Observational studies vs experiments
| Observational study | Experiment |
|---|---|
| researchers observe and record | researchers apply a treatment |
| can show association | can show cause and effect |
| e.g. compare health of coffee drinkers and non-drinkers | e.g. randomly assign people to drink coffee or not |
Only well-designed experiments can establish causation.
Key principles of experimental design
| Principle | Meaning | Why |
|---|---|---|
| Control | compare with a control group | to see the treatment’s real effect |
| Random assignment | chance decides groups | balances unknown factors |
| Replication | use enough subjects | reduces the role of chance |
| Blinding | subjects (and maybe researchers) don’t know who gets what | prevents expectations from affecting results |
The placebo effect
People often feel better just because they think they’re being treated. A placebo (like a sugar pill) lets researchers separate real effects from expectation.
- Single-blind: subjects don’t know their group.
- Double-blind: neither subjects nor those measuring results know.
Confounding variables
A confounding variable is linked to both the treatment and the outcome.
Example: People who carry lighters have higher rates of lung cancer. Lighters don’t cause cancer; smoking (the confounder) is linked to both. Random assignment helps prevent confounding in experiments.
Design example
Question: Does a new fertilizer increase tomato yield?
- Get 40 similar tomato plants.
- Randomly assign 20 to get the fertilizer (treatment) and 20 to get none (control).
- Keep sunlight, water and soil the same.
- Measure the yield of each plant.
- Compare the mean yields.
Variables
| Term | In the tomato experiment |
|---|---|
| Explanatory (independent) | fertilizer or not |
| Response (dependent) | yield |
| Controlled | sunlight, water, soil |
Other designs
- Matched pairs: pair similar subjects (or test each subject twice) and compare within pairs.
- Block design: group similar subjects (e.g., by age), then randomize within each block.
Ethics
Experiments on people need informed consent, minimal risk, and ethics review. Some questions (like whether smoking causes disease) can’t ethically be studied with experiments; observational studies must be used carefully.
Worked examples
Example 1: Researchers survey students’ sleep and grades. Study type? Observational; it can’t prove sleep causes better grades.
Example 2: Why use random assignment? To balance confounding variables between groups.
Example 3: In a drug trial, neither patients nor doctors know who gets the drug. This is double-blind.
Common mistakes
- Claiming causation from observational studies.
- No control group.
- Letting subjects choose their own group.
Try it yourself
- Is "comparing test scores of students who chose to attend tutoring and those who didn’t" an experiment?
- Name the control group in a test of a new study app.
- What does a placebo help measure?
- Name a possible confounding variable: "Students who eat breakfast get better grades."
Answers: 1) No; it’s observational (students chose). 2) students who use no app (or the old method) 3) the placebo effect, separate from the real effect 4) e.g. family routines, income, sleep habits
Quick recap
- An experiment applies a treatment; an observational study only observes.
- Use a control group, random assignment, replication and blinding (placebo).
- Watch for confounding variables.