ECON451 Maryland Effects of Drinking Too Much Fruit Juice Correlation and Causation This is a econ paper, need to related to correlation and causation, use

ECON451 Maryland Effects of Drinking Too Much Fruit Juice Correlation and Causation This is a econ paper, need to related to correlation and causation, use the article I choose and talks about its correlation and causation issues, there is one example essay check that one and write one similar format like the example essay. Total word count: 1200-1500the detailed writing instruction is post on the PPT, please check the last few slides, follow all these steps.There is one example essay, need to write one similar like that, but with different articles.The article I choose is this one: https://www.cnn.com/2019/05/17/health/fruit-juice-sugary-drink-early-death-study/index.html Econ 451 – Issues in Labor
Economics
Methods
Reading
• Read: “Do We Really Know What Makes Us
Healthy?”
– This is a substantive New York Times article on the
history of Hormone Replacement Therapy and
how causal research methods can be absolutely
crucial for understanding its effects.
– Yes, it’s very long. But it’s your only reading for
this section and a fantastic article.
What Do I Mean By “Methods”
• I’m talking about data analysis in modern
microeconomics.
– What is the effect of X on Y?
– Essentially all research in all science (both social
and natural) is dedicated to answering questions
like this.
How Do Other People Do It?
• Doctors – randomized, controlled trials
• Psychologists –controlled laboratory
experiments
• Qualitative Sociologists – get to know one X
and one Y really well
• Epidemiologists – observe a lot of X and Y
from a distance
How Do Other People Do It?
• Doctors – randomized, controlled trials
– Not often helpful with social policy
How Do Other People Do It?
• Psychologists – laboratory controlled
experiments
– Good starting place, but getting social policy right
based on a study in a lab is really hard
How Do Other People Do It?
• Sociologists – get to know one X and one Y
really well
– Great way to look for motivation to study an issue
further or understand the personal ramifications
of economic policy
– Not a great way to figure out what the effect of
changing X for everybody is
How Do Other People Do It?
• Epidemiologists – observe X and Y from a
distance
– In many circumstances, essentially useless
How Do Economists Do It?
• We used to be like the epidemiologists.
– A lot of economic research still is
– Especially when it produces results that are
politically convenient or reinforce “gut opinions”
about the way the world should work.
How Do Economists Do It?
• Eventually, labor economists figured out that
we’d learn a lot more if we tried to be as
similar as possible to doctors or psychologists.
– We can’t perform experiments with social policy,
but we can search for “natural experiments”,
“quasi-experimental variation”, and all sorts of
things in the real world that resemble or can be
made to resemble experiments.
Econometrics Review
• Linear regression:
y = a + b*x + u
– Estimating this tells us the observed relationship
between x and y:
• When x goes up by 1 unit, how many units does y go up?
– When does this tell us the effect of x on y?
• When x is uncorrelated with the unobserved error term, u.
– No omitted variable bias, no reverse causality
Facebook Makes You Depressed
“No one joins Facebook to be sad and lonely. But a new study
from the University of Michigan psychologist Ethan Kross
argues that that’s exactly how it makes us feel. Over two
weeks, Kross and his colleagues sent text messages to eightytwo Ann Arbor residents five times per day. The researchers
wanted to know a few things: how their subjects felt overall,
how worried and lonely they were, how much they had used
Facebook, and how often they had had direct interaction with
others since the previous text message. Kross found that the
more people used Facebook in the time between the two
texts, the less happy they felt—and the more their overall
satisfaction declined from the beginning of the study until its
end. The data, he argues, shows that Facebook was making
them unhappy.”
Facebook Makes You Depressed
• Reverse Causality
– Does being depressed make you use Facebook?
• Omitted Variable Bias
– Does using Facebook correlate with using
MySpace, and it’s actually MySpace that makes
people depressed?
Facebook Makes You Depressed
• Other types of OVB:
– Selection
• Facebook makes people happier. Therefore, the people who
use it most are unhappy people in need of a pick-me-up.
These people are still unhappier than average despite the
effects of Facebook.
– Measurement Error (less interesting)
• The study does a bad job of measuring how much you use
Facebook. Because of this, the estimates actually understate
how much Facebook makes you depressed.
Facebook Makes You Happy
“But, as with most findings on Facebook, the opposite
argument is equally prominent. In 2009, Sebastián Valenzuela
and his colleagues came to the opposite conclusion of Kross:
that using Facebook makes us happier. They also found that it
increases social trust and engagement—and even encourages
political participation. Valenzuela’s findings fit neatly with
what social psychologists have long known about sociality: …
Social networks are a way to share, and the experience of
successful sharing comes with a psychological and
physiological rush that is often self-reinforcing… The mere
thought of successful sharing activates our reward-processing
centers, even before we’ve actually shared a single thing.”
How To Fix These Problems
• Control for stuff:
– Imagine the following experiment. People are
recruited randomly from the population to perform a
task for money.
– Men are randomly given a wage between $0 and $5.
Women are randomly given a wage between $5 and
$10.
• What is the effect of the wage on output?
– Output = a + b*wage + u
How To Fix These Problems
• Control for stuff:
– Wage and gender are correlated. If output on the
task is correlated with gender also, then there’s
omitted variable bias.
– But, controlling for gender perfectly fixes this
because we know the wage is random, conditional
on gender:
• Output = a + b*wage + c*gender + u
How To Make Things Look Like
Experiments
• Diff-in-Diff Study
– We’ve seen a bunch of these already.
– Idea: State 1 changes from Policy A to Policy B,
State 2 always has policy A
• Options: compare State 1 before to State 1 after
– Outcome = a + b*after + u
• Compare State 1 after to State 2 after
– Outcome = a + b*state + u
How To Make Things Look Like
Experiments
• Diff-in-Diff Study
– Compare the before-after difference in State 1 to
the before-after difference in State 2
– Outcome after – Outcome before = a + b*state + u
• Only works if there are no “state-specific time trends”
How To Make Things Look Like
Experiments
• Diff-in-Diff Study: better version
– All of the states change from Policy A to Policy B at
one point or another
– All states with Policy A are the control group, all
states with Policy B are the treated group
• These groups shift composition over time, meaning the
estimate is not state-specific
How To Make Things Look Like
Experiments
• Instrumental Variable
– Experiment: subjects are recruited randomly from the
population to perform a task related to their class.
They can choose whether to perform the task for
money or for extra credit in the class. You want to see
which payment type motivates them more.
• Output = a + b*payment type + u
• Problem: imagine all the students that choose the extra
credit are doing poorly in class
How To Make Things Look Like
Experiments
• Instrumental Variable
– Solution: randomly vary the exchange rate between
money and extra credit.
• Payment type = c + d*exchange rate + e
• c + d*exchange rate is an estimate of payment type that is
uncorrelated with ability (or the error term generally) since
the exchange rate is random
– Use this estimate as the payment type variable:
• Output = a + b*(c + d*exchange rate) + u
How To Make Things Look Like
Experiments
• Instrumental Variable: Example
– Angrist & Krueger: What’s the value of a college
education?
– Wage = a + b*college + u
• Deciding to get a college degree is clearly correlated
with other things that affect wage
How To Make Things Look Like
Experiments
• Instrumental Variable: Example
– Need something random that affects the choice to get
a college education: Vietnam draft lottery
• Birthdays were randomly drawn and assigned priorities
• After the drawing, you knew, based on your birthday, how
likely it was that you would be called up
• If you could get admitted to college, you were exempt
• The lottery randomly generated variation in the incentive to
go to college
– College = c + d*lottery position + e
– Wage = a + b*(c + d*lottery position) + u
How To Make Things Look Like
Experiments
• Discontinuity
– Experiment: you want to know the effect of
tutoring on class grades. Imagine that everyone
with a GPA below x is assigned a tutor. Everyone
with a GPA above x is not.
• Grade = a + b*tutor + u
• Having a tutor is correlated with education ability
• Can’t control for GPA (why?)
How To Make Things Look Like
Experiments
• Discontinuity
– Solution: exploit the sharp discontinuity at GPA =
x.
• Someone with a GPA of x – 0.05 isn’t really that
different from someone with a GPA of x + 0.05
• But, one of them gets a tutor and the other doesn’t
• Just run the regression, looking only at people in the
neighborhood of x:
How To Make Things Look Like
Experiments
• Discontinuity
– Needs to meet one condition to work: people
cannot perfectly manipulate their position relative
to the cutoff:
• If I can purposefully lower my GPA by exactly 0.01 to
get a tutor, maybe I will
• Why is this a problem?
How To Make Things Look Like
Experiments
• Discontinuity
– Needs to meet one condition to work: people
cannot perfectly manipulate their position relative
to the cutoff:
• If I can purposefully lower my GPA by exactly 0.01 to
get a tutor, maybe I will
• Why is this a problem?
– Selection bias
How To Make Things Look Like
Experiments
• Discontinuity: Example
– Bharadwaj et al: How important is the healthcare
you receive as a newborn for you development?
• Grades = a + b*healthcare + u
• Problem: wealthier families and families with more
involved and motivated parents get access to better
quality hospitals
• Discontinuity: below certain precise cutoffs, newborns
can be classified as Low or Very Low Birthweight
How To Make Things Look Like
Experiments
• Discontinuity: Example
– LBW and VLBW newborns are assigned a boatload
of extra resources
– This is despite the fact that a newborn that weighs
1490 grams is essentially identical to a newborn
that weighs 1500 grams
– And if some doctors round 1490 up to 1500 while
others don’t, then they really are identical.
How To Make Things Look Like
Experiments
• Discontinuity: Example
How To Make Things Look Like
Experiments
• Discontinuity: Example
– Good things happen to newborns that get more
resources
• Both in the short term (infant mortality down)
• And in the long run (grades are better, even through
high school)
How To Make Things Look Like
Experiments
• Discontinuity: Example
How To Make Things Look Like
Experiments
• Synthetic control method
– Perhaps only one state (or group) changes their
policy. And perhaps they only do it for a short
period of time.
How To Make Things Look Like
Experiments
• Synthetic control method
– Perhaps only one state (or group) changes their
policy. And perhaps they only do it for a short
period of time.
– Problem: estimates will be incredibly sensitive to
which other states (or groups) you use as a
control.
How To Make Things Look Like
Experiments
• Synthetic control method
– The synthetic control method seeks to construct a
control group that matches the pre-policy trends
in the “treated” state as closely as possible.
How To Make Things Look Like
Experiments
• Synthetic control method
– The synthetic control method seeks to construct a
control group that matches the pre-policy trends
in the “treated” state as closely as possible.
• Eg: one person in the class gets randomly assigned a
tutor for the final exam. This person might be a normal
student, and outlier, who knows?
How To Make Things Look Like
Experiments
• Synthetic control method
• Eg: one person in the class gets randomly assigned a
tutor for the final exam. This person might be a normal
student, or an outlier, who knows?
• Use all the other students in the class to create an ideal
comparison student.
– Define a set of characteristics (often past test performance)
and find the set of weights that when applied to everyone else
in the class, best matches the one treated student.
– Compare the treated student to that weighted average of
other student score.
How To Make Things Look Like
Experiments
• Synthetic control method
– Cunningham and Shah (2017)
– Rhode Island accidentally and briefly legalized
prostitution in 2003
– Researchers want to estimate the impact this had
on crime and health outcomes
How To Make Things Look Like
Experiments
How To Make Things Look Like
Experiments
• Synthetic control method
– Cunningham and Shah (2017)
• They find more prostitution happens
• They find very large declines in reported rapes
• They find very large declines in female gonorrhea rates
How To Make Things Look Like
Experiments
• Run an Experiment
– Laboratory
– Field
– Large scale field
• Developing countries are starting to let economists do
real experiments with social policy
How To Make Things Look Like
Experiments
• Run an Experiment
– Muralidharan and Sundararaman (2011)
• What is the impact of performance bonuses and
classroom resources on teacher performance?
• Get permission (and $) to randomize this stuff in 300
schools in India
How To Make Things Look Like
Experiments
• Run an Experiment
How To Make Things Look Like
Experiments
• Run an Experiment
– Muralidharan and Sundararaman (2011)
• Find that performance bonuses are pretty effective
(more so when they’re individual)
• Find that providing inputs has a small positive impact
(about a third the size of the impact of bonuses)
despite being more expensive than the bonuses
Methods Project
• Project on research methods
– Due on Wednesday, May 29 (11:59pm) emailed to
me (mkuhn@uoregon.edu)
– Worth 20% of the overall grade
– Can work in groups of up to 6
Methods Project
• Basic Idea
– Economists clearly have an axe to grind with other
disciplines about research methods
– It’s my job to indoctrinate you
Methods Project
• Basic Idea
– Labor economists have taken over many other
fields because these methods are such an
improvement over the epidemiology approach
– Having a good understanding of when you can
conclude that X causes Y is valuable regardless of
what you do in life. Not just for your career (think
of the health example).
Methods Project
• Basic Idea
– Step 1: Go out and find a piece of news written
about research that you think fails to deal with
causality correctly.
• This isn’t actually that hard: I heard about the Facebook
makes you depressed thing just by listening to the news
on NPR
Methods Project
• Source suggestions (not limitations):
• NPR’s syndicated programs (RadioLab, TED Talks,
Marketplace (especially the “Science” Desk)
• Blog-post style news (Huffington Post, Drudge Report,
BuzzFeed)
• CNN Health or Science
• Links your parents post to you on Facebook
• Pretty much anything in Epidemiology (nutrition and
physiology are similar)
• Newspaper sections on “Lifestyle”
Methods Project
• Send me a link to the article that you want to
research and your group members for
approval.
– Try and do this at least a week before the due
date (so I don’t get overwhelmed with them) and
definitely before you do any work (so you don’t
waste your time)
– I’ll basically make sure your choice of article
indicates that you understand the assignment
Methods Project
• If I get an assignment written about an article
that doesn’t have a problem with causality, or
has nothing to do with causality, it will with
99% certainty get a bad grade.
– Last term: I got articles about opinions,
randomized controlled experiments and stuff that
was obviously repurposed from other courses.
Methods Project
• Step 2 – Write About the Article (in 2 parts)
– Explain the article and its causal claim
– Explain your critique of that claim, in terms of
what we’ve talked about (reverse causality,
selection, omitted variable bias)
• Give some specific examples (stories) detailing the
process by which causality could fail (recall the
Facebook example)
Methods Project
• Step 3 – Propose a Better Study (in 2 parts)
– First: the ideal experiment you would run if you
could control everything and didn’t have to worry
about morality
– Next: a feasible approximation of the ideal study
• This could consist of a small modification of the design
or a complete re-vamping
• It could be an experiment, or just use a different
strategy with real data (the same as or different from
the article). Use these lecture slides as a guide.
Methods Project
– So the whole thing (intro, body, conclusion)
should be at most 5 2-spaced pages. It would be
awesome if it were less (for the grader), but below
3 pages, I would start to worry if you can fit a
thorough analysis into that space (but if you are
really confident that you can, that’s totally fine).
– Bottom line: I don’t care if it’s short if it’s good. I
care about it being long either way.
Methods Project
• Grading:
– I don’t really care about style, interested in content,
but good style makes good content better
• How well do you understand the difference between
correlation and causation in the context of the article?
• Is the argument for a failure of causality well-supported and
linked to issues identified in class?
• Does the solution proposed actually get around the
criticism? Is the feasible solution feasible? Does the ideal
solution indicate understanding of the problem?
Methods Project
• Feel free to e-mail me with questions on your
specific article/project whenever you want.
1
Effects of Marijuana Use
On November 11, 2014 CNN Health released an article on the effects of marijuana use
on the brain. The article comments on a new study from the Proceedings of the National
Academy of Sciences of the United States of America (PNAS); which is a multidisciplinary
scientific journal. The article by CNN claims that this study proves that long term use of the drug
can cause problems with the brain, including lowering someone’s IQ.
The article starts by stating that users of the drug will have worse effects if they started
using at an early age. Their explanation for this is that developing brains that are exposed to the
drug will mature in a different way than normal. The study used 48 individuals who used the
drug between 14 and 30, and 62 individuals who did not use the drug of the same distribution.
All individuals tested by giving a urine sample, getting an MRI scan, and taking an IQ test.
One of the more conclusive results was, on average the marijuana users had 5 IQ points
less than non-users. The researchers state that the most effected part of the brain was the
orbitofrontal cortex, which is responsible for things such as decision making. It is important to
note that the researchers themselves stated that the study does not make a definite
conclusion, but does allow for some important insight.
The last small portion of the article provided a statement from Mason Tvert, who is the
director of communications for the Marijuana Policy Project. His comments emphasized the
comments from the researchers on how no definitive conclusion was made. He then went on to
comment on how these findings still do not support the argument in keeping the drug illegal.
His claim was that even though the drug may possibly have negative side effects, it is still better
than the effects of alcohol.
The largest problem that I find with the article was reverse causality of the claim that
marijuana lowers individuals’ IQ. I think that there is a difference in the users due to the users
choosing to use marijuana or not….
Purchase answer to see full
attachment