Experimental Design

1

• Experimental Design

2
Terminology

• Subjects
• Treatment
• Factors
• Placebo
• Control Group

3
Advantages of Experiments over Observation

• Evidence for Causation
• Study specific variables
• Controlling effects of lurking variables
• Combined effects of several factors

4
Principles of Experimental Design

• Control the effects of lurking variables on the
  response, most simply by comparing two or more
  treatments.
• Randomizeuse impersonal chance to assign
  subjects to treatments.
• Replicate each treatment on many units to reduce
  chance variation in the results.

5
Statistically Significant

• Precise meaning to statisticians.
• Is there a large enough difference in the
  response variable among groups to conclude that
  the difference is not merely due to chance?

6
Types of Randomization

• Completely Randomized Design
• Matched Pairs Design
• Compare two treatments
• Choose blocks of two units as closely matched as
  possible
• Possible to have the same subject receive two
  different treatments one after the other
• (order may have an effect so randomize that, too)

7
Block Design

• Group of experimental units or subjects that are
  known to be similar in some way that is expected
  to affect the response.
• Random assignment of units to treatments is
  carried out separately within each block.

8
Designing an Experiment

• Goal State what you want to know
• Response Specify the response variable
• Treatments Specify the factor levels and the
  treatments
• Experimental Units Specify subjects
• Experimental Design Control, Randomize,
  Replicate
• Specify any other experiment details.

9
Experiment Diagram
Group 1 8 plants
Treatment 1 No fertilizer
24 tomato plants from a garden store
Compare juiciness and tastiness
Random Assignment
Treatment 2 ½ dose
Group 2 8 plants
Group 3 8 plants
Treatment 3 Full fertilizer
10
Analyzing Your Experiment

• Side-by-side boxplots to compare the treatment
  groups
• Do the differences among treatment groups seem
  meaningful?

11
Cautions about Experimentation

• Confounding
• Lack of a control group
• Bias (systematic favoritism toward one outcome)
• Subjects must receive identical treatments
• Double blind.
• Lack of realism