Transcript Methodological Digression

Methodological
Digression
Overview
• Clinical Trials
• Regression Analysis
• Structure
• Terminology
Clinical Trials
• Gilliam and Iyengar (pp. 127-137)
• Clinical Trials seek to measure the
impact of some variable on a population
• To do so, the population is divided into
two subgroups who are subject, as
nearly as possible, to identical conditions
except for the variable under review
Clinical Trials
• Then, one group -- the experimental
group -- is subject to the variable under
review
• The other group -- the control group -- is
subject to a placebo
• The “placebo” is meant to mimic the use
of the experimental variable
Clinical Trials
• For example, Gilliam and Iyengar are
trying to investigate whether news
coverage impact public opinion about
crime
• To do so, they manipulated the lead-ins
to the same news story
• In one, the lead in was changed to
indicate that the crime story to follow
related to gang violence
Clinical Trials
•
•
In this case, then
the control group
would be the
viewers seeing an
ostensibly
“objective” version
of the story
The experimental
group would then
be the group
watching the
contrived “gang
story” lead in
Clinical Trials
•
•
Ideally, the subjects are
randomly assigned to each
group, and the
experimenter/s (the
person/s conducting the
experiment) are unaware of
which group is which
If that is the case then we
characterize that as a
double-blind clinical trial.
Clinical Trials
•
•
A trial in this sense is
an experiment, and a
clinical trial means the
experiment involves
human subjects.
A preclinical trial is an
experiment involving
non human subjects.
Regression Analysis
• Regression Analysis is a statistical tool
used to look for relationships between
variables
• The basic idea (skipping a ton of math)
is that we can take a test population
and measure the impact of a different
variables on another variable
Regression Analysis
• Variables come in two varieties:
• The dependent variable is the variable
the experimenter is attempting to
manipulate
• Independent variables are variables
thought to influence the dependent
variable
Regression Analysis
• For example, if we’re trying to determine
the relationship between television news
coverage of crime and attitudes towards
crime, which is the dependent and which
is the independent variable?
Regression Analysis
• Public Opinion (IV) = News Media (Dv)
• But there’s likely lots of variables that
also might influence PO, so before we
can say that news media shapes PO,
we’d want to try to measure the impact of
these other variables and see if our
dependent variable remains the main
variable shaping public opinion
Regression Analysis
•
•
•
•
In which case, our equation would look like:
Dv = ß1 + ß2 + ß3 + e
where ß = dependent variable
e = error
Dv = R ß1 +R ß2 +R ß3 + e
where R = change (“+” or “-”) and is what we
refer to as the correlation coefficient
We can measure not only the direction of
change, but also the intensity of the relationship
between the variables
Regression Analysis
• With Regression Analysis we can
mathematically control for all those
variables and measure the influence of
each on the independent variable
• The R will always be a number between
-1 and +1.
• Influence increases (either positively or
negatively) as the variable approaches
either pole
Regression Analysis
• Note, though, that relationships that get
too close to the poles are probably not
that important (or spurious)
• Also note that because the underlying
mathematics is statistically based and
therefore probabilistic, we can never be
100% certain of our results
Regression Analysis
•
•
•
•
Which raises the question of how certain can we
be that we just haven’t found a fluke?
The variable used to measure the likelihood of
that fluke is traditionally designated by the letter p
Skipping lots of math, we consider something
statistically significant (meaning not an accident)
when p ≤ .05
The f-test is another tool to measure the variance
of our sample and the likelihood that we have a
fluke
Public Opinion
• Public Opinion: Definition
• Measuring Public Opinion
• Survey Design
• Scientific vs. “Unscientific” polls
• Variables to be measured
• Factors Shaping Public Opinion
• Importance of Public Opinion
Definition
• Polling and Politics
• Iraq War
Definition
• Public Opinion: Aggregate of individual
attitudes or beliefs shared by some
portion of the adult population
Measuring Public
Opinion
• Need to add and combine these
individual opinions so that we can then
determine what the public as a whole
believes
• Collect data in a scientifically rigorous
fashion
Survey Design
• Identify target population
• Select Random Sample
• Write Questions
• Conduct Poll
• Analyze Data
Survey Design
• Identify target population
• Who’s opinion are you interested in
measuring?
• Select Random Sample
• Write Questions
• Conduct Poll
• Analyze Data
Survey Design
•Identify target population
•Select Random Sample
• every person in the target population has
an equal and known probability of being
included in the survey
•Write Questions
•Conduct Poll
•Analyze Data
Survey Design
• Identify target population
• Select Random Sample
• Write Questions
• ensure that questions are fair, nonleading, and clear
• Conduct Poll
• Analyze Data
Survey
Design
• Identify target population
• Select Random Sample
• Write Questions
• Conduct Poll
• contact those selected in the random
sample
• mail in, computer, in person
“accidental” contact measures do not
“work”
• Analyze Data
Survey Design
• Identify target population
• Select Random Sample
• Write Questions
• Conduct Poll
• Analyze Data
• Intepret what the numbers mean
Scientific vs
“Unscientific” Polls
• Key is in the random sample
• “random”: every person in the target
population has an equal and known
probability of being included in the
survey
• Allows us to calculate the margin of
error and the confidence interval
Scientific vs
“Unscientific” Polls
• Margin of Error:
How much the sample
reports differ from the total population
• +/- 3.5% to about +/- 6%
• 45% with a 4% margin of error
41%
45%
49%
Scientific vs
“Unscientific” Polls
• Confidence Interval:
How sure we are in
the results
• p should be somewhere in the
.01 to .05 range
Scientific vs
“Unscientific” Polls
• Need to be able to determine how much
your sample differs from the total
population, and how sure you are in the
results
• If no random sample, no way to
determine that
Variables to Measure
• Intensity
• Salience
• Consensus
• Divisiveness
• Change
Strongly
Agree
Agree Disagree Strongly
Disagree
Variables to Measure
• Intensity
• Salience
• Consensus
• Divisiveness
• Change
Yes
No
Don’t
Know
Variables to Measure
• Intensity
• Salience
• Consensus
• Divisiveness
• Change
Agree
Disagree
Variables to Measure
• Intensity
• Salience
• Consensus
• Divisiveness
• Change
Yes
No
Variables to Measure
• Intensity
• Salience
• Consensus
• Divisiveness
• Change
t1
t2
t3
t4
t5