In this course, you will develop and test hypotheses about your data. You will learn a variety of statistical tests, as well as strategies to know how to apply the appropriate one to your specific data and question. Using your choice of two powerful statistical software packages (SAS or Python), you will explore ANOVA, Chi-Square, and Pearson correlation analysis. This course will guide you through basic statistical principles to give you the tools to answer questions you have developed. Throughout the course, you will share your progress with others to gain valuable feedback and provide insight to other learners about their work.
A Question of Causation (Used with Permission from Annenberg Learner)
Loading...
From the course by Wesleyan University
Data Analysis Tools
331 ratings
From the lesson
Exploring Statistical Interactions
In this session, we will discuss the basic concept of statistical interaction (also known as moderation). In statistics, moderation occurs when the relationship between two variables depends on a third variable. The effect of a moderating variable is often characterized statistically as an interaction; that is, a third variable that affects the direction and/or strength of the relation between your explanatory (X) and response (Y) variable. Your task will be to test your own research question in the context of one or more potential moderating variables.
Meet the Instructors
Jen RoseResearch Professor
Psychology
Lisa DierkerProfessor
Psychology
[MUSIC]
Funding for this program is provided by Annenberg Learner.
[SOUND]
[BLANK
AUDIO]
[MUSIC]
Hello, I'm Pardis Sabeti and
this is Against All Odds where we make statistics count.
Causation in statistics can be a tricky thing because appearances
can often be deceiving.
Although we might observe a strong association between two variables
that doesn't automatically mean there's a cause and effect relationship.
One of the biggest challenges in attempting to prove causation
comes from hidden factors, they might not be immediately apparent,
they just lurk in the background.
And that's exactly what we call them.
Lurking variables.
Take for example, a study that found people who owned two cars
live longer than people who owned only one car.
Those who owned three cars were even more likely to live a longer life, and so on.
Does this mean buying more cars will allow you to avoid death?
No, of course not.
In this case, the lurking variable is the car buyers affluence.
Richer individuals own more cars and tend to live longer.
Probably because they have access to better medical care and healthier food.
The cars have nothing to do with it.
Of course, sometimes causation seems to be the only reasonable explanation of
the relationship between an explanatory variable and a response variable.
We can see this in the case of smoking and lung cancer.
>> Why do you buy Winston?
>> Believe it or not, there was a time when cartoons sold cigarettes on TV.
>> A Winston tastes good, like a cigarette should.
[SOUND] >> People smoked in hospitals and
smokers didn't give a second thought to the health risks they might be taking.
A far cry from today, when anyone can tell you smoking causes lung cancer.
[SOUND] It's even printed on cigarette packs.
But how can we say for sure that smoking causes lung cancer?
Yes, there's a strong association between the two.
But how do we know there are no lurking variables affecting our conclusions?
The best way to make a case for causation is to conduct an experiment.
An experiment imposes a treatment in order to observe it's effects.
In the case of smoking we could a local hospital and
randomly assign newborn babies to one of two groups.
Those we forced to smoke and those we prevent from smoking.
We could then keep our subjects in isolation and
as they get older compare the cancer rates between the two groups.
The only difference between the groups would be their smoking habits.
Of course, this particular experiment is highly unethical and
logistically impossible.
So if we can't conduct an experiment,
how do we prove a causal relationship between smoking and lung cancer?
The answer to that question was long incoming.
But offers a fascinating look at biostatistical research.
[MUSIC]
Cigarette smoking became increasingly popular in America after World War I.
When cigarettes were handed out to soldiers to boost morale.
Per capita consumption rose from 49 cigarettes in 1900 to
611 cigarettes in 1920.
Along with the rise in smoking came a disturbing rise in lung cancer rates.
[MUSIC]
A handful of doctors began to raise early warnings about the dangers of smoking.
One was Richard Overholt.
A young lung surgeon who noticed that his non-smoking patients
had quicker surgical recoveries and higher survival rates than the smokers.
He tried to get other doctors to encourage their patients to quit.
>> A good share of the doctors were smokers.
They didn't believe this and
it was very difficult to convince people that this was doing damage to them,
because they felt that they were getting pleasure from smoking.
[MUSIC]
>> During the 1930s and 40s,
the popularity of cigarette smoking soared as a new image of the sophisticated,
independent, and fun-loving smoker took hold.
Little thought was given to the risks.
Then the early 1940s, new studies sounded a louder alarm on the dangers of smoking.
[MUSIC]
One of the earliest and most compelling was a retrospective study
conducted by Ernst Wynder and his teacher Evarts Graham.
Subjects were separated into two groups.
Those with cancer, and those without.
Their health status was treated as a categorical variable.
With one of two possible values.
Cancer patient, or non-cancer patient.
Both groups were asked the same questions about their past habits.
Responses were compared to see which habits distinguished people with cancer
from those without.
Smoking stood out.
>> Dr. Wynder then a student.
St. Louis, under Dr. Evarts Graham, came to me and he said,
Camel cigarettes cause cancer of the lung.
I couldn't believe it.
You see what you look for and you look for what you know and
it never occurred to me that cigarettes caused cancer.
So we went to see my patient and then that time I had quite a large practice.
We discovered to our amazement that patients who had cancer of
the lung were 17 times to one as apt to be two pack-a-day smokers.
So here was a fact trying to tell us something.
>> Despite the remarkable discrepancy in smoking habits between the two groups of
patients this so-called retrospective study was not good enough,
because the study looked at past behavior.
Behavior it could not control.
It's possible the the lung cancer was due to any number of lurking variables.
What if, for
example, something in the smokers' genetics predisposed them to smoke.
And also made them more inclined to develop lung cancer.
In this case, it wouldn't be causation but rather a common cause, the smoker's DNA.
Or what if the smokers lived in heavily polluted environments
that might cause lung cancer.
Their smoking habits would be confounded or tangled up with pollution.
Making it impossible to discern the true cause of the disease.
Or perhaps it was a coincidence that their cancer patients
just happened to be smokers.
After all, it was a popular habit at the time.
Common cause, confounding factors, coincidence.
These are all alternative explanations for
a strong association between two factors that don't have a causal relationship.
The next step in solving this epidemiological mystery was setting up
prospective studies.
Doctors Hammond and Horn of the American Cancer Society gave about 200,000 people
a smoking questionnaire and followed them for 4 years.
Unlike a retrospective study, which begins with sick people, cancer patients, and
works backwards to examine their habits.
A prospect of study looks ahead, following healthy people.
Both smokers and nonsmokers.
Forward through time to see which ones develop lung cancer.
It's still not quite an experiment,
since people make their own choices about whether or not to smoke.
Rather than being randomly assigned to take up the habit.
But it brings science one step closer to nailing down evidence of causation
between smoking and lung cancer.
>> The preliminary study published in 1954 caused quite a sensation.
It was the largest study on smoking that had been done, it was the first
perspective study that had been published, and it showed that the people who smoked
cigarettes have a lung cancer rate ten times as high as people who never smoke.
And that the risk goes up proportionately according to the amount that,
number of cigarettes that people smoke per day.
>> This high correlation was consistent with the findings of the retrospective
studies.
But there is still disagreement about how to explain the correlation.
Remember those pesky,
lurking variables could be present making the association only appear strong.
To tackle this problem doctors designed a follow up study of one million people.
The smokers and nonsmokers were matched on such variables as age, geography, and
occupation.
The goal was to make smoking habits the only major difference
between the two groups.
>> We matched them on 19 different variables, so
they were as alike as we could make it.
And when we're finished here with this analysis the death rate of the smokers
was higher than the death rate of the nonsmokers.
>> Even though the two groups were as similar as possible
they were not identical.
Sir Ronald Fisher, known by many as the father of modern statistics,
argued that without a controlled experiment, there is no way to know for
sure whether the lung cancer was due to smoking or to another unknown variable.
So despite the accumulating evidence from dozens of studies, for many scientists and
statisticians, the labeling of causation was still premature.
So they turned to the lab for help.
Throughout the 50s and 60s, biochemical and
animal studies provided compelling supporting evidence.
Researchers painted tar on mouse skin and
rabbit ears, causing cancerous tumors to develop.
They also identified carcinogens in tobacco smoke.
And lab work confirmed another aspect of the prospective studies.
The does response relationship.
By exposing cancers to cigarette smoke, it was possible to determine how higher
numbers of cigarettes smoked is related to higher incidences of lung cancer.
>> So you have carcinogen tobacco smoke.
It is proof to be carcinogenic to animal tissue.
It is highly collated in retrospective and postspective studies in respect to cancer
of the lung in humans and that was more than sufficient to establish tobacco
smoke as a cause of lung cancer and subsequently other cancers in humans.
>> For some, the evidence was sufficient.
But the struggle to prove causation to a nation of smokers was not over.
Rich, powerful tobacco companies did not ignore the growing evidence
that cigarettes posed a serious risk to health.
Instead, they said out to reassure the American public that their brands
were safe.
Through strategic advertising, the featured doctor's endorsements and
claims of being less irritating to consumers' throats.
In 1962, the Surgeon General assembled an advisory committee on smoking and health.
The goal was to take on the question of causation.
Its task was formidable.
To review more than 6,000 studies on smoking.
Based on predetermined criteria the committee concluded
that the association between cigarette smoking and
lung cancer was found in many studies with many different groups of people.
The association was very strong.
Cancer regularly followed smoking in time.
Smoke did contain cancer causing substances.
And no explanation other than causation was plausible.
>> The committee states on page 61 of the report and I quote,
in view of the continuing and mounting evidence from many sources,
it is a judgement of the committee, that cigarette smoking contributes
substantially to mortality from certain specific diseases,
and to the overall death rate >> In the wake of the Surgeon General's
report, millions of Americans quit smoking.
Congress passed a law requiring warnings on cigarette packs, and
the anti-smoking movement as we know it was born.
The scientific consensus was in, smoking does cause lung cancer.
Now, you can see just how hard it is to prove causation when direct experiments
aren't possible.
Here the non-experimental evidence is about as strong as it gets.
Remember, not all associations can be explained by cause and effect, but
some can.
With enough evidence from enough different sources and
the help of statistics, we can be confident in our conclusions.
The public seems to be getting the message too.
Under 20% of Americans smoke these days.
That's less then half the proportion who did at smoking's peak
back in the 20th century.
I'm Pardis Sabeti for Against All Odds.
Stay tuned.
[MUSIC]
[SOUND]
Funding for
this program
is provided b
y Annenberg Learner.
[SOUND] [BLANK AUDIO] For information
about this and other Annenberg Learner
programs call 1-800-LEARNER and
visit us at www.learner.org.
[BLANK AUDIO]
Explore our Catalog
Join for free and get personalized recommendations, updates and offers.
Coursera provides universal access to the world’s best education,
partnering with top universities and organizations to offer courses online.
© 2019 Coursera Inc. All rights reserved.
