Short on time? This page is the recap. For explanations and worked examples, use the full study guide.
Recap
- An experiment imposes treatments on experimental units; an observational study does not.
- A well-designed experiment has comparison (at least two treatments), random assignment, replication (more than one unit per treatment) and direct control (fixed conditions for known extraneous variables).
- A control group gives a baseline: no treatment, a placebo or the standard treatment.
- Single-blind: only one side does not know the treatments: either the participants or the researchers who meet them. Double-blind: neither side knows.
- Random assignment spreads extraneous variables roughly evenly across groups, which reduces confounding.
Key relationships
| Idea | What to remember |
|---|---|
| Placebo effect | mean response to placebo − mean response to no treatment |
| Extraneous variable | affects the response but is not the explanatory variable studied |
| Confounding variable | linked to the explanatory variable, so you cannot tell which one changed the response |
| Completely randomized design | all units assigned to treatments at random; groups need not be equal |
| Randomized block design | group similar units into blocks, then randomise every treatment within each block |
| Matched pairs design | block design with two treatments: similar pairs, or each unit gets both in random order |
| Purpose of blocking | removes the blocking variable’s variation, so treatment comparisons are more precise |
| Scope of conclusions | random assignment → cause and effect; random selection → whole population; volunteers → units like them |
Assumptions and conventions
- Describe random assignment so someone else could repeat it: label units, say how labels are chosen, how repeats are handled and which labels go to which treatment.
- “Replication” here means more than one unit per treatment, not repeating the whole study.
- Every treatment must appear in every block.
Mistakes to avoid
- Mixing up random assignment and random sampling.
- Calling direct control a “control group”.
- Placebo effect = treatment − placebo. It is placebo − no treatment.
- Blocking without randomising inside the blocks.
- Blocking on a variable unrelated to the response. It adds work but no precision.
- Claiming the results apply to everyone when the units were volunteers.
- Vague random assignment, such as “randomly split them”, with no method.
Quick self-check
- Mean reduction in pain score: placebo 4.5, no treatment 3.0. What is the placebo effect? (4.5 − 3.0 = 1.5 points)
- Forty plants, four treatments, completely randomized and equal groups. How many plants per treatment? (10)
- Volunteers are randomly assigned to two diets. Can you conclude cause and effect? For whom? (Yes, for people similar to the volunteers)
Next: practice questions.