1) Adolescent cognitive ability and non-cognitive traits are associated with better adult physical and mental health and fewer depressive symptoms.
2) Adding family background factors attenuates but does not eliminate the associations, suggesting both shared family environments and individual attributes matter.
3) Further adding education and health behaviors further reduces the associations, indicating cognitive/non-cognitive traits may influence health through these mediating factors.
1. Adolescent Cognitive and Non-cognitive
Correlates of Adult Health
Robert Kaestner
Institute of Government and Public Affairs
University of Illinois
Department of Economics
University of Illinois at Chicago
Presentation
XXX Jornadas de Economia de la Salud
Asociacion de Economia de la Salud
Valencia June 25, 2010
2. Why study adolescent correlates of adult health?
Childhood (family) environments matter empirically.
Mazumder (2008) and others reported that approximately
50% of the variation in wages, earnings and household
income is due to differences between family (childhood)
environments.
Studies of health and longevity found significant sibling
correlations in life expectancy, incidence of heart disease,
and mental illness (Stunkard et al. 1986; Marenberg et al.
1994; Christensen and Vaupel 1996; Kiessepa et al. 2004;
vB Hjelmborg et al. 2006; Petersen et al. 2008).
Christensen and Vaupel (1996) reported that
approximately 25% of variation in life expectancy is
attributable to shared childhood environments.
3. Families Matter in My Sample
Sibling Correlations for Socioeconomic Status and Health
754 Same-sex Sibling Pairs in NLSY79
Coefficient of Correlation
Outcome Variation Coefficient P-value
SF-12 Physical Score 0.17 0.07 0.05
SF-12 Mental Score 0.16 0.09 0.02
Self-reported Health 0.43 0.19 <0.01
Self-reported Good Health 0.13 <0.01
Self-reported Poor Health 0.15 <0.01
CESD Score 1.16 0.11 <0.01
Height (1985) 0.05 0.71 <0.01
Daily Smoker (1998) 0.27 <0.01
Binge Drinker Past Month 0.10 <0.01
AFQT Percentile Score 0.72 0.68 <0.01
Annual Earnings 0.98 0.33 <0.01
4. Why study adolescent correlates of adult health?
Not yet known why childhood (family) matters?
Families provide resources (e.g., medical care) and invest in cognitive and
non-cognitive abilities of children that influence adult well being.
It is not clear what is the causal mechanism that links family (childhood) to
adult outcomes.
It could be:
Shared genetic factors
Shared socioeconomic family environment (e.g., family income,
parental education, family structure, number of siblings)
Shared community environment (e.g., quality of child care, quality of
elementary and secondary schooling. quality of public health
infrastructure)
5. Why study adolescent correlates of adult health?
Relatively little research focused on health.
Most research in this area has focused on adult social and economic
outcomes.
Relatively little research on the influence of childhood environment on
adult health
Prior research examining adult health focused on the childhood health
environment
Barker hypothesis (in utero determinants); studies of famines and
disease during prenatal period
Studies of effects of low-birth weight
Fogel’s work on nutrition, height and health
Preston’s work on urban/rural environment
6. Why study adolescent correlates of adult health?
The special importance of childhood.
Different forms of human capital, or what Heckman refers to as capabilities, are
complementary (Becker 2007; Cunha and Heckman 2007; Heckman 2007).
Investments in one form of human capital, for example a non-cognitive factor such as
rate of time preference, cause further investments in other forms of human capital (or
capabilities) that also improve adult outcomes.
Heckman and colleagues expand model of complementary investments in human capital by
incorporating a developmental aspect that recognizes that the timing of investments is also
important.
Temporal investments in human capital are also complementary. Investments in
cognitive ability during late adolescence (e.g., high school) are more productive (less
costly) when earlier investments in cognitive ability have been made.
In sum, childhood, and particularly early childhood, investments in human capital may be
extremely important in determining adult outcomes.
7. Potentially High Rates of Return on Investments
During Childhood
Rate of Return to Investments
Age
Heckman (2008)
8. Purpose and Contributions of Current Research
Add to the relatively small literature studying the early determinants of adult
health by providing a descriptive analysis of the relationship between adult health
and factors measured at end of childhood (ages 14 to 15)
Include cognitive, non-cognitive (e.g., self esteem), and family background
factors in the analysis—cognitive and non-cognitive factors have generally been
ignored.
Influence of non-cognitive factors on socioeconomic outcomes has become
an important research area (Heckman et al. 2006)
Differences in human capital accumulation and adult outcomes are too large
to be explained by differences in monetary costs
Non-cognitive abilities may significantly affect the non-monetary costs of
investment and may therefore provide an explanation for the lack of
investment
9. Summary of Previous Literature
Auld and Sidhu (2005)
Large effects of cognitive ability on health limitations
Family background had little effect on whether a person had a health limitation.
Elias (2005)
Small effects of cognitive ability on self-rated health
Family background had little effect on self-rated health
Cutler and Lleras-Muney (2007)
Found that cognitive ability and family background were significantly related to
health behaviors such as smoking
Influence of family background was larger than cognitive ability.
Hartog and Osterbeek (1998)
Reported that family background and cognitive ability are associated with adult health
50% of the effect of cognitive ability on health and nearly all of the effect of family
background on health work through completed schooling.
10. Specific Contributions of this Study
I focus on attributes at age 14 and 15 (end of childhood) and relate these
to adult health 25 years later at age 41
I focus on the direct effects of cognitive and non-cognitive ability—not
how the effect of education is mediated by addition of cognitive ability.
I consider an extensive set of factors.
I assess how much of the effect of cognitive ability and non-cognitive
ability can be explained by family background—the family environment
is the precursor to cognitive and non-cognitive ability.
I assess whether the associations between adolescent factors and adult
health are mediated by completed schooling and health behaviors—two
factors known to affect health
11. Illustrative Empirical Model
Cunha and Heckman (2007) and Heckman (2007)
Hit = g (α0 H ,α0C ,α0 NC , I H1,..., I H (t −1), IC1,..., IC (t −1), I NC1,..., I NC(t −1) , f H , fC , f NC )
i = 1,...,N (persons)
t = 0,...,t (age)
Model assumes three distinct forms of human capital (H, C, and NC)
Human capital accumulation at age t depends on initial endowments of human
capital
Human capital accumulation at age t depends on history of investments in different
forms of human capital
Production technology should incorporate complementarities between investments at
different ages and between forms of human capital at same age
12. My Ad-hoc Approach
Intended as a Descriptive Analysis
H i = β 0 + ∑ δ k DEMOGik + ∑ λk COG _ 14ik + ∑ γ k NONCOG _ 14ik
k k k
+ ∑ δ k FAMILYik + ei
k
A comparison of this approach with correct approach reveals that this approach omits important
determinants of adult health and fails to incorporate any of the complementarities described by
Becker (2007) and Heckman (2007).
Omissions are likely to result in estimates of the associations between cognitive and non-cognitive
attributes at age 14, and adult health that are too large (from a structural point of view).
The likely upward bias (in terms of their interpretation as structural estimates) of the estimated
associations is important information for determinants that do not have statistically significant
associations because failing to reject the null hypothesis in this case is relatively strong evidence
that these factors and earlier investments in these factors, are not likely to be important determinants
of adult health
13. Causal Mechanisms Linking Non-cognitive
Attributes and Health
Time preference—here measured by use of tobacco, alcohol and drugs by age 14
Ability to appreciate future increases likelihood of investment in health
Locus of control
Those who have an internal locus of control may be more likely to seek and
appreciate health information
Self Esteem
Self esteem may affect health by allowing a person to communicate better
with his or her doctor about symptoms, diagnoses and treatment regimes.
In sum, factors that affect a person’s appreciation of future benefits, ability to
communicate (e.g., agreeableness) and motivation to take action and follow
through (conscientiousness) may all affect health.
These factors are distinct from cognitive factors
14. Data
NLSY79
Health measured at around age 41
Johnson and Schoeni (2003) reported sibling correlations
(PSID) in health (0.6—very large compared to anything
reported elsewhere) that remain constant from age 25 to 55
Other estimates of sibling correlations for health at older ages
such as life expectancy are around 0.25
Correlations reported here are somewhat lower—0.1 to 0.2
Using health at age 41 may be informative for older ages when
health begins to deteriorate
Adolescent characteristics measured at ages 14 to 15
15. Data
Health
Short Form-12 (SF12) mental and physical health
Center for Epidemiological Studies Depression Scale (CES-D).
Self-rated general health: good health defined as self-rated health that is
excellent or good, and poor health defined as self-rated health that is poor
or fair.
Cognitive ability at age 14 or 15
Armed Forces Qualification Test (AFQT) percentile taken at age 14 or 15
adjusted for differences in age at time of test.
used the sample distribution of adjusted percentile scores to classify
people into quartiles of cognitive ability
16. Data
Non-cognitive traits at age 14 or 15
Rosenberg self-esteem scale (measured in 1980 at ages 15 and 16);
divided into low-, moderate- or high-self esteem
Rotter locus of control scale; divided into low- moderate or –high-
external locus of control
church attendance (never/rarely, sometimes, often)
history of stealing (never, sometimes, often)
use of tobacco, alcohol and marijuana by age 14
17. Data
Family Background
mother’s education (<9, 9-11 years, 12 years, 13 to 15 years, 16 or more
years, missing)
number of siblings (none, one, two to three, four or more)
family structure (two biological parents, two parents, mother only, other)
1978 family income (0-4,999, 5-9,9999, 10-19,999, 20-29,999, 30,000 or
more, missing)
whether childhood household had library card, magazines or newspapers
whether influential person would approve of not going to college
18. Data
Years of Completed Education (<12 inc GED, 12, 13-15, 16+)
Health Behaviors
daily smoker in 1998 (last year available)
binged drank in past month in 2002 (last year available)
obese (self-reported BMI>30)
engaged in any vigorous physical activity recently
Initial Health
age 14 or 15 health limitation
height and height squared
father deceased by age 40
Demographics
age (measured in six-month intervals)
race/ethnicity (non-Hispanic Black, non-Hispanic While, Hispanic, other)
respondent and mother’s natality (foreign-born)
whether foreign-language was spoken in the home
19. Sample Means of Outcomes
Females Males
Mean Std.Dev. Mean Std.Dev.
SF-12 Physical 51.2 8.8 52.8 6.7
SF-12 Mental 52.1 8.5 54.3 7.0
CESD Score 3.7 4.3 2.7 3.6
Good Health 0.56 0.61
Poor Health 0.15 0.12
20. Sample Means Selected Variables
Females Males
Mean Dev. Mean Dev.
Locus of Control Score (1979) 9.4 2.0 9.3 2.1
Mother-Father Most Influential 0.69 0.71
Approve Not Going College 0.22 0.26
Likely go on Food Stamps 0.47 0.42
Attend Church Sometime 0.21 0.24
Attend Church Often 0.49 0.40
Stole Sometimes 0.20 0.23
Stole Often 0.11 0.23
21. Sample Means Selected Variables
Females Males
Mean Dev. Mean Dev.
Two Biological Parents 0.62 0.62
Two Parents 0.08 0.08
Mother Only 0.27 0.23
Number of Sibling 2.2 1.6 2.4 1.8
Mother’s Education 9-11 0.27 0.22
Mother’s Education 12 0.35 0.39
Mother’s Education 13-15 0.08 0.11
Mother’s Education 16+ 0.08 0.08
Family Income in 1978 15958 12164 15497 11990
Daily Smoker 0.27 0.29
Binge Drinker 0.11 0.26
Obese 0.30 0.29
Engaged Vigorous Activity 0.68 0.81
22. Regression Sequence
First estimate a model including only cognitive and non-cognitive factors.
Add family background
Include what I refer to as initial health or health determined by family under
assumption that cognitive and non-cognitive factors do not influence health
at this age
Family background is pre-cursor to (origin of) cognitive and non-cognitive
factors—evidence to assess whether it is all family or whether there is scope
for intervention (raise cognitive ability)
Add completed years of schooling and health behaviors
Evidence that effects of cognitive and non-cognitive factors are working
through education and health behaviors
27. Conclusions
Cognitive ability and one non-cognitive trait—self esteem—
have significant, direct associations with adult health.
For males, direct (net of other factors) associations are of
same relative magnitude for cognitive ability and self
esteem.
For females, direct associations are larger (relatively) for
self esteem than cognitive ability, and cognitive ability has
small direct associations with health.
28. Conclusions
Completed education continues to be significantly
associated with adult health net of adolescent and family
background characteristics.
Among males, there was more evidence that obtaining an
educational threshold was associated with adult health than
there was evidence of an education gradient in health.
For females, there was consistent evidence of an education
gradient for health and associations are of the same order of
magnitude as the (direct) associations of cognitive ability
and self esteem.
29. Conclusions
Finally, we assessed whether adolescent cognitive and non-
cognitive factors are potential explanations of gender and
racial disparities in health.
Overall, we found little evidence that these factors can
explain much of the differences in health we observe
between men and women, and black and white persons.