Experiments vs observational studies: what you can conclude
What this lesson covers
- Only a randomised experiment supports a causal claim; an observational study supports an association.
- Random allocation balances confounders; random selection lets results extend to a population.
First, name the study type
- Experiment: the researcher assigns.
- Observational: the researcher measures.
- Always state which one.
An experiment
- An invented netball club, 40 players.
- Explanatory: the one that is changed.
- Response: the one that is measured.
| Netball study | |
|---|---|
| Study | experiment |
| Explanatory | the drink given |
| Response | sprint time |
| Groups | energy drink, placebo |
An observational study
- An invented café, 300 customers.
- Nobody assigned the coffee.
- Coffees a day, then hours of sleep.
| Café study | |
|---|---|
| Study | observational |
| Explanatory | coffees a day |
| Response | hours of sleep |
| Chosen by | the customers |
Confounders
- Linked to both variables.
- Cannot be separated from the coffee.
- So: an association only.
| Confounder | Linked to | And to |
|---|---|---|
| Shift work | more coffee | less sleep |
| Young children | more coffee | less sleep |
Random allocation
- Groups formed at random.
- Confounders tend to balance out.
- A difference can be put down to the drink.
Random selection
- How people get into the study.
- Chosen at random: extend to the club.
- Selection: population. Allocation: cause.
What each design lets you conclude
- Allocation decides cause or association.
- Selection decides population or sample.
| Allocated | Selected | Conclusion |
|---|---|---|
| yes | yes | cause, population |
| yes | no | cause, the sample |
| no | yes | association, population |
| no | no | association, the sample |
A causal claim too far
- "Coffee causes poor sleep": too far.
- No random allocation, confounders remain.
- State the type, claim an association.
| Too far | Justified | |
|---|---|---|
| Study type | not stated | observational |
| Claim | coffee causes it | an association |
| Reason | none given | no random allocation |
Why a placebo
- Looks and tastes the same, nothing active.
- Expecting a boost can change the result.
- Both groups share the expectation.
Single and double blind
- Single: players do not know.
- Double: the timers do not know either.
- Explain it in this study, not just name it.
| Means | |
|---|---|
| Placebo | identical, inactive |
| Single blind | players do not know |
| Double blind | timers do not know too |
Order effects, not "patterns"
- Both treatments, one after the other.
- Fatigue, practice, boredom.
- Name it and say how it changes the response.
| Order effect | Result |
|---|---|
| Fatigue | slower second run |
| Practice | faster second run |
| Boredom | less effort later |
Randomise the order
- Half, at random, have the drink first.
- Order effects hit both drinks alike.
- Reduced, not removed.
| Plan | |
|---|---|
| Half, at random | energy drink first |
| The other half | placebo first |
| Order effects | reduced, not removed |
Where the marks are
- Achieved: study type and variables.
- Merit: why the design choice matters here.
- Excellence: why and how, and the inference.
What to take away
- Allocation decides cause; selection decides who it applies to.
| Experiment | researcher assigns |
| Observational | just observes |
| Allocation | balances confounders |
| Selection | extends to population |
| Cause needs | random allocation |
| Order effects | randomise the order |
Your turn: four questions
- Pause here and answer each one.
- Name the study type every time.
- Full sentences, in context.
| Study | Find | |
|---|---|---|
| 1 | sport v attendance | study type |
| 2 | volunteers, allocated | conclusion |
| 3 | taste test, two drinks | order effect |
| 4 | pain tablet trial | why blind? |
Show answer
1. Observational: students chose whether to play sport. Explanatory: plays sport or not; response: attendance. Only an association can be claimed. 2. An experiment, so a causal claim for these volunteers, but not beyond people like them: they were not randomly selected. 3. The first drink can affect the taste of the second (or tasters tire); randomise which drink comes first for each person. 4. A placebo gives both groups the same expectation; blinding stops knowledge of the treatment changing the pain reported or recorded.
The answers
- Observational: an association only.
- Cause, but only for these volunteers.
- Carry-over taste: randomise the order.
- Placebo and blinding even out expectations.
Is your child on track?
Take the free placement test