Ch. 13

Observational Studies vs. Experiments

Experiments

Experiments impose a treatment on subjects.

Observational Studies have no treatment and just observes subjects (record what has happened or will happen).

A well designed experiment can allow statisticians to infer conclusions based on the results.

Observational Studies

2 types!Retrospective Study – a study to determine

what has happened. Looking at previous conditions.

Prospective Study – a study to determine what will happen. Follow a group of subjects for a period of time and record the variable(s) of consideration.

The individuals/objects on which the experiment is done are called experimental units (if people, called subjects). The specific experimental condition applied is called the treatment.

Explanatory Variable – the treatments (cause)

Response Variable – the variable that is being studied (effect)

More Vocabulary for Experiments

Factor – a variable that is controlled by experimenter.

Levels – different values of factors

Example: An experimenter wants to see the effects of aspirin and exercise on blood pressure. He will administer 0 mg, 400 mg, and 800 mg of aspirin and moderate and light exercise programs. What are the Factors and what are the levels?

Factors: Aspirin and exercise

Levels: 0 mg, 400 mg, 800 mg and light and moderate.

Treatments: combination of all levels and factors in an experiment

Referring back to the experiment on the previous slide, the treatments for that experiment are as follows:

Light exercise and 0 mg of aspirin

Light exercise and 400 mg of aspirin

Light exercise and 800 mg of aspirin

Moderate exercise and 0 mg of aspirin

Moderate exercise and 400 mg of aspirin

Moderate exercise and 800 mg of aspirin

Determine if each is an observational study or experiment. For experiments, identify the factors, levels, response variables, and treatments.

1) Over a 4 month period, among 30 people with bipolar disorder, patients who were given a high dose of omega-fats from fish oil improved more than those given a placebo.

2) In a test of roughly 200 men and women, those with moderately high blood pressure did worse on tests of memory and reaction time than those with normal blood pressure.

The design of an experiment describes the treatments and how the experimental units/subjects are assigned to the treatments. Poorly designed experiments can lead to incorrect conclusions. If experiments are not carefully designed, we cannot see the effects of explanatory variables because they are confounded with other variables in the environment

Confounding: suppose you want to compare laundry detergent A versus detergent B. You wash a bunch of loads using A and B. But you always put A in washer #1 and always put B in #2. Now you're confounded—you don't know if it's the detergent or the washing machine that made one load cleaner than the other.

Poorly designed experiment. How can this experiment be improved?

Principles of Experimental Design

1. Control of the effects of lurking variables on the response. Have a control group that either receives no treatment or a placebo.

2. Randomization, the use of a system to assign subjects to treatments.

3. Replication of the experiment on many subjects to reduce chance variation in the results. Repeat the experiments often!

4. Block Population: Apply treatments to similar strata. Reduce the chance of confounding the results.

Placebo

A placebo is a dummy treatment. For example if an experiment was designed to see if aspirin reduced blood pressure, the placebo would be a sugar pill. You could compare two groups, one taking aspirin and one taking the placebo (sugar pill) to see if there is a difference.

Placebo Effect: Many subjects show or report change even though they are receiving the placebo treatment.

Example #1A researcher wants to determine if regular

exercise reduces the risk of a heart attack? The researcher has recruited 4000 men over 42 who have not had a heart attack and are willing to participate in the study. The researcher has determined that the study will incorporate different levels of exercise, no exercise program, a 3 day a week program (moderate), and a 6 day a week program (heavy). Design a 5-year experiment to answer the question.

Make a diagram (flowchart) of the experimental design.

Blocking (Stratifying) is an experimental design component in which the researcher assumes that there are natural differences between categories within a block (gender, age, weight, etc.) and therefore wants to eliminate the natural variation caused by this possible confounding variable. A block design is used to control the effect of these outside variable(s) by bringing them into the experiment as blocks (Blocks are homogenous groups). Treatments are then randomly assigned within the blocks. Blocks are often the same size as the number of treatments.

Example #2

A researcher wants to determine if a new pill reduces cholesterol. The researcher as found 20 volunteers, 10 male and 10 female. Use a block design, explain how you would conduct this experiment.

Matched Pair Experimental Design

A matched pair is a special case of blocking—experimental units are grouped into blocks of size two. Some pairs are naturally occurring (siblings, husbands/wives, etc.) Sometimes subjects are put into pairs according to like characteristics (two highest cholesterol level subjects paired, then the next two highest, etc.) Sometimes it means pairing on an individual (one person takes two drug treatments in random order, separated by a “washout period”; volunteers wear two different brands of protective gloves, randomly assigned left/right).

A boot manufacturer wants to determine which of 2 types of boots lasts longer. The manufacturer has secured 40 volunteers. Design an experiment to test which pair of boots, A or B, lasts longer.

A firm wishes to test the durability of four tire types that we'll call A, B, C, and D for convenience.  Here are four possible studies they might perform.  In all cases, the cars are to be driven on a track under controlled conditions until its tires are deemed "worn out".  The response variable for each experimental unit (a car) is the number of miles the car drove with the tires.  Each of the first three designs contains at least one serious weakness.  Comment briefly on them.  The fourth design is called a blocked design.  State what the blocks are and explain what the advantage is of this design over design number 3.

1.  Four Cadillacs of the same type are purchased new from four dealers.  One gets tire A (i.e., gets outiftted with four type A tires), one gets B, one gets C, and one gets D.

2.  Twelve Cadillacs of the same type are purchased new from four dealers.  Three get tire A, three get B, three get C, and three get D.

3.  Twelve vehicles of different types are randomly selected from a list of many vehicle types and then are randomly allocated into four groups of three.  One group gets tire A, one group gets tire B, one group gets tire C, and one group gets tire D.

4.  Four Cadillacs, four Fords, and four Volkswagens are purchased.  One of each type of car gets tire A, one gets tire B, one gets tire C, and one gets tire D.

Double Blind Experiment – neither the subjects nor the people who have contact with them know which treatment a subject received.

Double Blind experiments are used to eliminate bias.

Single Blind Experiment – either the subjects or the people who record the results do not know which treatment a subject has received (only one of them!)

In Conclusion, what is (are):

1) The differences between observational studies and experiments?

2) The principles of experimental design?

3) Single-Blind and Double-blind?

4) the placebo effect?

5) the differences between a completely randomized design and a randomized block design?