Centre for Market and Public Organisation Parental income and child outcomes Paul Gregg, Carol Propper and Elizabeth Washbrook Avon Local Group of the.

Centre for Market and Public Organisation

Parental income and child outcomesPaul Gregg, Carol Propper and Elizabeth Washbrook

Avon Local Group of the Royal Statistical Society Meeting

Bristol, 26th May 2009

Family income and child outcomes Drivers of the developmental deficits of low income children are of interest to academics and

policymakers Lifecycle models of human capital formation (Cunha and Heckman 2007; Carneiro and Heckman

2003) Labour reforms to tax credits, child care and early education, child benefit, maternity leave (e.g.

Brewer 2007)

Developmental outcomes in childhood are related to multiple aspects of adult well-being and have long term social consequences Cognitive and educational outcomes – IQ, test scores, grades Non-cognitive and socio-emotional outcomes – behavior, self-concept, motivation, attention Physical health – obesity and poor nutrition, asthma, injuries, illnesses

Adult outcomes associated with at least one class of child outcome Years of schooling, qualifications, employment and earnings, mental health, life expectancy and

morbidity, antisocial and risky behavior, crime, fertility

This paper

Uses data from an unusually rich birth cohort dataset to compare the income gradients in six developmental outcomes in mid-childhood The income gradient is the unconditional association between income and the outcome – one

broad measure of social inequality

Develops a descriptive decomposition method to give an overview of the underlying associations that give rise to the observed income gradients We estimate the portion of the observed gradients predicted by income-related differences in a

wide range of potential explicators

Estimates can be interpreted in the light of two approaches taken in the literature Reduced form OLS studies of the relationship between child poverty and outcomes (e.g. Duncan

and Brooks-Gunn 1997). [Precursors to studies on the causal effects of income, e.g. Blau 1999; Dahl and Lockner 2005.]

Correlational SEM studies of the mediators between family income and child outcomes (Guo and Harris 2000; Yeung et al., 2002)

The contribution of descriptive estimates The associations identified in our estimates are not causal. They do not, for example, adjust for

reverse causation or the influence of unobservable third factors such as inherited ability

Causal approaches provide crucial evidence on parts of the puzzle of why low income children fall behind The effect of increasing cash benefits for low income families The effects specific factors on outcomes (intervention programmes, smoking, birth order, inherited

characteristics)

But causal studies relying on specific mechanisms cannot give an overview of relative importance of all the potential factors that drive the intergenerational persistence of poverty. Single and teen parenthood, low parental education, worklessness and deprived neighborhoods –

control variables Parental stress and depression, social connections, child care experiences, unhealthy environments,

parenting behaviors - mediators

Contribution of our paper What are the upper bounds on the effects of interventions targeted to specific

factors in terms of reducing social inequality in child well-being?

Conclusions depend on whether we focus on a single or multiple classes of outcomes. Some factors are associated with all three types of outcome. Examples are breast

feeding and child nutrition, discipline and maternal locus of control. Some factors are strongly associated with some outcomes but not others. Maternal

social networks and parental smoking explain the non-cognitive and health gradients but not the cognitive.

Some factors have opposing associations with certain aspects of development. Lack of car ownership and poor housing are associated with lower risk of obesity; lower attendance at center-based child care is associated with fewer behavioral problems

Data: The ALSPAC cohort

9476 children born in Bristol and the surrounding regions (Avon) in 1991/2

Population 1 million, mixture of rural, suburban and urban, broadly nationally representative

Census of pregnant mothers rather than random sample, very high frequency

Mother-completed postal questionnaires Teacher-completed postal questionnaires Hands-on assessments in clinics at ages 7, 8 and 9 Matched to Key Stage national school test results from the National Pupil Database

Outcome measuresCognitive

IQ at age 8. WISC-III UK. Academic achievement at age 7. Key Stage 1 national school tests (KS1). Reading, writing and mathematics.

Non-cognitiveLocus of control at age 8. Child completed. Short form of Nowicki-Strickland Internal-External scale for children.Self esteem at age 8. Child completed. Short form of Harter’s Self Perception Profile for Children.Behavioral problems at age 7. Teacher rated. Strengths and Difficulties Questionnaire. Hyperactivity, peer relations, conduct problems, emotional problems.

HealthFat mass at age 9. Total body dual energy X-ray absorptiometry (DXA scans).

The income gradient

Oij = δYi + eij eij ┴ Yi

Income(Y)

ChildOutcome j

(Oj)

δ

Oij is the jth outcome of the ith childYi is the log of average disposable equivalised household income at child age 3 and 4 in 1995 priceseij is an orthogonal error termAll outcomes standardised to mean 100, SD 10.In presentation, the signs of the coefficients are adjusted, such that positive coefficients represent more beneficial outcomes in all cases.

Income gradients in outcomes in middle childhood

88

90

92

94

96

98

100

102

104

106

108

30 80 130 180 230 280 330 380 430 480 530 580 630

Equivalised disposable weekly household income age 3/4 (1995 GBP)

Sco

re (

mea

n 1

00,

SD

10)

IQ (5.85)

KS1 (5.46)

Locus of control (3.30)

Self esteem (1.71)

Behavior (2.01)

Fat mass (1.34)

p < 0.01 for all gradients

Control variablesHousehold demographics: Single parenthood, siblings, mother’s ageLabor market status: Mother’s and partner’s employment and occupational classEducation: Mother’s, partner’s and maternal grandparents’ qualificationsNeighborhood: Local deprivation (IMD for ward at birth), social housing

Mediating variablesMaternal psychosocial functioning: Anxiety/depression, weighted life events, financial difficulties, parental relationship, frequency of smacking, social networks, locus of controlPreschool childcare: Type and intensity, between birth and age 3, between age 3 and school entryHealth & health behaviours: Birth weight and gestation, parental smoking, breastfeeding, diet at age 3Home learning environment: Books and toys, maternal teaching, educational outings, mother’s and partner’s reading and singing with childPhysical home environment: Car ownership, garden, noise, crowding, damp/mouldSchool quality and mix: Fixed effects

Decomposing the income gradient

Oij = δYi + eij eij ┴ Yi

Income(Y)

ChildOutcome j

(Oj)

δ

Modelling framework

Income(Y)

Controls(C)

e.g. parental education,

family structure

Mediators(M)

e.g. home learning

environment, diet

ChildOutcome j

(Oj)γj

θj

πj

β

λα

(1) Oij = γjMi + θjCi + πjYi + μij μij ┴ Mi, Ci, Yi

Modelling framework

Income(Y)

Controls(C)

e.g. parental education,

family structure

Mediators(M)

e.g. home learning

environment, diet

ChildOutcome j

(Oj)γj

θj

πj

β

λα

(2) Mi = βCi + λYi + ηi ηi ┴ Ci, Yi

Modelling framework

Income(Y)

Controls(C)

e.g. parental education,

family structure

Mediators(M)

e.g. home learning

environment, diet

ChildOutcome j

(Oj)γj

θj

πj

β

λα

(3) Ci = αYi + νi νi ┴ Yi

Modelling framework

Income(Y)

Controls(C)

e.g. parental education,

family structure

Mediators(M)

e.g. home learning

environment, diet

ChildOutcome j

(Oj)γj

θj

πj

β

λα

(4) Oij = (γjβα + γjλ + θjα + πj) Yi + error

= δjYi + eij

Modelling framework

δj = γjβα + γjλ + θjα + πj

The unconditional income gradient can be written as the sum of a set of path coefficients.

Each path coefficient is the product of the partial effects of one variable on another. If any link in the chain is zero, the path coefficient will be zero.

Path coefficients calculated by multiplying and summing the OLS coefficients from the underlying regressions. Standard errors estimated by bootstrapping with 200 repetitions.

Path coefficients can be combined in different ways to give alternative decompositions of the income gradient.

Income gradient decompositions - summary

Cognitive Non-cognitive Health

Path coefficient IQ KS1 Locus control

Self esteem

Behav- ior

Fat mass

A. Total mediators (sum i to vi) 1.87** 1.86** 1.78** 0.78* 1.27* 0.79* i. Maternal psychosocial functioning 0.66** 0.56** 0.48* 0.68** 1.23** 0.50** ii. Pre-school child care 0.16 0.07 0.44** -0.16 -0.28 0.20 iii. Health & health behaviors 0.38** 0.59** 0.43** 0.03 0.66** 0.72** iv. Home learning environment 0.54** 0.60** 0.33* 0.37** 0.21 -0.26* v. Physical home environment 0.05 0.10 -0.01 -0.26 -0.31 -0.40* vi. School fixed effects 0.08 -0.07 0.10 0.11 -0.23 0.05 B. Total unmediated controls (sum vii to x) 3.08** 2.84** 1.43** 0.62* 0.67* 0.66* vii. Household demographics 0.29† 0.12 0.25 0.06 -0.09 -0.17 viii. Labor market status 0.17 0.51** 0.17 -0.01 0.06 -0.03 ix. Education 2.10** 1.44** 0.94** 0.56** 0.48† 0.54** x. Neighborhood 0.54** 0.79** 0.07 0.00 0.23 0.32† C. Total unmediated income 0.90* 0.76** 0.09 0.31 0.07 -0.11 Unconditional income gradient (A+B+C) 5.85** 5.46** 3.30** 1.71** 2.01** 1.34** Observations 5708 8727 5390 5857 3294 6113

Income gradient decompositions - summary

Cognitive Non-cognitive Health

Path coefficient IQ KS1 Locus control

Self esteem

Behav- ior

Fat mass

A. Total mediators (sum i to vi) 1.87** 1.86** 1.78** 0.78* 1.27* 0.79* i. Maternal psychosocial functioning 0.66** 0.56** 0.48* 0.68** 1.23** 0.50** ii. Pre-school child care 0.16 0.07 0.44** -0.16 -0.28 0.20 iii. Health & health behaviors 0.38** 0.59** 0.43** 0.03 0.66** 0.72** iv. Home learning environment 0.54** 0.60** 0.33* 0.37** 0.21 -0.26* v. Physical home environment 0.05 0.10 -0.01 -0.26 -0.31 -0.40* vi. School fixed effects 0.08 -0.07 0.10 0.11 -0.23 0.05 B. Total unmediated controls (sum vii to x) 3.08** 2.84** 1.43** 0.62* 0.67* 0.66* vii. Household demographics 0.29† 0.12 0.25 0.06 -0.09 -0.17 viii. Labor market status 0.17 0.51** 0.17 -0.01 0.06 -0.03 ix. Education 2.10** 1.44** 0.94** 0.56** 0.48† 0.54** x. Neighborhood 0.54** 0.79** 0.07 0.00 0.23 0.32† C. Total unmediated income 0.90* 0.76** 0.09 0.31 0.07 -0.11 Unconditional income gradient (A+B+C) 5.85** 5.46** 3.30** 1.71** 2.01** 1.34** Observations 5708 8727 5390 5857 3294 6113

Income gradient decompositions - summary

Cognitive Non-cognitive Health

Path coefficient IQ KS1 Locus control

Self esteem

Behav- ior

Fat mass

A. Total mediators (sum i to vi) 1.87** 1.86** 1.78** 0.78* 1.27* 0.79* i. Maternal psychosocial functioning 0.66** 0.56** 0.48* 0.68** 1.23** 0.50** ii. Pre-school child care 0.16 0.07 0.44** -0.16 -0.28 0.20 iii. Health & health behaviors 0.38** 0.59** 0.43** 0.03 0.66** 0.72** iv. Home learning environment 0.54** 0.60** 0.33* 0.37** 0.21 -0.26* v. Physical home environment 0.05 0.10 -0.01 -0.26 -0.31 -0.40* vi. School fixed effects 0.08 -0.07 0.10 0.11 -0.23 0.05 B. Total unmediated controls (sum vii to x) 3.08** 2.84** 1.43** 0.62* 0.67* 0.66* vii. Household demographics 0.29† 0.12 0.25 0.06 -0.09 -0.17 viii. Labor market status 0.17 0.51** 0.17 -0.01 0.06 -0.03 ix. Education 2.10** 1.44** 0.94** 0.56** 0.48† 0.54** x. Neighborhood 0.54** 0.79** 0.07 0.00 0.23 0.32† C. Total unmediated income 0.90* 0.76** 0.09 0.31 0.07 -0.11 Unconditional income gradient (A+B+C) 5.85** 5.46** 3.30** 1.71** 2.01** 1.34** Observations 5708 8727 5390 5857 3294 6113

Income gradient decompositions - summary

Cognitive Non-cognitive Health

Path coefficient IQ KS1 Locus control

Self esteem

Behav- ior

Fat mass

A. Total mediators (sum i to vi) 1.87** 1.86** 1.78** 0.78* 1.27* 0.79* i. Maternal psychosocial functioning 0.66** 0.56** 0.48* 0.68** 1.23** 0.50** ii. Pre-school child care 0.16 0.07 0.44** -0.16 -0.28 0.20 iii. Health & health behaviors 0.38** 0.59** 0.43** 0.03 0.66** 0.72** iv. Home learning environment 0.54** 0.60** 0.33* 0.37** 0.21 -0.26* v. Physical home environment 0.05 0.10 -0.01 -0.26 -0.31 -0.40* vi. School fixed effects 0.08 -0.07 0.10 0.11 -0.23 0.05 B. Total unmediated controls (sum vii to x) 3.08** 2.84** 1.43** 0.62* 0.67* 0.66* vii. Household demographics 0.29† 0.12 0.25 0.06 -0.09 -0.17 viii. Labor market status 0.17 0.51** 0.17 -0.01 0.06 -0.03 ix. Education 2.10** 1.44** 0.94** 0.56** 0.48† 0.54** x. Neighborhood 0.54** 0.79** 0.07 0.00 0.23 0.32† C. Total unmediated income 0.90* 0.76** 0.09 0.31 0.07 -0.11 Unconditional income gradient (A+B+C) 5.85** 5.46** 3.30** 1.71** 2.01** 1.34** Observations 5708 8727 5390 5857 3294 6113

Income gradient decompositions – disaggregated mediators

Cognitive Non-cognitive Health

Path coefficient IQ KS1 Locus control

Self esteem

Behav- ior

Fat mass

i. Maternal psychosocial functioning 0.66** 0.56** 0.48* 0.68** 1.23** 0.50** Of which: Maternal anxiety/depression 0.05 0.06† 0.12† 0.14* 0.12† -0.07 Life event shocks -0.06 -0.02 0.00 0.05 0.22** 0.09† Subjective financial distress 0.14 0.13 -0.05 0.18 0.02 0.13 Quality of parental relationship -0.03 -0.14** -0.09 -0.06 -0.09 0.01 Frequency of smacking at 3 0.11** 0.05* 0.11** 0.18** 0.13* 0.00 Maternal social networks 0.08 0.07 0.06 0.08 0.40** 0.15† Maternal locus of control 0.35** 0.40** 0.34** 0.11 0.43** 0.19† Unconditional income gradient 5.85** 5.46** 3.30** 1.71** 2.01** 1.34** Observations 5708 8727 5390 5857 3294 6113

Income gradient decompositions – disaggregated mediators

Cognitive Non-cognitive Health

Path coefficient IQ KS1 Locus control

Self esteem

Behav- ior

Fat mass

iii. Health & health behaviors 0.38** 0.59** 0.43** 0.03 0.66** 0.72** Of which: Health at birth 0.04† 0.04* 0.01 0.01 0.00 -0.03 Smoking -0.17* 0.01 0.07 0.01 0.39** 0.29** Breast feeding 0.21** 0.14** 0.17* -0.01 0.12 0.17** Eating patterns at 3 0.30** 0.39** 0.17 0.02 0.14 0.28* Unconditional income gradient 5.85** 5.46** 3.30** 1.71** 2.01** 1.34** Observations 5708 8727 5390 5857 3294 6113

Income gradient decompositions – disaggregated mediators

Cognitive Non-cognitive Health

Path coefficient IQ KS1 Locus control

Self esteem

Behav- ior

Fat mass

iv. Home learning environment 0.54** 0.60** 0.33* 0.37** 0.21 -0.26* v. Physical home environment 0.05 0.10 -0.01 -0.26 -0.31 -0.40* Of which: Car ownership -0.08 0.06 -0.04 -0.11 0.09 -0.29** Has garden 0.02 -0.05 0.00 -0.07 -0.07 0.02 Noise 0.00 0.02 0.02 0.02 0.08 -0.02 Crowding 0.14 0.08 0.14 -0.05 -0.23 0.02 Damp/condensation/mould -0.04 -0.01 -0.12† -0.05 -0.18† -0.14* Unconditional income gradient 5.85** 5.46** 3.30** 1.71** 2.01** 1.34** Observations 5708 8727 5390 5857 3294 6113

Income gradient decompositions – mediators of income and controls

Cognitive Non-cognitive Health

Path coefficient IQ KS1 Locus control

Self esteem

Behav- ior

Fat mass

A. Total mediated controls 1.40** 1.38** 1.25** 0.14 0.53 0.69** B. Total mediated income 0.47* 0.48** 0.53* 0.63** 0.73† 0.09 C. Total mediators (A+B) 1.87** 1.86** 1.78** 0.78* 1.27* 0.79*

D. Total unmediated controls 3.08** 2.84** 1.43** 0.62* 0.67* 0.66* E. Total unmediated income 0.90* 0.76** 0.09 0.31 0.07 -0.11 F. Total unmediated (D+E) 3.99** 3.61** 1.52** 0.93† 0.75 0.55

G. Total controls (A+D) 4.49** 4.23** 2.68** 0.77* 1.21** 1.36** H. Total income (B+E) 1.36** 1.23** 0.62 0.94* 0.80 -0.02 I. Unconditional income gradient (G+H) 5.85** 5.46** 3.30** 1.71** 2.01** 1.34** Observations 5708 8727 5390 5857 3294 6113

Income gradient decompositions – mediators of income and controls

Cognitive Non-cognitive Health

Path coefficient IQ KS1 Locus control

Self esteem

Behav- ior

Fat mass

A. Total mediated controls 1.40** 1.38** 1.25** 0.14 0.53 0.69** B. Total mediated income 0.47* 0.48** 0.53* 0.63** 0.73† 0.09 C. Total mediators (A+B) 1.87** 1.86** 1.78** 0.78* 1.27* 0.79*

D. Total unmediated controls 3.08** 2.84** 1.43** 0.62* 0.67* 0.66* E. Total unmediated income 0.90* 0.76** 0.09 0.31 0.07 -0.11 F. Total unmediated (D+E) 3.99** 3.61** 1.52** 0.93† 0.75 0.55

G. Total controls (A+D) 4.49** 4.23** 2.68** 0.77* 1.21** 1.36** H. Total income (B+E) 1.36** 1.23** 0.62 0.94* 0.80 -0.02 I. Unconditional income gradient (G+H) 5.85** 5.46** 3.30** 1.71** 2.01** 1.34** Observations 5708 8727 5390 5857 3294 6113

Income gradient decompositions – mediators of income and controls

Cognitive Non-cognitive Health

Path coefficient IQ KS1 Locus control

Self esteem

Behav- ior

Fat mass

Total income (conditional on controls) 1.36** 1.23** 0.62 0.94* 0.80 - Of which mediated by:

Maternal psychosocial functioning 0.24† 0.14 0.15 0.41** 0.77** 0.35* Pre-school childcare - - 0.21* - -0.22 - Health & health behaviors - 0.14** - 0.10* 0.13 - Home learning environment 0.13† 0.13* - 0.13† 0.11 -0.14* Physical home environment - - - -0.10 -0.11 -0.17* School fixed effects - - - 0.14 - -

Unmediated 0.90* 0.76** - 0.31 - -0.11 Unconditional income gradient 5.85** 5.46** 3.30** 1.71** 2.01** 1.34** Observations 5708 8727 5390 5857 3294 6113 Path coefficients <0.1 in magnitude and insignificant at the 10% level not shown.

Conclusions

Our accounting exercise produces a number of findings in line with previous research Income gradients are steeper for cognitive outcomes than for non-

cognitive or health outcomes The estimated effect of income drops steeply when other forms of socio-

economic disadvantage are controlled Mediators between income and outcomes are many and diffuse Less cognitive stimulation in the home helps to account for the cognitive

deficits of low income children; poorer maternal psychosocial functioning helps to account for their behavioral deficits

Conclusions

Our comparative approach provides new insights that may be missed in more narrowly-focused studies Maternal psychosocial functioning and health-related behaviors appear

as important as the home learning environment in accounting for the cognitive deficits of low income children

Some factors have a modest role to play in explaining multiple gradients (e.g. breast feeding, discipline)

A focus only on cognitive outcomes may miss the adverse consequences of certain factors for other dimensions of development (e.g. smoking, social networks)

Not everything that high income parents do is necessarily good for their children. Behaviors that promote cognitive development (learning-focused environments) could have adverse consequences for physical health