Post on 12-Jan-2016
transcript
Types of Experimental Types of Experimental DesignsDesigns
EQ: How can I make a BETTER EQ: How can I make a BETTER experiment?experiment?
Observational Study: a study based on Observational Study: a study based on data in which no manipulation of data in which no manipulation of factors has been employedfactors has been employed
Experiment
Manipulates factor levels to create treatments
Randomly assigns subjects to the treatments
Compares the responses of the subjects across treatment levels
TerminologyTerminology
Experimental Units: individuals on which the experiment is done
Subjects: Human experimental units
Treatment: specific experimental condition applied to the units
Control group: Group that receives no treatment or a placebo
Placebo: A treatment known to have no effect
Blinding: Subject is unaware of what group they are in
Double Blinding: Researcher and subject are unaware of which treatment is being used.
Factor: An explanatory variable that is manipulated
Response:Variable that is measured in response to the factor
Study or Experiment?
Researchers have linked an increase in the incidence of breast cancer in Italy to dioxin released by an industrial accident in 1976. The study identified 981 women who lived near the site of the accident and were under age 40 at the time. Fifteen of the women had developed breast cancer at an unusually young average age of 45. Medical records showed that they had heightened concentrations of dioxin in their blood and that each tenfold increase in dioxin level was associated with a doubling of the risk of breast cancer.
Observational study
Study or Experiment?
Is diet or exercise effective in combating insomnia? Some believe that cutting out desserts can help alleviate the problem, while others recommend exercise. Forty volunteers suffering from insomnia agreed to participate in a month-long test. Half were randomly assigned to a special no-desserts diet; the others continued desserts as usual. Half of the people in each of these groups were randomly assigned to an exercise program, while the others did not exercise. Those who ate no desserts and engaged in exercise showed the most improvement.
Experiment
Principles of Experimental DesignPrinciples of Experimental DesignControl aspects of the experiment that we know may have an
effect on the response, but that are not the factors being studied.
Randomize to even out effects that we cannot control
Replicate over as many subjects as possible.
Analyzing ExperimentsAnalyzing ExperimentsAspirin Study
Replication:
Control:
Treatment :
Blinding:
Randomization:
Subjects/Units: 1000 male volunteers
Aspirn
Patients not know which pill they are taking
A group will take a placebo pillThe men will be randomly assigned to either the treatment group or placebo group.
Each treatment will be replicated 500 times
Factor: Aspirin Response: Number of heart attacks
Levels: Low dose and none (Placebo)
Randomized Experiment DesignRandomized Experiment DesignAspirin Study
Low dose aspirin
500 men Placebo
Compare number of heart attacks in each group
500 men
1000 volunteers
Groups TreatmentsR
andom
Ass
ignm
ent
CautionsCautionsLack of Blindness
•If the experimenter assessing the responses to treatments If the experimenter assessing the responses to treatments knows which treatment was given, he/she may bias the knows which treatment was given, he/she may bias the resultsresultsif a subject knows that they got the placebo treatment, they if a subject knows that they got the placebo treatment, they may under-report their responsemay under-report their response
Lack of realism
subjects or treatments or setting of an experiment may not subjects or treatments or setting of an experiment may not realistically duplicate the conditions we want to studyrealistically duplicate the conditions we want to study
OptiGro fertilizer claims that with their OptiGro fertilizer claims that with their product you will grow “Juicier, product you will grow “Juicier, tastier” tomatoes. You want to test tastier” tomatoes. You want to test the claim and determine if you need the claim and determine if you need the full dose for these results. How the full dose for these results. How would you design an experiment would you design an experiment involving 24 tomato plants and a involving 24 tomato plants and a panel of judges to judge their panel of judges to judge their tastiness.tastiness.
Optigro ExperimentOptigro Experiment
Replication:
Control:
Treatment and levels:
Blinding:
Randomization:
Subjects/Units: 24 plants
Fertilizer( none, ½, full)
Not needed
Using a group with no fertilizer allows us to compare the effect of the fertilizer
The treatment applied to each plant is random
Each treatment will be replicated 8 times
Factor: Fertilizer Response: Plant Growth
Completely randomized design
24 plants
8 plants
8 plants
8 plants
No fertilizer
½ dose of fertilizer
Full Fertilizer
Compare quality
Sources of variationSources of variation
What if we could only get 12 plants What if we could only get 12 plants for store A and had to buy 12 plants for store A and had to buy 12 plants from another store, store B. Should from another store, store B. Should we consider this in the experiment we consider this in the experiment design?design?
BlockingBlocking
units are separated in to blocks units are separated in to blocks (groups of units that are known (groups of units that are known before the experiment to be similar before the experiment to be similar in some way that is expected to in some way that is expected to affect the response to the affect the response to the treatments, e.g. blocking by gender treatments, e.g. blocking by gender or race); random assignment of units or race); random assignment of units to treatments is carried out to treatments is carried out separately within each blockseparately within each block
BlockingBlocking
12 plantsFrom store A
4 plants
4 plants
4 plants
No fertilizer
½ dose of fertilizer
Full Fertilizer
Compare quality
4 plants
4 plants
4 plants
No fertilizer
½ dose of fertilizer
Full Fertilizer
Compare quality
12 plantsFrom store B
24 plants
Blocking and controlBlocking and control
Blocking increases our controlBlocking increases our control• Eliminated the effect of origin of the Eliminated the effect of origin of the
plants as a cause for the growth plants as a cause for the growth
Matched PairsMatched Pairs
Creat blocks of units that are as Creat blocks of units that are as closely alike as possible and closely alike as possible and randomly select which one receives randomly select which one receives the treatment OR one subject/unit the treatment OR one subject/unit who receives two treatments in who receives two treatments in random order (a person serves as random order (a person serves as their own control) their own control)
A manufacturer of boots plans to conduct an experiment to compare a new method of waterproofing to the current method. The appearance of the boots is not changed by either method. The company recruits 100 volunteers in Seattle, where it rains frequently , to wear the boots as they normally would for 6 months. At the end of the 6 months, the boots will be returned to the company to be evaluated for water damage.
Diagram a completely randomized experiment for this situation
DiagramDiagram
100 Volunteers
Waterproof Boot
Regular boot
Compare wear
50 people
50 people
What are some sources of What are some sources of variation in this experiment?variation in this experiment?
Some people work inside and others Some people work inside and others outsideoutside
Some people may have more active Some people may have more active lifestyles than otherslifestyles than others
Some people may forget to wear Some people may forget to wear their bootstheir boots
Some people may be more or less Some people may be more or less careful if they think they have the careful if they think they have the new bootnew boot
Matched Pairs DesignMatched Pairs Design
100 Volunteers
Randomly choose one foot to wear Waterproof Boot and the other to wearRegular boot.
Compare wear