Experiments and cause-and-effect
Why run an experiment?
- An experiment is the only study that can show cause and effect — that a treatment makes an outcome change.
- The researcher actively applies a treatment and controls the conditions, rather than just watching what happens.
- An observational study only records what already occurs. It can show an association, but not causation, because a confounding variable might explain the link.
The idea of a controlled experiment
- Split the units (people, plants, plots) into a treatment group and a control group — or two different treatments.
- Apply the treatment to one group and compare the outcome between the groups.
- A difference in outcome can be put down to the treatment — but only if the groups were otherwise alike to begin with.
Random allocation is the key
- Randomly allocate units to the groups (coin toss, random numbers, drawing lots).
- Random allocation makes the groups similar on average for every variable — both known and unknown.
- That balances out confounding variables, so any real difference in outcome is caused by the treatment, not by a pre-existing difference between the groups.
- Random allocation (splitting units into groups) is not the same as random sampling (choosing who takes part): an experiment needs allocation to claim cause; sampling decides who you can generalise to.
The four design principles
- Comparison — always compare the treatment group with a control (or another treatment).
- Random allocation — assign units to groups by chance.
- Replication — enough units in each group that the result is not a fluke of one or two.
- Control — hold all other conditions the same for both groups (timing, environment, how you measure).
A gym tests whether a new stretching routine reduces muscle soreness. 40 volunteers are randomly split into two groups: one does the new routine, the other their usual warm-up. Each rates their soreness after training. Identify the treatment, the control, the response variable, and explain why the random split matters.
Treatment and control
The treatment is the new stretching routine; the control is the usual warm-up.
Response variable
The response is the soreness rating after training — the outcome being compared.
Why random allocation matters
Splitting the 40 volunteers at random makes the two groups similar on average for everything else (fitness, age, effort), so a difference in soreness can be attributed to the routine rather than to a pre-existing difference between the groups.
Test yourself
Practice by grade
One question each at Achieved, Merit and Excellence. Have a go, then compare with the model answer.
A researcher records the exercise habits and resting heart rates of 200 adults and finds fitter adults have lower heart rates. Is this an experiment or an observational study? Can she conclude that exercise lowers heart rate?
Explain how randomly allocating units to the treatment and control groups allows an experiment to make a cause-and-effect claim.
A company tests a new keyboard by giving it to staff who volunteer, and keeps the old keyboard for everyone else, then compares typing speed. Identify a flaw that prevents a causal conclusion and describe how to fix it.