Goal of this segment: Write a report on your findings
Main focus: Question/Topic you focused on in the research segment
Topic: How family economic inequality affects the future generation (focus on the impact on their education and income).
What should always be covered:
Motivation:
Why did you pick your topic? Why is it interesting? (Refer back to magazine article and empirical evidence.)
– Magazine article: https://www.theatlantic.com/business/archive/2015/06/what-matters-inequality-or-opportuniy/393272/
– Topic: how family economic inequality affects the future generation (focus on the impact on their education and income).
Analysis of Existing Research:
What is the state of research on the topic?
-Research article: (I upload on the attachment—content server). Intergenerational income immobility in Finland: contrasting roles for parental earnings and family income.
Description of research methods, datasets etc.
Discussion:
What is the main takeaway from the literature? Compare and contrast findings. What is your take on the topic? What is missing in the existing literature?J Popul Econ (2013) 26:1057–1094
DOI 10.1007/s00148-012-0442-8
ORIGINAL PAPER
Intergenerational income immobility in Finland:
contrasting roles for parental earnings and family
income
Robert E. B. Lucas · Sari Pekkala Kerr
Received: 17 October 2011 / Accepted: 13 August 2012 /
Published online: 2 October 2012
© Springer-Verlag 2012
Abstract An intergenerational model is developed, nesting heritable earning
abilities and credit constraints limiting human capital investments in children.
Estimates on a large, Finnish data panel indicate very low transmission from
parental earnings, suggesting that the parameter of inherited earning ability
is tiny. Family income, particularly during the phase of educating children,
is shown to be much more important in shaping children’s lifetime earnings.
This influence of parental incomes on children’s earnings rises as the children
age because the returns to education rise. Despite Finland’s well-developed
welfare state, persistence in economic status across generations is much higher
than previously thought.
Keyword Intergenerational explanations panel
JEL Classification J62 · J13 · I21
Responsible editor: Alessandro Cigno
Electronic supplementary material The online version of this article
(doi:10.1007/s00148-012-0442-8) contains supplementary material, which is available to
authorized users.
R. E. B. Lucas (B)
Economics Department, Boston University, 270 Bay State Road, Boston MA 02215, USA
e-mail: rlucas@bu.edu
S. P. Kerr
Helsinki Center of Economic Research (HECER), Wellesley Centers for Women,
University of Helsinki, P.O. Box 17, 00014 Helsinki, Finland
S. P. Kerr
Wellesley College, 106 Central Street, Wellesley, MA 02481, USA
1058
R.E.B. Lucas, S.P. Kerr
Omena ei putoa kauas puusta. (The apple doesn’t fall far from the tree).
Syntynyt kultalusikka suussa. (Born with a golden spoon in his mouth).
Traditional Finnish adages.
Three distinct explanations have been postulated for the commonly observed, positive correlation between incomes of children and those of their
parents. One hypothesis rests upon the intergenerational transmission, either
genetically or through the home environment, of unobserved earning abilities.
This variant suggests an auto-regressive process in earnings across generations.
A second approach emphasizes budget constraints, limiting parents’ investments in human capital of their children. A third alternative models parental
preferences with respect to their children as endogenous to their own parents’
incomes, resulting in a three-generation auto-regressive process.1
Parental resources may potentially shape economic success of children
via a number of channels, in addition to the duration of education, such as
through the quality of education, extracurricular opportunities, neighborhood
influences, and postschool training (Restuccia and Urrutia 2004). Yet, in
comparison to the numerous analyses of schooling achievement outcomes,2
attempts to discern the relative contributions of auto-regression in earning
ability versus family incomes in the process of lifetime earnings’ formation
remain few. Mulligan (1997, 1999) offers an important exception, estimating
little difference in the intergenerational regression toward the mean in wages
between US adult children who have (or expect to) inherit more than $25,000
relative to those who do not anticipate such bequests, supporting Mulligan’s
conclusion, “…that the observed intergenerational dynamics of measures of
economic status are not the result of borrowing constraints” Mulligan (1999,
p. S215). However, Gaviria (2002) finds Mulligan’s result to be sensitive to
the specific inheritance break-point imposed and reaches the opposite conclusion. Corak and Heisz (1999) suggest that a slower regression toward mean
earnings among the Canadian sons of middle-income families, than among
their wealthier counterparts, may reflect liquidity constraints on middle-class
parents. Grawe (2004) disagrees and applies a quantile regression approach
to the same data and finds no evidence to support the importance of liquidity
constraints. Björklund et al. (2005) examine an extraordinary Swedish data set,
which permits comparisons in earnings of several sibling types, indicating a
dominant role for a genetic component rather than an environmental effect.
1 Becker
and Tomes (1979, 1986), Behrman et al. (1995). Easterlin et al. (1980) expressed the
third alternative in terms of fertility outcomes, though it may be applied equally to quality rather
than just quantity of children. Excellent surveys are provided in Behrman (1997), Solon (1999),
Björklund and Jäntti (2009), Björklund and Salvanes (2010), and Black and Devereux (2010).
2 Contributions include Cameron and Heckman (1998, 2001), Behrman and Rosenzweig (2002),
Carneiro and Heckman (2002), Plug and Vijverberg (2003, 2005), Black et al. (2005), Belley and
Lochner (2007), Lochner and Monge-Naranjo (2011), and Pronzato (2012).
Intergenerational income immobility in Finland
1059
Yet, as Björklund et al. emphasize, identifying the effects of nature versus
nurture from birth-type comparisons requires some strong assumptions.3
In Section 1, a model nesting both intergenerational auto-regression in
earning ability and a family budget constraint is outlined. The nested model
is estimated, drawing upon a very remarkable, and largely unexplored, data
panel encompassing the entire Finnish population from 1970 to 1999. The
transition to an empirical, regression model, related prior work and some
general issues in estimation are considered in Section 2. The data, and prior
work on intergenerational transmission in Finland, are described in Section 3.
Initial estimates for both sons and daughters, as well as some sensitivity tests,
follow in Section 4. Both inherited ability and the influence of family budget
constraints may, in principle, shape earnings of the next generation, and this is
recognized in the nested model. However, all of our estimates point to small
and fairly insignificant values for the Becker–Tomes parameter of inherited
earning ability, no matter whether the father, mother, or both are considered
as sources. Yet the results suggest throughout that family incomes are an
important factor shaping intergenerational transmission in Finland. Section
4 also includes some evidence, from a second data set, on the influence of
both parents’ and grandparents’ incomes on earnings in the third generation.
Following Behrman and Taubman (1989), these estimates permit a test of the
endogenous preference hypothesis of Easterlin et al. (1980).
Previous evidence on Finland, and on the Nordic countries more generally,
suggests significantly greater intergenerational mobility than in the USA or
UK.4 Cross-country comparisons of estimates are fraught with differences
in specification and data (Solon 2002; Corak 2006; Bratsberg et al. 2007;
Jäntti et al. 2007). Nonetheless, the estimates of intergenerational transmission
elasticities presented here approximate some of the existing estimates for the
USA and UK. This suggests that, although Finland clearly has a more equal
distribution of income than the USA, mean intergenerational mobility across
the spectrum in Finland may be quite limited.
In the US context, sons’ earnings appear to become more like those of
their fathers as the son ages (Reville RT, Intertemporal and life cycle variation in measured intergenerational earnings mobility, unpublished). Section
5 confirms a significant, parallel tendency among both sons’ and daughters’
earnings in the Finnish context. Haider and Solon (2006) note that this
property may derive from age dependence in the association between observed
and lifetime earnings. In Section 5, a specific form of this association is
investigated, stemming from rising returns to education with age. Supporting
results on the age dependence of returns to education in Finland, including
cross-sibling comparisons, are also presented in Section 5. After accounting for
the interaction between age and the returns to education, no further significant
3 See
also Duncan et al. (2005), Sacerdote (2007), Liu and Zeng (2009), and Behrman et al. (2011).
for example, Gustaffson (1994), Björklund and Jäntti (1997), Österbacka (2001), Björklund
et al. (2002), Jäntti et al. (2007).
4 See,
1060
R.E.B. Lucas, S.P. Kerr
rise in the intergenerational transmission elasticity is observed, as Finnish sons
and daughters age.
The seminal contribution of Becker and Tomes (1986, p. 10) incorporates,
“…the effect of imperfect access to debt contracted for children.” There
remains the possibility of credit constraints on parental borrowing against their
own future income, and hence the effect of timing of income receipts upon
intergenerational transmission. In Section 6, it is demonstrated that parental
income when the child is of school age, even after instrumenting for these
income measures, has a larger impact on the child’s subsequent earnings than
does family income thereafter.5
Section 7 closes by attempting to put these fresh results in perspective.
Whether lessons may be derived directly for other countries from these results
on Finland remains to be seen; similar nested models appear not to have been
estimated elsewhere. Nonetheless, it is compelling to note the importance
of family incomes, even in this Finnish context where the welfare state has
attained such importance.
1 Models of intergenerational transmission: with and without
credit constraints
The Becker and Tomes (1979, 1986) model of intergenerational mobility is in
the class of consensus, parental preference models, in which children play no
role in decision making (Behrman 1997). Mulligan (1999) lends specificity to
this framework by assuming a single child family in which the utility (U p ) of
the parent takes a CES form:
1/η
(1)
U p = Cpη + α Ycη
where Cp represents consumption of the parent, Yc is permanent income of
the child, α is the weight placed on altruism, and (1 − η)−1 is the elasticity of
substitution. The parent chooses an amount (Hc ) to invest in the child’s human
capital, bequests and inter vivos transfers (Bc ) to be passed to the child, and
the parent’s own consumption level, so as to maximize Eq. 1 subject to two
constraints. The first is a budget constraint:
Cp + Hc + Bc ≤ Yp
(2)
where Yp represents wealth of the parent. In the second constraint, each child’s
wealth is generated from the child’s human capital and transfers such that
Yc = Gc Hcρ + Bc (1 + ι)
5 In
(3)
a related paper, Oreopoulos et al. (2008) show that earnings are lower among a group of
Canadians, whose fathers lost their jobs in plant closings, than among a measurably comparable
group whose fathers were not retrenched. See also Eide and Showalter (2000).
Intergenerational income immobility in Finland
1061
where Gc denotes inherent capacities of the child to generate earnings, ρ is the
return on human capital (with 0 < ρ < 1), and ι is the rate of interest obtained
on transfers. In the event that an interior solution exists for both Hc and Bc ,
the first-order conditions may be solved to provide
1/(1−ρ)
Hc = ρGc (1 + ι)−1
.
(4)
Provided that ι is independent of the amount of bequest, the budget constraint
is linear and the parent’s choice of Hc is independent of the elasticity of
substitution; hence, η does not appear in Eq. 4. The parent simply equates the
returns on investment in the child’s human capital with that on bequests (see,
however, Jürges 2000).
On the other hand, negative transfers (borrowing against the child’s wealth
at the same rate ι) may well be infeasible (Becker and Tomes 1986). If credit
constraints restrict transfers to be non-negative, then the optimal choice of the
parent is given from the first-order condition on maximizing Eq. 1 with respect
to Hc , subject to Eqs. 2 and 3 and the additional constraint that Bc ≥ 0. When
this last constraint is binding, the solution has the implicit form:
1/(η−1)
Yp = Hc + ρα Gηc Hcρη−1
(5)
in which ∂ H c /∂Yp > 0, given Gc . In the instance of a Cobb–Douglas utility
function, Eq. 5 reduces to the more tractable, explicit form:
Hc = Yp ρα (1 + ρα)−1 .
(6)
For convenience, the case represented by Eq. 4 is referred to as the wealth
model in the remainder of this paper, while Eq. 5 is termed the creditconstrained case.6 A critical distinction may be drawn between these two cases.
In the wealth model, investments in the child’s human capital are affected only
by the child’s inherent earning capacity Gc and, in particular, parental wealth
does not appear in Eq. 4. In contrast, in the credit-constrained case, parental
wealth indeed affects the child’s human capital directly in both Eqs. 5 and
6. Thus, in the credit-constrained case, parental wealth already impacts that
of the child through the budget constraint, but in the wealth model, another
factor must be introduced to generate an auto-regressive process in earnings
across generations. In Becker and Tomes (1979, 1986), this additional factor
is a presumption of auto-regression in inherent earnings capacity, akin to the
seminal ideas of Galton (1869), embodying an unknown mix of genetic and
cultural transmissions, which may be written here as
Gc = Gψp γc
(7)
where Gp is the inherent earning capacity of the parent, γc is a stochastic
component in this inheritance, and 0 < ψ < 1.
6 These
correspond to the high-resource and low-resource family cases in Behrman et al. (1995).
1062
R.E.B. Lucas, S.P. Kerr
Let Ec represent that component of the child’s wealth which we may call
lifetime earnings, such that
Ec = Gc Hcρ .
(8)
In the case of the wealth model, one can then substitute in Eq. 8 for Hc from
Eq. 4 to provide
ρ (ρ−1) 1/(1−ρ)
Gc
.
(9)
Ec = (1 + ι) ρ −1 /
If investments in the parent’s human capital were also dictated by the wealth
model, then a relationship equivalent to Eq. 9 would hold for the parent’s
earnings too and may be inverted as
−1 ∗ −ρ ∗
∗
Gp = Ep1−ρ 1 + ι∗
ρ
(10)
where Gp , Ep , ι*, and ρ* correspond in the parent’s generation to Gc , Ec , ι,
and ρ in the case of the child. From Eqs. 10, 7, and 4, investment in the child’s
human capital in the wealth model is then given by
−1 ∗ −ψρ ∗ 1/(1−ρ)
∗
Hc = Epψ(1−ρ ) ργc (1 + ι)−1 1 + ι∗
ρ
.
(11)
Moreover, combining Eqs. 7, 9, and 10 yields the intergenerational, autoregressive earnings structure in this wealth model
ψ(1−ρ ∗ )/(1−ρ)
E c = κ0 E p
1 (1−ρ)
γc /
(12)
in which κ0 depends only upon the parameters ι, ρ, ι*, ρ*, and ψ. Note
the implication that, if the returns on human capital remain approximately
constant across generations (that is ρ ≈ ρ*), then the elasticity of the child’s
lifetime earnings with respect to earnings of the parent simply represents the
elasticity of transmission of inherent earnings ability, ψ, in this model.
In the case of the wealth model, a pure income effect does not change
investments in the child’s human capital, so Yp does not affect the child’s
earnings in Eq. 12. The child’s wealth (Yc ) increases with Yp , even in this case,
but this is entirely through bequests rather than enhanced earnings. On the
other hand, more able parents with higher earnings invest more in the child’s
human capital, as in Eq. 11. Positive auto-regression in intergenerational
earnings derives from the direct effect of inherited ability upon the child’s
earnings and from the additional human capital of the more able child.
For any credit-constrained family with homothetic preferences, investment
in human capital expands with wealth, even when earning ability is not
inherited (ψ = 0). Whether human capital investments are amplified if ψ > 0
depends upon the shape of preferences for these credit-constrained families.7
In the case of a Cobb–Douglas utility function, human capital investments are
Hc with respect to Gc has the same sign as η, thus being negative
(positive) for elasticities of substitution 1).
7 In Eq. 5, the partial derivative of
Intergenerational income immobility in Finland
1063
independent of Gc and hence independent of inherent ability transmission, as
may be seen from Eq. 6. Nonetheless, the child’s earnings remain enhanced
directly by inherent abilities in Eq. 8, so combining Eq. 8 with Eqs. 6, 7, and 10
provides
∗
Ec = κ1 Ypρ Epψ(1−ρ ) γc
(13)
where κ1 is a function of the parameters α, ρ, ι*, ρ*, and ψ. In Eq. 13,
parental wealth enhances the child’s earnings through a budget constraint
effect. Moreover the parent’s earnings are correlated with parental ability and
hence the child’s ability and earnings, provided ψ > 0.
2 Empirical counterparts
Permanent incomes are unobservable and data on entire lifetime earnings
are nowhere available to date. Estimation of equations such as Eq. 12 or 13
must rely, instead, upon observed incomes or earnings of individuals and of
their parents, tracked over some limited duration within the life cycle. For the
moment, a simple, quite traditional approach is adopted, to be extended in
Section 5. Let the natural logarithm of child c’s earnings observed at time t
(ect ) be written
ect = λc + Act + ξct
(14)
where λc is a fixed effect for person c, Act is a vector of polynomial terms in c’s
age at time t (act ), is a vector of coefficients, and ξct is a stochastic disturbance
term. Similarly, let the logarithms of earnings and income for parent p (of child
c), observed at time τ , be given respectively by
epτ = λp + ∗ Apτ + ξpτ
(15i)
ypτ = πp + ∗ Apτ + υpτ
(15ii)
where Apτ is a vector of polynomial terms in age of parent p at time τ , λp and
πp denote personal fixed effects, * and * are vectors of parameters, and ξpτ
and υpτ are stochastic disturbance terms.
Since (for now) age profiles are assumed common across individuals in Eqs.
14 and 15, the logarithms of the lifetime earnings and permanent income terms,
Ec , Ep , and Yp , may be defined to equal the individual fixed effects λc , λp ,
and πp plus terms common across observations. Substituting in Eqs. 12 and 13,
respectively, in the wealth model and credit-constrained cases leaves
ect = ω0 + ω1 epτ + Act + ∗ Apτ + μct
(16)
ect = χ0 + χ1 epτ + χ2 ypτ + X Act + X ∗ Apτ + νct .
(17)
Both models are nested within
ect = β0 + β1 epτ + β2 ypτ + B Act + B∗ Apτ + εct
(18)
1064
R.E.B. Lucas, S.P. Kerr
with β2 = 0 if the wealth model (Eq. 16) holds. In addition, since ω1 = ψ(1 −
ρ ∗ )(1 − ρ)−1 and χ1 = ψ(1 − ρ ∗ ), provided (1 − ρ ∗ ) > 0, the inheritance of
earning ability (ψ) approaches zero as β1 goes to zero. In other words, this
offers a key to exploring the role of inherited earning ability as modeled
by Becker and Tomes (1979, 1986). Presuming the rate of return to parent’s
education (ρ*) < 1, if the estimate of β1 approaches zero, this would indicate
an inherited earning ability parameter, ψ in Eq. 7, that also approaches zero.
Issues in estimation In instances where multiple time-period observations
on the earnings of the child are available, a number of options have been
explored to estimate intergenerational transmission elasticities. In his initial
approach, Zimmerman (1992) treats each year of data on sons’ earnings in
cross section and then considers the mean of these independent estimates.
Abul Naga (2002) notes that a more efficient estimator of this mean across
time periods is obtained from a between-individual estimator, adopting the
mean earnings of the child as dependent variable, provided that the transitory
components of the child’s and parent’s earnings are stationary. Such betweenindividual estimators are, perhaps, the most common choice (Behrman and
Taubman 1990; Mulligan 1999). A third alternative is to take fuller advantage
of the panel features of the data.8 The results presented in Section 4 are derived
from between-individual, time-average estimators. However, in the extensions
considered in Section 5, panel estimates are also presented.
The disturbance term in Eq. 18 is composed of
εct = ρ0 nγc − β1 ξ pt − β2 υ pt + ξct
(19)
−1
where ρ0 = (1 − ρ) in the wealth model and ρ0 = 1 in the credit-constrained
case. In general, εct thus includes time-dependent errors in measurement
from representing the parent’s lifetime earnings or permanent income by a
short panel of observations (ξ pt or υ pt ). The well-known consequence is that
ordinary least squares (OLS), applied to equations such as Eq. 18, generates
inconsistent estimates. Two main approaches have been adopted to address
this inconsistency.
The first though less common approach to the errors in measurement in Eq.
18 has been to instrument a single year of data on a parent’s earnings.9 The
second approach has been called the method of averages, in which parental
earnings or incomes are measured by the mean of time-series observations on
parents (early applications include Behrman and Taubman (1990), Altonji and
Dunn (1991), and Solon (1992)). Abul Naga (2002) considers some properties
8 For
example, Altonji and Dunn (1991) and Zimmerman (1992) both apply GMM estimators to
their panel data, while Lillard and Kilburn (1997) maximize a joint likelihood function derived
from an ARMA structure in transitory earnings.
9 Exploratory results in the present context suggest that the use of instrumental variables increases
the estimated transmission from family income but not from parental earnings, though identifying
appropriate instruments is, as always, a difficult task and this route is not followed in the present
paper.
Intergenerational income immobility in Finland
1065
of such a method of averages estimator in a wealth model, in which β2 = 0 and
the age terms (Act and Apτ ) are suppressed for simplicity. Assuming that the
three remaining stochastic components, nγc , ξ pt and ξct , are stationary, homoskedastic, and mutually uncorrelated, Abul Naga notes that the probability
limit for the OLS estimator of β1 in Eq. 18 is then
plim β1 =β1 −
β1 σξ2p Tp
(1 − ρ ∗ )2 σGp2 + σξ2p Tp
(20)
where σG2 p is the variance of nGp , σξ2p is the variance of ξ pt , and Tp represents
the number of periods over which the parent’s earnings are averaged. The
estimator in Eq. 20 is biased toward zero, but the extent of attenuation bias
diminishes with Tp . In his survey, Solon (1999) stresses the importance of
larger values of Tp and some testing along these lines is reported in Section 4.
3 The Finnish context: data and prior work on Finland
In the mid-1960s, personal identity codes were introduced in Finland. These
identity codes enable Statistics Finland to access information on individuals
across administrative registers, such as the Central Population Register and
Tax Register. Since 1970, Statistics Finland has compiled a population census
every 5 years and by 1990 the census was entirely register-based. By matching
the unique personal identifiers across the censuses, Statistics Finland has
constructed a Longitudinal Census Data File with panel data on the entire
population of Finland at 5-year intervals from 1970 to 1995. In addition,
since 1987, Statistics Finland has maintained the Longitudinal Employment
Statistics file which is updated annually. Since the same personal identifier
is adopted in both the Census and the Longitudinal Employment Statistics,
the two data sets can be merged, providing panel data on each resident of
Finland for 1970, 1975, 1980, 1985, and then annually from 1987 through 1999.
Throughout the entire database, cohabiting families are assigned a common
family identification number.10 Thus, it is possible to identify the parent(s)
living with a child in 1970 then to trace the child and parent(s) through to
1999.11
10 “A
family consists of a married or cohabiting couple and their children living together; or a
parent and his or her children living together; or a married or cohabiting couple without children.
Persons living in the household-dwelling unit who are not members of the nuclear family are not
included in the family population, even if they are related” (Statistics Finland 1995, p. 16).
11 Whether the adults with whom the child was living in a 1970 nuclear family are the biological
parents is not known. Children recorded in the family unit comprise biological and adopted
children of either spouse, though foster children and children in the care of the family are not
classified as part of the family.
1066
R.E.B. Lucas, S.P. Kerr
Table 1 Descriptive statistics
Sons and daughters
Sons
Daughters
Age range in 1970
Number of individuals
Percent with single parent in 1970
Number age 10+ in 1970
Number age 10+ with positive earnings observations
0–16
32,714
7.5
14,964
14,394
0–16
31,504
7.7
14,543
13,891
Sons and daughters age 10+ with positive earnings
Mean
SD
Mean
SD
Number of positive earnings observations per individual
Mean age during positive earnings
Time mean of log earnings
Time mean ratio of wages to earnings
Number of family income observations per individual
Time mean log family income
13.54
33.95
11.33
0.90
15.84
11.46
3.34
2.78
0.80
0.23
3.22
0.72
13.26
34.03
10.95
0.93
15.94
11.47
3.45
2.86
0.80
0.19
3.05
0.72
SD
9.94
8.03
5.60
1.22
0.39
Parents
Fathers
Mothers
Number of individuals
Number with positive earnings obs.
Percent died by 1999
Percent emigrated by 1999
26,727
26,266
31.7
3.0
Mean
39.43
52.12
9.44
10.78
0.69
28,749
26,377
12.5
3.0
Mean
36.75
51.08
9.94
10.39
0.79
Age in 1970
Mean age during positive earnings
Number of positive earnings observations per individual
Time mean of log earnings
Time mean ratio of wages to earnings
SD
9.24
7.37
5.33
1.29
0.35
To provide a family-based sample from these data, Statistics Finland drew
observations in two stages: the first stage is a 1 %, random sample of individuals from the 1970 Population Census and the second stage comprises all
family members cohabiting, at the time of the 1970 census, with the first-stage
individuals. For present purposes, observations are drawn from this sample on
all children aged 0–16 and living in a family in 1970.12 This provides panel data
on 32,714 sons, 31,504 daughters, and their parents.
Descriptive statistics on this sample are provided in Table 1. Note that
nearly 8 % of the sons and daughters are in single-parent families. In addition,
almost a third of the initial fathers had died by 1999, though only some 12.5
% of mothers died. The other source of attrition from these register-based,
panel data is by emigration, which claimed about 3 % of parents during this
interval.
12 Among the children ages 0–16 in our database, less than three quarters of 1 % are not living with
their family and hence omitted from the family-based sample. This very low fraction is in accord
with a report by The Population Research Institute (1995), which shows that in the mid-1980s
Finnish children continued to live with their parents even beyond age 16 more frequently than did
children in the other Nordic countries: 95 % of boys ages 16–19 were still at home in Finland as
were 88 % of girls.
Intergenerational income immobility in Finland
1067
Measures of earnings and incomes Estimation of Eq. 18 calls for measures
to represent earnings of parents and of their offspring, and to represent family
income. Our data include three measures of annual income for each individual:
(1) wages and salaries; (2) entrepreneurial (self-employment) income from
agriculture, business, and partnerships; and (3) total income subject to state
taxation, which includes most social security benefits.13 The data on wages
and salaries for each individual are reported directly to the Government
of Finland by employers. The data on self-employment and other taxable
income are compiled from the Tax Register. In Finland, taxable income is
recorded separately for each person; joint-filing of income taxes does not exist.
Throughout, all income measures are expressed in 2000 Finnish Marks, using
the annual, national cost-of-living index to deflate.
Two representations of the earnings of sons and of daughters (ect ) are
initially adopted as alternative, dependent variables: annual wages (including
salaries) and total annual earnings (the sum of wages and self-employment
earnings).14 The same measures on wages and earnings are also available for
parents. Self-employment earnings are a more important income source to
the parents than to the younger generation: almost a third of the average
father’s earnings, and about 20 % of the average mother’s earnings, are
derived from self-employment in years in which they earn (see Table 1).
The standard theoretical framework, as outlined in Section 1, refers to the
earnings of one parent (epτ ).15 Most of the empirical literature assumes this
13 “Income subject to state taxation does not include scholarships and grants received from the
public corporations for study or research, earned income from abroad if the person has worked
abroad for at least six months, part of the social security benefits received from the public sector
and tax-exempt interest income” (Statistics Finland 1995, p. 18). Since 1985 taxable income in
Finland does, however, include child, maternity, and unemployment benefits.
14 Most estimates of the intergenerational correlation in incomes are based on sample survey data,
though a few studies have also extracted income measures from tax registers (Corak and Heisz
1999, on Canada; Österberg 2000, on Sweden; and Mazumder 2005, who uses the US Social
Security Administration’s Summary Earnings Records). Register-based data, whether derived
from employer or tax records, have some advantages over survey responses, particularly where the
survey respondent is not the person employed. Yet register-based data also have some limitations.
The Finnish data lack information on hours worked, for example, and hence are inadequate to
examine hourly rates of pay and preclude computation of full income. The information on selfemployment earnings also has some drawbacks. The usual caveats apply with respect to the role of
capital income components within self-employment earnings. Moreover, given progressive taxes
on each individual, an incentive exists to spread family self-employment income across family
members where possible. Nonetheless, self-employment is not a major source of earnings for
young people. For only 5.2 % of sons and 5 % of daughters are more than 90 % of their earnings,
in an average year, derived from self-employment. For the remaining sons and daughters, selfemployment earnings represent only 7 and 4.4 % of their earnings, respectively.
15 This literature also presumes a single child. An extension to multiple children is to be the subject
of a separate paper, though preliminary findings indicate that the results in the present paper prove
robust to including controls for family size.
1068
R.E.B. Lucas, S.P. Kerr
refers to the father.16 To the extent that such parental earnings are meant to
embody the transmission of ability traits from parent to child, through genetic
or environmental effects as in Eq. 7, the assumption that fathers are the solitary
source may be questioned. Moreover, an exclusive focus on earnings of fathers
censors the children of single mothers from the sample (and vice versa). The
labor force participation rate of women is high in Finland. For example, in
1980, 78 % of mothers in our sample received earnings and 91 % generated
earnings at some stage in the panel. Overall, the time average earnings of
mothers represent 42 % of the parental total earnings in two-parent families. In
Section 4, a number of options are therefore considered, adopting the earnings
of fathers only, of mothers only, including earnings of both mother and father
separately, and earnings of whichever parent has the higher mean log earnings.
To represent family income, the taxable incomes accruing to both parents
are summed. To continue, such aggregation, even after instances in which
parents separate, may not be entirely appropriate. On the other hand, in
Finland, payments of alimony and child support are very common following separation, so the continued pooling of both parents’ incomes remains
eminently reasonable. Although the impacts of parental separation on child
outcomes are beyond the scope of the present paper, some results on the
sensitivity of intergenerational transmission estimates to alternative definitions
of family income, adjusted for separation, are noted in the following section.
In the estimation of the generic form (Eq. 18), the distinction between
family income and parent’s earnings is critical. In Finland, most transfers
from the state are included in the definition of taxable income. Under the
Finnish welfare state system, these transfers are a significant component of
income for a wide range of families. Our data do not permit a breakdown of
unearned income by source. However, separate data, reported by KELA (the
social insurance institution of Finland), show transfers, on average across all
households in Finland as a percentage of disposable income, rising from about
22 % in 1980 to over 43 % by 1995. Largely as a result of these contributions,
there is a substantial variation in the contribution of parental earnings at all
levels of family income in our data. For instance, within each percentile of
families, ranked by time-averaged family incomes during periods in which the
major earner is employed, the interquartile range in the contribution of the
father to family income never falls below 29 % of the median contribution,
and for mothers, the comparable measure is 37 %. The result is that the mean
coefficient of variation in fathers’ earnings is nearly 0.5 within each percentile
of family income. Moreover, there is a strong negative correlation between
combined earnings of the parents and receipt of other income at each level of
16 Certainly, the majority of empirical studies, both of the USA and elsewhere, seek to relate sons’
earnings to earnings of their fathers. A much smaller number of studies correlate the earnings
of daughters and of fathers, and a handful of studies have considered the mothers’ earnings.
See Solon (1999), Tables 3, 4, 5 and 6 for a summary of estimates for both sons and daughters.
Subsequent estimates of intergenerational income transmission for daughters include Österbacka
(2001) and Chadwick and Solon (2002).
Intergenerational income immobility in Finland
1069
family income. This unearned income falls fairly steadily as a fraction of family
income, from about the 4th to the 98th percentile of family income, suggesting
the importance of the welfare state system in generating these incomes (for
greater detail, see Electronic supplementary material—Appendix A).17
Prior work on the Finnish data In a series of papers, Markus Jäntti and
Eva Österbacka adopt the quinquennial census portion of the Finnish data to
estimate the intergenerational correlation in earnings. Estimates are provided
for both sons and daughters, related to either mothers’ or fathers’ earnings
separately. For example, Österbacka (2001) applies ordinary least squares to
equations relating the mean annual earnings of sons and of daughters in 1985,
1990, and 1995 to the mean annual earnings of fathers or mothers in 1970
and 1975, controlling for age and age squared of the child and parent. These
specifications provide the following estimates of intergenerational earnings
transmission elasticities:18
Father’s earnings
Mother’s earnings
Sons
0.129 (25.8)
0.037 (9.3)
Daughters
0.100 (16.7)
0.023 (4.6)
In comparison to somewhat similar estimates for other countries,
Österbacka (2001, p. 480) concludes “that intergenerational mobility is relatively high in Finland.” Jäntti et al. (2007) estimate higher transmission
elasticities, at least for sons:19
Father’s earnings
Family income
Sons
0.173 (8.9)
0.220 (11.1)
Daughters
0.080 (4.1)
0.112 (11.4)
In these latter estimates, earnings observed in 1993 and 2000 are truncated
at a lower bound. These measures are then deployed to project what earnings
would be at age 40, for sons and daughters born between 1958 and 1960.
Fathers’ earnings are measured in 1975 when their mean age was 47, again
projecting their earnings at age 40 for the right-hand variable. Family income
is measured by the sum of both parents’ earnings in 1975, at which stage the
difference between earnings and total taxable income remained quite small.
17 The
appendices to this paper are available as Boston University Institute for Economic Development Discussion Paper No. 221 at http://www.bu.edu/econ/files/2011/09/Lucas_Kerr.pdf.
18 Source: Österbacka (2001, Table 3). t statistics for a zero null hypothesis are shown in parentheses, calculated from the reported standard errors. In addition, Österbacka explores estimates
within quintiles of parents’ earnings and the correlations between siblings’ earnings, while also
undertaking a decomposition of the intergenerational correlations. See also Jäntti and Österbacka
(How much of the variance in income can be attributed to family background? Empirical Evidence
from Finland, unpublished).
19 See Tables 2 and 9 in Jäntti et al. (2007). t statistics for a zero null hypothesis are shown in
parentheses, calculated from the reported confidence intervals.
1070
R.E.B. Lucas, S.P. Kerr
In large part, these projections and restrictions on the sample are imposed to
render results as comparable as possible to those generated by Jäntti et al.
for the UK, USA, and other Nordic countries. By these measures, Finland
again displays greater intergenerational mobility than the UK and far greater
mobility than the USA, though exhibiting a mixed ranking among the Nordic
states.
In their study of the impact of the comprehensive school reform enacted
in Finland between 1972 and 1977, Pekkarinen et al. (2009) estimate much
higher transmission elasticities from fathers’ to sons’ earnings, on the order
of 0.30 prior to the reform and 0.24 after treatment. Here, sons’ earnings are
measured in 2000, for cohorts born between 1960 and 1966, of which the reform
impacted portions of those born from 1961 to 1965 and all born thereafter.
Fathers’ earnings are represented by the average log of earnings from 1970
through 1990. It is important to note, however, that in this study, the measure
of “earnings” for both sons and fathers actually refers to our taxable income
measure, rather than wages or wages plus self-employment earnings. Note also
that in the following Section 4 the focus is on a set of sons and daughters all of
whom were too old to have been affected by this comprehensive school reform.
All of our samples of sons and daughters do, however, fall in an age
range that potentially benefitted from the substantial expansion in government
guaranteed loans for tertiary-level students, offering subsidized interest rates,
which occurred in 1969. Yet, despite this expansion, surveys among students
at the time indicate that these government loans and grants represented only
about 50 % of subsistence costs (Blomster 2000).
4 Initial estimates
The results in this section are from between-individual regression models that
adopt a time-series average of log earnings or wages of sons and daughters as
the dependent variable. Initially, only years of positive earnings (wages) are
included when computing the mean log earnings (wages) of both child and
parent. As a further restriction in this section, only sons and daughters who
are at least age 25 in 1985 are included.20
Table 2 first reports estimates, obtained by OLS, of transmission from
parental earnings (wages) to those of sons and daughters. As mentioned in
Section 3, separate estimates are reported with several alternative measures
20 The age range varies considerably over which earnings of children are included in prior studies.
Solon (1992) imposes a lower bound at age 25 in his study of sons’ earnings, while the lower bound
in Zimmerman (1992) is 29, and Dearden et al. (1997) look at UK sons and daughters when they
are 33. Although only children who are at least 25 in 1985 are included in our initial analysis in
Section 4, any earlier positive earnings (or wages) of these children are included in computing
mean earnings, provided the son or daughter was at least 20 at the time of observation in the 1975
or 1980 census. To the role of age in these estimates, Section 5 returns.
11,691
0.124
(19.61)
12,957
0.058
(10.13)
Sons
11,024
0.102
(14.00)
12,528
0.051
(9.20)
9,338
0.096
(12.72)
0.067
(8.27)
11,346
0.048
(7.82)
0.043
(7.39)
13,377
0.141
(22.23)
14,139
0.062
(11.00)
0.144
(20.49)
13,377
0.054
(9.96)
14,139
11,340
0.042
(7.71)
12,500
0.044
(7.10)
Daughters
t statistics for a zero null hypothesis in parentheses. Standard errors from heteroskedastic-consistent matrix
No. of observations
Major earner’s mean log wages
Mean log combined wages
Mother’s mean log wages
Mean log wages
Father’s mean log wages
No. of observations
Major earner’s mean log earnings
Mean log combined earnings
Mother’s mean log earnings
Mean log earnings
Father’s mean log earnings
Table 2 Transmission from parental earnings and wages (between-individual estimates on positive earnings)
10,732
0.051
(8.36)
12,138
0.055
(9.53)
9,118
0.027
(4.07)
0.038
(5.42)
10,993
0.035
(5.23)
0.044
(7.23)
12,954
0.055
(10.25)
13,645
0.059
(9.74)
0.061
(10.11)
12,954
0.055
(9.55)
13,645
Intergenerational income immobility in Finland
1071
1072
R.E.B. Lucas, S.P. Kerr
of parental earnings. The only other explanatory variables included in these
specifications are the age of the child and of the parent(s), each expressed
as a separate polynomial to the fourth power. More particularly, the ages of
both the child and parent are measured as the mean age during which positive
earnings are reported.21 The correspondence of a mother’s observed earnings
to her lifetime earnings may well differ, at any given age, from that of a father.
Consequently, when looking at the major earner’s earnings, a dummy for their
gender is also included, both separately and interacted with their polynomial
age profile. For brevity, coefficients on the age vectors of the child and parents
are suppressed in Table 2. The estimates based on father’s earnings omit
single-mother families, and vice versa, while the estimates including father’s
and mother’s earnings separately include only two-parent families in which
both parents work at some point. The last two estimates include all instances
in which at least one parent works at some point (which encompasses more
than 98 % of both sons and daughters with positive earnings) and combined
earnings are the sum of both parents’ earnings in any period in which at least
one works.
In each variant in Table 2, the transmission elasticities from parental earnings or wages to those of their sons and daughters are precisely estimated but
low. No matter which measure of parental earnings is adopted, transmission
to the next generation remains small by comparison with estimates for other
countries. In other words, these initial results are quite consistent with prior
findings on Finland, summarized in Section 3.22
Table 3 shows OLS estimates of transmission from family income, with
and without parental earnings included.23 In all three specifications, the transmission elasticity from family income proves larger than the transmissions
from parental earnings in Table 2. Moreover, the coefficient on family income
proves relatively insensitive to the inclusion of parents’ earnings in Table 3.
For instance, in the last specifications in Table 3, the coefficient on family
income exceeds that on earnings of the major earner with more than 99.9 %
confidence for both sons and daughters. These estimates of the transmission
elasticity from family income to earnings of the next generation are larger for
21 Any
distinction between this mean age measure, the mean age over which the individual is
observed, or age at a specific point in time, is not always drawn in this context. However, in
practice, the distinction has relatively little impact on the estimates.
22 Projecting to age 40 and censoring earnings data, Jäntti et al. (2007) obtain slightly higher
estimates for sons. The following section returns to the issue of an age interaction.
23 Given the strong similarity in the results for earnings and wages in Table 2, for brevity, only
results for the case of earnings are reported in the remainder of the paper. The specifications in
family income include mean age of the father during periods when he reports positive income
and a similar measure for the mother, both in polynomials through the fourth power, plus dummy
variables for cases where no father or mother is initially present.
Intergenerational income immobility in Finland
1073
Table 3 Transmission from family income (between-individual estimates on positive earnings)
Mean log earnings
Sons
Father’s mean
log earnings
Mother’s mean
log earnings
Major earner’s mean
log earnings
Family mean
log income
No. of observations
Daughters
0.012
(1.94)
0.006
(0.98)
0.225
(20.77)
14,364
0.234
(16.06)
11,346
0.002
(0.30)
0.015
(2.42)
0.019
(2.62)
0.212
(17.26)
14,139
0.173
(14.49)
13,860
0.174
(11.20)
10,993
0.025
(3.19)
0.153
(11.00)
13,645
t statistics for a zero null hypothesis in parentheses. Standard errors from heteroskedasticconsistent matrix
boys than for girls, but for both genders these transmission elasticities from
family income are significantly larger than from parental earnings.24
Sensitivity analysis: alternative approaches to family income As anticipated in
Eq. 20, the estimates of transmission from family income are sensitive to the
number of observations on parents’ incomes. This is explored more closely in
the Electronic supplementary material—Appendix B, where it is demonstrated
that the transmission elasticity for sons rises asymptotically and significantly
with the number of observations on family income, with smaller increments
the wider apart are these observations in time.
In Section 3, it was noted that incomes of both parents are aggregated in
each time period even if they are not cohabiting at the time. It is shown in
the Electronic supplementary material—Appendix B, however, that this is not
a major concern; estimates of the transmission elasticity from family income
prove quite insensitive to distinguishing states of cohabitation.
The distinction between parental earnings and family income is rendered
possible in this study because of the importance of unearned income to a wide
range of families. However, if this unearned income is driven by prior earnings,
then the linear specification in Eq. 18 may not fully control for the effect of
parental earnings (and hence inherited earning abilities) upon earnings of the
next generation, which might act indirectly through the family income term.
24 Several US studies also estimate transmission from family income to be larger than from parental
earnings measures (see the review in Solon 1999). However, these studies uniformly concentrate
upon transmission from parents’ family income to child’s family income, the latter presumably
reflecting assortative mating in addition to the considerations in Section 1. As far as we are aware,
no study incorporates both family income and parental earnings as regressors. A test for equality of
all coefficients between the two stages of sampling within our data, based on the last specification
in Table 3, gives F(17, 14,105) = 0.937 for sons and F(17, 13,611) = 0.564 for daughters. Pooling
the subsamples appears not to be problematic.
1074
R.E.B. Lucas, S.P. Kerr
The estimated weak associations with parental earnings even in the absence of
family income measures, as reported in Table 2, suggest that this is not the case.
However, it is possible to explore the issue further. Asset income may well
reflect savings out of prior earnings, but it is state transfers rather than property
income that are the major issue in this context; as noted previously, unearned
income declines monotonically as a fraction of family income across almost the
entire spectrum of income levels, with property income assuming importance
only among the top two percentiles. However, receipts of state transfers in
the forms of unemployment, maternity, and paternity benefits are also tied to
prior earnings (see the notes on state transfers in the Electronic supplementary
material—Appendix D). Before 1985, these transfers were quite tiny and none
was taxable income and hence not part of our measure of family income. From
1987 onwards, our data report these transfers to each individual. Excluding
these measures from family income diminishes the estimated transmission
elasticity from family income, but only very slightly, by about 0.01 for both
sons and daughters (see the results tabulated in the Electronic supplementary
material—Appendix B). Retirement incomes are also tied to prior earnings,
but separate data on these are not available. However, if the measure of
family income includes only periods in which both parents earn (and hence
are not retired), as well as subtracting the aforementioned unemployment and
maternity benefits, the estimated transmission elasticities remain essentially
unaffected.
Besides the method of averages, an alternative approach in estimating
permanent family income is as a family fixed effect from panel estimation of
log family income on the profile of ages reached by parents. The implication of
adopting this alternative approach, within the first specification from Table 3,
is actually to increase the estimated transmission elasticities from family
income for both sons and daughters (see the results tabulated in the Electronic
supplementary material—Appendix B).
Lastly, a number of studies have explored aspects of nonlinearity in intergenerational transmission elasticities (see, for instance, Corak and Heisz 1999;
Grawe 2004; Bratsberg et al. 2007). The first linear specification from Table 3
is displayed in Fig. 1, along with an estimate that replaces the continuous
measure of family mean log income with a vector of dummy variables for each
percentile of family income. For neither gender is any clear departure from
linearity apparent in Fig. 1.25
Sensitivity analysis: three generations Two issues are taken up here with
respect to the influence of grandparents’ incomes. The first arises with respect
25 Restricting the specification to be linear, as opposed to the far more parsimonious form, generates an F statistics for sons of F(97, 14,251) = 1.646 and for daughters F(97, 13,747) = 1.114. A
least absolute deviation estimate at the median also generates only a slightly lower value for transmission from family income than does the estimate at the mean. See Electronic supplementary
material—Appendix B.
Intergenerational income immobility in Finland
1075
Linearity Check
Sons
11.8
Percentile dummies
Linear specification
11.7
Sons' mean log earnings
11.6
11.5
11.4
11.3
11.2
11.1
11
10.9
10.8
9.5
10
10.5
11
11.5
Family mean log income
12.5
13
12
12.5
13
Daughters
11.4
Percentile dummies
Linear specification
11.3
Daughters' mean log earnings
12
11.2
11.1
11
10.9
10.8
10.7
10.6
9.5
10
10.5
11
11.5
Family mean log income
Fig. 1 Linearity check
to the endogenous taste hypothesis of Easterlin et al. (1980). Behrman and
Taubman (1989) achieve a testable form of the fertility expression of this hypothesis; in the event that parents’ taste for smaller families is positively shaped
by higher incomes of their own parents, incomes of grandparents should be
negatively associated with number of children in the third generation. To
establish a parallel in the present context, suppose that the parental altruism
parameter, α, depends upon grandparents’ incomes. Since α is embodied in the
intercept in the generic earnings equation for the child, at least in the creditconstrained context, incomes of grandparents should appear as an additional
argument in shaping earnings of the third generation. In particular, if richer
1076
R.E.B. Lucas, S.P. Kerr
grandparents induce a taste for quality rather than quantity of children, the
association with grandparents’ incomes would be positive.
The second potential source of influence from grandparents’ incomes arises
from the use of Eq. 10 in deriving Eq. 13. This involves a tacit presumption
that human capital investments in the parent, made by grandparents, are not
credit-constrained. If, instead, the human capital of the parent were formed
in a credit-constrained process, then using parental equivalents to the human
capital choice (Eq. 6) and earning definition (Eq. 8) and inverting the latter
yields
−1 ρ ∗
∗
Gp = Ep Yg−ρ 1 + ρ ∗ α ∗ ρ ∗ α ∗
(21)
where Yg represents the wealth of the grandparents. Using this in conjunction
with Eqs. 6 and 8 then provides a second-order, auto-regressive, intergenerational transmission process:
∗
Ec = κ2 Ypρ Epψ Yg−ψρ γc
(22)
where κ2 is a function of α, ρ, α*, ρ*, and ψ. This formulation suggests that
the logarithm of grandparents’ income should have a negative influence upon
log earnings of children in the third generation if ability transmission (ψ) is
important. The intuition behind such a negative association derives from the
proposition that richer grandparents would have invested more in the human
capital of their children, given the abilities of these offspring to earn (Gp ).
Given Ep , Gp (and hence Gc and in turn Ec ) would then be lower, the wealthier
are the grandparents.26
Beyond interest in these two hypotheses themselves, an additional concern
warrants noting. To the extent that parents’ and grandparents’ incomes are
positively correlated, omission of the latter in examining intergenerational
transmission would result in omitted variable bias, though of opposite direction
depending upon which hypothesis dominates.
To explore this issue here, a separate data set on Finland is adopted.
Statistics Finland manually matched a random 10 % sample of the individuals
from the 1950 census with their later-established identity numbers. Combined
with the same sampling frame of data described in Section 3, information on
three generations of Finns is thus derived, with grandparents observed in 1950
and 1970 onwards, plus parents and their children observed from 1970 to 1999
(for greater detail on these data, see Pekkala and Lucas 2007).
Table 4 shows the results of OLS estimates on these data. An ambiguity
arises with the measurement of Yg , which may refer to income of the father’s
parents, the mothers’ parents, or both. Accordingly, Table 4 includes each
of these three variants. Given that only a 10 % sample was drawn from the
26 Behrman
and Taubman (1985) test a three-generational, schooling equivalent to Eq. 22 on a US
sample. The same data set is used in Behrman and Taubman (1989) for a three-generational test
of the Easterlin et al. (1980) fertility hypothesis. See also Robertson and Roy (1982), Warren and
Hauser (1997), and Jeon and Shields (2005).
25,744
0.192
(21.60)
27,289
0.249
(27.91)
0.195
(21.29)
−0.011
(1.64)
0.254
(27.46)
−0.019
(2.68)
0.008
(0.77)
25,735
0.184
(11.04)
−0.011
(1.65)
0.032
(2.94)
27,281
0.210
(12.43)
−0.019
(2.69)
30,829
0.180
(23.15)
31,853
0.243
(29.82)
0.008
(1.22)
0.176
(22.11)
−0.006
(0.93)
0.244
(28.84)
t statistics for a zero null hypothesis in parentheses. Standard errors from heteroskedastic-consistent matrix
No. of observations
Parent’s mean log earnings
Mother’s parents’ mean log income
Father’s parents’ mean log income
Daughters’ mean log earnings
Family mean log income
No. of observations
Parent’s mean log earnings
Mother’s parents’ mean log income
Father’s parents’ mean log income
Sons’ mean log earnings
Family mean log income
Table 4 Three generations
0.008
(1.20)
0.025
(3.14)
30,680
0.144
(10.63)
−0.006
(0.95)
0.026
(3.29)
31,730
0.206
(15.11)
0.165
(6.93)
0.296
(11.02)
2,830
0.152
(5.94)
−0.013
(0.58)
0.058
(2.74)
2,905
0.290
(9.94)
−0.013
(0.60)
0.011
(0.53)
0.087
(1.77)
−0.012
(0.54)
0.056
(2.69)
0.049
(1.59)
0.209
(4.01)
−0.012
(0.55)
0.010
(0.51)
0.060
(1.76)
Intergenerational income immobility in Finland
1077
1078
R.E.B. Lucas, S.P. Kerr
1950 census, the chances of identifying the parents of both father and mother,
living in 1970, are low, and the resulting sample is therefore much smaller when
incomes of both sets of grandparents are included. In each of the three variants,
three specifications are reported. The first is a baseline, excluding incomes
of grandparents, for comparison with Table 3. If anything, the estimated
transmission from family income is slightly higher in the three-generation
sample, though any differences are quite small. The second specification
introduces the several alternative measures of grandparents’ incomes. The
only hint of a positive association is from mother’s parents’ incomes to earnings
of daughters, when both sets of grandparents are included. Otherwise there
seems little to suggest a positive association to support the Easterlin–Pollak–
Wachter hypothesis. The third specification is a logarithmic transformation of
Eq. 22, adding in parent’s earnings (here interpreted as earnings of the major
earner as in Table 3). Several of the coefficients on grandparents’ incomes
are negative, but only in the case of son’s earnings is father’s parents’ income
significantly below zero.
The three generation tests clearly do not prove particularly powerful in
distinguishing the two hypotheses; indeed, it is feasible that the two simply
offset each other. It is, however, important to note that inserting grandparents’
incomes does little or nothing to diminish the estimated transmission from
parental family income. Omitting grandparents’ incomes does not appear to
incur significant bias.
Sensitivity analysis: inclusion of zero earning observations A number of alternative representations of parental earnings have already been explored in connection with Table 2. However, Couch and Lillard (1998) raise an additional
doubt about the validity of estimated intergenerational earnings elasticities
that exclude observations when either the parent or child has zero earnings, an
exclusion also imposed in our results so far. In the Electronic supplementary
material—Appendix C, it is demonstrated that including zero earning observations in the Finnish context actually raises the intergenerational transmission
estimates from family income but not from parental earnings. Although higher
income families’ children are more likely not to work while continuing their
education, this effect is outweighed by the propensity of sons and daughters
from lower income families to be unemployed or to drop out of the labor force
(see also Ekhaugen 2009).
5 On the age dependence of intergenerational transmission
In the transition to an empirically tractable specification, it is assumed in
Eq. 14 that the age profile of sons’ and daughters’ earnings is common to
all individuals or at least that any variations are uncorrelated with parental
earnings and incomes. Consider, however, the possibility of an interaction
Intergenerational income immobility in Finland
1079
between age and the child’s acquired human capital in shaping earnings.27
Specifically, let the private returns on human capital asymptotically approach
a limit as the person ages, such that
ect = λc + λ hc act + Act + ξct
(23)
where hc / nHc . Given a common discount factor (δt ) and life horizon among
individuals, lifetime earnings may then be defined as the present value of ect
such that
nEc = δt ect = λc + λ hc + λ0
(24)
where δt ≡ 1, = δt act , and λ0 comprises the discounted elements of Act
and ξct . Substituting for λc from Eq. 24 in Eq. 23 yields
ect = −λ0 + nEc − λ hc + λhc /act + Act + ξct .
(25)
Substituting in Eq. 25 from Eqs. 11 and 12 in the case without credit constraints, or Eqs. 6 and 13 in the credit-constrained case, together with Eq. 15
then leaves an expanded version of the generic regression model (Eq. 18):
ect = β0 + β1 epτ + β2 ypτ + β3 epτ act + β4 ypτ act + β act + BAct
+ B∗ Apτ + Ba Apτ act + εct .
(26)
In which,
Without credit constraint
β1
β2
β3
β4
= ψ (1 − ρ∗) (1 − ρ)−1 (1 − λ )
=0
= ψ (1 − ρ∗) (1 − ρ)−1 λ
=0
With credit constraint
β1
β2
β3
β4
= ψ (1 − ρ∗)
= ρ − λ
=0
= λ.
If λ < 0, then the intergenerational transmission, from parent’s earnings in the
wealth model and income in the credit-constrained case, should rise with age
at which the child’s earnings are observed.28
Age interaction estimates In a between-individuals estimator, any distinction
between an age effect and a cohort (or time) effect cannot be discerned.
Yet there are reasons to suspect that intergenerational economic mobility in
Finland could differ across cohorts of children too, even within the time span
27 See
the discussion in Card (1999) for instance.
28 Such an effect has previously been detected in the US context by Reville RT (Intertemporal
and
life cycle variation in measured intergenerational earnings mobility, unpublished), though Lee
and Solon (2009) find less clear-cut patterns with respect to an age interaction. Equation 26 may
be thought of as a specific form of the model developed in Haider and Solon (2006) which notes
the potential age dependence of parameters such as β1 and β2 in Eq. 18. See also Jenkins (1987).
1080
R.E.B. Lucas, S.P. Kerr
of our sample.29 A panel estimation approach is therefore initially adopted to
explore these potential age and cohort interaction effects.30
Only observations on positive earnings are included, in years in which the
person is at least 21 years of age. The data refer to the continuous, annual
portion of the sample from 1987 to 1999 and fixed effects are included for
the year of observation. An error-components estimator is applied to these
panel data in the first two results shown in Table 5, while the third adopts
a maximum-likelihood estimator with an AR1 process in the errors. The
estimated transmissions from the earnings of the father and mother in the
first estimate, and of the major earner in the next two estimates, remain tiny
and do not rise with age of the son or daughter. On the other hand, for both
sons and daughters, the estimates indicate a significant and substantial rise in
transmission elasticity from family income as the child ages. For both genders,
similar results are found, though not shown, if a minimum age cutoff of 25 is
adopted rather than 21.
To explore whether the interaction between family income and age reflects
a cohort effect, rather than age, a term in family income relative to cohort
is included in these panel estimates. Cohort is defined to equal 1 for sons
and daughters who were 16 in 1970, through 17 for those aged less than 1
year in 1970. This cohort term proves to be tiny and virtually orthogonal to
the other terms of interest.31 The fixed effects for each year of observation
(which are not tabulated) clearly reflect the substantial drop in real earnings
during the massive recession after 1991. However, exclusion of these time fixed
effects makes virtually no difference to the parameter estimates shown. Not
surprisingly, given the orthogonality of the terms in family income relative to
cohort and time fixed effects, the between-individual, OLS estimates in Table 5
also prove quite close to those obtained from the pooled, panel data.
Figure 2 illustrates the implied age profiles of the family income transmission elasticities from three of these estimates: random effects and OLS
when earnings of both parents are included and the random effects case
with earnings of the major earner (in which single-parent-earning families
are included). Figure 2 also plots two additional error-components estimators,
comparable to the first estimates in Table 5. In both additional plots, family
29 For example, although all of the sons and daughters were potential beneficiaries of the guaranteed student loan scheme introduced in 1959 and explicitly subsidized after 1969, only the younger
cohorts tended to benefit from the massive expansion in this system after the mid-1980s. Moreover,
the various cohorts encountered the effects of the deep recession of the early 1990s at different
ages (Hämäläinen and Hämäläinen 2005).
30 Tests, using the t-bar statistics suggested by Im et al. (2003), reject with very high levels
of confidence unit roots in the panel earnings data. See Electronic supplementary material—
Appendix E.
31 Pekkala and Lucas (2007) adopt more flexible forms of the cohort interaction and confirm the
absence of any clear trend in family income transmission across cohorts born between 1954 and
1970 in Finland, though a downward decline is apparent among earlier cohorts born from 1930 to
1950.
299,637
−21.478
(37.36)
−0.007
(2.98)
0.924
(44.02)
0.017
(4.02)
0.008
(1.91)
349,022
0.057
(4.05)
0.811
(37.89)
−0.984
(2.34)
−18.419
(29.43)
−0.004
(1.95)
Sons’ log earnings
Panel
Random effects
0.029
(28.77)
0.645
(34.30)
−0.178
(1.63)
−13.845
(23.97)
−0.005
(3.46)
0.573
(409.5)
349,022
AR1
31,182
−10.691
(5.74)
0.024
(3.94)
0.546
(8.92)
26,644
−14.288
(6.74)
0.678
(9.67)
0.014
(3.04)
0.004
(0.96)
Time mean
OLS
280,082
−13.591
(20.55)
−0.005
(2.09)
0.591
(24.98)
0.008
(1.89)
0.015
(3.64)
327,165
0.057
(3.62)
0.477
(19.61)
−0.829
(1.72)
−10.783
(15.02)
−0.004
(1.92)
0.036
(31.90)
0.406
(19.78)
−0.202
(1.70)
−9.086
(14.46)
−0.004
(2.32)
0.526
(346.4)
327,165
Daughters’ log earnings
Panel
Random effects
AR1
t statistics for a zero null hypothesis in parentheses. Standard errors from heteroskedastic-consistent matrix
No. of observations
AR1 parameter
Family mean log income/cohort
Family mean log income/child’s age
Major earner’s mean log earnings/child’s age
Family mean log income
Major earner’s mean log earnings
Mother’s mean log earnings
Father’s mean log earnings
Table 5 Age interaction effects
29,905
−7.848
(4.32)
0.026
(4.15)
0.383
(6.27)
25,556
−10.183
(4.97)
0.477
(6.89)
0.006
(1.28)
0.016
(3.44)
Time mean
OLS
Intergenerational income immobility in Finland
1081
1082
R.E.B. Lucas, S.P. Kerr
Age Profile of Family Income Transmission Elasticity
Sons
RE Both
Family Income Transmission Elasticity
0.4
Spline
0.3
OLS
RE Major
0.2
0.1
FY
FY as FE
0
25
30
35
40
45
Age
0.3
RE Both
Daughters
Family Income Transmission Elasticity
0.25
Spline
0.2
OLS
0.15
RE Major
0.1
0.05
FY
FY as FE
0
25
30
35
40
45
Age
Fig. 2 Age profile of family income transmission elasticity
mean log income and family mean log income relative to child’s age are
replaced by family mean log income multiplied by a vector of dummies for
each age at which the sons and daughters are observed. In one of these plots,
Intergenerational income immobility in Finland
1083
the time mean of family log income is adopted, while in the other the estimate
of family log income as a fixed effect, as described in the previous section,
is deployed. These flexible formulations indicate a fairly monotonic rise in
family income transmission elasticity, at least through about age 40, for both
sons and daughters. Indeed, the hyperbolic specification adopted in Eq. 26
approximates this far more flexible age profile reasonably well.
Age and the returns to education It is hypothesized, in Eq. 25, that an
interaction between human capital and age, in shaping earnings, underlies
the rising transmission from family incomes as sons and daughters age. To
explore this hypothesis a little more closely, let the logarithm of investment
in the child’s human capital (hc ) be approximated by the number of years of
education completed. Interacting this measure of schooling with age at which
earnings are observed, the first approach reported in Table 6 demonstrates
that at least OLS estimates of returns to schooling indeed rise with age, for
both sons and daughters in Finland.32
Such OLS estimates of the returns to education are, however, generally
biased and inconsistent to the extent that schooling is correlated with omitted
family background or ability effects. In an attempt to address this, a number
of studies instrument years of schooling with some measure of educational
reform or distance to school facilities (see Card 1999, for a critical review). The
possibility of instrumenting years of schooling with institutional information
about regional schooling in Finland was investigated. However, the available
set of instruments proved too weak to offer meaningful results.33 Instead,
Table 6 reports OLS estimates on the first differences between the eldest and
next oldest son and daughter pairs. To the extent that siblings share a common
family environment and possibly similar abilities, any bias may be reduced in
such comparisons. In fact, the results on first differences prove remarkably
similar to the initial estimates. Moreover, if the set of observations is extended
to include all sibling pairs of same gender, in descending age order, the results
are hardly affected.
Both the influence of family income upon a child’s earnings and the rate of
return to schooling are thus estimated to rise with age of the child. To explore
are included, but not tabulated, for the terms a−1
ct , Act , Apτ , and Apτ /act from Eq. 26.
instruments available include the following: the proportions of school-age children in a
region living 0–3, 3–5, or more than 5 km from a school in 1959 and in 1969; a dummy variable for
whether the child was affected by the conversion to a comprehensive system, which was introduced
on a rolling basis throughout the country, roughly from north to south, between 1972 and 1976
(cf. Black et al. 2005, on Norway; on the Finnish reform, see Pekkarinen et al. 2009); a dummy
variable for whether the child’s region of residence at age 18 contained a university town and, if
not, distance to such a region (see Conneely and Uusitalo 1998). The partial F statistics for the
seven available measures, in a first-stage regression including controls for a−1
ct and Act , are F(7,
31,501) = 3.98 for sons and F(7, 30,227) = 2.81 for daughters.
32 Controls
33 The
31,514
0.280
(20.83)
−6.867
(16.20)
7,712
0.259
(9.15)
−6.354
(7.25)
0.562
(9.32)
−11.29
(6.13)
31,182
Recursive
model
0.272
(19.36)
−6.884
(15.97)
0.027
(4.50)
0.243
(3.85)
−3.209
(1.66)
31,182
OLS
30,240
0.234
(16.11)
−4.949
(11.23)
7,099
0.211
(6.54)
−4.365
(4.38)
0.367
(6.10)
−7.509
(4.19)
29,905
Daughters’ mean log earnings
OLS
Sisters
Recursive
difference
model
t statistics for a zero null hypothesis in parentheses. Standard errors from heteroskedastic-consistent matrix
No. of observations
Family mean log income/child’s age
Family mean log income
Major earner’s mean log earnings
Child’s years of education/child’s age
Child’s years of education
Sons’ mean log earnings
OLS
Brothers
difference
Table 6 Age interaction and the returns to education
0.223
(15.02)
−4.830
(10.72)
0.030
(4.87)
0.143
(2.32)
−2.675
(1.45)
29,905
OLS
1084
R.E.B. Lucas, S.P. Kerr
Intergenerational income immobility in Finland
1085
further, the connection between these patterns considers the following simple,
recursive model:
ect = θ0 + θ1 epτ + θ2 sc + θ3 sc act + θ4 ypτ + θ5 ypτ act + θ6 act + Act
+ ∗ Apτ + a Apτ act + ζct
sc = ϕ0 + ϕ1 ypτ + ϕ2 act + Act + ∗ Apτ + a Apτ act + ζc
(27i)
(27ii)
where sc represents years of schooling for child c, ωct and πc are stochastic
disturbances, and other terms are previously defined. Why should Eq. 27i
include terms in parental income (ypτ ), given the child’s schooling? At least
two explanations have been offered in the literature. Mulligan (1999), upon
finding that US children’s wages are positively correlated with their parent’s
permanent income, given schooling of the child, interprets this in terms of a
model with two dimensions of ability. The first dimension shifts the earnings
function (as in our terms Gc and Gp ); the other affects the returns to human
capital (ρ* for the parent). In the wealth model context, Mulligan then notes
“Because {Gc } is positively correlated with {Gp and ρ*}..., the earnings of adult
children are positively related to parental income within a group of adult
children whose parents made the same human capital investments.”34 Given
that both Gp and ρ* affect parental earnings (Ep ), the additional dimension to
ability then offers further justification for inclusion of epτ in Eq. 27i but not for
inclusion of ypτ , given epτ . Becker (1972), in commenting on related findings by
Bowles (1972) for the USA, offers a second interpretation. Specifically, Becker
suggests that the residual effect of parental income upon their child’s earnings,
controlling for the education level of the child, may simply reflect aspects of
human capital not well represented by years of schooling. For instance, if
the child’s log human capital were defined by hc ≡ sc + qc , where qc might
represent quality of schooling or postschool training, then the ypτ terms in
Eq. 27i would stem from the role of the omitted variable qc in the creditconstrained case.
Estimates of Eq. 26, such as those reported in Table 6, may be thought of as
reduced forms of Eq. 27 (Levine and Mazumder 2002) such that
β2 = (θ2 ϕ1 + θ4 )
β4 = (θ3 ϕ1 + θ5 ) .
(28)
After estimating both equations in Eq. 27 as seemingly unrelated regressions,
the third columns in Table 6 report the estimates of the coefficients on ypτ
and ypτ /act implied by the formulae in Eq. 28. These derived estimates clearly
resemble those obtained from direct estimation of the reduced form by OLS
in Table 6. In fact, neither of the cross equation constraints in Eq. 28 can be
rejected at a 99 % confidence level, either for sons or daughters.
34 Mulligan (1999: S197). The portions in brackets are converted into notation of the present paper.
1086
R.E.B. Lucas, S.P. Kerr
The final columns in Table 6 report between-individual, OLS estimates of
Eq. 27i. The estimates of the returns to schooling and its age profile, θ2 and
θ3 , prove essentially orthogonal to inclusion of the additional terms in family
income and major earner’s earnings. The estimated coefficients on the major
earner’s mean log earnings remain tiny, suggesting that Mulligan’s second
dimension of ability may not be very important in this context. The estimated
coefficients on family mean log income would, however, be consistent with
Becker’s suggested interpretation, in a credit-constrained case, that the quantity of schooling is an inadequate representation of human capital investments.
More importantly, for present purposes, while the estimated returns to education continue to rise significantly with age, the estimated effect of family mean
log income relative to child’s age is statistically indistinguishable from zero, at
least on a 99 % confidence test, both for sons and daughters. In other words,
within this specification, after controlling for rising returns to schooling with
age, no significant rise remains in the intergenerational transmission parameter
as the child ages.
6 Within-generation credit constraints
The conventional focus on a budget constrained solely by permanent income
implicitly assumes an absence of any credit constraint within the parental
lifetime. For simplicity in reconsidering this presumption, distinguish two time
periods: during and after the parents invest in the child’s human capital. Let
the parents’ preferences be
U p = C1 C2 Ycα
(29)
where Ci = parents’ consumption in time period i. The budget constraints for
the two time periods are
C1 ≤ Y1 − Hc + L
C2 ≤ Y2 − L (1 + ι)
(30)
where Yi = parents’ income in time period i, L are loans taken out by
parents during period 1, and ι is the interest rate as before. If there is
no constraint
on L, then Hc depends
only upon permanent income of the
parents Yp = Y1 + Y2 (1 + ι)−1 , as in Eq. 6. At the opposite extreme, if L
is constrained to be zero, then Y1 replaces Yp in Eq. 6. With heterogeneous
households, the mean family may exhibit a mix of these two extremes.
To explore this possibility empirically, a measure of the mean log family
income during the interval when the child is age 0–17 is added to the generic
regression model. In the first two OLS estimates in Table 7, the intergenerational transmission elasticity from mean log family income is found to be
greater on family income received during this interval when the child is young,
26,501
0.623
(8.75)
−13.939
(6.55)
0.050
(4.83)
0.008
(1.70)
0.003
(0.70)
30,854
0.013
(2.05)
0.489
(7.84)
−10.238
(5.47)
0.050
(5.31)
Sons’ mean log earnings
OLS
0.016
(2.55)
0.602
(7.90)
−13.994
(5.97)
0.082
(4.10)
−0.057
(2.25)
30,854
30,854
944.5
−0.013
(1.21)
0.345
(4.43)
−9.049
(4.76)
0.176
(4.07)
IV
t statistics for a zero null hypothesis in parentheses. Standard errors from heteroskedastic-consistent matrix
Family mean log income when child 0–17/
no. of periods observed
No. of observations
First-stage partial F
Family mean log income when child 0–17
Family mean log income/child’s age
Family mean log income
Major earner’s mean log earnings
Mother’s mean log earnings
Father’s mean log earnings
Table 7 Family income during schooling years
25,420
0.428
(6.02)
−9.490
(4.61)
0.026
(2.35)
0.003
(0.68)
0.016
(3.43)
29,546
0.022
(3.24)
0.354
(5.61)
−7.512
(4.09)
0.020
(2.06)
0.022
(3.40)
0.480
(6.66)
−11.661
(5.40)
0.080
(4.25)
−0.092
(3.89)
29,546
Daughters’ mean log earnings
OLS
29,546
795.7
−0.011
(0.97)
0.157
(1.92)
−5.767
(3.06)
0.184
(3.91)
IV
Intergenerational income immobility in Finland
1087
1088
R.E.B. Lucas, S.P. Kerr
though these initial estimates are statistically weaker for daughters relative to
sons. Family income is observed from one to five times before the children
achieve age 18, depending upon the initial age of the child, with three or
less observations for about 87 % of the sons and daughters. In the appendix,
it is demonstrated that our prior estimates of intergenerational transmission
elasticities increase significantly and asymptotically with the number of time
observations over which family income is averaged, as suggested by Eq.
20. Accordingly, the third estimates in Table 7 append the time average of
log family income when the child is less than 18 relative to the number of
observations in forming this average (as well as a control for this number of
observations). In doing so, the asymptotic transmission from additional income
during schooling years of the child indeed proves larger for both genders.
Parents who are anxious to invest more in their child’s human capital
may attempt to load their income into the earlier part of the child’s life.
Accordingly, in the final results in Table 7, mean family income when the
child is age 0–17 is instrumented, adopting a variable based on the average
unemployment rates of fathers and mothers living in the same region as the
parents of the observed son or daughter. For whichever parent is the major
earner, the gender-specific, regional rate is averaged over periods in which
the particular parent contributes to family income. The instrument is then the
difference in this mean unemployment rate during the period when the child
is age 0–17 as opposed to the average over the entire period of observation
(Erola and Moisio 2002). As shown in Table 7, this instrument proves very
powerful in the first stage. To the extent that the instrumental variable is
valid, the results suggest that ordinary least squares underestimates the addition to intergenerational transmission from family income during schooling
years.
In a US context, Haider and Solon (2006) and Grawe (2006) note that the
attenuation inconsistency, from estimating intergenerational earnings transmission by ordinary least squares on fathers’ current earnings as a proxy for
lifetime earnings, is minimized if fathers are observed in the age range from
their early 30s to mid-40s. Böhlmark and Lindquist (2006) find comparable
results for Sweden (see also Baker and Solon 2003, on Canadian men). Since
the average age of our Finnish major earners is 42 when their children are
of school age, the larger transmission elasticity from family income during
these schooling years may simply reflect an effect of diminished attenuation
inconsistency during this period of observation, rather than intertemporal borrowing constraints. Towards exploring this distinction, the prior specifications
may therefore be extended to include an interaction between family mean log
income when the child is ages 0–17 and the square of the difference from 40
in the major earner’s mean age during these years (plus a control for this age
differential term itself). In both the ordinary least squares and instrumental
variables estimates, the estimated coefficients on the additional interaction
term are indistinguishable from zero, even at a 90 % confidence level. In
other words, this supports the notion that additional family income when both
Intergenerational income immobility in Finland
1089
sons and daughters are passing through school is indeed associated with higher
subsequent earnings for this next generation.
7 In perspective: a summing up
A nested model has been developed, incorporating an auto-regressive, intergenerational process in earnings that stems from heritability in earning
abilities, and a budget constraint on human capital investments in children
imposed by the permanent incomes of their parents. Any transmission from
parental earnings to those of their children is found to be very weak in the
rich Finnish data, while the transmission from family incomes to earnings of
both sons and daughters is significantly and substantially larger. Moreover,
the income available to parents during the phase when their children are
passing through school is found to have an additional effect on the future
earnings of both sons and daughters, even after instrumenting this additional
income. The influence of permanent income is consistent with constraints on
the capacity of parents to borrow against future incomes of their children; the
additional influence of family income during the child’s early life is consistent
with constraints on parents’ ability to raise credit against their own subsequent
incomes.
Exploration of a three-generation data set points to three things: a lack of
support for the Easterlin–Pollak–Wachter hypothesis of endogenous tastes,
which might have offered an alternative explanation for the transmission from
family income to children’s earnings; further support for a weak ability transmission parameter; and confidence that omission of grandparents’ incomes
does not substantially bias estimates of transmission from parents’ incomes.
The lack of any substantial, auto-regressive process from parent’s earnings
to those of their offspring in Finland proves insensitive to whether both father’s
and mother’s earnings are considered in two-earner families, or the earnings
of the higher earning parent are adopted across both single-parent and twoparent families, and insensitive to several other combinations of these parents’
earnings too. The significant intergenerational transmission from parents’
family income to earnings of their sons and daughters also proves robust to
a range of permutations considered: the estimated elasticity of transmission
increases significantly the more observations are available on family income,
though it proves even greater if permanent income is estimated as a fixed
effect instead; the elasticity is fairly insensitive to alternative definitions of
family income when parents separate and to exclusion of earnings-related
maternity and unemployment benefits and of retired persons; moreover, a less
parsimonious specification suggests that the elasticity does not vary greatly
across the income spectrum. In addition, the elasticity is estimated to increase
if zero-earning observations of sons and daughters are incorporated because,
although higher income families’ children are more likely not to work while
continuing their education, this effect is outweighed by the propensity of sons
1090
R.E.B. Lucas, S.P. Kerr
and daughters from lower income families to be unemployed or to drop out of
the labor force.
Sensitivity of intergenerational mobility estimates to age of the second generation has been noted in prior work on other countries. In the Finnish data,
at least through about age 40, a fairly monotonic rise in the intergenerational
transmission elasticity from family incomes is observed, the older the age at
which both sons’ and daughters’ earnings are observed. Moreover, both in
cross section of sons and daughters and by comparisons of siblings, the returns
to education are shown to rise with age in Finland. After decomposing the
intergenerational transmission elasticity to account for this rise in the rate of
return to years of schooling, as sons and daughters mature, any residual rise
in transmission elasticity with respect to age proves statistically insignificant.
Having wealthy parents increasingly pays off as one becomes older, and this is
substantially because the returns to parental investments in years of schooling
rise with age.
Because the income distribution is fairly even in Finland, there are relatively
few extremely wealthy families. This may diminish the likelihood of observing
families able to leave positive bequests, as required for the wealth model to
hold. Given the prevalence of collective bargaining in Finland, earning compression, particularly in the earlier years of our data, may render earnings less
likely to reflect idiosyncratic abilities (Eriksson and Jäntti 1997). Nonetheless,
the evidence on the importance of family incomes is compelling. In other
countries, unearned income might reflect abilities to invest, which could be
passed on as an ability to earn in the next generation. In Finland, however,
most of unearned income is derived from state transfers rather than investment
income. Yet, even in the presence of Finland’s comprehensive welfare state
system, families’ permanent incomes, as well as their incomes during children’s
schooling years, appear important in understanding the substantial intergenerational income intransigence.
Finally, as noted in the introduction, prior work on the Nordic countries
(including Finland) suggests low transmission rates from parent’s earnings to
those of their children. The results in this paper certainly concur with these
impressions; intergenerational auto-regression in earnings is low in Finland,
but transmission from family income to children’s earnings is not. International
comparisons are difficult. Nonetheless, our estimates of intergenerational
transmission in Finland are not obviously lower than those obtained outside
of the Nordic region. The privileges that come with higher incomes, not only
schooling but the many constituents of human capital, appear to be important
factors shaping the earnings of the next generation in Finland. In contrast, all
of our estimates point to small and fairly insignificant values for a parameter
of inherited earning ability. Apples indeed fall close to Finnish trees; having
either a silver or golden spoon handy is what breeds success.
Acknowledgements The authors are most grateful to the Academy of Finland for the partial
funding of this study under project number 52198. The paper has benefitted considerably from
Intergenerational income immobility in Finland
1091
the constructive comments and suggestions of an anonymous referee as well as those of Joshua
Angrist, David Autor, Markus Jäntti, Kevin Lang, and Dilip Mookherjee.
References
Abul Naga RH (2002) Estimating the intergenerational correlation of incomes: an errors-invariables framework. Economica 69(273):69–91
Altonji JG, Dunn TA (1991) Relationships among the family incomes and labor market outcomes
of relatives. In: Ehrenberg R (ed) Research in labor economics. JAI Press, Greenwich, pp
269–310
Baker M, Solon G (2003) Earnings dynamics and inequality among Canadian men, 1976–1992:
evidence from longitudinal income tax records. J Labor Econ 21(2):289–321
Becker GS (1972) Schooling and inequality from generation to generation: comment. J Polit Econ
80(3):S252–S255
Becker GS, Tomes N (1979) An equilibrium theory of the distribution of income and intergenerational mobility. J Polit Econ 87(6):1153–1189
Becker GS, Tomes N (1986) Human capital and the rise and fall of families. J Labor Econ 4(3):s1–
s39
Behrman JR (1997) Intrahousehold distribution and the family. In: Rosenzweig MR, Stark O (eds)
Handbook of population and family economics, vol 1A. Elsevier, Amsterdam, pp 125–187
Behrman JR, Kohler H-P, Jensen VM, Pedersen D, Petersen I, Bingley P, Christensen K (2011)
Does more schooling reduce hospitalization and delay mortality? New evidence based on
Danish twins. Demography 48(4):1347–1375
Behrman JR, Rosenzweig MR (2002) Does increasing women’s schooling raise the schooling of
the next generation? Am Econ Rev 92(1):323–334
Behrman JR, Taubman P (1985) Intergenerational earnings mobility in the United States: some estimates and a test of Becker’s intergenerational endowments model. Rev Econ Stat 67(1):144–
151
Behrman JR, Taubman P (1989) A test of the Easterlin fertility model using income for two
generations and a comparison with the Becker model. Demography 26(1):117–123
Behrman JR, Taubman P (1990) The intergenerational correlation between children’s adult earnings and their parents’ income: results from the Michigan panel survey of income dynamics.
Rev Income Wealth 36(2):115–127
Behrman JR, Pollak RA, Taubman P (1995) The wealth model: efficiency in education and
distribution in the family. In: Behrman JR, Pollak RA, Taubman P (eds) From parent to child:
intrahousehold allocations and intergenerational relations in the United States. University of
Chicago Press, Chicago, pp 113–138
Belley P, Lochner L (2007) The changing role of family income and ability in determining educational achievement. J Hum Cap 1(1):37–89
Björklund A, Jäntti M (1997) Intergenerational income mobility in Sweden compared to the
United States. Am Econ Rev 87(5):1009–1018
Björklund A, Jäntti M (2009) Intergenerational income mobility and the role of family background. In: Salverda W, Nolan B, Smeeding TM (eds) The Oxford handbook of economic
inequality. Oxford University Press, Oxford, pp 491–521
Björklund A, Eriksson T, Jäntti M, Raaum O, Österbacka E (2002) Brother correlations in
earnings in Denmark, Finland, Norway and Sweden compared to the United States. J Popul
Econ 15(4):757–772
Björklund A, Jäntti M, Solon G (2005) Influences of nature and nurture on earnings variation: a
report on a study of various sibling types in Sweden. In: Bowles S, Gintis H, Osborne Groves M
(eds) Unequal chances: family background and economic success. Princeton University Press,
Princeton, pp 145–164
Björklund A, Salvanes KG (2010) Education and family background: mechanisms and policies.
In: Hanushek E, Machin S, Woessmann L (eds) Handbook of economics of education, vol 3.
Elsevier, Amsterdam, pp 201–247
Black SE, Devereux PJ (2010) Recent developments in intergenerational mobility. In: Ashenfelter
O, Card D (eds) Handbook of labor economics, vol 4B. Elsevier, Amsterdam, pp 1487–1541
1092
R.E.B. Lucas, S.P. Kerr
Black SE, Devereux PJ, Salvanes KG (2005) Why the apple doesn’t fall far: understanding intergenerational transmission of human capital. Am Econ Rev 95(1):437–449
Blomster P (2000) Yliopisto-opiskelijoiden Toimeentulo ja Opintotuki 1900-luvun Suomessa (The
livelihood of university students and access to financial aid in 20th century Finland). Sosiaali-ja
Terveysturvan Tutkimuksia: 56. Kansaneläkelaitos, Helsinki
Böhlmark A, Lindquist MJ (2006) Life-cycle variations in the association between current and
lifetime income: replication and extension for Sweden. J Labor Econ 24(4):879–896
Bowles S (1972) Schooling and inequality from generation to generation. J Polit Econ 80(3):S219–
S251
Bratsberg B, Røed K, Raaum O, Naylor R, Jäntti M, Eriksson T, Österbacka E (2007) Nonlinearities in intergenerational earnings mobility: consequences for cross-country comparisons. Econ
J 117(519):C72–C92
Cameron SV, Heckman JJ (1998) Life cycle schooling and dynamic selection bias: models and
evidence for five cohorts of American males. J Polit Econ 106(2):262–333
Cameron SV, Heckman JJ (2001) The dynamics of educational attainment for Black, Hispanic and
White males. J Polit Econ 109(3):455–499
Card D (1999) The causal effect of education on earnings. In: Ashenfelter O, Card D (eds)
Handbook of labor economics, vol 3A. Elsevier, Amsterdam, pp 1277–2097
Carneiro P, Heckman JJ (2002) The evidence on credit constraints in post-secondary schooling.
Econ J 112(482):705–734
Chadwick L, Solon G (2002) Intergenerational income mobility among daughters. Am Econ Rev
92(1):335–344
Conneely K, Uusitalo R (1998) Estimating heterogeneous treatment effects in the Becker schooling model. Paper presented at the labor economics seminar series, industrial relations section.
Princeton University, Princeton
Corak M (2006) Do poor children become poor adults? Lessons from a cross country comparison
of generational earnings mobility. Res Econ Inequal 13(1):143–188
Corak M, Heisz A (1999) The intergenerational earnings and income mobility of Canadian men:
evidence from longitudinal income tax data. J Hum Resour 34(3):504–533
Couch KA, Lillard, DR (1998) Sample selection rules and the intergenerational correlation of
earnings. Labour Econ 5(3):313–329
Dearden L, Machin S, Reed H (1997) Intergenerational mobility in Britain. Econ J 107(440):47–66
Duncan G, Kalil A, Mayer SE, Tepper R, Payne MR (2005) The apple does not fall far from the
tree. In: Bowles S, Gintis H, Osborne Groves M (eds) Unequal chances: family background
and economic success. Princeton University Press, Princeton, pp 23–79
Easterlin RA, Pollak RA, Wachter ML (1980) Toward a more general economic model of fertility
determination: endogenous preferences and natural fertility. In: Easterlin RA (ed) Population
and economic change in developing countries. University of Chicago Press, Chicago, pp 81–
150
Eide ER, Showalter MH (2000) A note on the rate of intergenerational convergence of earnings.
J Popul Econ 13(1):159–162
Ekhaugen T (2009) Extracting the causal component from the intergenerational correlation in
unemployment. J Popul Econ 22(1):97–113
Eriksson T, Jäntti M (1997) The distribution of earnings in Finland: 1971–1990. Eur Econ Rev
41(9):1763–1779
Erola J, Moisio P (2002) Did Finland stagnate? Social mobility and long-term unemployment in
Finland 1970–1995. Sosiologia 39(3):185–199
Galton F (1869) Hereditary genius: an inquiry into its laws and consequences. Macmillan, London
Gaviria A (2002) Intergenerational mobility, sibling inequality and borrowing constraints. Econ
Educ Rev 21(4):331–340
Grawe ND (2004) Reconsidering the use of nonlinearities in intergenerational earnings mobility
as a test for credit constraints. J Hum Resour 39(3):813–827
Grawe ND (2006) Lifecycle bias in estimates of intergenerational earnings persistence. Labour
Econ 13(5):551–570
Gustaffson B (1994) The degree and pattern of income immobility in Sweden. Rev Income Wealth
40(1):67–86
Haider S, Solon G (2006) Life-cycle variation in the association between current and lifetime
earnings. Am Econ Rev 96(4):1308–1320
Intergenerational income immobility in Finland
1093
Hämäläinen K, Hämäläinen U (2005) A lost generation? State dependence in unemployment
among labour market entrants in different phases of business cycle. Paper presented at the
European Society for Population Economics Conference, Paris
Im KS, Pesaran MH, Shin Y (2003) Testing for unit roots in heterogeneous panels. J Econ
115(1):53–74
Jäntti M, Bratsberg B, Røed K, Raaum O, Naylor R, Österbacka E, Björklund A, Eriksson
T (2007) American exceptionalism in a new light: a comparison of intergenerational earnings mobility in the Nordic countries, the United Kingdom and the United States. The
Warwick Economics Research Paper Series, Department of Economics, University of
Warwick, Warwick
Jenkins S (1987) Snapshots versus movies: ‘lifecycle biases’ and the estimation of intergenerational
earnings inheritance. Eur Econ Rev 31(5):1149–1158
Jeon Y, Shields MP (2005) The Easterlin hypothesis in the recent experience of higher-income
OECD countries: a panel-data approach. J Popul Econ 18(1):1–13
Jürges H (2000) Of rotten kids and Rawlsian parents: the optimal timing of intergenerational
transfers. J Popul Econ 13(1):147–157
Lee C-I, Solon G (2009) Trends in intergenerational income mobility. Rev Econ Stat 91(4):766–
772
Levine DI, Mazumder B (2002) Choosing the right parents: changes in the intergenerational
transmission of inequality between 1980 and the early 1990s. Federal Reserve Bank of Chicago
Working Paper 2002-8, Chicago
Lillard LA, Kilburn RM (1997) Assortative mating and family links in permanent earnings. Rand
Corporation Report DRU/1578/NIA, Santa Monica
Liu H, Zeng J (2009) Genetic ability and intergenerational earnings mobility. J Popul Econ
22(1):75–95
Lochner LJ, Monge-Naranjo A (2011) The nature of credit constraints and human capital. Am
Econ Rev 101(6):2487–2529
Mazumder B (2005) Fortunate sons: new estimates of intergenerational mobility in the United
States using social security earnings data. Rev Econ Stat 87(2):235–255
Mulligan CB (1997) Parental priorities and economic inequality. University of Chicago Press,
Chicago
Mulligan CB (1999) Galton versus the human capital approach to inheritance. J Polit Econ
107(6):S184–S224
Oreopoulos P, Page M, Stevens AH (2008) The intergenerational effects of worker displacement.
J Labor Econ 26(3):455–83
Österbacka E (2001) Family background and economic status in Finland. Scand J Econ
103(3):467–484
Österberg T (2000) Intergenerational income mobility in Sweden: what do tax-data show? Rev
Income Wealth 46(4):421–436
Pekkala SA, Lucas REB (2007) Differences across cohorts in Finnish intergenerational income
mobility. Ind Relat Symp Trends Intergenerational Mobil 46(1):81–111
Pekkarinen T, Uusitalo R, Kerr SP (2009) School tracking and intergenerational income mobility:
evidence from the Finnish comprehensive school reform. J Public Econ 93(7–8):965–973
Plug E, Vijverberg W (2003) Schooling, family background, and adoption: is it nature or is it
nurture? J Polit Econ 111(3):611–641
Plug E, Vijverberg W (2005) Does family income matter for schooling outcomes? Using adoptees
as a natural experiment. Econ ...
Purchase answer to see full
attachment
Why Choose Us
- 100% non-plagiarized Papers
- 24/7 /365 Service Available
- Affordable Prices
- Any Paper, Urgency, and Subject
- Will complete your papers in 6 hours
- On-time Delivery
- Money-back and Privacy guarantees
- Unlimited Amendments upon request
- Satisfaction guarantee
How it Works
- Click on the “Place Order” tab at the top menu or “Order Now” icon at the bottom and a new page will appear with an order form to be filled.
- Fill in your paper’s requirements in the "PAPER DETAILS" section.
- Fill in your paper’s academic level, deadline, and the required number of pages from the drop-down menus.
- Click “CREATE ACCOUNT & SIGN IN” to enter your registration details and get an account with us for record-keeping and then, click on “PROCEED TO CHECKOUT” at the bottom of the page.
- From there, the payment sections will show, follow the guided payment process and your order will be available for our writing team to work on it.