The Impact of Family Income on Child Achievement: Evidence from the Earned Income Tax Credit

Chevalier et al. IZA Journal of Labor Economics 2013, 2:8 http://www.izajole.com/content/2/1/8

ORIGINAL ARTICLE Open Access

The impact of parental income and education on the schooling of their children Arnaud Chevalier1, Colm Harmon2, Vincent O’ Sullivan3* and Ian Walker4

* Correspondence: [email protected] 3TILDA, Trinity College Dublin and ESRI, Dublin Full list of author information is available at the end of the article

© A m

Abstract

We investigate the relationship between early school-leaving and parental education and paternal income using UK Labour Force Survey data. OLS estimation reveals modest effects of income, stronger effects of maternal education relative to paternal, and stronger effects on sons than daughters. Using IV to simultaneously model the endogeneity of parental education and income, the maternal education effect disappears, while paternal education remains significant but only for daughters. In our favourite specification, which proxy for permanent income, paternal income becomes insignificant. Thus policies alleviating income constraints to alter schooling decisions may not be as effective as policies which increase permanent income. JEL codes: I20; J62

Keywords: Early school leaving; Intergenerational transmission

1. Introduction A considerable literature has focused on the effects of parental background on out-

comes for their children such as cognitive skills, education, health and subsequent in-

come (for a review, see Black and Devereux (2011)). Parents may affect the behavior

and decisions taken by their children through genetic transmission, preferences, or/

and environment – put simply, more educated and richer parents can provide a “better”

environment for their children, which creates an inequity which is the focus of sizable

policy attention (see for example McLachlan et al. (2013)).

The nature of the policy interventions to address this inequity depends critically on

the nature of the intergenerational transmission mechanism, and the extent to which

the relationship is causal. In particular, it has proven difficult to determine whether the

transmission mechanism works through inherited genetic factors or environmental fac-

tors and, to the extent that it is the latter, what is the relative importance of parental

education and income? Moreover, the link between the schooling of parents and their

children could be due to unobserved inherited characteristics rather than a causal ef-

fect of parental education or income per se in household production. This issue is ex-

plored in detail in the review by Björklund and Salvanes (2011).

An example of the importance of maternal education on child’s outcome is provided

in Currie and Moretti (2003), who found a positive relationship between maternal edu-

cation and their child’s birth weight, which is a strong predictor of child health. The

existence of such effects provides an important argument for subsidizing education, es-

pecially in households with low income and/or low educated parents. Indeed there

Chevalier et al.; licensee Springer. This is an Open Access article distributed under the terms of the Creative Commons ttribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any edium, provided the original work is properly cited.

2013

Chevalier et al. IZA Journal of Labor Economics Page 2 of 222013, 2:8 http://www.izajole.com/content/2/1/8

may be multiplier effects since policy interventions that increase educational attainment

for one generation may create spillovers to subsequent generations. The literature is not

entirely consistent but it is generally believed that, while raising the education of mothers

and fathers has broadly similar effects on household income, the external effects on chil-

dren associated with parental education are larger for maternal education than for pater-

nal, because mothers tend to be the main provider of care within the household.

Understanding the mechanisms by which parental education may affect children’s

outcomes is more difficult than establishing that there is an effect. Firstly, parental edu-

cation may be a direct input into the production function that generates the quality of

the endowments that children have in various domains (health, ability etc.). Secondly, it

may also affect the choice of other inputs. Thirdly, it may indirectly facilitate a higher

quantity and/or quality of other inputs through its effect on household income. The

use of policy instruments such as income transfers to attempt to break the cycle of dis-

advantage presumes this latter route is important.

This paper addresses an important issue in the existing literature: the causal effect of

parental education on children, allowing for the separate effects of maternal and pater-

nal education; and for the causal effect of household income, controlling for education.

To date, no study has simultaneously tried to account for the endogeneity of both par-

ental education and parental income. The distinction between education and income is

important since differences in policy approaches hang on their relative effects. Using a

series of British cross-sectional datasets, we begin by confirming the usual finding,

using least squares, that parental education levels and paternal income are positively as-

sociated with good child outcomes: in particular, with later school leaving1. This out-

come measure is important because recent UK governments have targeted a reduction

in the proportion of pupils leaving at 16, and committed to a phased increase in the

minimum age at which youths can leave education and training. We go on to use in-

strumental variable methods to take account of the endogeneity of both parental in-

come and education. We exploit a variety of ideas for identification that have been

used in other research, including changes in the minimum school leaving age for the

parents; month of birth of the parents which captures early school tracking that af-

fected the parental cohorts; and parental union status and its interactions with occupa-

tion that affect paternal income independently from education.

2. Previous literature It is widely shown that children brought up in less favorable conditions obtain less edu-

cation, despite the large financial returns to schooling (Heckman and Masterov (2005).

Indeed there is a large correlation between the education level of parents and their chil-

dren (Björklund and Salvanes (2011). However the transmission mechanism behind

such intergenerational correlations has never been made clear. Krueger (2004) reviewed

various contributions supporting the view that financial constraints significantly impact

on educational attainment. However, Carneiro and Heckman (2004) suggest that

current parental income does not explain child educational choices, but that family

fixed effects that contribute to permanent income, such as parental education levels,

have a much more positive role. This is the central conclusion of Cameron and Heckman

(1998), using US data, and Chevalier and Lanot (2002), using the UK National Child

Development Study data.

Chevalier et al. IZA Journal of Labor Economics Page 3 of 222013, 2:8 http://www.izajole.com/content/2/1/8

To date, researchers have attempted to identify the exogenous effect of either paren-

tal education or of parental income, but not both effects simultaneously. The literature

on estimating the causal effect of parental education on the child’s educational attain-

ment has relied on three identification strategies: instrumental variables, adopted chil-

dren, and twins.

The first identification strategy, and the one most relevant to this paper, uses instrumen-

tal variables methods based on ‘natural’ experiments or policy reforms that change the

educational distribution of the parents without directly affecting children. Black et al.

(2005) exploit Norwegian educational reforms which raised the minimum number of

years of compulsory schooling over a period of time, and at differential rates between re-

gions of the country. Their IV estimates show little evidence of causal effects except for

evidence of maternal education on son’s education. However, Oreopoulos et al. (2006)

using the same approach, but using US Census data from 1960, 1970 and 1980, report

that an increase in parental education by one year decreases the probability of a child re-

peating a schooling year (or grade) by between two and seven percentage points. Investi-

gating some possible mechanisms for this education transmission, Dixon et al. (2013)

confirm the causal effect of paternal education on literacy and numeracy skills from as

early as age four, and estimate that by age sixteen, the children of parents affected by a

British school leaving age reform, gained an additional 0.1 of a standard deviation in test

scores over other children. Lundborg et al. (2011) report a positive effect on cognitive,

non-cognitive skills and health for Swedish children whose parents were affected by a

similar reform. Of course, the minimum school leaving age is likely to affect the bottom of

the schooling distribution more than the top so there is a clear case for thinking, in a het-

erogeneous effects model, that such estimates provide Local Average Treatment Effect es-

timates that are not strictly comparable to OLS. However, to the extent that policymakers

are particularly concerned about early school leavers, such estimates are still of interest.

Alternative strategies to account for the endogeneity of parental education is to

account for genetic effects by comparing adopted and natural children (Sacerdote (2004)

or Plug (2004)), or the children of twins (Behrman and Rosenzweig (2002). The conclu-

sions from these identification strategies tend to support positive effects of paternal edu-

cation and no effect of maternal education in twin studies, and positive effects of both

parents in adoption studies. For example, Bjorklund et al. (2006) uses a register of Swedish

adoptees, which allows controls for both natural and adoptive parents’ education. After

correcting for the potential bias caused by non-randomness in this population, they find

that genetics account for about 50% of the correlation in education between generations

but also that the causal effect of adoptive parents’ education remains highly significant; i.e.

nurture is an important factor.

Holmlund et al. (2011) investigate whether the disparities in results in the literature

are due to differences in the sample used or in the identification strategies. Using Swedish

Population Register data, they implement all three methods, i.e. twins, adoptees and IV.

Their results are consistent with the weight of the existing literature. In twin studies, the

maternal effect is small and about half of the paternal education effect. This conclusion is

reversed when using adoptee samples. When relying on IV to estimate the causal effect of

parental education, the paternal effect is never significant but the maternal effect is quite

large. They also find that there are non-linearities in the effect of education with the effect

of parental education being larger at higher levels of education. Pronzato (2012) notes that

Chevalier et al. IZA Journal of Labor Economics Page 4 of 222013, 2:8 http://www.izajole.com/content/2/1/8

the estimates for maternal education are sensitive to size and selection, in a way that can

reconcile the results from different studies.

The literature on the causal effects of parental earnings or incomes on educational

outcomes is not as extensive as the literature on parental education. Blanden and Gregg

(2004) review US and UK evidence on the effects of policy changes which largely focus

on improving short-term family finances (see also Almond and Currie (2010). These in-

clude initiatives such as variation in welfare-to-work reforms which change the extent

of financial support available, or the Moving to Opportunity (MTO) experiments in the

US, which provided financial support for higher housing costs associated with moving

to more affluent areas2. Confounding factors create sizable challenges in assessing the

impact of income on child educational attainment. For example welfare reforms may

create exogenous income ‘shocks’ but they also are aimed at increasing parental labor

supply, which may also affect child educational attainment.

In the UK, the Educational Maintenance Allowances (EMA’s) provided a sizeable

means tested cash benefit conditional on participation in education (Dearden et al.

2009). Evaluations of the EMA policy trials showed enrollments increased by up to 6%

in families eligible for full subsidies. However, this transfer was conditional on staying

in school and so this reform is not directly informative about the effects of uncondi-

tional variations in income.

Other studies have relied on intertemporal variation in incomes between siblings, but

it is unclear that parents do not take compensatory actions in the face of differential fi-

nancial resources associated with each sibling. If they do, then sibling studies estimate

the effects net of those actions. Other studies look at value added in the form of

changes in outcomes associated with changes in income over time to difference out un-

observed heterogeneity. Similar studies use early measures of outcomes as controls for

unobserved heterogeneity. However, estimation of such lagged dependent variable

models are, in general, inconsistent in the presence of fixed child or family effects. Nor

are they very satisfactory ways of dealing with endogeneity because income may, itself,

respond to lagged outcomes – for example, a failing child may stimulate a parent to

work harder, to provide more financial resources to allow the child to improve.

In the absence of convincing experimental evidence and because of doubts over the

validity of sibling-based studies, instrumental variables have been used to identify the

effect of parental income effects on child outcomes. Shea (2000) uses union status (and

occupation) as an instrument for parental income. The identifying assumption is that

unionized fathers are not more ‘able’ parents than nonunionized fathers with similar

observable skills. Mayer (1997) uses variation in family income caused by state welfare

rules, income sources and income before and after the education period of the child, as

well as changes in income inequality. While strong identification assumptions are used

in both these studies, they both find that changes in parental long-run income have

only modest and sometimes negligible effects on the human capital of the children3.

Finally, Jenkins and Schluter (2002) is notable for being one of the few studies to con-

trol for both income, at various ages, and education. They study the type of school

attended (vocational or academic), using a small German dataset, and find income late

in the educational period is more important than early stage income. However, income

effects are small relative to education effects. However, the analysis assumes exogenous

income and parental education.

Chevalier et al. IZA Journal of Labor Economics Page 5 of 222013, 2:8 http://www.izajole.com/content/2/1/8

3. Data, sample selection and sources of exogenous variation Our analysis is based on the Labour Force Survey (LFS) - a quarterly survey of house-

holds in the U.K. In each quarter there are roughly 120,000-160,000 respondents (more

in earlier surveys) from the approximately 50,000-65,000 households surveyed. House-

holds are surveyed for five consecutive quarters. We pool the data from households in

the fifth quarter over the period 1993-20124. Children aged 16 to 18 living at home are

interviewed in the LFS, so parental information can be matched to the child’s record

only when children are living with their parents. Our sub-sample consists of those chil-

dren observed in LFS at ages 16 to 18 inclusive (and therefore have made their decision

with respect to post compulsory education participation) which is approximately

56,000 observations, or 4% of all LFS respondents. This corresponds closely to the

population of 16 to 18 year olds in census data.

The age range we consider is limited because we need to observe respondents while

they are still living at home in order to observe parental background (respondents are

not asked directly about their parents). An examination of British Household Panel

Survey data suggests that only 6% of children aged 16–18 have already left home. How-

ever, this censoring in the LFS data becomes more severe with older teenagers - whilst

98% of 16-year-old children are observed living with both parents, this proportion falls

to 88% for those aged 18 years old5. We therefore drop 2,581 teenagers who do not live

with their parents and a further 12,995 who live with only one of their parents, since

we need to observe both paternal and maternal education.

Due to these age restrictions, the key outcome of interest in this paper is the decision

to participate in post-compulsory schooling, defined as a dummy equal to one if the 16

to 18 year old child is either in post compulsory education at present or was in educa-

tion between 16 and 18 but had left school by the time of interview6. Only 16 year olds

who are surveyed between September and December are included to ensure informa-

tion on their post- compulsory education decision is available. To arrive at our final

sample we also drop observations from Scotland and Northern Ireland. Although these

regions changed their minimum school leaving ages at different times than England

and Wales, they also have quite distinct education systems. This drops another 5,342

observations.

Moreover, for our instruments to be observable in the data we impose the following

additional restrictions: we select teenagers whose father is working and reporting his in-

come; both parents were born after 1933 (and so were not affected by the earlier raising

of the school leaving age from 14 to 15, and whose school leaving is unlikely to have been

directly affected by World War II); both parents were born in the United Kingdom and

are currently resident in England or Wales7. We also drop observations with missing data

on the variables of interest. The details of the original LFS data and the impact of the

selection criteria on the variables of interest can be seen in Table 1. Focusing on children

living with both parents means that our sample is disproportionally composed of children

staying in. While the final set of restrictions severely cut the sample size (16,798

observations left), this does not alter the mean age and educational attainment of those

16–18 year-olds living with both parents. As such, our final sample appears representative

of the population of 16–18 year old living with both parents.

Figures 1 and 2 show the participation rate in post-compulsory schooling in our final

sample broken down by paternal and maternal education. The education of the

Table 1 Sample selection from pooled LFS 1992-2012

All those aged 16, 17 or 18

Living away from parents

Living with one parent

Living with both parents

Final sample

Age distribution:

% aged 16 23.08 10.34 24.26 23.5 24.01

% aged 17 39.84 29.87 41.12 40.06 39.92

% aged 18 37.09 59.78 34.62 36.44 36.07

% staying on at 16 75.08 50.02 72.37 77.46 78.12

Observations: 56,650 2,581 12,995 41,074 16,798

Note: The following are dropped from the penultimate column to form the final sample in the following sequence: families residing in heavily oversampled Northern Ireland (1,893 observations), families residing in Scotland (3,449), families where the parents were not born in England and Wales but who arrived after finishing their schooling (6,372), very old or very young parents (317), those missing basic demographic information such as date of birth (280), where father is not working (4,004) or self employed (4,353) or unknown (328), union status of father unknown (1,316) or has no or missing reported earnings (1,964).

Chevalier et al. IZA Journal of Labor Economics Page 6 of 222013, 2:8 http://www.izajole.com/content/2/1/8

children appears closely correlated with the education of their parents, particularly up

to a school leaving age of 18. Having parents with more education than this level does

not substantially affect the staying-on probability of children, which is then almost

100%8. The fraction of sons with post compulsory schooling is 30 points higher for

father with a degree (leaving age of 21) compared to those who left school at 15. There

are some sizable gaps between the participation of girls and boys from lower educated

parents. Among this group sons are more than 10 percentage points less likely to par-

ticipate than daughters However, these gender gaps in educational attainment narrow

with parental education. Despite the restriction to construct it, the sample displays the

expected relationship of interest.

Table 2 shows some selected statistics for the sample used in our analysis. The post-

compulsory schooling participation rate is 74% for boys and 83% for girls9. There are

large, statistically significant, differences in the parental education and household in-

come levels between those that remain in school compared to those that leave: almost

one year extra of parental education on average and more than 20% higher paternal

earnings. Additionally, early leavers typically have parents who are two years younger.

Parental income is potentially endogenous either because it is correlated with unob-

servable characteristics which are correlated with the child’s educational attainment, or

because the parental education effect is transmitted through income. Following Shea

(2000), we assume that union membership status creates an exogenous change in in-

come, which is independent of parenting ability and the child’s educational choice.

0.5

0.6

0.7

0.8

0.9

1

15 16 17 18 19 20 21 22 23 24 25

P ro

po rt

io n

S ta

yi ng

O n

Age father left education

Sons Daughters

Figure 1 Post compulsory participation by paternal education.

0.5

0.6

0.7

0.8

0.9

1

15 16 17 18 19 20 21 22 23 24 25 P

ro po

rt io

n S

ta yi

ng O

n

Age mother left education

Sons

Daughters

Figure 2 Post compulsory participation by maternal education.

Chevalier et al. IZA Journal of Labor Economics Page 7 of 222013, 2:8 http://www.izajole.com/content/2/1/8

Indeed the raw data, presented in Table 2, shows that children who stayed on were just

as likely to have unionized fathers as children who do not stay on in education.

Lewis (1986) and much subsequent work, demonstrates that wages vary substan-

tially with union status, even after controlling for observable skills. This is also the

case in our data. Figure 3 shows the kernel densities of the earnings of union

member fathers and non-union fathers. The raw union/non-union earnings gap for

fathers in our selected sample is 8%. If union wage premia reflect rents rather than

unobserved ability differences it seems plausible to make the (stronger) identifying

assumption, used in this paper; that union status, controlling for occupation, is un-

correlated directly with the parental influence on educational outcomes of the chil-

dren10. In any event, we are assuming, as in Shea (2000), that unionized fathers

(and their spouses) are not more ‘productive’ as parents than non-union fathers

with similar observable skills. This is supported by evidence from the British Co-

hort Study (BCS) where we find little to suggest that parenting behavior differs

across the union status of fathers11.

Parental education is likely to be endogenous. Here we rely on two sources of

exogenous variation. First, we identify the effect of parental education on children’s

education using the exogenous variation in schooling caused by the raising of the

minimum school leaving age (abbreviated as RoSLA: Raising of the School Leaving Age).

Specifically, individuals born before September 1957 could leave school at 15, while those

born after this date had to stay for an extra year of schooling. This policy change creates a

discontinuity in the years of education attained by the parents. Figures 4 and 5 illustrate

this by showing mean years of schooling by birth cohort (in 4 month intervals) around

the reform date. That is, we take a narrow window of birth cohorts around the reform

(+/− four years) to minimize the influence of any long-term trends across birth cohorts. There is a marked jump for parents born after September 1957 that coincides with

the introduction of the new higher school leaving age. Note that, pre-reform be-

tween 30% and 40% of the parents used in this paper left school at the minimum

age. So the reform did bite, and changed the behavior of a substantial fraction of

individuals in the affected cohorts. Individuals affected by the new school leaving

age had, on average, completed half a year more schooling than those born just be-

fore the reform. Chevalier et al. (2004) showed that the effect of this reform was

almost entirely confined to the probability of leaving at 15 relative to 16 – there

was little effect on educational attainment higher up the years of education

Table 2 Descriptive statistics LFS 1992–2012 estimation sample

Paternal log earnings

Paternal school

leaving age

Maternal school

leaving age

Paternal age

Maternal age

Father affected by RoSLA

Mother affected by RoSLA

Father born Sept – Dec.

Mother born Sept – Dec.

Paternal union membership

Age of respondent

Sons: N = 8661

Did not stay in FT education (26%) 5.91 15.83 15.86 45.39 43.20 0.43 0.32 0.30 0.34 0.42 17.13

(0.49) 1.09) (0.98) (5.80) (4.92) (0.49) (0.47) (0.46) (0.47) (0.49) (0.76)

Did stay in FT education (74%) 6.18 16.72 16.79 47.21 45.09 0.41 0.31 0.31 0.32 0.41 17.12

(0.61) (1.49) (1.42) (5.35) (4.72) (0.49) (0.46) (0.46) (0.47) (0.49) (0.76)

*** *** *** *** *** *** ***

Daughters: N = 8137

Did not stay in FT education (17%) 5.91 15.85 15.94 45.27 43.12 0.46 0.33 0.31 0.33 0.41 17.14

(0.53) (1.06) (1.01) (5.48) (4.70) (0.50) (0.47) (0.46) (0.47) (0.49) (0.75)

Did stay in FT education (83%) 6.20 16.85 16.87 47.18 45.11 0.40 0.29 0.31 0.32 0.42 17.11

(0.61) (1.54) (1.43) (5.28) (4.67) (0.49) (0.45) (0.46) (0.47) (0.49) (0.77)

*** *** *** *** *** *** *** * Note: Selected means, standard deviation in brackets. Significance level of differences between those who stay and those who do not stay. ***at 1% level, **at 5% level, *10% level.

C h evalier

et a l. IZ A Jo u rn a l o f La b o r Eco

n o m ics

P ag

e 8 o f 2 2

2 0 1 3 , 2

:8 h ttp

://w w

w .iza

jo le

.co m

/co n te

n t/2

/1 /8

Paternal Weekly Earnings

union non-union

0 500 1000 1500 2000

0

.001

.002

Figure 3 Distribution of Father’s weekly earnings (in GBP 2005) by union status.

Chevalier et al. IZA Journal of Labor Economics Page 9 of 222013, 2:8 http://www.izajole.com/content/2/1/8

distribution. Hence, this reform only identifies a LATE for individuals with low

levels of education. Table 2 shows that the proportion of fathers who were born

before the RoSLA reform is higher than for mothers, reflecting their slightly

greater age. Note also that early school leavers have younger parents who are thus

more likely to have been affected by the minimum school leaving age reform.

A second source of variation in parental schooling, that we exploit, derives from

parental month of birth, as in Crawford et al. (2007). There are several ways in

which month of birth can affect the parents’ education levels: through entry tim-

ing, leaving timing, whole group teaching, developmental differences and through

peer effects. The academic year starts in September but the traditional admissions

policy that reigned in the 1950’s and 1960’s, when most of the parents in our data

were young, allowed entry at the start of the term that the child turned 5 so that

there were three points of entry each year: September, January and April/May.

Thus the August born would start in April/May and have two fewer terms in

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

1 9

5 3 1

1 9

5 4 1

1 9

5 5 1

1 9

5 6 1

1 9

5 7 1

1 9

5 8 1

1 9

5 9 1

1 9

6 0 1

1 9

6 1 1

Age 19+ Age 18 Age 17 Age 16 Age 15

Birth Cohort in Four Month Intervals

Figure 4 Distribution of paternal school leaving age by third of year of birth.

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

1 9 5 3 1

1 9 5 4 1

1 9 5 5 1

1 9 5 6 1

1 9 5 7 1

1 9 5 8 1

1 9 5 9 1

1 9 6 0 1

1 9 6 1 1

Age 19+

Age 18

Age 17

Age 16

Age 15

Birth Cohort in Four Month Intervals

Figure 5 Distribution of maternal school leaving age by third of year of birth.

Chevalier et al. IZA Journal of Labor Economics Page 10 of 222013, 2:8 http://www.izajole.com/content/2/1/8

primary school than their classmates. These cohorts had two possible exit dates:

children born between September and January, inclusive, could leave school at

Easter, and thus not sit exams (Easter leaving rule) whether others had to stay

until the end of June. In the 1950’s and 60’s whole class teaching was the domin-

ant teaching method and developmental differences associated with month of birth

might imply that the youngest and the oldest might fare worse than the average.

Since students sit examinations at same time, they would be at different develop-

ment ages when facing the same examination, penalizing younger students. Finally,

peer effects might also arise because the youngest might be dominated or intimi-

dated by the oldest.

This developmental effect is particularly important since most of the parents in

these cohorts would have faced a selective schooling system where children were

segregated into academic or vocational schools at the age of 11 based on a single

test conducted on the same day across the whole country - known as the 11+

exam. Based on the results of this test, children were educated either in vocational

or academic tracks. Children in the vocational track were much more likely to

leave school at the minimum compulsory age, while those in the academic track

could go on to higher secondary school and university, see Harmon and Walker

(1995). These two different types of schools placed quite different expectations on

the children and there was very little movement between school types after the age

of 11. Figure 6 shows, by year of birth, the average age at which the parents in

our data left full-time education for those who were September born, the eldest in

their class cohort, compared to those who were July born, the youngest12. The

youngest individuals typically obtained around ¼ of a year less education than their

oldest classmates. Note that the gap closed completely for cohorts born in the

early 1960’s when the 11+ examination was abandoned in most areas of the coun-

try. Thus, the month of birth effect in educational achievement seems to be mostly

driven by the early tracking faced by the earlier cohorts.

16

16.2

16.4

16.6

16.8

17

17.2

1 9

4 5

1 9

4 6

1 9

4 7

1 9

4 8

1 9

4 9

1 9

5 0

1 9

5 1

1 9

5 2

1 9

5 3

1 9

5 4

1 9

5 5

1 9

5 6

1 9

5 7

1 9

5 8

1 9

5 9

1 9

6 0

1 9

6 1

1 9

6 2

1 9

6 3

1 9

6 4

1 9

6 5

September Born

Figure 6 Average school leaving age by year of birth: England only.

Chevalier et al. IZA Journal of Labor Economics Page 11 of 222013, 2:8 http://www.izajole.com/content/2/1/8

4. Estimates Our basic model of the impact of parental background on the post-compulsory school-

ing participation of their children is:

PCc ¼ β0 þ β1Sm þ β2Sf þ β3Y f þ X0hΛ þ f DBm; DBf ; DBc � �

þ ωc ð1Þ

where the c, m and f subscripts refer to the child, mother’s and father’s characteristics

within a particular household h13. The dependent variable PCc is a dummy variable de-

fining participation in post compulsory education. The effect of parental education

levels, as measured in years of schooling of both the mother and father (Sm, Sf), is in-

cluded in the model linearly. In Figures 1 and 2 there is clearly a diminishing effect of

parental education, however a linear specification more easily facilitates instrumental

variable estimation in later estimates. Log parental income Yf is measured by father’s

real log gross weekly earnings from employment. DB refers to date of birth (year and

month) so that f(.) controls for cohort trends in paternal, maternal and child education.

With respect to child’s date of birth, we use a set of dummy variables for each year.

With respect to parental date of birth (both year and month for both parents), we con-

vert the date to a continuous number and then enters the model in cubic form for each

parent. Xh contains characteristics common to all three members of the family - year of

survey dummies as well as region of residence at time of survey. Equation 1 is esti-

mated as a linear probability model to subsequently facilitate the use of instrumental

variables14.

In later extensions to our basic model, we additionally condition on paternal occupa-

tion, so that the difference in unionization between occupations does not identify the

IV model. We categorize paternal occupation into blue-collar/manual versus white-

collar/non-manual. We can repeat the entire analysis using seven different occupation

categories – unskilled, semi-skilled, etc. but the results are essentially the same. In

other extensions to our basic model we add paternal union status, so that the identifi-

cation in the IV model only comes from the interaction terms between union status

and occupation. This then captures any differences in parenting behavior that union-

ized fathers may have.

Table 3 summarizes our OLS estimates of the effects of parental income and parental

education levels on the probability of post-compulsory schooling of the child15. Specifica-

tion (1) only controls for parental years of schooling. A year of maternal education is

Table 3 OLS estimates of parental education and income on the probability of post-compulsory schooling of children

Specification: (1) (2) (3) (4) (5) (6) (7) (8) (9)

Sons: N = 8661

Maternal school leaving age

0.029*** 0.028*** 0.026*** 0.026*** 0.026*** 0.025***

(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)

Paternal school leaving age

0.025*** 0.020*** 0.017*** 0.015*** 0.016*** 0.015***

(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)

Paternal log earnings 0.118*** 0.066*** 0.064*** 0.040*** 0.062*** 0.038***

(0.008) (0.008) (0.008) (0.008) (0.008) (0.009)

Daughters: N = 8137

Maternal school leaving age

0.024*** 0.023*** 0.021*** 0.020*** 0.021*** 0.020***

(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)

Paternal school leaving age

0.015*** 0.012*** 0.009*** 0.007*** 0.009*** 0.007***

(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)

Paternal log earnings 0.088*** 0.050*** 0.045*** 0.027*** 0.044*** 0.027***

(0.007) (0.008) (0.008) (0.008) (0.008) (0.008)

Control for paternal union member

no no no no no no yes yes yes

Control for paternal blue collar occ

no no no yes yes yes yes yes yes

Note: LFS 1992–2012. Standard errors in parentheses. Specifications also include year of survey dummies, regional dummies, dummies of child’s year of birth, cubics in parental dates of birth. * p < .1. ** p < .05. *** p < .01.

Chevalier et al. IZA Journal of Labor Economics Page 12 of 222013, 2:8 http://www.izajole.com/content/2/1/8

associated with a 2.9 percentage points increase in the probability of post-16 participation

for boys and 2.4 percentage points for girls. The impact of paternal education is somewhat

lower especially for daughters. Specification (2) examines the impact of paternal income

but excludes the parental education controls and suggests an income elasticity with respect

to post compulsory schooling participation of 11.8% boys and for 8.8% for girls. Finally spe-

cification (3) includes both education and income controls. The direct effects of maternal

education estimated in Specification (1) are reduced very slightly. The effect of paternal

education is reduced by 20% and the income effects are reduced by 40% compared to (1)

and (2) respectively, highlighting the correlation between paternal education and income.

The second set of estimates (4, 5 and 6) in Table 3 adds the paternal occupation status

(blue collar dummy variable). This is potentially an endogenous variable, but since the

unionization rate differs by occupation, when not controlling for occupation the union

instrument might partially capture occupational income, which would invalidate its use.

Since paternal occupation is correlated both with paternal education and (permanent)

income, the estimates for these parameters are reduced. In specifications (5) and (6) income

is best interpreted as deviation within blue/white collar status. As such, the income effects

are reduced by about 30% for boys and about 50% for girls. Note, however, that when

controlling for occupation, adding paternal income only marginally reduces the effect

of paternal education on the educational attainment of children. Thus indicating that

the correlation between paternal education and income mostly captures the permanent

component of income, rather than income shocks.

The third panel of Table 3 shows the effects of parental education and paternal

earnings when controlling for the effects (not interacted) of paternal union

Chevalier et al. IZA Journal of Labor Economics Page 13 of 222013, 2:8 http://www.izajole.com/content/2/1/8

membership and paternal blue-collar dummy (specifications (7), (8) and (9)). The ef-

fects of parental education are virtually unchanged compared to the estimates pre-

sented in columns (4) to (6); supporting the view that paternal union membership has

no direct effect on the education decision of his children. For girls, the effect of pater-

nal income also remains unchanged compared to specifications (5) and (6) while for

boys it decreases marginally.

To summarize these results: one additional year of maternal education is associated

with an increase in the probability of staying on of around 2–2.5 percentage points for

maternal education (larger effects on boys) and a smaller effect of paternal education

between 1 and 2.5 percentage points, especially for girls. The gap between the effect of

maternal and paternal education increases when measures of, or proxies for, income

are introduced since paternal education is correlated with paternal income. Note also

that the current income effects are greatly reduced when a proxy for permanent in-

come (the father occupation variable) is controlled for. The income effects are in the

same range as those obtained by Dearden et al. (2009) in their evaluation of the Education

Maintenance Allowance. However, the EMA was a (conditional) cash transfer of only £30

per week, so the estimated income elasticities reported here appear quite small. The ef-

fects of parental education are in the higher range of those reported by Oreopoulos et al.

(2006) for drop-out in the US.

To simultaneously control for the potential endogeneity of paternal income and par-

ental schooling, we specify the following system of first stage equations for maternal

education, paternal education and the log of paternal earnings:

Sm ¼ g DBm; DBf ; DBc � �

þ X0hΓ þ Z0f;mΛ þ ψm ð2Þ Sf ¼ h DBm; DBf ; DBc

� � þ X0hΠ þ Z0f;mΗ þ χf ð3Þ

Y f ¼ k DBm; DBf ; DBc � �

þ X0hΞ þ Z0f;mΘ þ ϕf ð4Þ

The functions g(.), h(.), k(.) are polynomials of order three in parental dates of birth

and include a set of dummy variables in child year of birth. X contains the control

variables. As in the OLS estimates, we consider two specifications: controlling for occu-

pation and controlling for occupation and union status. Z contains the instrumental

variables. ψm, χf, ϕf are idiosyncratic errors terms. The instruments are as discussed

above: a dummy variable RoSLA, which takes the value one for parents born after

September 1957, a linear function in month of birth16, which takes the value of one for

September born through to twelve for August born, and either the interaction of paternal

union membership (PUM) and paternal blue collar status (i.e. manual worker) or this

interaction and the direct effect of paternal union status. Given that we control for

smooth trends in parental date of birth, the RoSLA variable acts as a regression discon-

tinuity and picks up the effects of the reform alone.

We estimated a wide variety of first stages and corresponding second stage equations

to examine the sensitivity of the second stage estimates to the set of exclusion restrictions

used to define the instrumental variables. Our IV estimates have the property, which is

also a feature of the OLS estimates, that the addition of income to a model containing

just parental education levels makes little difference to the parental education estimates.

Thus, we refrain from presenting specifications that contain just parental income or just

parental education levels and Table 4 only reports our favorite specification.

Table 4 Instrumental variable estimates: first stage regressions

Sons Daughters

Paternal schooling

Maternal schooling

Log paternal earnings

Paternal schooling

Maternal schooling

Log paternal earnings

Paternal RoSLA 0.228** −0.011 0.010 0.263** 0.082 −0.024

(0.104) (0.091) (0.024) (0.105) (0.092) (0.024)

Paternal MoB −0.019** 0.007 −0.003 −0.009 −0.004 0.000

(0.008) (0.007) (0.002) (0.008) (0.007) (0.002)

Maternal RoSLA 0.022 0.231*** 0.017 −0.049 0.242*** 0.035

(0.102) (0.089) (0.102) (0.104) (0.091) (0.024)

Maternal MoB 0.005 −0.016** 0.001 0.016* −0.014** 0.006***

(0.008) (0.007) (0.002) (0.008) (0.007) (0.002)

Paternal Union Membership (PUM) 0.477*** 0.484*** 0.010 0.462*** 0.509*** −0.001

(0.067) (0.059) (0.016) (0.069) (0.061) (0.016)

PUM*Blue Collar −0.540*** −0.520*** 0.190*** −0.508*** −0.495*** 0.209***

(0.100) (0.087) (0.023) (0.102) (0.089) (0.023)

Using PUM as an IV

F-test of all instruments in each individual equation

10.51 13.11 23.71 9.7 13.85 25.23

Angrist and Pischke F test 6.24 7.77 35.45 5.75 8.25 37.48

F-test of all instruments in all equations jointly

15.27 15.78

Using PUM in second stage as exogenous variables

F-test of all instruments in each individual equation

8.39 9.42 15.82 7.78 8.74 17.59

Angrist and Pischke F test 7.28 7.44 13.81 6.23 4.34 7.17

F-test of all instruments in all equations jointly

13.70 12.80

All instruments appear in all first stage equations. All second stage exogenous controls appear in all first stage equations but estimates for these are omitted from these tables in the interest of brevity. * p < .1. ** p < .05. *** p < .01.

Chevalier et al. IZA Journal of Labor Economics Page 14 of 222013, 2:8 http://www.izajole.com/content/2/1/8

All of the instruments are used in each of the first stage equations. There are some

cases where an instrument is having a significant effect in an equation that a priori it

would not be expected to have an effect on. For example, paternal union membership

has a significant effect in the maternal schooling equation. This is probably a byproduct

of correlation between our instruments caused by assortative mating.

The instruments for education have almost identical effects on father’s and mother’s

education. Being affected by the raising of the school leaving age increased the affected

parent’s education by 0.23 of a year, while an August born child, on average, left school

one fifth of a year earlier than a September born child. Note that, instrumenting educa-

tion by only one of the instruments at the time makes little difference to the size and

significance of the estimates (not reported) suggesting that the two instruments identify

the effects of exogenous shocks to parental education through different mechanisms.

So while both instruments identify a population of marginal students, these are not

identical populations.

For paternal earnings we consider two specifications either union membership is in-

cluded as an instrument as well as its interactions with blue collar status or it is

Chevalier et al. IZA Journal of Labor Economics Page 15 of 222013, 2:8 http://www.izajole.com/content/2/1/8

included in the second stage and only the interaction is used as an instrument (blue

collar status is always included in the second stage). These two specifications of the

model result in identical first stages. Only the interaction of union membership and oc-

cupation is significant in the earning equation showing that blue-collar workers drive

the union premium. Union membership has no effect on white collars earnings but in-

crease blue-collar earnings by up to 20%.

Weak instruments can result in biased estimates in the second stage equation so it is

important that the instruments’ relevance is tested. Since we have three endogenous

variables there are various ways for testing the weakness of the instruments. We thus

present different F-tests to test the joint significance of the instruments. First, we test

for the joint significance of all the instruments in all the equations separately. For sons,

the F-statistic for the joint significance of all the instruments in the Paternal Schooling

Equation is 10.51. The F-statistic for maternal schooling and paternal earnings are

13.11 and 23.71 respectively. Additionally, we present an F-test of the joint significance

of the instruments in the three equations simultaneously, which for the sons’ equation

is 15.27. These statistics compare reasonably favorably with the Stock and Yogo (2002)

suggested cut-off of around 10. Additionally, we present the modified F-statistic of

Angrist and Pischke (2009) which is suggested for cases of multiple endogenous vari-

ables, and which partials out the effects of the instruments on the other endogenous

variables. These F-tests range from 6.24 to 35.45 for sons. There are no suggested crit-

ical values of this test but using the Stock and Yogo’s rule of thumb would suggest that

the instruments for education are somehow weak. We repeat the exercise for our fa-

vored specification, i.e. treating union membership as a control variable. We reach

similar conclusions regarding the validity of the instruments under this more strict ex-

clusion restriction. These conclusions are similar for the estimates based on daughters.

Table 5 shows the second stage estimates and reports the probability of staying on sep-

arately for sons and daughters, and for the two specifications, i.e. excluding union status

or not. The effect of maternal education is considerably reduced and always imprecisely

estimated. The estimate itself varies widely from specification to specification. The effect

of paternal education on sons is never significant and varies widely, but for daughters the

IV point estimate, at 0.07, is ten times the size of the OLS estimate. The lack of signifi-

cance of parental education is similar to the findings of Black et al. (2005) for Norway.

However, we find a reasonably strong effect of paternal education on daughter’s educa-

tion, as in twin studies, which is much larger than the OLS estimate. The IV estimates

should be interpreted as LATE. The fathers affected by the school reform had large finan-

cial returns to that decision (Harmon and Walker 1995) which may have changed how

they value education. It remains unclear why this affects daughters’ behavior more

than boys.

The effect of paternal earnings varies with the set of instruments used. There is never

a statistically significant effect of paternal earnings on girls and the coefficients vary

widely depending on the specification. The estimated effect of paternal earnings on

sons also varies between specifications. It is large and significant when earnings are in-

strumented by both union status and its interaction with blue collar status. However

when only the interaction of union and occupation is used as instruments and both

union and occupations are controlled for in the second stage (columns 2 of Table 5),

the estimates become very imprecise and unstable. Again, those results are similar to

Table 5 Instrumental variable estimates: LFS 1992-2012

Specification: (1) (2) (3) (4)

Population ALL ALL Blue White

Sons:

Maternal school leaving age 0.011 −0.030 0.016 −0.016

(0.037) (0.045) (0.082) (0.044)

Paternal school leaving age 0.028 0.003 −0.051 0.077*

(0.036) (0.040) (0.082) (0.044)

Paternal log earnings 0.157** −0.025 0.192** −0.221

(0.066) (0.126) (0.082) (0.234)

N 8661 8661 3,833 4,828

Daughters:

Maternal school leaving age 0.001 0.021 −0.009 0.017

(0.034) (0.055) (0.060) (0.036)

Paternal school leaving age 0.070** 0.078* 0.015 0.055

(0.035) (0.041) (0.077) (0.038)

Paternal log earnings −0.031 0.039 −0.030 0.059

(0.060) (0.165) (0.074) (0.188)

N 8137 8137 3,630 4,507

Instruments RoSLA RoSLA RoSLA RoSLA

MoB MoB MoB MoB

PUM PUM PUM

PUM*Blue PUM*Blue

Second stage controls PUM

Blue Blue

Notes: Standard errors in parentheses. Specifications include year of survey dummies, regional dummies, dummies of child’s year of birth, cubics in parental dates of birth. RoSLA is a dummy for the Raising of School Leaving Age, MoB stands for Month of Birth (linear), PUM for Paternal Union Status, and Blue indicates if the father workers in a manual job. * p < .1. ** p < .05.

Chevalier et al. IZA Journal of Labor Economics Page 16 of 222013, 2:8 http://www.izajole.com/content/2/1/8

those obtained by Shea (2000) and suggest that (unconditional) cash transfer might not

alter the transmission of education choice from one generation to the next.

We now discuss some robustness checks of our results. In the last two columns of Table 5,

we present estimates for two sub-populations: blue and white collars, since the first stage re-

sults presented in Table 4, suggest that our instrument for parental earnings has power only

for the blue-collar population17. The paternal union status dummy variable, which captures

unionized white-collar fathers, has no effect on paternal earnings whereas the interaction of

paternal union status and paternal blue-collar status has a positive (coefficient of around 0.2

depending on gender of the child) and statistically significant effect. Looking at the second

stages for these sub-groups, we again find no effect of parental characteristics on the out-

comes of daughters. For sons, we see that there are heterogeneous effects of paternal

earnings – there is a significant effect of blue-collar fathers’ earnings on sons but no effect

of white collar fathers’ earnings on sons. This suggests that there is heterogeneity in the

effect of parental income on participation, with poorer sons likely to benefit from financial

transfers. For white-collar fathers, there is a weak effect of paternal education on sons.

We now investigate the validity of our estimates when using the Raising of the School

Leaving Age as an IV for parental school leaving age. There may be concerns that in-

cluding date of birth of parents (converted to a continuous variable) as a cubic

Table 6 Effect of “fake” and actual raising of school leaving Age on parental education

Sons Daughters

Year of “Fake” reform Effect of father’s RoSLA dummy variable on age

father left education

Effect of mother’s RoSLA dummy variable on age mother left education

Effect of father’s RoSLA dummy variable on age

father left education

Effect of mother’s RoSLA dummy variable on age mother left education

1953 −0.078 −0.162* −0.036 −0.132

(0.100) (0.090) (0.104) (0.093)

1954 0.007 −0.232** −0.100 −0.169*

(0.101) (0.090) (0.104) (0.093)

1955 0.214** −0.046 −0.064 0.031

(0.101) (0.090) (0.104) (0.093)

1956 0.193* 0.090 0.021 0.134

(0.102) (0.090) (0.104) (0.092)

1957 True reform 0.241** 0.247*** 0.276*** 0.258***

(0.104) (0.089) (0.104) (0.091)

Estimates used in this table based on first stage equations that control for year of survey dummies, regional dummies, dummies of child’s year of birth, cubics in parental dates of birth, paternal union status, paternal blue collar status and interaction of union status and blue collar worker. * p < .1. ** p < .05. *** p < .01.

Chevalier et al. IZA Journal of Labor Economics Page 17 of 222013, 2:8 http://www.izajole.com/content/2/1/8

polynomial and using a dummy variable for being affected by RoSLA (being born after

September 1957) is not correctly modeling the discontinuity in educational attainment

and only picks up a general upward trend in parental educational attainment. In Table 6,

we show our estimates of the effect of “fake” RoSLA reforms on parental education.

We simulate the effects of miscoding RoSLA to dates prior to its actual introduction.

For example, in the first row of Table 6, we see that when RoSLA is set to have hap-

pened in 1953, there was no significant effect of the RoSLA dummy variable on father’s

education and a negative effect (albeit only significant at the 10% level) on maternal

education. Comparing the effects of the fake reforms with the actual reform (the bot-

tom row of Table 6) we see that the actual reform is picking up the discontinuity in at-

tainment rather than just upward trends because the dummy variables are positive and

highly statistically significant for both fathers and mothers for both the sample of sons

and the sample of daughters, which is never the case in any of the fake reform cases.

Another concern about our results may be that the general lack of significance of the IV

results is driven by a high correlation between the instruments, which would make the

model under-identified. Indeed, the correlations between the parent’s ROSLA status is high

(0.67) but the correlation between parental month of birth is close to zero. To investigate

this issue further, we simplified our model and estimated the effect of paternal income and

maternal education, or paternal income and paternal education, only. Using only one par-

ent at the time also means that the estimate will capture the education effect of the other

parent. Since parents are positively matched, we would expect the estimates obtained

with a single parent to be larger than those obtained when including both parent’s educa-

tion. The point estimates obtained are almost identical to those reported for the full

model but surprisingly are less precisely estimated18. As such it seems unlikely that our

results are driven by the correlation between the instruments.

One final concern is that the various criteria introduced to select our sample, i.e.

living at home with both parents with father’s income being correctly reported,

Chevalier et al. IZA Journal of Labor Economics Page 18 of 222013, 2:8 http://www.izajole.com/content/2/1/8

resulted in an unrepresentative sample. To address this, we re-estimated our OLS

results on a sample of two-parent households that included Northern Ireland and

Scotland residents, non-English born parents, or very old/young parents. This produces a

very similar set of results as in Table 3, although the maternal education effect is much

smaller. Running the model on a sample of single mothers raises questions of female

labour supply that are beyond the scope of this paper to correct, but the crude regressions

show that both maternal education and maternal earnings have positive and significant

effects on the child remaining in full-time education19. As such we are confident that our

conclusions are not driven by sample selection.

5. Conclusion This paper has addressed the intergenerational transmission of education and investigated

the extent to which early school leaving (at age 16) may be due to variations in permanent

income and parental education levels. Least squares revealed conventional results - stronger

effects of maternal than paternal education and stronger effects on sons than daughters. We

also found that the education effects remained significant even when household income

was included. Current income remains significant even when some measures of permanent

income are included which indicates that some children could be financially constrained in

their decision to attend post-compulsory education.

When using paternal union status and the interaction of paternal union status and pa-

ternal occupation as instrumental variables, the IV results strengthen the role of paternal

education, but for daughters - one year of paternal education increases the probability of

his daughter staying on by seven percentage points. In contrast, maternal education has

no statistically significant impact on the probability of remaining in education for either

sons or daughters – a result which is robust to the range of instrument sets used. In

treating paternal income as endogenous, the elasticity of income on schooling decisions

increases by a multiple of four for sons, but there is no effect on daughters. The results

with respect to income are more sensitive to the instrument set used, and are less robust

than the results relating to parental education.

The results reinforce the challenges faced by policymakers in shaping policy options to

encourage greater educational participation. Policymakers often view the problem of low

participation as an issue of credit constraints at the minimum school leaving age acting as

a barrier to further participation. For the United Kingdom, evaluations of the pilot imple-

mentation of a means tested conditional cash transfer (Educational Maintenance Allow-

ances (EMAs)) paid to the student show relatively large changes in participation in post-

compulsory education in the ‘treated’ group, particularly for those with the lowest prior

achievement ((Dearden et al. (2009)). This may indicate credit constraint issues, but also

is consistent with a price effect from the lowering of the ‘opportunity cost’ of staying on.

In this latter case, we cannot be sure about the extent to which this extra education is

valuable to these participants. A conditional cash transfer policy may therefore carry siz-

able deadweight potential despite the success of the evaluations of initiatives like EMA.

Indeed, recent UK evidence (Chowdry et al. (2010)) that linked administrative data for a

cohort from age 11 to age 19 finds that poor attainment in secondary school is more im-

portant in explaining lower participation rates (in higher education) amongst students

from disadvantaged backgrounds than income/credit constraints.

Chevalier et al. IZA Journal of Labor Economics Page 19 of 222013, 2:8 http://www.izajole.com/content/2/1/8

In this paper we show that a policy of increasing family permanent income through in-

creasing parental education would also have some positive effects, especially for daughters.

More importantly, a focus on increasing participation can have ‘multiplier’ effects through

the intergenerational impact - the recently proposed increase of the de facto school leaving

age in the UK to 18 would also benefit future generations through direct intergenerational

transmission of educational choice, especially through the father.

Endnotes 1We also investigate the relative effects of parental education levels and household in-

come on educational achievement at age 16. High school students in England and

Wales usually study up to ten subjects until the age of 16 which are then examined at

the end of compulsory schooling in the school year that they reach 16. These are scored

as A* to G with A* -C being regarded as passing grades. The government’s objective is

that 60% of all 16 year olds pass in at least five subjects. This level of achievement is usu-

ally required to progress into senior high school. Not surprisingly our estimated results

for the effect on junior high school achievement closely parallel the results for early school

leaving and so are not reported here, but are available on request. 2MTO programs are associated with noticeable improvements in child behavior and

test scores, but whether these are caused by the financial gain, changes in the physical

environment, school effects and/or peer-group changes remains unclear. Work on

MTO by Sanbonmatsu et al. (2011) suggests that MTO-driven neighbourhood effects

on academic achievement were not significant. 3Acemoglu and Pischke (2001) use similar arguments to Mayer (1997) and exploit

changes in the family income distribution between the 1970’s and 1990’s. They find a

10 percent increase in family income is associated with a 1.4% increase in the probabil-

ity of attending a four-year college. Loken (2010) studies the long-term effect of family

income on children’s educational attainment using the Norwegian oil shock in the

1970s as an instrument. They find no causal relationship. 4Prior to 1993 there was no earnings data in LFS. Between 1993 and 1998, earnings

data is available only for fifth wave respondents; from 1998 the earnings data is col-

lected in the first and in the final wave. 5Re-estimating without the 18 year olds showed no economically or statistically sig-

nificant differences in results. 6We may thus underestimate participation if individuals postpone attending post-

compulsory education until they are over 18. We believe this is a rather rare event. 7Whilst this may create some selection bias it would be difficult to overcome this in

our data. Since parental separation is probably more likely for children with large (but

unobservable) propensities to leave school early, and it is also likely to be negatively

correlated with parental education and income we might expect to underestimate the

effect of income and education on the dependent variables. We also examined the ef-

fects of living with a stepparent but found that while there was a negative stepparent ef-

fect, the interaction between this and education or income proved insignificant. Results

are available on request. 8Note that due to institutional characteristics there are only a few parents with school

leaving age of 17, 19, 20 and above 23.

Chevalier et al. IZA Journal of Labor Economics Page 20 of 222013, 2:8 http://www.izajole.com/content/2/1/8

9Official statistics from the Department for Children, Schools and Families show 67% of

boys and 75% of girls (averaged over the years examined in this paper) choosing to stay so

our own staying-on figures from LFS are a little higher. This is likely to be a reflection of

the selections that we have made. 10Support for the view that unionization picks up differences in labor market

productivity is mixed. Murphy and Topel (1990) find that individuals who switch

union status experience wage changes that are small relative to the corresponding

cross-section wage differences, suggesting that union premia are primarily due to

differences in unobserved ability. However Freeman (1994) counters this view, ar-

guing that union switches in panel data are largely spurious so that measurement

error biases the union coefficient towards zero in the panel. 11The British Cohort Study (BCS) data, of all children born in England and

Wales in a particular week in 1970, records, in considerable detail, the attitudes

and behaviours of fathers towards their children. This data suggests rather small

differences in attitudes and behaviours across union status. For example, 23% on

unionised fathers disagreed with the statement that “The needs of children are

more important than one’s own”, compared to 18% of the non-unionised; 60%

(62%) of children with unionised (non-unionised) fathers watched TV less than

2 hours per day on a typical weekend day; 83% (88%) of unionised (non-union-

ised) fathers read stories more than once per week; 57% (52%) of unionised (non-

unionised) fathers always (as opposed to often/sometimes/never) talked to his child even

when busy; 79% (79%) of unionised (non-unionised) fathers showed the child physical af-

fection at least once per day and 36% (37%) praised the child at least once per day; 94%

(95%) of unionised (non-unionised) fathers has helped young children learn numbers,

etc.; and 79% (80%) of unionised (non-unionised) fathers aspired for the child to continue

in full-time education at age 16. The children also reported behaviour that might well re-

flect parenting styles. For example, 56% (54%) of the children of unionised (non-union-

ised) fathers made their own bed and 49% (52%) cleaned their own room. 12We use July rather than August for this comparison since there is likely to be some

ambiguity with August-born children to the extent that schools exercised discretion at

the margin. 13Standard errors are clustered at the household level throughout the analysis. Only

about 13% of the sample are siblings. The analysis is repeated separately for sons and

daughters. Only about 7% of the sample are same-sex siblings and hence used to esti-

mate the same equations. 14The marginal effects from probit estimation are very close to the linear probability

model coefficient estimates and are available on request. 15We control for smooth cohort trends by including a cubic function of parents

months/years of birth. Full results are available on request. Similar estimates based on

probit models are also available. 16Greater flexibility could be sought but at the cost, of course, of potential weakness

in the instruments. 17The corresponding first stage equations, not reported here, were estimated separ-

ately for blue collar and white collar fathers. 18Results available on request. 19Results available on request.

Chevalier et al. IZA Journal of Labor Economics Page 21 of 222013, 2:8 http://www.izajole.com/content/2/1/8

Competing interests The IZA Journal of Labor Economics is committed to the IZA Guiding Principles of Research Integrity. The authors declare that they have observed these principles.

Acknowledgements Financial support from the HM Treasury Evidence Based Policy programme, the Nuffield Foundation Small Grant Scheme and the award of a Nuffield Foundation New Career Development Fellowship to Harmon is also gratefully acknowledged. Chevalier is also a Research Fellow at the University College Dublin Geary Institute. We are grateful to Pedro Carneiro, Kevin Denny, Robert Haverman, Lisa Farrell, Robin Naylor, Bobbi Wolfe and seminar participants at ANU Canberra, CEMFI in Madrid, RAND in Santa Monica, the University of Glasgow, School of Public Health at Harvard University, the Melbourne Institute, RMIT, University of Warwick and the Tinbergen Institute for comments. We also thank the Associate Editor and Referees for suggestions that have greatly improved the manuscript. The data used in this paper was made available by the UK Data Archive at the University of Essex. We will provide every assistance to other approved researchers who wish to use this data to replicate and extend our analysis. Responsible editor: Pierre Cahuc.

Author details 1Royal Holloway, University of London, London, UK and IZA, Bonn, Germany. 2University of Sydney, Sydney Australia and IZA, Bonn, Germany. 3TILDA, Trinity College Dublin and ESRI, Dublin. 4Lancaster University Management School, Lancaster, UK and IZA, Bonn, Germany.

Received: 21 September 2013 Accepted: 12 November 2013 Published: 09 Dec 2013

References

Acemoglu D, Pischke JS (2001) Changes in the wage structure, family income, and Children’s education. Eur Econ Rev

45(4–6):890–904 Almond D, Currie J (2010) Human capital development before age five. National Bureau of Economic Research, Inc,

NBER Working Papers. 15827, Cambridge, Massachusetts, USA Angrist JD, Pischke J-S (2009) Mostly Harmless Econometrics: An Empiricist’s Companion. Princeton University Press,

Princeton and Oxford Behrman JR, Rosenzweig MR (2002) Does increasing women’s schooling raise the schooling of the next generation?

Am Econ Rev 92(1):323–334 Bjorklund A, Lindahl M, Plug E (2006) The origins of intergenerational associations: lessons from Swedish adoption data.

Q J Econ 121(3):999–1028 Björklund A, Salvanes KG (2011) Chapter 3 - Education and Family Background: Mechanisms and Policies. In: Eric A,

Hanushek SM, Ludger W (ed) Handbook of the Economics of Education, vol Volume 3, 2011th edition. Elsevier, pp 201–247

Black SE, Devereux PJ (2011) Chapter 16 - Recent Developments in Intergenerational Mobility. In: David C, Orley A (ed) Handbook of Labor Economics, vol Volume 4, Part B. Elsevier, pp 1487–1541

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

Blanden J, Gregg P (2004) Family income and educational attainment: a review of approaches and evidence for Britain. Oxf Rev Econ Policy 20(2):245–263

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

Carneiro P, Heckman JJ (2004) Human Capital Policy. In: Heckman JJ, Krueger AB (ed) Inequality in America. MIT Press, Cambridge

Chevalier A, Harmon C, Walker I, Zhu Y (2004) Does education raise productivity, or just reflect it? Economic Journal 114(499):F499–F517

Chevalier A, Lanot G (2002) The relative effect of family characteristics and financial situation on educational achievement. Educ Econ 10(2):165–181

Chowdry H, Crawford C, Dearden L, Goodman A, Vignoles A (2010) Widening participation in higher education: analysis using linked administrative data. Institute for Fiscal Studies, IFS Working Papers: W10/04, London, UK

Crawford C, Dearden L, Meghir C (2007) When You Are Born Matters. The Impact of Date of Birth on Child Cognitive Outcomes In England. The Institute for Fiscal Studies, London, UK

Currie J, Moretti E (2003) Mother’s Education and the intergenerational transmission of human capital: evidence from college openings. Q J Econ 118(4):1495–1532

Dearden L, Emmerson C, Frayne C, Meghir C (2009) Conditional cash transfers and school dropout rates. J Hum Resour 44(4):827–857

Dixon M, Gregg P, Robinson H (2013) Early, Late or Never? When does Parental Education Impact Child Outcome? IZA Discussion Paper, Bonn, Germany

Freeman RB (1994) H. G. Lewis and the study of union wage effects. J Labor Econ 12(1):143–149 Harmon C, Walker I (1995) Estimates of the economic return to schooling for the United Kingdom. Am Econ Rev

85(5):1278–1286 Heckman JJ, Masterov DV (2005) Skill Policies for Scotland. In: Coyle D, Alexander W, Ashcroft B (ed) New Wealth for

Old Nations: Scotland’s Economic Prospects. Princeton University Press, Princeton and Oxford, pp 119–165 Holmlund H, Lindahl M, Plug E (2011) The causal effect of parents’ schooling on Children’s schooling: a comparison of

estimation methods. J Econ Lit 49(3):615–651

Chevalier et al. IZA Journal of Labor Economics Page 22 of 222013, 2:8 http://www.izajole.com/content/2/1/8

Jenkins S, Schluter C (2002) The Effect of Family Income during Childhood on Later-Life Attainment: Evidence from Germany. DIW Discussion Papers, Berlin, Germany

Krueger AB (2004) Inequality, Too Much of a Good Thing. In: Heckman JJ, Krueger AB (ed) Inequality in America: What role for human capital policies? Edited with an introduction by Benjamin M. Friedman. Alvin Hansen Symposium on Public Policy series. MIT Press, Cambridge and London, pp 1–75

Lewis HG (1986) Union Relative Wage effects: A Survey. University of Chicago Press, Chicago Loken KV (2010) Family income and children’s education: using the Norwegian oil boom as a natural experiment.

Labour Econ 17(1):118–129 Lundborg P, Nordin M, Olof Rooth D (2011) The Intergnerational Transmission of Human Capital: Exploring the Role of

Skills and Health Using Data on Adoptees and Twins. IZA Discussion Paper, Bonn, Germany Mayer SE (1997) What money can’t buy: family income and children’s life chances. Harvard University Press, Cambridge

and London McLachlan R, Gilfillan G, Gordon J (2013) Deep and Persistent Disadvantage in Australia. Australian Government

Productivity Commission Staff Working Paper. Australian Government Productivity Commission, Canberra, Australia Murphy KM, Topel RH (1990) Efficiency Wages Reconsidered: Theory and Evidence. In: Weiss Y, Fishelson G (ed)

Advances in the theory and measurement of unemployment. St. Martin’s Press, New York, pp 204–240 Oreopoulos P, Page M, Stevens AH (2006) The intergenerational effects of compulsory schooling. J Labor Econ

24(4):729–760 Plug E (2004) Estimating the effect of mother’s schooling on Children’s schooling using a sample of adoptees.

Am Econ Rev 94(1):358–368 Pronzato C (2012) An examination of paternal and maternal intergenerational transmission of schooling. J Popul Econ

25(2):591–608 Sacerdote B (2004) What Happens When We Randomly Assign Children to Families? National Bureau of Economic

Research, Inc. NBER Working Papers, Cambridge, Massachusetts, USA, 10894 Sanbonmatsu L, Ludwig J, Katz LF, Gennetian LA, Duncan GJ, Kessler RC, Adam E, McDade TW, Lindau ST (2011)

Moving to Opportunity for Fair Housing Demonstration Program - Final Impacts Evaluation. US Department of Housing and Urban Development Office of Policy Development and Research, Washington DC, USA

Shea J (2000) Does parents’ money matter? J Public Econ 77(2):155–184 Stock JH, Yogo M (2002) Testing for Weak Instruments in Linear IV Regression. National Bureau of Economic Research,

Inc. NBER Technical Working Paper, Cambridge, Massachusetts, USA, p 284

Cite this article as: Chevalier et al.: The impact of parental income and education on the schooling of their children. IZA Journal of Labor Economics

10.1186/2193-8997-2-8

2013, 2:8

Submit your manuscript to a journal and benefi t from:

7 Convenient online submission 7 Rigorous peer review 7 Immediate publication on acceptance 7 Open access: articles freely available online 7 High visibility within the fi eld 7 Retaining the copyright to your article

Submit your next manuscript at 7 springeropen.com