application 1: health and earnings
DESCRIPTION
Application 1: Health and Earnings. Methods of Economic Investigation Lecture 3. Why are we doing this?. Want to apply what we’ve talked about this week to real-life situation Better able to understand academic papers - PowerPoint PPT PresentationTRANSCRIPT
Application 1: Health and Earnings
Methods of Economic Investigation
Lecture 3
Why are we doing this? Want to apply what we’ve talked about this
week to real-life situation
Better able to understand academic papers Even if you go to industry (finance, consulting,
etc.) or government/policy—academic work is often used
Being able to see why something works and doesn’t is critical
What are we doing today? Think about the causal relationship
between health and earnings
Review: How to define a research question How to develop a way to distinguish correlation
from causation Thinking about problems with measurement
and data Applying econometrics to real-life
A little background: Facts Strong relationship between income and
health (health gradient)
Lots of correlates to income and health Education Race/Ethnicity
Need to know relationship for determining actual policies
How can we show a relationship? In a cross-section: Do richer people (or
countries) have better average health? In a time-series: As people (or countries) get
richer, does the average level of health increase?
In a panel: Do people, after getting more money, become healthier?
In a repeated cross-section: Do cohorts (groups of people in the same year) who appear to have more money, have better health?
In a time series…
7274
7678
Nu
mbe
r o
f Yea
rs
1020
3040
50In
com
e (
in 1
000
s o
f US
$)
1975 1980 1985 1990 1995 2000 2005year
Median Household Income Avg. Life Expectancy
Source: Source: National Center for Health Statistics, National Vital Statistics Reports, vol. 54, no. 19, June 28, 2006; Source: U.S. Census Bureau, Current Population Survey, Annual Social and Economic Supplements.
In a cross-section
Source: Lynch, et al (1998) AJPH, p 1078
In a Panel…
Source: Smith, JEP 1999, pp 147
What is the research question? What is it we want to know?
If we gave some people money—would that make them healthier?
Accumulation effects over time? Does it matter who (which income group) we give the
money to? Does it matter when we give them the money?
If we improve the income of some group of people, with that improve their health outcomes?
Long-term or short-term? Holding other stuff fixed? Do we care how that extra
income affects other things that later affect health?
What is the fundamental identification problem?
HEALTH OF INDIVIDUAL iToo sick for
school?
No Access to Medical Care?
Parents didn’t know about med/docs/etc?
INCOME of INDIVIDUAL i
Where is the bias? –I Reverse causation
Health might affect inputs (like education) which then affect income
Health might make it hard to work
In that example—good health is positively correlated with income. What does that imply for the bias?
Poor health can cost money
Where’s the bias? –II Third Factors
Education might make people better able to earn and better at taking up health protective behaviors
Ability might make people more educated, have higher income and healthier?
Underlying genetics might make people more able, higher educated, higher income, etc.
What is health anyway? Want to see if income causes better health
but how to we quantify better health?
This about what it is we’re after? Quality of life? Extreme outcomes (death, dismemberment)? Things that are costly for society (infectious
diseases, eg.)
What is the outcome?-I Maybe there’s something in existing data…
Mortality Extreme outcome: Small changes might be hard to
see Maybe not what we care about (if everyone lives but
some are very sick…) Illness/disease
Hard to get info on Diagnosis bias LOTS of third factor causation here…
What about doctor/hospital visits?
What is the outcome?-II Maybe we could just ask people…
How good is your health (1 being excellent 4 being poor)
This is called “Self-reported health status” Commonly used measure
Survey Response Can we compare answers across people? Bias especially bad if response type varies by
SES/Income characteristics
What experiment would we design? Thought experiment: If we took a random
sample of the population and divided them in half, arbitrarily gave half an extra income and measured their health, would they be different than the others? Can we do this (maybe?!!) Does something like this happen in real life? What will change and can we measure it?
What should we estimate? If we could do the experiment we just
described: we would want to test:E(Health | Treatment) > E(Health | Control)
Very simple econometric specification:
Regression is indexed by:i: the individual c:the group (e.g. either treatment or control)(This works because our sample is randomly
drawn and assigned, next class…)
icicic GroupTreatmentHealth ) (*
How do we interpret estimates? Recall that our OLS estimate is:
Our estimate is very simple:
We can put a dollar value on this since we designed the change!
X
XyE
)|(
GroupControlofIncomeGroupTreatmentofIncome
GroupControlofHealthGroupTreatmentofHealth
-
-
Research Design If we can give people extra income, then
we can measure their health afterwards and see what happens
If we can’t do this experiment, can we think of sometimes when people completely randomly get money? Example: The Lottery (Paper by Mikael Lindhal, Journal of Human Resources, 2005)
Identifying assumption: People who win the lottery look like people who play the lottery but don’t win
Data problems The survey didn’t ask if you played the
lottery—so can we compare lottery players to non-players? Can we do anything else? New Identifying assumption: Playing the lottery
is NOT correlated with characteristics that are correlated with health
Can we prove this? Compare lottery players and non-players
Comparing Players to Non-Players
What does he estimate? Want to see what the effect of a lottery
prize is on health
Health Measure in 1981
Lottery Winnings between 1969-1981
SES Characteristics: Age, gender family background, etc. in 1968
Results Evidence of
reduced health (other results index these measures and get significant results)
Evidence of reduced mortality
Interpreting results If we know the change in income for lottery
—we can estimate an effect size
Results imply: 10 percent increase in income increased general health by 0.04 standard deviations
How to put this into context? What do other interventions find? Is this a replicatable policy?
Internal Validity Do we believe the identifying assumption?
Maybe not People who win the lottery might play a lot so they
have more disposable income or may be more risk-loving and that affects other characteristics too
If he only looked at players and compared big winners to little winners—Effects are EVEN bigger (40 percent) what does that tell you?
External Validity If we believe the identifying assumption,
can we generalize this? Who plays the lottery? We might only be
identifying this for a certain point on the distribution
Do we think the effect size would be the same for very rich people?
It happened in Sweden—with compressed income distribution and good health system/safety net
Is this comparable to the US? Is this comparable to a developing country?
Next Steps If we had done our thought experiment,
we might have had some of these problems Who participates in our experiment? Is the change in health the same for all people
on the income distribution? What is the mechanism by which this works? (if
it’s access to health care—better make sure that’s in place in our experiment too…)
Next week: Experimental Evaluations…