ACCOUNTING FOR INTERGENERATIONAL INCOME
PERSISTENCE: NONCOGNITIVE SKILLS, ABILITY
AND EDUCATION*
Jo Blanden, Paul Gregg and Lindsey Macmillan
We analyse in detail the factors that lead to intergenerational persistence among sons, where this is
measured as the association between childhood family income and later adult earnings. We seek to
account for the level of income persistence in the 1970 BCS cohort and also to explore the decline in
mobility in the UK between the 1958 NCDS cohort and the 1970 cohort. The mediating factors
considered are cognitive skills, non-cognitive traits, educational attainment and labour market
attachment. Changes in the relationships between these variables, parental income and earnings are
able to explain over 80% of the rise in intergenerational persistence across the cohorts.
Intergenerational mobility is the degree of fluidity between the socio-economic status of
parents (usually measured by income or social class) and the socio-economic outcomes of
their children as adults. A strong association between incomes across generations indi-
cates weak intergenerational income mobility, and may mean that those born to poorer
parents have restricted life chances and do not achieve their economic potential.
Recent innovations in research on intergenerational mobility have been concentra-
ted on improving the measurement of the extent of intergenerational mobility, and on
making comparisons across time and between nations. The evidence suggests that the
level of mobility in the UK is low by international standards (Ja
¨
ntti et al., 2006; Corak,
2006; Solon, 2002). Comparing the 1958 and 1970 cohorts indicates that mobility has
declined in the UK (Blanden et al., 2004).
This article takes this research a stage further by focusing on transmission mecha-
nisms; those variables that are related to fa mily incomes and have a return in the labour
market. First we evaluate the relative importance of education, ability, noncognitive (or
ÔsoftÕ) skills and labour market experience in generating the extent of intergenerational
persistence in the UK among the 1970 cohort. In the second part of the article we seek
to appreciate how these factors have contributed to the observed decline in mobility in
the UK. We focus here on men for reasons of brevity.
Education is the most obvious of these transmission mechanisms. It is well established
that richer children obtain better educational outcomes, and that those with higher
educational levels earn more. Education is therefore a prime candidate to explain
mobility and changes in it. Indeed, Blanden et al. (2004) find that a strengthening rela-
tionship between family income and participation in post compulsory schooling across
cohorts can help to explain part of the fall in intergenerational mobility they observe.
Cognitive ability determines both educational attainment and later earnings, making
it another likely contributor to intergenerational persistence. We might expect a strong
link between parental income and measured abil ity, both because of biologically
inherited intelligence and due to the investments that better educated parents can
* This work was funded by the Department for Education and Skills through the Centre for the Economics
of Education. We are grateful for helpful comments from three referees.
The Economic Journal, 117 (March), C43–C60. Ó The Author(s). Journal compilation Ó Royal Economic Society 2007. Published by
Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.
[ C43 ]
make in their children. We seek to understand the extent to which differing achieve-
ments on childhood tests across income groups can exp lain differ ences in earnings,
both directly, and through their relationship with final educational attainment.
Galindo-Rueda and Vignoles (2005) demonstrate that the role of cognitive test scores
in determining educational attainment has declined between these two cohorts.
A growing literature highlights that noncognitive personality traits and personal
characteristics earn rewards in the labour market and influence educational attainment
and choices (Feinstein, 2000; Heckman et al., 2006; Bowles et al., 2001; Carneiro et al.,
2006). If these traits are related to family background then this provides yet anot her
mechanism driving intergener ational persistenc e. Osborne-–Groves (2005) considers
this possibility explicitly and finds that 11% of the father–son correlation in earnings
can be explained by the link between personalities alone; where personality is meas-
ured only by personal effi cacy.
Finally, labour market experience and employment interruptions have long been
found to influence earnings (Stevens, 1997). Gregg and Tominey (2005) highlight, in
particular, the negative impacts of spells of unemployment as young adults; we there-
fore analyse labour market attachment as another way in which family background
might influence earnings.
In the next Section we lay out our modelling approach in more detail. Section 2
discusses our data. Section 3 presents our results on accounting for the level of inter-
generational mobility while Section 4 describes our attempt to understand the change.
Section 5 offers conclusions.
1. Modelling Approach
In economics, the e mpirical work on intergenerational mobility is generally concerned
with the estimation of b in the following regression;
lnY
child
i
¼ a þ b lnY
parents
i
þ e
i
ð1Þ
where lnY
child
i
is the log of some measure of earnings or income for adult children, and
lnY
parents
i
is the log of income for parents, i identifies the family to which parents and
children belong and e
i
is an error term. b is therefore the elasticity of children’s income
with respect to their parentsÕ income and (1 b) can be thought of as measuring
intergenerational mobility.
Conceptually, we are interested in the link between the permanent incomes of
parents and children across generations. However, the measures of income available in
longitudinal datasets are likely to refer to current income in a period . In some datasets
multiple measures of current income can be averaged for parents and children, m oving
the measure somewhat closer to permanent income. Additionally it is usual to control
for the ages of both generations.
1
In the cohort datasets we use, substantial measure-
ment error is likely to remain, meaning that our estimates will be biased downwards as
measures of intergenerational per sistence. The issue of measurement error becomes
particularly important when considering the changes in mobility across cohorts and
this will be returned to when discussing our findings.
1
Solon (1999) provides a review of the evolution of the intergenerational mobility literature.
C44 [MARCHTHE ECONOMIC JOURNAL
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
We report the intergenerational partial correlation r, alongside b because differences
in the variance of lnY between generations will distort the b coefficient. This is ob-
tained simply by scaling b by the ratio of the standard deviation of parentsÕ income to
the standard deviation of children’s income, as shown below.
r ¼ Corr
ln Y
parents
;ln Y
child
¼ b
SD
lnY
parents
SD
lnY
child
: ð2Þ
The main objective in this article is to move beyond the measurement of b and r, and to
understand the pathways through which parental income affects children’s earnings.
The role of non-cognitive skills can be used as an example, assuming for the moment
that these are measured as a single index. We can measure the extent to which these
skills are related to parental income Noncog
i
¼ a
1
þ k lnY
parents
i
þ e
1i
, and estimate
their pay-offs in the labour market lnY
child
i
¼ -
1
þ qNoncog
i
þ u
1i
This means that the overall intergenerational elasticity can be decomposed into the
return to non-cognitive skills multiplied by the relationship between parental income
and these skills, plus the unexplained persistence in income that is not transmitted
through non-cognitive traits.
b ¼ qk þ
Covð u
1i
; lnY
parents
i
Þ
VarðlnY
parents
i
Þ
: ð3Þ
In our analysis we consider non-cognitive skills among several other mediating
factors: cognitive test scores, educational performance and early labour market
attachment.
Our decomposition approach requires the estimation of the univariate relationships
between the transmission variables and parental income. These are then combined
with the returns found for those variables in an earnings equation. We build up the
specifications of our earnings equations gradually, as we believe that many of the
associations operate in a sequential way. For example, Heckman et al. (2006) show that
part of the advantage of higher non-cognitive skill s works through enabling children to
reach a higher education level. In the previous example we have shown the uncondi-
tional influence of non-cognitive skills on intergenerational persistence. To see
how non-cognitive skill works through education levels, we can add education to the
earnings equation.
lnY
child
i
¼ -
2
þ dNoncog
i
þ pEd
i
þ u
2i
: ð4Þ
Then estimate the relationship between educational attainment and parental income.
Ed
i
¼ a
2
þ c lnY
parents
i
þ e
2i
: ð5Þ
The conditional decomposition is then:
b ¼ dk þ pc þ
Covðu
2i
; lnY
parents
i
Þ
VarðlnY
parents
i
Þ
: ð6Þ
Where dk is the conditional contribution of non-cognitive skill and pc is the contri-
bution of age 16 examination results. Therefore the difference between qk and dk
2007] C45
ACCOUNTING FOR INTERGENERATIONAL INCOME
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
shows the extent to which the non-cognitive skills contribute to intergenerational
persistence by enabling more affluent children to achieve better qualifications at 16.
In the second part of this study we use the same approach to account for the change
in intergenerational persistence. If we continue with the simple exa mple shown above,
we can write
b
70
b
58
¼ d
70
k
70
d
58
k
58
þ p
70
c
70
p
58
c
58
þ
Covð u
2i70
;lnY
parents
i70
Þ
VarðlnY
parents
i70
Þ
Covð u
2i58
;lnY
parents
i58
Þ
VarðlnY
parents
i58
Þ
:
ð7Þ
Or in words, the difference in persistence is formed of two parts; the difference be-
tween the explained persistence across the cohorts plus the difference between the
unexplained persistence. If th e explained part of b is larger in the second cohort than
in the first then this indicates that the factors we explore are responsible for part of the
increase in intergener ational persistence.
2. Data
We use information from the two mature publicly accessible British cohort studies, the
British Cohort Study (BCS) of those born in 1970 and the National Child Development
Study (NCDS) of those born in 1958. Both cohorts began with around 9,000 baby boys,
although as we shall see our final samples are considerably smaller than this. We shall
first provide a discussion of how we use the 1970 cohort, before considering how the
data are used in the comparative Section of the article.
2.1. British Cohort Study
The BCS originally included all those born in Great Britain between 4
th
and 11
th
April
1970. Information was obtained about the sample members and their families at birth
and at ages 5, 10, 16 and 30. We use the earnings informa tion obtained at age 30 as the
dependent variable in our intergenerational models. Employees are asked to provide
information on their usual pay and pay period. Data quality issues mean we must drop
the self-employed. Parental income is derived from information obtained at age 10 and
16; where parents are asked to place their usual total income into the appropriate band
(there were seven options at age 10 and eleven at age 16). We generate continuous
income variables at each age by fitting a Singh-Maddala distribution to the data using
maximum likelihood estimation. This is particularly helpful in allocating an expected
value for those in the open top category.
2
We adjust the variables to net measures and
impute child benefit for all families.
3
The explanatory variable used in the first part of
the article is the average of income over ages 10 and 16.
2
Singh and Maddala (1976). Many thanks to Christopher Crowe for providing his stata program smint.ado
which fits Singh-Maddala distributions to interval data.
3
The distribution of the income variables obtained compares reassuringly with incomes for similarly
defined families in the same years of the Family Expenditure Surveys, figures showing this are available from
the authors on request.
C46 [MARCHTHE ECONOMIC JOURNAL
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
In the childhood surveys parents, teachers and the children themselves are asked to
report on the child’s behaviour and attitudes. These responses are combined to form
the noncognitive m easures as described in Box 1. Information on cognitive skills is
obtained at age 5 from the English Picture Vocabulary test (EPVT) and a copying test.
At age 10 the child took part in a reading test, maths test and British Ability Scale test
(close to an IQ test). Exam results at age 16 were obtained from information given in
the age 26 sub-sample and the age 30 sample. This includes detailed information on
the number of exams passed (both GCE O-level and CSE). Information on educational
achievements beyond age 16 is also available from the age 30 sample, as is information
on all periods of labour market and educati onal activity from age 16 to 30. This
information is used to generate the measures of labour market attachment which are
the proportion of months from age 16 to 30 when the individual is out of education,
out of the labour force and unemployed.
2.2. Comparative Data on the Two Cohorts
Some modification s must be made to the variables used when comparing the BCS with
the earlier National Child Development Study (NCDS). The NCDS obtains data at birth
and ages 7, 11, 16, 23, 33 and 42 for children born in a week in March 1958. Parental
income data is available only at age 16, meaning that the comparative analysis of this
data is based only on income at this age. The questions that ask about parental income
in the two cohorts are not identical and adjustments must be made to account for
differences in the way income is measured; see Blanden (2005, Chapter 4) for full
details. Intergenerational parameters for the NCDS are obtained by regressing earnings
Box 1
Non-cognitive Variables in BCS
Mother and teacher-reported scales are formed from principal components analyses of the
following behavioural ratings. The respondent grades the incidence of the behaviour in the child
along a 1–100 scale, where the definitions of 1 and 100 vary according to the behaviour being described.
Mother reported at age 5:
Anti-social: disobedient, destructive, aggressive, irritable, restless and tantrum
Neurotic: miserable, worried, fearful, fussy and complains of aches and pains
Teacher reported variables from age 10: scales are formed according to the suggestions made in
Osborn and Milbank (1987).
Application: 15 items, including the child’s concentration and perseverance and his/her ability to
understand and complete complex tasks.
Clumsiness: 12 items, includes items on bumping into things, and the use of small objects such as scissors.
Extroversion: 6 items concerning talkativeness and an explicit question about extroversion.
Hyperactivity: 6 items, includes the items squirmy, excitable, twitches, hums and taps.
Anxious: 9 items, includes items very similar to those which generate the mother reported anxiety scale.*
Child reported variables at age 10:
Locus of control: CAROLOC score for locus of control (Gammage, 1975).
Self-esteem: LAWSEQ score for self-confidence (Lawrence, 1973, 1978).
Mother-reported variable at age 16:
Anxiety: Derived from a principal components analysis of the mother’s reports of the applicability to the
child of the following descriptions: worried; solitary; miserable; fears new; fussy; obsessed with trivia;
sullen; and cries for little cause.
*Osborn and Milbank (1987) include two further scales; peer relations and conduct disorder, but we do not
include these in our analysis as we find they have no relationship with earnings.
2007] C47ACCOUNTING FOR INTERGENERATIONAL INCOME
Ó The Author(s). Journal compilation Ó Royal Economic Society 2007
