Working Paper
Series2019:20
Brenner, J. (2010). Life-cycle variations in the association between current and lifetime earnings: evi-
dence for German natives and guest workers. Labour Economics, 17(2), 392–406. https:// doi. org/
10. 1016/j. labeco. 2009. 03. 006
Bukodi, E. & Goldthorpe, J. (2018). Social Inequality and Social Mobility: Is there an Inverse Relation?
SocArXiv (mimeo)
Carneiro, P. (2008). Equality of opportunity and educational achievement in Portugal. Portuguese Eco-
nomic Journal, 7(1), 17–41. https:// doi. org/ 10. 1007/ s10258- 007- 0023-z
Carranza, R. (2023). Upper and Lower Bound Estimates of Inequality of Opportunity: A Cross-National
Comparison for Europe. Review of Income and Wealth, 69(4), 838–860. https:// doi. org/ 10. 1111/ roiw.
12622
Causa, O., & Johansson, Å. (2010). Intergenerational social mobility in OECD countries. OECD Jour-
nal: Economic Studies, 2010(1), 1–44. https:// doi. org/ 10. 1787/ eco_ studi es- 2010- 5km33 scz5r jj
Cervini-Plá, M. (2015). Intergenerational earnings and income mobility in Spain. Review of Income and
Wealth, 61(4), 812–828. https:// doi. org/ 10. 1111/ roiw. 12130
Chadwick, L., & Solon, G. (2002). Intergenerational income mobility among daughters. American Eco-
nomic Review, 92(1), 335–344. https:// doi. org/ 10. 1257/ 00028 28027 60015 766
Chetty, R., & Hendren, N. (2018a). The Impacts of neighborhoods on intergenerational mobility I:
Childhood exposure effects. The Quarterly Journal of Economics, 133(3), 1107–1162. https:// doi.
org/ 10. 1093/ qje/ qjy007
Chetty, R., & Hendren, N. (2018b). the impacts of neighborhoods on intergenerational mobility II:
County-level estimates. The Quarterly Journal of Economics, 133(3), 1163–1228. https:// doi. org/
10. 1093/ qje/ qjy006
Chetty, R., Friedman, J., Hendren, N., Jones, M., & Porter, S. (2020a). The Opportunity Atlas: Mapping
the Childhood Roots of Social Mobility. National Bureau of Economic Research Working Papers
No. 25147. https:// doi. org/ 10. 3386/ w25147
Chetty, R., Friedman, J., Saez, E., Turner, N., & Yagan, D. (2020b). The Determinants of Income Seg-
regation and Intergenerational Mobility: Using Test Scores to Measure Undermatching. National
Bureau of Economic Research Working Papers No. 26748. https:// doi. org/ 10. 3386/ w26748
Chetty, R., Grusky, D., Hell, M., Hendren, N., Manduca, R., & Narang, J. (2017). The fading American
dream: Trends in absolute income mobility since 1940. Science, 356(6336), 398–406. https:// doi.
org/ 10. 1126/ scien ce. aal46 17
Chetty, R., Hendren, N., & Katz, L. (2016). The effects of exposure to better neighborhoods on children:
New evidence from the moving to opportunity experiment. American Economic Review, 106(4),
855–902. https:// doi. org/ 10. 1257/ aer. 20150 572
Chetty, R., Hendren, N., Jones, M., & Porter, S. (2020c). Race and economic opportunity in the United
States: an intergenerational perspective. The Quarterly Journal of Economics, 135(2), 711–783.
https:// doi. org/ 10. 1093/ qje/ qjz042
Chetty, R., Hendren, N., Kline, P., & Saez, E. (2014). Where Is the land of opportunity? The geography
of intergenerational mobility in the United States. The Quarterly Journal of Economics, 129(4),
1553–1623. https:// doi. org/ 10. 1093/ qje/ qju022
Chu, Y., & Lin, M. (2020). Intergenerational earnings mobility in Taiwan: 1990–2010. Empirical Econom-
ics, 59(1), 11–45. https:// doi. org/ 10. 1007/ s00181- 019- 01637-0
Clements, B. (1999). The Efficiency of Education Expenditure in Portugal. International Monetary Fund
Staff Working Paper 99/179. https:// doi. org/ 10. 5089/ 97814 51859 010. 001
Comi, S. (2004). Intergenerational mobility in Europe: evidence from ECHP. CHILD - Centre for House-
hold, Income, Labour and Demographic economics Working Papers wp18_04.
Corak, M. (2019). The canadian geography of intergenerational income mobility. The Economic Journal,
130(631), 2134–2174. https:// doi. org/ 10. 1093/ ej/ uez019
Dahl, M., & DeLeire, T. (2008). The Association Between Children’s Earnings and Fathers’ Lifetime Earn-
ings: Estimates Using Administrative Data. Institute for Research on Poverty Discussion Paper No.
1342–08
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
104
L.Clemente-Casinhas et al.
Daruich, D. (2018). The Macroeconomic Consequences of Early Childhood Development Policies. Federal
Reserve Bank of St. Louis Working Papers 2018–29.
Daude, C., & Robano, V. (2015). On intergenerational (Im) mobility in latin america. Latin American Eco-
nomic Review, 24(9), 1–29. https:// doi. org/ 10. 1007/ s40503- 015- 0030-x
Dearden, M., & Reed, H. (1997). Intergenerational mobility in Britain. Economic Journal, 107(440), 47–66.
https:// doi. org/ 10. 1111/ 1468- 0297. 00141
Deutscher, N., & Mazumder, B. (2023). Measuring intergenerational income mobility: A synthesis of
approaches. Journal of Economic Literature, 61(3), 988–1036. https:// doi. org/ 10. 1257/ jel. 20211 413
Dunn, C. (2007). The Intergenerational Transmission of Lifetime Earnings: Evidence from Brazil. The B.E.
Journal of Economic Analysis & Policy, 7(2). https:// doi. org/ 10. 2202/ 1935- 1682. 1782
Emran, M., Ferreira, F., Jiang, Y. & Sun, Y. (2019). Intergenerational Educational Mobility in Rural Econ-
omy: Evidence from China and India. Munich Personal RePEc Archive No. 94121
Eriksen, J. & Munk, M. (2020). The Geography of Intergenerational Mobility – Danish Evidence. Economic
Letters, 189. https:// doi. org/ 10. 1016/j. econl et. 2020. 109024
Ferreira, F., & Gignoux, J. (2014). The measurement of educational inequality: Achievement and opportu-
nity. The World Bank Economic Review, 28(2), 210–246. https:// doi. org/ 10. 1093/ wber/ lht004
Ferreira, S., & Veloso, F. (2006). Intergenerational mobility of wages in Brazil. Brazilian Review of Econo-
metrics, 26(2), 181–211. https:// doi. org/ 10. 12660/ bre. v26n2 2006. 1576
Fletcher, J., & Han, J. (2019). Intergenerational mobility in Education: Variation in geography and time.
Journal of Human Capital, 13(4), 585–634. https:// doi. org/ 10. 1086/ 705610
Fletcher, J., Lu, Q., Mazumder, B. & Song, J. (2023). Understanding Sibling Correlations in Education:
Molecular Genetics and Family Background. IZA Discussion Paper No. 15862
Fortin, N. & Lefebvre, S. (1998). Intergenerational Income Mobility in Canada. In M. Corak (Ed.), Labor
Markets, Social Institutions and the Future of Canada´s Children. Statistics of Canada, Ottawa
Friedman, M. (1957). A Theory of the consumption function. Princeton University Press.
Gallagher, R., Kaestner, R., & Persky, J. (2019). The geography of family differences and intergenerational
mobility. Journal of Economic Geography, 19(3), 589–618. https:// doi. org/ 10. 1093/ jeg/ lby026
GDIM (2018). Global Database on Intergenerational Mobility. Development Research Group, World Bank.
Washington, D.C.: World Bank Group
Haider, S., & Solon, G. (2006). Life-cycle variation in the association between current and lifetime earnings.
American Economic Review, 96(4), 1308–1320. https:// doi. org/ 10. 1257/ aer. 96.4. 1308
Heidrich, S. (2017). Intergenerational Mobility in Sweden: a regional perspective. Journal of Population
Economics, 30, 1241–1280. https:// doi. org/ 10. 1007/ s00148- 017- 0648-x
Helsø, A. (2020). Intergenerational Income Mobility in Denmark and the United States. The Scandinavian
Journal of Economics, 1–24. https:// doi. org/ 10. 1111/ sjoe. 12420
Hilger, N. (2016). The Great Escape: Intergenerational Mobility in the United States Since 1940. National
Bureau of Economic Research Working Papers No. 21217. https:// doi. org/ 10. 3386/ w21217
Jenkins, S. (1987). Snapshots versus movies: “lifecycle biases” and the estimation of intergenerational earn-
ings inheritance. European Economic Review, 31(5), 1149–1158. https:// doi. org/ 10. 1016/ S0014-
2921(87) 80010-7
Kyzyma, I., & Groh-Samberg, O. (2020). Estimation of Intergenerational mobility in small samples: Evi-
dence from German survey data. Social Indicators Research, 151(4), 621–643. https:// doi. org/ 10.
1007/ s11205- 020- 02378-9
Lam, K., & Liu, P. (2019). Intergenerational educational mobility in Hong Kong: Are immigrants more
mobile than natives? Pacific Economic Review, 24(1), 137–157. https:// doi. org/ 10. 1111/ 1468- 0106.
12215
Latif, E. (2017). The relationship between intergenerational educational mobility and public spending: Evi-
dence from Canada. Economic Papers, 36(3), 335–350. https:// doi. org/ 10. 1111/ 1759- 3441. 12177
Latif, E. (2018). Trends in Intergenerational educational mobility in Canada. The Australian Economic
Review, 52(1), 61–75. https:// doi. org/ 10. 1111/ 1467- 8462. 12297
Lee, C., & Solon, G. (2009). Trends in intergenerational income mobility. The Review of Economics and
Statistics, 91(4), 766–772. https:// doi. org/ 10. 1162/ rest. 91.4. 766
Lee, H., & Lee, J. (2020). Patterns and determinants of intergenerational educational mobility: evidence
across countries. Pacific Economic Review, 26(1), 70–90. https:// doi. org/ 10. 1111/ 1468- 0106. 12342
Lefranc, A., & Trannoy, A. (2005). Intergenerational earnings mobility in France: is France more mobile
than the US? Annales D’economie Et De Statistique, 78, 57–77. https:// doi. org/ 10. 2307/ 20079 128
Leigh, A. (2007). Intergenerational Mobility in Australia. The B.E. Journal of Economic Analysis & Policy,
7(2). https:// doi. org/ 10. 2202/ 1935- 1682. 1781
Marrero, G., & Rodríguez, J. (2013). Inequality of opportunity and growth. Journal of Development Eco-
nomics, 104, 107–122. https:// doi. org/ 10. 1016/j. jdeve co. 2013. 05. 004
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
105
Using Survey Data toEstimate Intergenerational Mobility in…
Mazumder, B. (2005). Fortunate sons: new estimates of intergenerational mobility in the United States using
social security earnings data. Review of Economics and Statistics, 87(2), 235–255. https:// doi. org/ 10.
1162/ 00346 53053 970249
Mendolia, S., & Siminski, P. (2019). New estimates of intergenerational mobility in Australia. Economic
Record, 92(298), 361–373. https:// doi. org/ 10. 1111/ 1475- 4932. 12274
Mincer, J. (1974). Schooling, Experience and Earnings. National Bureau of Economic Research, New York
Mocetti, S. (2007). Intergenerational Earnings Mobility in Italy. The B.E. Journal of Economic Analysis &
Policy, 7(2). https:// doi. org/ 10. 2202/ 1935- 1682. 1794
Muñoz, E.& Siravegna, M. (2023). When measure matters: Coresidence bias and intergenerational mobility
revisited. Inter-American Development Bank Working Paper No. IDB-WP-01469. https:// doi. org/ 10.
18235/ 00048 81
Murray, C., Clark, R., Mendolia, S., & Siminski, P. (2018). Direct measures of intergenerational income
mobility for Australia. Economic Record, 94(307), 445–468. https:// doi. org/ 10. 1111/ 1475- 4932.
12445
Narayan, A., Weide, R., Cojocaru, A., Lakner, C., Redaelli, S., Mahler, D., Ramasubbaiah, R. & Thewis-
sen, S. (2018). Fair Progress? Economic Mobility Across Generations Around the World. Equity and
Development. Washington, D.C.: World Bank
Neidhöfer, G., & Stockhausen, M. (2018). Dynastic inequality compared: Multigenerational mobility in the
United States, the United Kingdom, and Germany. Review of Income and Wealth, 65(2), 383–414.
https:// doi. org/ 10. 1111/ roiw. 12364
Nicoletti, C. & Ermisch, J. (2008). Intergenerational Earnings Mobility: Changes Across Cohorts in Britain.
The B.E. Journal of Economic Analysis and Policy, 7(2), 1–38. https:// doi. org/ 10. 2202/ 1935- 1682.
1755
Nicoletti, C., & Francesconi, M. (2006). Intergenerational mobility and sample selection in short panels.
Journal of Applied Econometrics, 21(8), 1265–1293. https:// doi. org/ 10. 1002/ jae. 910
Nimubona, A., & Vencatachellum, D. (2007). Intergenerational education mobility of black and white South
Africans. Journal of Population Economics, 20, 149–182. https:// doi. org/ 10. 1007/ s00148- 006- 0120-9
Nuñez, J. & Miranda, L. (2010). Intergenerational Income Mobility in A Less-Developed, High-Inequality
Context: The Case of Chile. The B.E. Journal of Economic Analysis & Policy, 10(1). https:// doi. org/
10. 2202/ 1935- 1682. 2339
Nybom, M., & Stuhler, J. (2016). Heterogeneous Income profiles and life-cycle bias in intergenerational
mobility estimation. Journal of Human Resouces, 51(1), 239–268. https:// doi. org/ 10. 3368/ jhr. 51.1.
239
Nybom, M., & Stuhler, J. (2017). Biases in standard measured of intergenerational income dependence.
Journal of Human Resouces, 52(3), 800–825. https:// doi. org/ 10. 3368/ jhr. 52.3. 0715- 7290R
Nybom, M. (2018). Intergenerational Mobility: A Dream Deferred? International Labour Organization
Research Paper Series
OECD. (1995). OECD Economic Surveys: Portugal 1995. OECD Publishing.
OECD. (2018). A broken social elevator? OECD Publishing, Paris.
OECD. (2019). Education at a Glance 2019: OECD Indicators. OECD Publishing.
Pagan, A. (1984). Econometric Issues in the Analysis of Regressions with Generated Regressors. Interna-
tional Economic Review, 25(1), 221–247.
Pereira, P. (2010). Higher Education Attainment: The Case of Intergenerational Transmission of Education
in Portugal. IZA Discussion Paper No. 4813. https:// doi. org/ 10. 2139/ ssrn. 15704 31
Piketty, T. (2000). Theories of persistent inequality and intergenerational mobility. In A. Atkinson & F.
Bourguignon (Eds.), Handbook of income distribution (pp. 429–476). Elsevier.
Piraino, P. (2007). Comparable Estimates of Intergenerational Income Mobility in Italy. The B.E. Journal of
Economic Analysis & Policy, 7(2). https:// doi. org/ 10. 2202/ 1935- 1682. 1711
Piraino, P. (2015). Intergenerational earnings mobility and equality of opportunity in South Africa. World
Development, 67, 396–405. https:// doi. org/ 10. 1016/j. world dev. 2014. 10. 027
Raaum, O., Bratsberg, B., Røed, K., Österbacka, E., Eriksson, T., Jäntti, M. & Naylor, R. (2008). Marital
Sorting, Household Labor Supply, and Intergenerational Earnings Mobility Across Countries. The
B.E. Journal of Economic Analysis & Policy, 7(2). https:// doi. org/ 10. 2202/ 1935- 1682. 1767
Ray, D. (2006). Aspirations, Poverty, and Economic Change. In A. Banerjee, R. Bénabou & D. Mookherjee
(Eds.), Understanding Poverty (pp. 409–422). Oxford University Press. https:// doi. org/ 10. 1093/ 01953
05191. 003. 0028
Restuccia, D., & Urrutia, C. (2004). Intergenerational persistence of earnings: the role of early and college
education. The American Economic Review, 94(5), 1354–1378. https:// doi. org/ 10. 1257/ 00028 28043
052213
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
106
L.Clemente-Casinhas et al.
Ross (2019). Occupation Aspirations, Education Investment, and Cognitive Outcomes: Evidence from
Indian Adolescents. World Development, 123. https:// doi. org/ 10. 1016/j. world dev. 2019. 104613
Schneebaum, A., Rumplmaier, B. & Altzinger, W. (2014). International Educational Persistence in Europe.
Vienna University of Economics and Business Department of Economics Working Papers wuwp174
Schneebaum, A., Rumplmaier, B., & Altzinger, W. (2016). Gender and migration background in intergen-
erational educational mobility. Education Economics, 24(3), 239–260. https:// doi. org/ 10. 1080/ 09645
292. 2015. 10061 81
Solon, G. (1992). Intergenerational income mobility in the United States. American Economic Review,
82(3), 393–408.
Solon, G. (2002). Cross-country differences in intergenerational earnings mobility. Journal of Economic
Perspectives, 16(3), 59–66. https:// doi. org/ 10. 1257/ 08953 30027 60278 712
Solon, G. (2004). A Model of intergenerational mobility variation over time and place. In M. Corak (Ed.),
Generational income mobility in north America and Europe (pp. 38–47). Cambridge University
Press.
Torche, F. (2015). Analyses of intergenerational mobility: an interdisciplinary review. Annals of the Ameri-
can Academy of Political and Social Science, 657(1), 37–62. https:// doi. org/ 10. 1177/ 00027 16214
547476
Urbina, D. (2018). Intergenerational educational mobility during expansion Reform: Evidence from Mexico.
Population Research and Policy Review, 37, 367–417. https:// doi. org/ 10. 1007/ s11113- 018- 9466-4
Van de Gaer, D., Schokkaert, E., & Martinez, M. (2001). Three meanings of intergenerational mobility.
Economica, 68(272), 519–538. https:// doi. org/ 10. 1111/ 1468- 0335. d01- 15
Weintraub, R., Lucia, M., Fernald, L., Adler, N., Bertozzi, S. & Syme, S. (2015). Perceptions of Social
Mobility: Development of a New Psychosocial Indicator Associated with Adolescent Risk Behaviors.
Frontiers in Public Health, 3(62). https:// doi. org/ 10. 3389/ fpubh. 2015. 00062
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
1.
2.
3.
4.
5.
6.
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center
GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers
and authorised users (“Users”), for small-scale personal, non-commercial use provided that all
copyright, trade and service marks and other proprietary notices are maintained. By accessing,
sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of
use (“Terms”). For these purposes, Springer Nature considers academic use (by researchers and
students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and
conditions, a relevant site licence or a personal subscription. These Terms will prevail over any
conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription (to
the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of
the Creative Commons license used will apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may
also use these personal data internally within ResearchGate and Springer Nature and as agreed share
it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not otherwise
disclose your personal data outside the ResearchGate or the Springer Nature group of companies
unless we have your permission as detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial
use, it is important to note that Users may not:
use such content for the purpose of providing other users with access on a regular or large scale
basis or as a means to circumvent access control;
use such content where to do so would be considered a criminal or statutory offence in any
jurisdiction, or gives rise to civil liability, or is otherwise unlawful;
falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association
unless explicitly agreed to by Springer Nature in writing;
use bots or other automated methods to access the content or redirect messages
override any security feature or exclusionary protocol; or
share the content in order to create substitute for Springer Nature products or services or a
systematic database of Springer Nature journal content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a
product or service that creates revenue, royalties, rent or income from our content or its inclusion as
part of a paid for service or for other commercial gain. Springer Nature journal content cannot be
used for inter-library loans and librarians may not upload Springer Nature journal content on a large
scale into their, or any other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not
obligated to publish any information or content on this website and may remove it or features or
functionality at our sole discretion, at any time with or without notice. Springer Nature may revoke
this licence to you at any time and remove access to any copies of the Springer Nature journal content
which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or
guarantees to Users, either express or implied with respect to the Springer nature journal content and
all parties disclaim and waive any implied warranties or warranties imposed by law, including
merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published
by Springer Nature that may be licensed from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a
regular basis or in any other manner not expressly permitted by these Terms, please contact Springer
Nature at
onlineservice@springernature.com
