STAT 1001 - Introduction to
Statistical Reasoning
Professor Shaw-Hwa Lo
Content
Chapter 1 - Controlled Experiments Chapter 15 - The Binomial Formula
Chapter 2 - Observational Studies Chapter 16 - The Law of Averages
Chapter 3 - Histogram Chapter 17 - The Expected Value and SD
Chapter 4 - The Average and Standard Deviation Chapter 18 - The Normal Approximation
Chapter 5 - The Normal Approximation for Data Chapter 20 - Chance Errors in Sampling
Chapter 6 - Measurement Error Chapter 21 - Accuracy of Percentages
Chapter 7 - Plotting Points and Lines Chapter 23 - Accuracy of Averages
Chapter 8 - Correlation Chapter 24 - A Model for Measurement Error
Chapter 9 - More about Correlation Chapter 26 - Tests of Significance
Chapter 10 - Regression Chapter 27 - More Tests for Averages
Chapter 11 - RMSE of Regression
Chapter 12 - The Regression Line
Chapter 13 - What Are the Chances?
Chapter 14 - More about Chances
Chapter 1 - Controlled Experiments
Back to Table of Content.
A new drug is introduced. How should we design an experiment to
test whether the drug is effective or now. The basic method is
comparison.
To make comparisons, the drug is given to subjects in a treatment
group, which are a group of subjects that received the drug. The
other subjects who do not receive the drugs are called controls.
The subjects are assigned to the two groups, treatment and control,
at random. This type of experiment is called double-blind
experiment.
Next, we take a look at an example.
Example 1:
The first polio epidemic hit the United States in 1916. During the next
40 years polio claimed many hundreds of thousands of lives,
especially hidden. By 1950’s, several vaccines against this disease
had been discovered. One vaccine developed by Jonas Salk seemed
promising. In lab trials, it had proved safe and had caused the
production of antibodies against polio.
Let us take a closer look at this experiment.
The experiment of polio vaccine (developed by Jonas Salk) was conducted by
Public Health Service and the National Foundation for Infantile Paralysis
(NFIP).
Suppose NFIP had just given the vaccine to large numbers of children. If the
incidence of polio in 1954 dropped sharply from 1953, that would seem to
prove the effectiveness of the vaccine.
In 1952, there were about 60,000 cases; and in 1953, there were about half
as many. Low incidence would mean that the polio vaccine was effective. The
polio turned out to be an epidemic disease with incidences vary year to year.
Hence, without randomized trials, we cannot say for sure whether the vaccine
was effective. What should we do?
We need to conduct a controlled experiment.
But before we conduct controlled experiment, we need to think a little
bit more.
This is because we are dealing with a situation that requires us to
give children drugs, which has a lot of moral and ethical concerns.
The subjects were children in the age groups most vulnerable to polio. These
are children in grades 1, 2, and 3.
Two million children were involved.
Half a million children received the vaccine.
One million children did not receive the vaccine.
Half a million children refused vaccination.
The organization only compared the two groups: treatment and control. The
children refused the vaccination are not counted towards the control group. In
this case, we are dealing with treatment with a different size than control
group.
With regarding to the ethical problem: we cannot force children to take the
vaccine. The organization asked the parents for decision. If the parents grant
permission, the children would then be placed in treatment group.
This seemed to lead to another issue. It was known that high-income parents
would more likely consent to treatment than lower-income parents. This
created a bias against the vaccine, because children of high-income parents
are more vulnerable to polio.
This is because polio is a disease of hygiene. A child who lives in less
hygienic surroundings may have picked up mild polio early in their
childrenhood while they are still under the protection of their parents, which
gave them their own [Link] is less likely the case for children with
high-income parents.
The statistical lesson:
The treatment and controls should be as similar as possible. Only
then can we say that the difference between the responses from the
experiment is due to the treatment rather than something else.
If the two groups differ with respect due to some factor other than the
treatment, the effect of this other factor might be confounded with
the effect of treatment. Separating these effects can be difficult, and
confounding is a major source of bias.
The final solution:
The organization ended up only working with children whose parents
have granted the permission to the treatment. Then the control group
is randomly chosen from the children in the above group. In this
scenario, we can ensure to our best knowledge that the procedure is
objective and impartial.
When an impartial chance procedure is used to assign the subjects
to treatment or control, the experiment is said to be randomized
controlled.
Another form of precaution is placebo:
Children in the control group were given an injection of salt dissolved
in water. During the experiment the subjects did not know whether
they were in treatment or in control, so their response was to the
vaccine, not the idea of treatment. It may seem unlikely that subjects
could be protected from polio just by the strength of an idea. While
this is true, researchers have found that hospital patients suffering
from severe postoperative pain have been given a ‘pain killer’ which
was made of a completely neutral substance: about one-third of the
patients experienced prompt relief.
Additional precaution is double-blindedness:
The diagnosticians had to decide whether the children contracted
polio during the experiment. Many forms of polio are hard to
diagnose, and in borderline cases the diagnosticians could have
been affected by knowing whether the child was vaccinated. The
doctors were not told which group the child belonged to. This is
called double blinding: the subjects did not know whether they got
the treatment or the placebo, and neither did those who evaluated
the responses.
So what does the result look like?
Disregard what we discussed above, the organization conducted
experiment that is not completely impartial. Next, we will be looking at
the result where the experiment is conducted with a bias against the
vaccine.
Notice that the experiment is not completely impartial and unbiased.
This is because the treatment group included only children with
parents consented with the vaccine but the control group are mixed
with children with parents would have not consented.
Readings from Table 1:
The rate is much lower for the treatment group than for the control
group, which is a decisive evidence of the effectiveness of the Salk
vaccine.