eco question

Chart1

1900	1900
1901	1901
1902	1902
1903	1903
1904	1904
1905	1905
1906	1906
1907	1907
1908	1908
1909	1909
1910	1910
1911	1911
1912	1912
1913	1913
1914	1914
1915	1915
1916	1916
1917	1917
1918	1918
1919	1919
1920	1920
1921	1921
1922	1922
1923	1923
1924	1924
1925	1925
1926	1926
1927	1927
1928	1928
1929	1929
1930	1930
1931	1931
1932	1932
1933	1933
1934	1934
1935	1935
1936	1936
1937	1937
1938	1938
1939	1939
1940	1940
1941	1941
1942	1942
1943	1943
1944	1944
1945	1945
1946	1946
1947	1947
1948	1948
1949	1949
1950	1950
1951	1951
1952	1952
1953	1953
1954	1954
1955	1955
1956	1956
1957	1957
1958	1958
1959	1959
1960	1960
1961	1961
1962	1962
1963	1963
1964	1964
1965	1965
1966	1966
1967	1967
1968	1968
1969	1969
1970	1970
1971	1971
1972	1972
1973	1973
1974	1974
1975	1975
1976	1976
1977	1977
1978	1978
1979	1979
1980	1980
1981	1981
1982	1982
1983	1983
1984	1984
1985	1985
1986	1986
1987	1987
1988	1988
1989	1989
1990	1990
1991	1991
1992	1992
1993	1993
1994	1994
1995	1995
1996	1996
1997	1997
1998	1998

Sheet1

	1998	875.8		1900	2,518.00
	1997	887.3		1901	2,473.10
	1996	902.4		1902	2,301.30
	1995	918.5		1903	2,379.00
	1994	920.2		1904	2,502.50
	1993	931.5		1905	2,423.70
	1992	910.9		1906	2,399.00
	1991	925.5		1907	2,494.40
	1990	938.7		1908	2,298.90
	1989	950.5		1909	2,249.20
	1988	975.7		1910	2,317.20
	1987	970		1911	2,245.40
	1986	978.6		1912	2,211.70
	1985	988.1		1913	2,206.50
	1984	982.5		1914	2,149.30
	1983	990		1915	2,174.80
	1982	985		1916	2,266.60
	1981	1,007.00		1917	2,275.90
	1980	1,039.00		1918	2,541.60
	1979	1,010.00		1919	2,057.20
	1978	1,043.70		1920	2,147.10
	1977	1,051.60		1921	1,958.20
	1976	1,084.10		1922	2,049.50
	1975	1,094.40		1923	2,141.40
	1974	1,151.80		1924	2,038.00
	1973	1,201.20		1925	2,068.70
	1972	1,214.80		1926	2,146.20
	1971	1,213.10		1927	1,989.50
	1970	1,222.60		1928	2,124.60
	1969	1,271.80		1929	2,081.20
	1968	1,304.50		1930	1,943.80
	1967	1,274.00		1931	1,895.10
	1966	1,309.00		1932	1,897.10
	1965	1,306.50		1933	1,850.10
	1964	1,303.80		1934	1,888.20
	1963	1,346.30		1935	1,860.10
	1962	1,323.60		1936	1,963.70
	1961	1,298.80		1937	1,882.60
	1960	1,339.20		1938	1,764.30
	1959	1,317.30		1939	1,766.90
	1958	1,343.40		1940	1,785.00
	1957	1,356.70		1941	1,694.60
	1956	1,333.70		1942	1,635.80
	1955	1,332.30		1943	1,702.40
	1954	1,314.80		1944	1,618.50
	1953	1,385.60		1945	1,575.40
	1952	1,394.60		1946	1,529.70
	1951	1,423.50		1947	1,532.00
	1950	1,446.00		1948	1,501.70
	1949	1,457.30		1949	1,457.30
	1948	1,501.70		1950	1,446.00
	1947	1,532.00		1951	1,423.50
	1946	1,529.70		1952	1,394.60
	1945	1,575.40		1953	1,385.60
	1944	1,618.50		1954	1,314.80
	1943	1,702.40		1955	1,332.30
	1942	1,635.80		1956	1,333.70
	1941	1,694.60		1957	1,356.70
	1940	1,785.00		1958	1,343.40
	1939	1,766.90		1959	1,317.30
	1938	1,764.30		1960	1,339.20
	1937	1,882.60		1961	1,298.80
	1936	1,963.70		1962	1,323.60
	1935	1,860.10		1963	1,346.30
	1934	1,888.20		1964	1,303.80
	1933	1,850.10		1965	1,306.50
	1932	1,897.10		1966	1,309.00
	1931	1,895.10		1967	1,274.00
	1930	1,943.80		1968	1,304.50
	1929	2,081.20		1969	1,271.80
	1928	2,124.60		1970	1,222.60
	1927	1,989.50		1971	1,213.10
	1926	2,146.20		1972	1,214.80
	1925	2,068.70		1973	1,201.20
	1924	2,038.00		1974	1,151.80
	1923	2,141.40		1975	1,094.40
	1922	2,049.50		1976	1,084.10
	1921	1,958.20		1977	1,051.60
	1920	2,147.10		1978	1,043.70
	1919	2,057.20		1979	1,010.00
	1918	2,541.60		1980	1,039.00
	1917	2,275.90		1981	1,007.00
	1916	2,266.60		1982	985
	1915	2,174.80		1983	990
	1914	2,149.30		1984	982.5
	1913	2,206.50		1985	988.1
	1912	2,211.70		1986	978.6
	1911	2,245.40		1987	970
	1910	2,317.20		1988	975.7
	1909	2,249.20		1989	950.5
	1908	2,298.90		1990	938.7
	1907	2,494.40		1991	925.5
	1906	2,399.00		1992	910.9
	1905	2,423.70		1993	931.5
	1904	2,502.50		1994	920.2
	1903	2,379.00		1995	918.5
	1902	2,301.30		1996	902.4
	1901	2,473.10		1997	887.3
	1900	2,518.00		1998	875.8

Sheet1

Sheet2

1959	1,317.30		1967	1,274.00		1978	1,043.70		1998	875.8
1958	1,343.40		1966	1,309.00		1977	1,051.60		1997	887.3
1957	1,356.70		1965	1,306.50		1976	1,084.10		1996	902.4
1956	1,333.70		1964	1,303.80		1975	1,094.40		1995	918.5
1955	1,332.30		1963	1,346.30		1974	1,151.80		1994	920.2
1954	1,314.80		1962	1,323.60		1973	1,201.20		1993	931.5
1953	1,385.60		1961	1,298.80		1972	1,214.80		1992	910.9
1952	1,394.60		1960	1,339.20		1971	1,213.10		1991	925.5
1951	1,423.50					1970	1,222.60		1990	938.7
1950	1,446.00					1969	1,271.80		1989	950.5
						1968	1,304.50		1988	975.7
									1987	970
									1986	978.6
									1985	988.1
									1984	982.5
									1983	990
									1982	985
									1981	1,007.00
									1980	1,039.00
									1979	1,010.00

Sheet3

Income and Health (Country Level)

Source: Angus Deaton (NBER, 2004); Note: Circular areas reflect relative population sizes

18 Education and Health (Country Level)

Source: Cutler and Lleras-Muney (NBER, 2006)

19Education and Health

 What are some possible pathways through which education might affect health?

1. Education  Health (Direct) 2. Education  Income  Health (Indirect) 3. Education  Lifestyle  Health (Indirect) 4. Ability  Education and Income  Health 5. Life Expectancy  Education (reverse causality)

20 Education and Mortality Risk  Completed Grades and Mortality Risk

NOTE: Analysis controls for race and gender only (not income); Source: Cutler and Lleras-Muney (NBER, 2006)

21 Education and Health Status  Completed Grades and Poor Health

NOTE: Analysis controls for race and gender only (not income); Source: Cutler and Lleras-Muney (NBER, 2006)

22 Education and Health Decisions

 Completed Grades and Smoking

NOTE: Analysis controls for race and gender only (not income); Source: Cutler and Lleras-Muney (NBER, 2006)

23 Education and Health Decisions

 Completed Grades and Seat Belt Use

NOTE: Analysis controls for race and gender only (not income); Source: Cutler and Lleras-Muney (NBER, 2006)

24Education and Health Prevention  Completed Grades and Cancer Screening

NOTE: Analysis controls for race and gender only (not income); Source: Cutler and Lleras-Muney (NBER, 2006)

25 Education and Health Prevention  Completed Grades and Smoke Detectors

NOTE: Analysis controls for race and gender only (not income); Source: Cutler and Lleras-Muney (NBER, 2006)

26 Causal Evidence

 All of these graphs show suggestive patterns of relationships between education and health, but are they causal effects, or just correlations?  Eg. Are people really being educated about seat belt use between 11th

and 17th years of school? Or are people who are naturally more careful also more likely to invest in education?

 Suppose we want to know the causal effect of education on health, and we estimate the model:

𝐻𝐻𝑖𝑖𝑖𝑖 = 𝑎𝑎 + 𝑏𝑏1 ∗ 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑖𝑖 + 𝑏𝑏2 ∗ 𝑎𝑎𝑎𝑎𝑒𝑒𝑖𝑖𝑖𝑖 + 𝑏𝑏3 ∗ 𝑒𝑒𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑖𝑖𝑖𝑖 + 𝑏𝑏4 ∗ 𝑖𝑖𝑖𝑖𝑒𝑒𝑐𝑐𝑖𝑖𝑒𝑒𝑖𝑖𝑖𝑖 + 𝑒𝑒𝑖𝑖𝑖𝑖

 Why might b1 be a biased estimate of the causal effect of education in this model?

27 Causal Evidence

 Suppose that people have unobserved ability Ai and that the rate of return to investing in education is higher for people that are born with high values of Ai

 Since we can’t observe ability, part of it ends up in the error term, so the true model is:

𝐻𝐻𝑖𝑖𝑖𝑖 = α + 𝑏𝑏1 ∗ 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑖𝑖 + 𝑏𝑏2 ∗ 𝑎𝑎𝑎𝑎𝑒𝑒𝑖𝑖𝑖𝑖 + 𝑏𝑏3 ∗ 𝑒𝑒𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑖𝑖𝑖𝑖 + 𝑏𝑏4 ∗ 𝑖𝑖𝑖𝑖𝑒𝑒𝑐𝑐𝑖𝑖𝑒𝑒𝑖𝑖𝑖𝑖 + (ε𝑖𝑖𝑖𝑖 + Ai)

 We know that educi is correlated with Ai, so one of the key assumptions of OLS is violated, and all of the estimated parameters could be biased, including b1

 To fix this problem, we can use instrumental variables  Why kind of IV do we want to look for? Examples?

Instrumental Variables Overview

• Suppose you wish to estimate a regression model of the form:

But face the problem that X is correlated with ε, which violates the OLS assumptions • Suppose also that there is another variable Z with the following

properties: • Z is correlated with X (correlation must be reasonably strong)

• Z must also be monotonically related to X (so the relationship between Z and X always goes in the same direction)

• Z is only correlated with y because of its correlation with X • Therefore, Z must be uncorrelated with ε • This is commonly termed the “exclusion restriction”

• Then we can get an unbiased estimate of β using Z as an “instrumental variable” for X

𝑦𝑦 = 𝛼𝛼 + 𝛽𝛽𝛽𝛽 + 𝜀𝜀

Instrumental Variables Overview

• Suppose Z satisfies the assumptions required to be a valid instrumental variable. What next?

• Conceptually, can think of estimating the regression in two steps • First stage is to predict X using Z

• Then we can use the predicted value �𝛽𝛽 in the second stage

• If the IV assumptions hold this will give an unbiased estimate of β

𝑦𝑦 = 𝛼𝛼 + 𝛽𝛽 �𝛽𝛽 + 𝜀𝜀

�𝛽𝛽 = 𝛼𝛼 + 𝛽𝛽𝛽𝛽 + ζ

𝑦𝑦 = 𝛼𝛼 + 𝛽𝛽𝛽𝛽 + 𝜀𝜀

Causal Evidence

 The idea behind using IVs here is that we want to find a variable that is correlated with education but uncorrelated with individual ability

 In labor economics there are several such variables commonly used:  One is to study the period in the mid 1900s when many

states changed their laws about the minimum number of years of schooling required for children  If some people are forced by law to stay in school longer, this

additional education is uncorrelated with ability

Causal Evidence

 In labor economics there are several such variables commonly used:  A second type of IV used is the number of new colleges

that were opened near a student just before they turned 17 years old  The idea is that if you turn 17 right before the a nearby college

opens you would likely have to move away to attend college

 If a similar person turns 17 right after the nearby college opens they can choose to go to the school nearby, and avoid having to move

 The opening of a college is viewed as a shift in the supply of education that is uncorrelated with the distributions of ability of people who graduate high school right before the college opens versus right after

Causal Evidence

 Currie and Moretti (2003) use the opening of new colleges as an instrument that affects the cost of attending, but is uncorrelated with confounding factors like unobserved ability

 The question they study is slightly different though:  Question: Do the children of women with more education have

better health outcomes?

 Test whether the children of women who turn 17 right after a college opens in their area have better health outcomes than children of women who turned 17 in the same area right before the college opened

2SLS Model

 First stage model is:

 Second stage model is:

 The idea is that ability Ai is likely uncorrelated with everything in the first stage, so it’s in the residual

 However, school openings are correlated with education, so they provide some information about education differences across people and over time

 If so, when we use the first stage to predict education, the predicted value will be uncorrelated with Ai, but informative about actual education

 We can use this predicted value of education in the second stage, and δ1 will be unbiased

Currie and Moretti: First Stage

Currie and Moretti: Results

• For women in counties where a college opened before they turned 17, their subsequent children were less likely to be premature or have low birth weight

Currie and Moretti: Results

• No gains for women in the same locations who were 25 years old when the college opened

• Suggests the effect is driven by change in education opportunities, rather than some other unobserved geographic factors

Currie and Moretti Results: Second Stage

One additional year of education at the high- school/college margin • Reduces probability of

low birth weight child by 1 percentage point (20% of mean)

• Reduces preterm births by 1 percentage point (~14% of mean)

• Increases prenatal care use

• Reduces probability of smoking while pregnant by 6 percentage points (30% of mean)

HEALTH SPENDING

Data on Spending

 Measuring health care costs and expenditures  Current levels

 Growth Components

 Reasons for health care cost growth

National Health Expenditures as a Percentage of Gross Domestic Product, 1980 – 2015

Source: Centers for Medicare & Medicaid Services, Office of the Actuary.

9. 1% 9. 4% 10

.2 %

10 .3

% 10

.2 %

10 .4

% 10

.6 %

10 .8

% 11

.2 %

11 .6

% 12

.3 %

13 .0

% 13

.4 %

13 .7

% 13

.6 %

13 .7

% 13

.7 %

13 .6

% 13

.6 %

13 .7

% 13

.4 %

14 .1

% 14

.9 %

15 .4

% 15

.5 %

15 .5

% 15

.6 %

15 .9

% 16

.4 %

17 .4

% 17

.4 %

17 .3

% 17

.2 %

17 .1

% 17

.4 %

17 .8

%

0%

2%

4%

6%

8%

10%

12%

14%

16%

18%

20%

80 82 84 86 88 90 92 94 96 98 00 02 04 06 08 10 12 14

P er

ce nt

ag e

of G

D P

International Spending Comparison

Source: Kaiser Family Foundation

International Spending Comparison

Source: PBS

Rates of Growth

 Growth rates: More useful metrics  Persistent meaningful reduction in spending would

require a decrease in rate of growth

 Many policies that aim to reduce spending have instead resulted in one-time reductions in spending, followed by the same high growth rate

Growth is the Difference …

Source: Kaiser Family Foundation

NHE Components: 1970-2013

Select Spending Categories

1970 1985 2005 2013

$B %NHE

$B %NHE

$B %NHE

$B %NHE

Hospitals $27.60 $165.40 $611.60 $936.90 36.80% 37.60% 30.80% 32.31%

Physicians $14.00 $89.80 $421.20 $586.70 18.70% 20.40% 21.20% 20.23%

Pharmaceuticals $5.50 $21.10 $200.70 $271.10 7.30% 4.80% 10.10% 9.35%

Administrative $2.80 $25.60 $143 Not

Available3.70% 5.80% 7.20% Home Health / Nursing Homes

$4.30 $37.30 $169.30 $235.60 5.70% 8.50% 8.50% 8.12%

Total NHE $74.90 $439.90 $1,987.70 $2,900.00

General Trends: Spending Components

 Hospitals share is slightly declining over time  Outpatient care and physician services slightly

increasing  Pharmaceutical share drops and then increases

strongly through 1990s  Medicare Part D implemented in 2006

 Administrative share is increasing  Main lesson: spending levels have increased

rapidly in every category