Professional Documents
Culture Documents
Key points:
• The ‘gradient’ in health
• Education is causally related to health
• The reason why is unclear, though may have to do with cognitive processing
I. Introduction
Health has a very strong gradient in education. Mortality and quality of life are each
related to education. ‘Gradient’ is actually a technical term, invented by (Sir) Michael Marmot
in the UK.
Start off with basic facts. Most are by education, some are by income, but think of them
the same way. Education is better b/c it reflects permanent income. Show data for US, but this
is true for the UK and other developed countries as well.
1
Quality of life is very
different by income. Low income
people have very rapidly declining
QOL compared to higher income
people.
2
II. Possible Theories
𝐻 = 𝛽0 + 𝛽1 𝐸 + 𝑋𝛽 + 𝜀;
𝐸 = 𝛼0 + 𝛼1 𝐻 + 𝑍𝛼 + 𝜇
3. Third factors:
Discount rates (Vic Fuchs)
Value of the future
Disentangling the relationships requires some plausible variation. Take a specific example: does
education lead middle aged people (45-64) to be in better health? One can easily gather data on
this, for example the Health and Retirement Study (people aged 51+, followed biennially from
1992 on). Education is well measured. Proxy for health with several variables:
- risk factors (obese, smoker)
- presence of disease (heart disease, cancer)
- functional limitations (difficulty walking, standing)
- death (conditional on being alive at the beginning of the survey)
3
III. Education and Health: Causal Determination
Some very good work has gone into the causality question. Two instruments in particular strike
me as very good.
Between 1915 and 1939, 30+ states changed compulsory school and child labor laws. Can learn
about the impact of education for the person at the margin of dropping out of school or not.
Look at people who were 14 between 1915 and 1939, matched to state of birth (ignores early life
mobility).
Data from 1960, 1970, and 1980 censuses. Calculate ten year death rates over the 1960s (1960-
70) and 1970s (1970-80) by birth year and state of birth.
Results:
This has been followed up by Oreopoulos (Canada and England and Wales, 2003), Arendt
(Denmark, 2005), and Spasojevic (Sweden, 2003).
4
But not all studies find this. Clark and Royer examine the impact of major education changes in
the UK in 1947 (stay in school to 15) and 1972 (stay in school to 16). Neither of these changes
seemed to lower mortality rates.
5
College openings (Currie and Moretti, QJE, 2003)
Sample: Data from Vital Statistics, 1970-99 – data is at the birth level
Women 24-45 (get when education completed)
First births (get rid of parity effects)
Regressions:
̂ + 𝛼3 𝐴𝐺𝐸 + 𝛼4 𝐶𝑂𝐻𝑂𝑅𝑇 +
Second stage: 𝑂𝑢𝑡𝑐𝑜𝑚𝑒 = 𝛼0 + 𝛼1 𝐸𝑑𝑢𝑐
𝛼5 𝐶𝑜𝑢𝑛𝑡𝑦 ∗ 𝑌𝑒𝑎𝑟 + 𝛼6 𝑌17 + 𝛼7 𝑈𝑟𝑏𝑎𝑛17
6
Results: 4 year college openings affect education Improved outcomes
Average
years of Pr[college
education education]
Pr[high school
Pr[community
only]
college]
7
Results:
• A year of education
reduces the incidence of
low birth weight by
approximately 10 percent,
and reduces the incidence
of preterm birth by 6
percent, on average.
• Education reduces smoking
while pregnant
• Education reduces fertility,
esp. non-marital fertility
𝐻 = 𝛽0 + 𝛽1 𝐸 + 𝑑𝑒𝑚𝑜𝑔 + 𝜀;
𝐻 = 𝛼0 + 𝛼1 𝐸 + 𝑑𝑒𝑚𝑜𝑔 + 𝑋𝛼 + 𝜀;
The ratio 𝛼1 /𝛽1 shows the impact of characteristic X on the education-health relationship. Note:
This is not kosher econometrically if X is endogenous.
8
Results: Controlling for ASVAB explains a part of the education-health gradient.
Controlling for life satisfaction, personality variables, and social integration doesn’t affect
education-health gradient. Discount rate controls also don’t matter.
Summary
9
One way to see the lack of importance of discount rates – behaviors are correlated with
education but not that correlated across individuals.
Interpretation
Depths of beliefs
Look at the Motor Vehicle Occupant Safety Survey (periodic from the US Department of
Transportation.
10
Complexity
Lichtenberg and Lleras-
Muney show that more educated
people take newer drugs. The effect
isn’t huge – average drug is -0.1
years newer – but could indicate a
willingness or access to newer,
better therapy.
11