Experimental Design

• 3 major components of experimental design.– Control—Block sizes.– Randomization– Replication

Control

• What are we trying to control?– Lurking (extraneous)

variables.• Example: To test which

tires work or wear best on a car, what things to we have to control?

– Tire pressure, weight of car, speed of car, road conditions…the list is endless.

Control--continued

• Example 2: New medicine to help with headaches. They want to compare it with the old.– Extraneous (lurking)

variables?• Gender, body size, age

of participant to name a few.

• Methods of control– Direct control---in car

example have the exact same car for each type of tire and drive the same speed, with the same road, etc. Tough to do!

Control--continued

• Methods of control.– Blocking: You block to

control a known extraneous variable.

• Example: Headache medicine—gender could be a factor. We will separate men and women into 2 blocks

– Then half are given new medicine and half are given the old medicine to compare results.

500 subjects

Men Women

Randomly assign half

To new medicine

Randomly Assign half to old

medicine

RandomlyHalf to newmedicine

RandomlyHalf to old medicine

Control--continued

• Same problem but medicine is for prostate cancer. • Direct control is used in that only men 50-60 are studied.• Blocking is used for more control because they believe

that these medicines work differently with various races.• Let’s say there are three medicines being tested (A,B,C)

they are randomly assigned after the blocks.• NOTE: BLOCKS ARE NOT RANDOM> YOU

CONTROL.

Men age 50-60

in study

Caucasians Blacks Asians Hispanics

A B C A B C A B C A B C

Control—Blocks—matched pair

• Blocks that are size two are called “matched pair.”– Twins are the best

example of a matched pair.

– If two people have the same blood pressure.

– Sometimes the matched pair is the same person.

Three major designs

• Completely randomized design (1 group)

• Block design

• Matched pair—(a form of a block.)

The reality is Blocking is used almost every time. Either you have 1 big block, blocks of your choosing, or blocks of size 2 called matched pairs.

Randomization

• Concept is treatment groups are randomly assigned.

• Randomization, or the theory behind it, is to eliminate selection bias.

• There are many, unlimited ways to randomize.

Other ways to randomize

• Dice• Names in a hat• Random numbers.• Count 1-3 and put in

groups.• Flip coin.

Replication

• There are two meanings of replication.– The First is that there are an adequate

number of samples in your study.– The next is that similar studies looking for the

same relationship done by different parties yield the same result consistently.

Notes

• Double blind– The subjects AND the

evaluators both do not know the treatments given.

– WHY?– Who must know the

treatments?

• Beware of observational studies that are passed on as experiments—see notes.

• AP Alerts:• Is this an observational study or an experiment? Explain.

– Typical answer: It’s an observational study because no treatment was imposed.– Or---It’s an experiment because a treatment was imposed.

• Should this experiment be double blind? Explain.– No, because the evaluation method is fact based (or objective). The subjects should not know which

treatment but the for the evaluator it doesn’t matter.– Or….yes, the subjects and evaluators should not know which treatment was given. The evaluation tool is

subjective (opinion based) so the evaluators shouldn’t know which treatment was given.

• What is the purpose of a control group?– To provide a basis of comparison to a situation where no treatment was given or a placebo.

• Does this mean smoking causes ADHD? Explain.– No, was cannot conclude cause/effect unless we have a well controlled experiment. In this case, this is an

observational study. Cause/effect can not be for sure determined by an observational study. Only a link can be established.

Things to know for AP test• There are three main ways to gather info:

– Survey– Observational study– Experiment– simulation

• Observational study vs. experiment—the difference– Observational study does not impose a treatment, so cause and effect are difficult to

determine.– Observational studies are often used for studies that are immoral or expensive to conduct.

• Example: effects of smoking • Example: effects of living by electrical wires for pregnant women.

• Stratified vs. Blocking– Stratified is the term for a survey when you are trying to get an equal number of people or a

percentage number of people in your survey guaranteed.– Blocking is to control a known extraneous variable in an experiment.