Good experimental design rests on comparison, random assignment, control and replication.
Step 1: Let's Learn
Read it, or press Listen and follow the words.
Random assignment is the key
It balances every other difference between groups, known and unknown, which is what licenses a causal conclusion.
The control group
A control group, often on a placebo, provides the comparison the treatment is measured against.
Blinding
Subjects blind to their group cannot bias their own response. Double blinding removes the measurers expectations too.
Blocking
Group similar subjects into blocks, then randomise within each. It removes a known source of variation and raises power.
Confounding
A confounding variable changes alongside the treatment, so their effects cannot be separated. Random assignment is what prevents it.
Control, randomise, replicate
Control holds other variables steady, randomisation balances what cannot be controlled, and replication provides enough units to see past chance. All three are required for a sound experiment.
Confounding
A confounding variable affects the response and is associated with the treatment, so their effects cannot be separated. Random assignment is what makes confounding unlikely, which is the whole reason for it.
Blinding and placebo
Subjects who know their treatment respond differently, and so do assessors. Double blinding removes both effects. A placebo group is what makes the comparison meaningful rather than merely descriptive.
Blocking
Grouping similar units and randomising within each block removes a known source of variation. It is the experimental analogue of stratified sampling and it increases the power of the study.
Step 2: Try It Yourself
Tap and try it out.
Treatment
Control
Treatment has the most. It has 6 more than Control.
Step 3: Watch an Example
One step at a time.
Watch Sana Improve a Design
A study tests a training programme on runners of widely differing fitness.
- Step 1
Fitness varies enormously and will swamp any treatment effect if ignored.
Step 4: Your Turn
Practice makes it stick.
The Licence
Problem 1 of 2
What licenses a causal conclusion? 1 random assignment, 2 a large sample.
The Blinding
Problem 2 of 2
Neither subjects nor measurers know the groups. Double blind? 1 yes, 0 no.
Design It Well
1 of 8
A group receiving a fake treatment is called what? 1 control, 2 treatment.
2 of 8
Grouping similar subjects before randomising is called what? 1 blocking, 2 blinding.
3 of 8
Does blocking raise power? 1 yes, 0 no.
4 of 8
Can an observational study alone establish causation? 1 yes, 0 no.
5 of 8
A variable changing alongside the treatment is called what? 1 confounding, 2 blocking.
6 of 8
How many principles of design were listed?
7 of 8
Which features belong to a well-designed experiment?
8 of 8
Subjects improve from believing they were treated. Placebo effect? 1 yes, 0 no.
Step 5: Quick Check
Show what you know.
Question 1 of 2
Grouping similar subjects before randomising within each group is called what? 1 blocking, 2 blinding.
Question 2 of 2
Why does random assignment permit a causal conclusion?
What You Learned
- Comparison, random assignment, control and replication are the principles.
- Random assignment is what licenses a causal conclusion.
- Blocking removes a known source of variation and raises power.