4.2.3.1 - Control

Why control matters

Psychologists use control to make a study fair. If an experiment finds a difference between two conditions, the researcher wants to argue that the difference was caused by the independent variable, not by some accidental difference in the way the study was run.

The variables below show what the researcher is trying to isolate and what else could interfere.

The effect of that change is recorded through the dependent variable.

Other influences may also change this outcome.

Reducing those alternative influences strengthens internal validity.

Imagine a researcher investigating whether caffeine improves reaction time. The independent variable is the drink: caffeine or no caffeine. The dependent variable is reaction time in milliseconds. If the caffeine group is tested in the morning and the no-caffeine group is tested after lunch, time of day becomes a problem. Any difference in reaction time might be due to alertness at different times of day rather than caffeine.

Control does not mean making every participant identical. That would be impossible. It means designing the study so that alternative explanations are less likely. A well-controlled experiment makes the comparison between conditions as fair as possible.

The AQA specification names five control ideas:

AO2 scenario: Priya tests whether white noise improves concentration. One group completes a puzzle with white noise playing and another completes it in silence. If the silent group uses an easier puzzle, the study is poorly controlled because puzzle difficulty could explain the result. If both groups receive the same instructions, same puzzle difficulty and same time limit, the researcher can make a fairer comparison.

AO3 judgement: control strengthens cause-and-effect conclusions, but control is never perfect. Participant differences, participants guessing the aim, mood, tiredness and real-world distractions can still affect behaviour. The best exam answers explain which specific threat a control procedure reduces in the scenario.

Random allocation and randomisation

Random allocation and randomisation both use chance, but they are not quite the same thing.

Random allocation is most obviously used in an independent groups design, where different participants take part in different conditions. For example, in a caffeine experiment, each participant's name could be entered into a random number generator so that they are assigned to either the caffeine group or the no-caffeine group.

A workable allocation procedure

For 20 available participants, assign unique numbers 1–20. Use a computer to shuffle those numbers without replacement. Put the first ten in the caffeine condition and the remaining ten in the comparison condition. Everyone is allocated once, the group sizes are equal, and the researcher does not choose the people who appear most alert.

Independent coin tosses are also random, but may give unequal group sizes. Alternating names on a class register is predictable rather than genuinely random. The procedure should be chosen before outcomes are known.

The point is to reduce systematic bias. If the researcher chooses who goes into each group, they might knowingly or unknowingly put more alert, motivated or sporty participants in one condition. Random allocation makes it less likely that one condition contains a particular type of participant.

Random allocation does not guarantee perfectly equal groups, especially in a small sample. By chance, one group may still contain more confident, tired or experienced participants. However, it is usually better than a researcher making the decision because it prevents human choice from systematically favouring one condition.

Random allocation is one type of randomisation. Randomisation is the broader term. A researcher might randomise:

Random sampling is distinct from experimental randomisation. Random sampling is about who gets into the study from the wider population. Randomisation is about how the study is set up after the participants or materials are available. In exams, do not define randomisation as "picking participants randomly" unless the question is specifically about sampling.

AO2 scenario: Dr Khan studies whether revising with music affects recall. If he randomly assigns students to a music or silence group, that is random allocation. If he uses a computer to shuffle the order of words in the recall list, that is randomisation of materials. If he uses a random number generator to select students from a complete Year 12 register, that is random sampling.

Exam trap: random allocation reduces participant-variable bias between groups. It does not control order effects in a repeated measures design by itself, and it does not make a sample representative of the population.

Counterbalancing

Counterbalancing is used when the same participants take part in more than one condition. This usually means a repeated measures design.

Repeated measures can be useful because the same participants are compared with themselves. This controls participant variables. However, it creates a different problem: the order in which conditions are completed may affect the results.

Suppose every participant completes a concentration task first in silence and then with music. If performance is worse with music, the researcher might think music reduces concentration. But participants may simply be more tired in the second condition. If performance is better with music, this might be because participants have practised the task once already.

For two conditions, counterbalancing often uses an AB/BA structure:

If practice improves performance in the second task, both silence and music benefit for some participants. If fatigue harms performance in the second task, both silence and music are harmed for some participants. The aim is not to remove order effects from every individual. The aim is to balance them across conditions so that they do not systematically favour one condition.