THE EDUCATION-HEALTH GRADIENT

Gabriella Conti, James Heckman, and Sergio Urzua

Author information Copyright and License information
See other articles in PMC that cite the published article.
Abstract
Much of the policy discussion about reducing health disparities across socioeconomic groups has focused
on improving health insurance coverage and access to health care. However, increasing attention is being
paid to the social determinants of health (Commission on Social Determinants of Health 2008). Several
scholars have noted that better health early in life is associated with higher educational attainment. More
educated individuals, in turn, have better health later in life and better labor market prospects (Janet
Currie 2009, David Cutler and Adriana Lleras-Muney 2010).
This paper summarizes our current research examining the early origins of health disparities by education
(Gabriella Conti, James J. Heckman & Sergio Urzua 2010, henceforth CHU). It contributes to a growing
literature that establishes a strong relationship between health and education and more generally between
early childhood conditions and adult outcomes. Gaps in both cognitive and noncognitive abilities of
children of different socioeconomic groups emerge at early ages (Flavio Cunha, Heckman, Lance J.
Lochner and Dimitriy V. Masterov 2006). So do gaps in health (Anne Case, Darren Lubotsky & Christina
Paxson 2002). Various studies suggest that it is possible to enrich adverse early environments and
promote child development.
CHU identify the causal effect of education on health and health-related behaviors. We determine the role
played by cognitive, noncognitive, and health endowments measured in the early years in explaining
numerous adult outcomes. Family background characteristics, and cognitive, noncognitive and health
endowments are all important determinants of health disparities at age 30. Our methodology allows us to
determine the fraction of health gaps by education that can be explained by selection into education on
early life endowments and the fraction that can be attributed to the causal effect of education. Not
accounting for personality traits overestimates the importance of cognitive ability in determining adult
health and participation in healthy behaviors. Selection into education on early life factors explains more
than half of the observed difference by educational level in poor health, depression and
obesity.
1
Education has an important causal effect in explaining differences in smoking rates, and
participation in many other health behaviors, as well as on a number of other outcomes. We find
significant gender differences in the effect of education on health. We go beyond the current literature
that typically estimates mean causal effects to compute distributions of treatment effects. We show how
the returns to education vary among individuals who are similar in their observed characteristics. Mean
effects hide gains and losses to treatment in the population. Our analysis highlights the important role
played by factors determined in the early years in promoting health.
A positive correlation between health and schooling is one of the most well-established findings in the
social sciences. Figure 1 shows mean educational differentials for a variety of outcomes in the British
Cohort Study (BCS70) data we analyze. The full length of the bar for each outcome is the raw educational
differential.
2
The extent to which this association reflects causality is still subject to much debate.
3
Three
explanations for this correlation are offered in the literature: that causality runs from schooling to health
(Grossman 2006), that it runs from health to schooling (Currie 2009), and that both are determined by a
third factor, such as time or risk preferences. Understanding the relative importance of each of these
mechanisms in generating observed differences in health by education is helpful in designing policies to
promote health.

Figure 1
Disparities in outcomes by education by gender in the British Cohort Study 1970 (outcomes measured at
age 30)
Our research joins the literatures in economics, epidemiology and psychology. We explore the
relationship between health and cognitive ability. We also explore the relationship between personality
traits and health. This research contributes to understanding the nonmarket returns to education.
Go to:
1 Models, Methods, and Data
CHU estimate a semiparametric structural model of the choice of schooling and the causal effect of
schooling on a variety of health outcomes and healthy behaviors. Agents may select into schooling based
on expected market and nonmarket returns. We have precise measures of a number of early childhood
environmental factors, X. In addition, we have proxies for a vector of latent capabilities for early life
cognition, personality, and health endowments, , that, in addition to the X, affect both the choice of
education and the adult outcomes studied. (SeePedro Carneiro, Karsten Hansen, and Heckman
2003 and Heckman, Jora Stixrud, and Urzua 2006 for descriptions of the methodology.) If we could
condition on observed characteristics X and unobserved characteristics , any remaining association
between education and adult outcomes would be causal.
While we cannot directly measure , we have a large number of proxies for the low-dimensional in our
data. These measures are taken early in life (at age 10). Using proxies for accounting for the
measurement error in the proxies and controlling for observables, X, we estimate causal effects of
education on adult health and healthy behaviors. Our method is a form of matching on both observables
and unobservables where the unobservables are proxied, and we account for the errors in the proxies for
the unobservables.
4
It can also be interpreted as a latent variable structural model with the key
unobservables measured up to error that is accounted for in the estimation procedure. We find substantial
evidence of measurement error.
5
We estimate the model two ways: using matching and using structural
methods. Both methods produce results that are in agreement for all parameters identified by both
methods. For the sake of brevity, we report only the results from the structural method.
6

The BCS70 data that we analyze is a survey of all babies born (alive or dead) in one week in April 1970.
We have panel records on schooling, family background, a variety of health and healthy behaviors, and
labor market outcomes. Schooling choices depend on expected market and nonmarket returns. In this
paper, schooling is a binary decision and refers to attendance beyond the compulsory school-leaving age.
We have multiple measures of cognitive and noncognitive skills and early health, taken at age 10. See the
Web Appendix for further details.
To avoid dependence of estimates on distributional assumptions, we use mixtures of multivariate normals
models to characterize the distributions of the latent capabilities. We can generate all treatment
effects.
7
We can also estimate the distribution of treatment effects following Carneiro, Hansen &
Heckman (2003) and Abbring & Heckman (2007). The estimated model passes tests of goodness of fit.
8

Go to:
2 Findings
We relate early measures of endowments to the adult outcomes measured at age 30 that are shown
in Figure 1. We find sorting of individuals by schooling in terms of cognitive, noncognitive, and early
health endowments. Early health endowments are weakly associated with schooling for women but not at
all for men. Thus, in our data, evidence on the link between early health and education emphasized
by Currie (2009) is at best weak. However, consistent with Currie (2009), we find that early health has a
statistically significant direct effect on adult outcomes. Noncognitive factors play a powerful role in
predicting participation in healthy behaviors. Introducing noncognitive factors substantially weakens the
predictive power of cognitive factors in promoting adult outcomes, but cognitive ability remains an
important determinant of education and labor market outcomes.
9

Education is estimated to have a strong causal effect on most outcomes examined. For each
outcome, Figure 1displays the fraction of the observed educational differential that can be attributed to the
causal effect of education the part of each bar labeled causal component.
10
Note the gender
differential. Education plays a much more important causal role for males than females in accounting for
gaps in obesity rates, exercise, and employment.
Our analysis moves beyond the traditional literature which only considers mean treatment effects and
estimates distributions of treatment effects. Knowledge of these distributions is fundamental in
uncovering what lies behind a zero estimated average treatment effect, and the proportion of individuals
who actually benefit from the treatment (education). We find substantial heterogeneity in treatment
responses. Consider the case of smoking for females.
11
The proportion of people who stop smoking is
much bigger than the proportion of people who start smoking, so the average treatment effect turns out to
be negative (see Figure 2).


Figure 2
Population distribution of the average treatment effect for females -health behaviors at age 30
Compare these results to the results for obesity for females. Underlying a statistically insignificant
average treatment effect of education on obesity there are gains and losses which balance each other out:
the same proportion of women (roughly 20 percent) lose and gain from the treatment. For males there is a
net negative effect.
Go to:
Treatment Effect Heterogeneity: the Role of Early Endowments
CHU study how the average treatment effect of education varies with the level of endowment of cognitive
and noncognitive skills, and early health. While there is a significant amount of heterogeneity in the effect
of education across outcomes by levels of endowments, some patterns emerge. On most outcomes for
males, the beneficial effect of education is much bigger at the bottom of the noncognitive ability
distribution and at the top of the cognitive ability distribution. See Figure 3, which plots the average
treatment effect of education on smoking for different quantiles of the cognitive, noncognitive, and health
endowments. The evidence on differential effects of education by level of cognitive skill is consistent
with the interpretation that the information content on the dangers of smoking provided by post-
compulsory education needs to be combined with the capacity to process that information in order for it to
be effective.

Figure 3
Treatment effect heterogeneity - Smoking (males) at age 30
Go to:
The Role of Cognitive and Noncognitive Ability
We find strong evidence that noncognitive traits promote health outcomes and healthy behaviors. If
noncognitive traits measured at age 10 are not included in the estimated model, early cognitive ability has
a strong and statistically significant effect for many outcomes. There is a smaller estimated effect of
cognitive traits in models where we analyze health endowments, cognition, and noncognitive traits
jointly. This comes as no surprise since the correlation between cognitive and noncognitive endowments
is 0.54.
12
Our analysis sounds a warning for research in the area of cognitive epidemiology that has not
given adequate attention to personality traits and focuses exclusively on the role played by intelligence
(see, e.g, Linda S. Gottfredson & Ian J. Deary 2004). If anything, noncognitive factors are relatively more
important determinants of health and healthy behaviors.
Go to:
Summary
The research reported in CHU examines the early origins of health disparities across education groups.
We determine the role played by early cognitive, noncognitive, and health endowments. We identify the
causal effect of education on health and health-related behaviors. We develop an empirical model of
schooling choice and post-schooling outcomes, where both schooling and the outcomes determined in
part by schooling are influenced by measured early family environments and latent capabilities (cognitive,
noncognitive and health). We show that family background characteristics, and cognitive, noncognitive,
and health endowments developed by age 10, are important determinants of labor market and health
disparities at age 30. Not properly accounting for personality traits overestimates the importance of
cognitive ability in determining adult health. Selection on factors determined early in life explains more
than half of the observed difference by education in poor health, depression, and obesity. Education has
an important causal effect in explaining differences in many adult outcomes and healthy behaviors. We
uncover significant gender differences. We go beyond the current literature which typically estimates
mean effects to compute distributions of treatment effects. We show how the health returns to education
can vary among individuals who are similar with respect to their observed characteristics, and how a
mean effect can hide gains and losses for different individuals. Our research highlights the important role
played by the early years in producing health.
Go to:
Acknowledgments
We thank the following funders for their support: The California Endowment, The Commonwealth
Foundation, The Nemours Foundation, the Buffett Early Childhood Fund, the J.B. and M.K. Pritzker
Foundation, NICHD (R01 HD054702) and an Anonymous funder. The opinions expressed in this paper
are those of the authors and do not necessarily reflect opinions of the funders.
Go to:
Footnotes
The website for this paper is http://jenni.uchicago.edu/gradient/.
1
Obesity is measured by BMI 30 for males and BMI 25 for females.
2
We discuss the decomposition into causal and selection components later.
3
See Michael Grossman (2008).
4
See Heckman, Susanne Schennach and Benjamin Williams 2010.
5
For further details, see CHU or our Web Appendix at http://jenni.uchicago.edu/gradient/.
6
The matching results are available in our Web Appendix.
7
Our identification strategy does not rely on identification at infinity or conventional exclusion
restrictions. Like matching, it relies on conditional independence assumptions. Unlike conventional
matching, we allow the conditioning variables to be measured with error. See the discussion in Carneiro,
Hansen & Heckman (2003), Jaap H. Abbring & Heckman (2007), Heckman, Stixrud & Urzua (2006),
or Heckman, Schennach & Williams (2010). In fact, the support of the estimated probability of schooling
is essentially the full unit interval so that identification of the model is over the full support of the
unobservables in the choice equation, so that an identification-at-infinity identification strategy would be
valid in our case.
8
See CHU and the Web Appendix for details.
9
See the Web Appendix for the full results.
10
These are Average Treatment Effects.
11
Results are the same for males.
12
We display the joint distributions of the endowments in the Web Appendix.
Go to:
Contributor Information
Gabriella Conti, Department of Economics, University of Chicago, 1126 East 59th Street, Chicago, IL
60637; phone, 773-702-7052; fax, 773-702-8490.
James Heckman, Department of Economics, University of Chicago, 1126 East 59th Street, Chicago, IL
60637; phone, 773-702-0634; fax, 773-702-8490.
Sergio Urzua, Department of Economics and Institute for Policy Research, Northwestern University,
Handerson Hall, 2001 Sheridan Road, Evanston, IL 60208; Phone, 847-491-8213.
Go to:
References
1. Abbring Jaap H, Heckman James J. Econometric Evaluation of Social Programs, Part III:
Distributional Treatment Effects, Dynamic Treatment Effects, Dynamic Discrete Choice, and
General Equilibrium Policy Evaluation. In: Heckman J, Leamer E, editors. Handbook of
Econometrics. 6B. Amsterdam: Elsevier; 2007. pp. 51455303.
2. Carneiro Pedro, Karsten Hansen, Heckman James J. Estimating Distributions of Treatment
Effects with an Application to the Returns to Schooling and Measurement of the Effects of
Uncertainty on College Choice. International Economic Review. 2003;44(2):361422.
3. Case Anne, Lubotsky Darren, Paxson Christina. Economic Status and Health in Childhood: The
Origins of the Gradient. American Economic Review. 2002;92(5):13081334.
4. Commission on Social Determinants of Health. Closing the Gap in a Generation: Health Equity
through Action on the Social Determinants of Health. World Health Organization Final Report;
Geneva: 2008.
5. Conti Gabriella, Heckman James J, Urzua Sergio. The Early Origins of the Education-Health
Gradient Under revision. Perspectives on Psychological Science 2010
6. Cunha Flavio, Heckman James J, Lochner Lance J, Masterov Dimitriy V. Interpreting the
Evidence on Life Cycle Skill Formation. In: Hanushek Eric A, Welch Frank., editors. Handbook
of the Economics of Education. Chapter 12. Amsterdam: North-Holland; 2006. pp. 697812.
7. Currie Janet. Healthy, Wealthy, and Wise: Socioeconomic Status, Poor Health in Childhood, and
Human Capital Development. Journal of Economic Literature. 2009;47(1):87122.
8. Cutler David M, Lleras-Muney Adriana. Understanding Differences in Health Behaviors by
Education.Journal of Health Economics. 2010;29(1):128. [PMC free article] [PubMed]
9. Gottfredson Linda S, Deary Ian J. Intelligence Predicts Health and Longevity, but Why? Current
Directions in Psychological Science. 2004;13(1):14.
10. Grossman Michael. Eric Hanushek and Finis Welch. Handbook of the Economics of
Education. Vol. 1. Chapter 10. Amsterdam: Elsevier; 2006. Education and Nonmarket Outcomes;
pp. 577633.
11. Grossman Michael. The Relationship between Health and Schooling: Presidential
Address. Eastern Economic Journal. 2008;34(3):281292.
12. Heckman James J, Stixrud Jora, Urzua Sergio. The Effects of Cognitive and Noncognitive
Abilities on Labor Market Outcomes and Social Behavior. Journal of Labor
Economics. 2006;24(3):411482.
13. Heckman James J, Schennach Susanne, Williams Benjamin. Unpublished manuscript. University
of Chicago, Department of Economics; 2010. Matching with Error-Laden Covariates.