Transcript day10 - University of South Carolina

STAT 110 - Section 5
Lecture 10
Professor Hao Wang
University of South Carolina
Spring 2012
Last time: Experiment
explanatory
variable
(e.g.,nicotine
patch)
Causes?
response
variable
(e.g, quit
smoking)
lurking variable
(e.g.,
determination,
family
background)
Solution: Randomized Comparative
Experiment
Chapter 6 – Experiments in the Real World
Experimental Designs
(I) completely randomized – all the experimental
subjects are allocated at
random among all the
treatments
• more than one explanatory variable can be used
• use treatment combinations to conduct the study
What are the effects of repeated exposure to an
advertising message? Effects of TV ads
• Subjects watch 40 minute TV program
including ads for a camera
• Length of the ads: 30sec or 90 secs
• Repetitions of the ads: 1times, 3times or
5times
• Answer questions about attitude towards the
camera, intention to purchase
What are the effects of repeated exposure to an
advertising message? Effects of TV ads
(II) Matched Pairs Design - Subjects are matched to
form pairs, or each subject receives both treatments.
Randomization occurs within each pair.
• Responses for the pairs are compared.
• Not complete randomization
• This allows matching to reduce the effect of
variation among the subjects.
Matched Pairs Design Example 1
• Let’s test the effectiveness of a sunscreen lotion.
• We want to study how well it works.
• The problem with a randomized comparative
experiment is that people have different skin
types and body chemistries.
• So, what’s a better method?
• Test one lotion on one arm and the other lotion
on the other arm of each person.
• Compare the effects on each arm for all subjects.
Matched Pairs Design Example 2
• Some studies have tried to determine how
genetics and environmental factors contribute
to intelligence, aggression or substance
addictions.
• Most of the twin’s studies compare identical
twins, having 100% genetic similarity
(III) Block Design
block – a group of experimental subjects that are
known before the experiment to be similar
in some way that is expected to affect the
response to the treatments
block design – the random assignment of subjects
to treatments is carried out
separately within each block
Block Design
• A separate randomized comparative experiment
is performed for each block.
• Matched pairs are an example of block designs.
• Each pair is a block.
• Blocks are another form of control since they
control the effects of some outside variables by
bringing those variables into the experiments to
form the blocks.
Block Design Example
Which TV commercial is most effective?
• Suppose we are interested in the effect of a weight
loss diet.
• There are a total of 2 treatment combinations (diet
or not) .
• Suppose we have available to us a total of N = 60
individuals to which we are going to apply the different
diets based on the 2 treatments.
• Prior to the experimentation the individuals were divided
into n = 10 homogeneous groups of size 6.
• The grouping was based on factors that may affect the
effects of diet (age, gender, initial weights)
Advantages of Using a Block Design
• Block designs are similar to stratified samples.
• Blocks allow us to draw separate conclusions
about each block.
• Blocks reduce confounding.
• We can include a potential lurking variable in the
design and its effects can now be accounted for.
(We call this a blocking variable.)
• In a randomized block design, where is the
randomization performed?
•
•
•
•
A.
B.
C.
D.
When placing subjects into a block.
When picking the response of interest
When assigning treatments w/in a block.
When picking the blocking variable.
single blind – an experiment is single blind if the
subjects are unaware of the exact
treatment being imposed on them
 controls for subject bias
double blind – an experiment is double blind if the
subjects and the experimenter are
unaware of the exact treatment
being imposed
 controls for subject and
experimenter bias
Example: Blindly Lowering Cholesterol
Which lowers cholesterol more?
Special diet versus drug?
Details:
• Three treatments: drug only, diet and drug, diet with
placebo.
• The 46 volunteers were randomized by a statistician.
• Blinding: researchers and participants both blind as to which
drug (or placebo) people in those two groups were taking.
However, participants and dieticians could not be blind to
what the participants were eating. Lab staff evaluating
cholesterol measurements were blinded to the treatment.
Problems with Subjects
nonadherers – subjects who participate but do not
follow the experimental treatment
refusals – some subjects that we want in our study
may refuse to participate
dropouts – subjects may start in the study and later
dropout
 especially true for experiments that last
over an extended period of time
Group task: An Experiment
A taste test is being conducted to compare Dr. K to
Dr. Pepper. You have 21 individual participants
Discuss with your group:
1) Design a completely randomized experiment
2) Identify lurking variables
3) Design a matched pairs experiment
4) How to blind and how to randomize?
Having each person try both soda’s makes this a:
A – Completely Randomized Design
B – Double Blind
C – Matched Pairs Design
D – Statistically Significant
14 people preferred Dr. K over Dr. Pepper and 7
preferred Dr. Pepper over Dr.K. The percentage
of people who preferred Dr. K was:
A – 14/7 = 2%
B – 14/21 = 66.7%
C – 7/14 = 50%
D – 7/21 = 33.33%
The experiment found that 14 people preferred Dr.
K over Dr. Pepper and 7 preferred Dr. Pepper
over Dr.K, for a total sample size of 21.
Approximately what is the margin of error for
95% confidence?
A – 1/14 = 7.1%
B – 14/21 = 66.7%
C – 1/sqrt(21) = 21.8%
D – 3%
Flipping a coin to see which soda the subject drank
first
A – Would hopefully remove confounding with the
lurking variable of which you tasted first
B – Would hopefully remove confounding with the
lurking variable of preferring things just because
they are biased towards name brands
C – Would hopefully remove confounding with the
explanatory variable of which you tasted first
D – Would hopefully remove confounding with the
explanatory variable of preferring things just
because they are biased towards name brands
The drinker not seeing which drink they were given
made the experiment
A – A Block Design
B – Double Blind
C – Single Blind
D – Randomized
The drinker not seeing which drink they were given
A – Would hopefully remove confounding with the
lurking variable of which you tasted first
B – Would hopefully remove confounding with the
lurking variable of preferring things just because
they are biased towards name brands
C – Would hopefully remove confounding with the
explanatory variable of which you tasted first
D – Would hopefully remove confounding with the
explanatory variable of preferring things just
because they are biased towards name brands