the relationship between health and …f - statistic 2,814.71 1,078.82 a each r egression contains...
TRANSCRIPT
THE RELATIONSHIP BETWEEN
HEALTH AND SCHOOLING: WHAT’S
NEW?
Michael Grossman
City University of New York Graduate Center and
National Bureau of Economic Research
Fourth Congress of the Colombian Health Economics
Association
Cali, Colombia
February 18-20, 2015
1
INTRODUCTION
Two most fundamental relationships in health economics
• Relationship between health insurance and medical care utilization
• Relationship between health and schooling
• Both generated by causality in both directions and by omitted “third variables”
• My keynote address: research on the second of these relationships that has
appeared in the past five years (2010-2014)
• Focus on empirical studies but will call your attention to a few theoretical
developments
• Will perhaps raise more questions than answers
2
WHY IS THE RELATIONSHIP BETWEEN
HEALTH AND SCHOOLING SO IMPORTANT?
• Demand for health model emphasizes that medical
care is only one of many determinants of health,
natural to explore others
• Demand for health model views health as a form of
human capital, natural to allow for and explore
complementarities between health capital and other
forms of human capital, the most important of which
is knowledge capital (schooling)
3
EMPIRICAL MOTIVATION
• “The one social factor that researchers agree is consistently linked to longer lives in every country where it has been studied is education. It is more important than race; it obliterates any effects of income.” Gina Kolata, “A Surprising Secret to Long Life: Stay in School,” New York Times, January 3, 2007
• “With the exception of black males, all recent gains in life expectancy at age twenty-five have occurred among better educated groups, raising educational differentials in life expectancy by 30 percent.” Meara, Richards, and Cutler, Health Affairs, March/April 2008
• Link does not necessarily imply causality from more schooling to better health
• Health may cause schooling or omitted “third variables” may cause health and schooling to vary in same direction
4
Table 1
Infant Mortality Rate, Age-Adjusted Mortality Rate, and Educational Attainment, United States, Selected Years, 1910-2000
Year
Infant Mortality Rate
(Deaths per 1,000 live births)
Age-Adjusted Mortality Rate (Deaths per
100,000 population based on year 2000 standard
population)
College Graduates (Percentage of persons aged 25 and older who
completed four years of college or more)
1910 131.8 2,317.2 2.7 1920 92.2 2,147.1 3.3 1930 69.0 1,943.8 3.9 1940 54.9 1,785.0 4.6 1950 33.0 1,446.0 6.2 1960 27.0 1,339.2 7.7 1970 21.4 1,222.6 11.0 1980 12.9 1,039.1 17.0 1990 9.7 938.7 20.3 2000 7.4 869.0 25.6
5
6
Table 2
Infant Mortality Rate and Educational Attainment,
United States versus Colombia
Year
Infant Mortality Rate
U.S.
Infant Mortality
Rate
Colombia
College
Graduates
U.S
College
Graduates
Colombia
1960 27.0 88.9 7.7 1.0
1970 21.4 69.7 11.0 1.8
1980 12.9 44.6 17.0 3.6
1990 9.7 29.0 20.3 7.3
2000 7.4 21.2 25.6 9.4
Source for Colombia data: http://databank.worldbank.org/data/databases.aspx. For Colombia,
college graduates are the percentage of persons 25 years of age and older with tertiary schooling
completed—includes universities as well as institutions that teach specific capacities of higher
learning such as colleges, technical training institutes, community colleges, nursing schools,
research laboratories, centers of excellence, and distance learning centers.
Table 3
Infant and Age-Adjusted Mortality Regressions, U.S.a
Infant Mortality Rate
Age-Adjusted Mortality Rate
Percentage with four years of college or more
-1.617 (-5.06)
-28.950 (-3.96)
R2 0.996 0.990
F-statistic 2,814.71 1,078.82
aEach regression contains an intercept and a cubic time trend. t-statistics are given in parentheses.
7
CAUSALITY FROM HEALTH TO
SCHOOLING
• Students in poor health miss more days of school due to
illness and learn less while in school
• Result: Negative effect on school achievement and years of
formal schooling completed
• Long-lasting effect if past health an input into current health
• Reduction in mortality increases number of periods over which
returns from investments in knowledge can be collected
8
CAUSALITY FROM SCHOOLING TO
HEALTH
• Productive efficiency: more educated obtain more
health output from given amounts of medical care
and other inputs
• Allocative efficiency: more educated pick a different
input mix to produce a certain commodity than less
educated; mix gives them more output of that
commodity than the mix selected by the less
educated
9
OMITTED THIRD VARIABLES
• Fuchs (1982) time preference hypothesis
• Persons who are more future oriented attend school
for longer periods of time and are more likely to
make investments in their own health and in their
children’s health
10
THEORETICAL DEVELOPMENTS
11
iHEA SESSION: REASSESSING
GROSSMAN
• Audrey Laporte: “Should the Grossman Model Retain Its Iconic
Status in Health Economics?”
• Yes, but in context of pure consumption model with savings
ruled out
12
T. GALAMA AND H. VAN KIPPERSLUIS, “A
THEORY OF EDUCATION AND HEALTH”
• “Currently, we still lack comprehensive theoretical models in which the
stocks of health and knowledge are determined simultaneously…. The rich
empirical literature treating interactions between schooling and health
underscores the potential payoff to this undertaking.” (Old Discredited
Economist, 2000, quoted by Titus and Hans, page 3 of their paper)
• How can we go beyond the following simple structural model based on
their paper?
H = H(E, Prices of Health Inputs, Assets, Time Preference)
E = E(H, Prices of Schooling Inputs, Assets, Time Preference)
• GV do comparative dynamics by assuming stock of HC is sum of
knowledge and health stocks and no sick time
13
JAMES HECKMAN AND COLLEAGUES, SKILL
PRODUCTION (Cuhna, Heckman, and
Schennach 2010)
• Conceptualization of technology of cognitive and noncognitive skill
formation in childhood and adolescence; key determinants of
completed schooling and health as an adult
• Introduce concepts of self-productivity (skills produced early in life
increase skills at later stages) and dynamic complementarity (early
investments raise marginal productivity of later investments), which
interact to generate multiplier effects
• Investments in adolescence have much larger payoffs when earlier
investments have been made
• Stress importance of noncognitive skills: “big five personality traits”
(conscientiousness , openness , extraversion, agreeableness, and
neuroticism), and related personality measures
14
E. FRIEDMAN AND R. MARE: NEW CAUSAL
RELATIONSHIP FROM SCHOOLING TO
HEALTH
• Typical assumption: Parents invest in their children’s
schooling because child quality enters their utility function
• Friedman and Mare (Demography, 2014): time of adult children
an input in the production of their elderly parents’ health,
efficiency of this time a positive function of schooling
• Evidence that adult offspring’s education positively associated
with elderly parents’ survival
15
EMPIRICAL STUDIES: INTRODUCTION
Three Types • Direct Inclusion of Third Variables
• Twin Studies
• Instrumental Variables
• Note: Include studies by Heckman and colleagues in first type, but
they are much more complicated, latent cognitive and personality
skills, measured and unmeasured components of these skills, factor
analysis to control for measured and unmeasured cognitive and non-
cognitive ability, simultaneous estimation of outcome equation and
measurement equations
16
INVENTORY OF EMPIRICAL STUDIES
Inclusion of Third Variables (8)
Conti and Hansman (2013); Conti and Heckman (2010); Conti,
Heckman, and Urzua (2010); De Walque (2010); Kaestner and
Callison (2011); Savelyev (2014); Savelyev and Tan (2014); Van
Der Pol (2011)
Twin Studies (9)
Amin and Behrman (2014); Amin, Behrman, and Spector (2013);
Amin, Lundborg, and Rooth (2011); Behrman et al. (2011);
Lundborg, Lyttkens, and Nystedt (2012); Lundborg (2013);
Lundborg, Nordin, and Rooth (2012); Madsen et al. (2010);
Webbink, Martin, and Visscher (2010)
17
INVENTORY-CONTINUED
Instrumental Variables (24)
Agüero and Bharadwaj (2013); Andalón, Williams, and Grossman (2014); Atella and Kopinska (2014); Braakmann (2011); Buckles et al. (2013); Caneiro, Meghir, and Parey (2013); Chou, Liu, Grossman, and Joyce (2010); Clark and Royer (2013); Cowan (2011); Dinçer, Kaushal, and Grossman (2014); Etilé and Jones (2011); Holmlund, Lindahl, and Plug (2011); Jensen and Lleras-Muney (2012); Jürges, Reinhold, and Salm (2011); Kempter, Jürges, and Reinhold (2011); Li and Powdthavee (2014) Lundborg, Nilsson, and Rooth (2014); McCrary and Royer (2011); Meghir, Palme, and Simeonova (2012); Mocan and Cannonier (2012); Powdthavee (2010); Tsai, Liu, Chou, and Grossman (2011); Van den Berg, Janys, and Christensen (2012); Van Kippersluis, O’Donnell, and Van Doorslaer (2011)
18
SELECTED EMPIRICAL RESULTS:
THIRD VARIABLES STUDIES
• All studies find positive and significant effects of completed schooling
on at least some key measures of adult health and beneficial health
behaviors
• Van Der Pol (2011): Controls for time preference in Dutch DNB
Household Survey; outcomes include self-rated health, cigarette
smoking, long-term illness, BMI, and obesity
• Conti and Heckman (2010): Control for cognitive and noncognitive
ability at age 10 and health at that age in examining outcomes at age
30 in 1970 British Cohort Study; outcomes include self-rated health,
daily smoking, and obesity; education effects bigger for those with
more cognitive ability and for those with less noncognitive ability
19
THIRD VARIABLES STUDIES-
CONTINUED
• Savelyev (2014): Terman Life Cycle Study of Children with High Ability,
IQ > 140, 11 years old in 1921, followed through 1991, all high school
grads
• For men, graduation from college increases life expectancy at age 30
by approximately 9 years compared to non-grads, with IQ, big five
personality traits, and health status (all measured at age 12) held
constant; no effect for women
• Heckman student, Heckman type estimation
20
EMPIRICAL RESULTS: TWIN STUDIES
Point-Counterpoint
• Behrman et al. (2011): No effects on adult mortality or hospitalizations
in Danish twin registry, 2,500 identical (monozygotic, MZ) twin pairs
• Amin, Behrman, and Spector (2013): No effects on obesity, smoking,
and physical health in UK twins, 741 female MZ twin pairs
• Madsen et al. (2010): No effects overall in same data as Behrman but
negative effects for males born before 1935 and negative effects for
large schooling differences within pairs
• Webbink, Martin, and Visscher (2010): Negative effect on male obesity
(350 MZ pairs), no effect on female obesity, 700 MZ pairs in Australia
21
TWIN STUDIES-CONTINUED
• Lundborg (2013): Effects in expected directions of high school completion on
self-rated health, chronic conditions , and exercise behavior in Midlife in the US
survey; no effect on smoking and BMI; 347 MZ pairs
• Lundborg, Lyttkens, and Nystedt (2012): Negative effects on mortality in
Swedish twin registry, 9,000 MZ pairs; individuals with at least 13 years of
schooling can expect to live an additional 24 years at age 60 compared to 21
years for those with less than 10 years of schooling; 84 % of low-educated
individuals lived to age 70, compared to 90% of high-educated individuals;
results control for birthweight and height
• Lundborg, Nordin, and Rooth (2012): Negative effect of MZ twin mother’s
schooling on son’s health in Swedish twin registry, pertains to sons who have
enlisted in the military, health measured by physical exam administered on
enlistment; no effect of MZ twin father’s schooling; approximately 300 twin
pairs
22
INSTRUMENTAL VARIABLES STUDIES:
INTRODUCTION
• All but one study use compulsory school reform or school entry cutoff
date as an instrument, sometimes combined with new school
openings at a differential rate among areas
• Group by outcome
• Adult mortality
• Adult health, health behaviors, health knowledge
• Infant health, adolescent health, and mechanisms (effects of parents’
schooling)
23
IV STUDIES: ADULT MORTALITY
• Clark and Royer (2013): No effect in Britain; “Our results…suggest that
economic models that assume a strong causal effect of education on health
should be carefully reconsidered.”
• Meghir, Palme, and Simeonova (2012): Similar lack of effect in Sweden
• Van Kippersluis, O’Donnell, and Van Doorslaer (2011): negative effect for men
but not women in the Netherlands; for men surviving to age 81, S by 1 year
probability of dying before age 89 by 3 percentage points relative to baseline
of 50%
• Buckles et al. (2013): negative effect in U.S. for men; college completion
reduces cumulative mortality from 1980-2007 by almost 30% relative to the
mean, men 38-49 in 1980; instrument is risk of induction during Vietnam War
• Conclusion by Clark and Royer somewhat premature
24
ADULT HEALTH AND HEALTH
BEHAVIORS
• Braakmann (2011): no effects on variety of self reported health measures, smoking, heavy drinking, and diet in Britain; similar findings reported by Clark and Royer (2013)
• Powdthavee (2011): negative effect on hypertension based on physical exam in Britain
• Etilé and Jones (2011): negative effect on smoking and quitting in France
• Buckles et al. (2013): negative effects on smoking, heavy drinking, and obesity, positive effect on exercise in US
• Atella and Kopinska (2014): negative effects on BMI and caloric intake, positive effect on calorie expenditure in Italy
25
HEALTH KNOWLEDGE AND RELATED
BEHAVIORS
• Agüero and Bharadwaj (2013): Positive effect on having more
knowledge about HIV and negative effect on number of sexual
partners for women in Zimbabwe
• Andalón, Williams, and Grossman (2014): Positive effects on
conceptive knowledge and use of contraception at sexual debut for
women in Mexico
• Dinçer, Kaushal, and Grossman (2014): Positive effects on knowledge
of the ovulation cycle and use of modern family planning methods for
women in Turkey
• Mocan and Cannonier (2012): Positive effects on use of modern
contraception and to be tested for AIDS for women in Sierra Leone
26
INFANT HEALTH, ADOLESCENT
HEALTH, AND MECHANISMS
• McCrary and Royer (2011): No effects of mother’s schooling on low
birthweight, infant mortality, maternal smoking and alcohol use during
pregnancy, and prenatal care use for pregnant women in California and Texas
• Chou, Liu, Grossman, and Joyce (2010): Negative effects of mother’s
schooling on low birthweight, neonatal mortality, postneonatal mortality, and
infant mortality in Taiwan; increase in schooling associated with school reform
saved almost 1 infant life in 1,000 live births
• Lundborg, Nilsson, and Rooth (2014): Positive effects of mother’s schooling
on son’s physical health and height in Swedish military enlistment register
• Dinçer, Kaushal, and Grossman (2014): Positive effects on age at first marriage
and at first birth, negative effect on number of pregnancies, weak evidence of
negative effect on infant mortality in Turkey
27
ISSUES/QUESTIONS
• School entry age used by McCrary and Royer and others as instrument may not be exogenous; parents can hold their children back or petition to have them start early; Shigeoka (2014) finds almost 2,000 births per year shifted from a week before to a week after the school entry cutoff date in Japan
• Why do results differ? ATE versus LATE
• Twin studies have small N. Differences in schooling between twins may be small. Why do identical twins obtain different amounts of schooling?
• Some IV studies find that OLS is consistent. Some do not test this. Schooling is endogenous, but is it possible that recursive model with uncorrelated errors is correct? Third variables influence schooling but have no impact on health with schooling held constant.
28
ISSUES/QUESTIONS CONTINUED
• Take seriously H S, life expectancynumber of periods
over which returns collectedS
• Estimate structural schooling equation and structural health
equation at same time
• How can theoretical advances (Galama and van Kippersluis;
Heckman and colleagues) help?
29