Section E: Research design critique
The research design, and why it is an experiment
The outline asks you to name the design "as clearly as possible", with a reason. Here is a full description, followed by an explanation of each part.
Why it is an experiment
An experiment is a study in which the researcher deliberately changes one thing, keeps everything else the same, and measures the effect. Two features make this study an experiment rather than an observational study.
- The researchers manipulated the cause. They decided which children saw an aggressive model, which saw a calm model and which saw none. The children did not choose.
- They assigned children by chance. Within each matched triplet, chance decided which child went to which condition. This is random assignment. It means that, apart from the condition, the groups should differ only by chance.
Because of these two features, the researchers can argue that differences between groups were caused by what the children watched. A survey finding that children who see more aggression at home are more aggressive could not support that causal claim, because many other things differ between families.
Why it is a laboratory experiment
A laboratory experiment takes place in a setting the researchers control. Here the rooms, toys, timing and the model's actions were the same for every child, down to the fixed arrangement of toys in the test room. The children were even taken to a room "set off from the main nursery school building". The alternative would be a field experiment, carried out in a natural setting such as the children's usual classroom, which gives up some control in return for realism.
Why it is between-subjects, with matched groups
In a between-subjects design, also called independent groups, each child takes part in only one condition. No child saw both the aggressive and the calm model. This avoids one child's experience in one condition affecting their behaviour in another.
The risk of a between-subjects design is that the groups differ before the study starts. The researchers reduced this risk by matching: they rated every child's existing aggressiveness, grouped children into triplets of similar scores, and then assigned one child from each triplet to each condition at random. This is a matched-groups design.
Why it is factorial, with a control group
A factorial design studies more than one independent variable at once, so that their combined effects can be seen. The experimental children were divided by three variables, each with two levels:
| Variable | Levels | Manipulated by the researchers? |
|---|---|---|
| Model's behaviour | Aggressive or calm | Yes |
| Sex of the model | Male or female | Yes |
| Sex of the child | Boy or girl | No, this is a participant variable |
Two levels of each of three variables gives 2 × 2 × 2 = 8 groups, with six children in each. The 24 control group children saw no model at all. The control group answers a question the eight groups cannot: compared with seeing nothing, does a calm model reduce aggression, and does an aggressive model increase it?
Strictly speaking, sex of the child was not manipulated, so comparisons between boys and girls are not experimental. Boys and girls differ in many ways at once, so the study can show that they responded differently, but not why.
How aggression was measured
The dependent variable was measured by structured observation. This means trained observers recorded behaviour using categories defined in advance, rather than writing free descriptions. The observers watched through a one-way mirror, so the children did not know they were being observed. This is covert observation. Behaviour was recorded in each five-second interval of the 20-minute session, a method called time sampling. A second observer scored half the children independently, to check inter-rater reliability, the extent to which two observers agree.
Measurement in plain terms
Structured observation turned something messy, a child's play, into numbers that could be compared across groups. The price is that only the behaviour inside the categories was recorded. What the child meant, felt or said in other ways was not part of the data, apart from a few remarks the authors quoted.
What the design can and cannot show
| Question | Can the design answer it? | Why |
|---|---|---|
| Does watching an aggressive adult cause children to copy specific aggressive acts soon afterwards? | Yes | Manipulated cause, random assignment, control group |
| Does the effect last for weeks or months? | No | Behaviour was measured once, minutes after watching |
| Will children act aggressively towards people? | No | The target was a toy designed to be hit |
| Do boys and girls respond differently, and why? | Only the "whether", not the "why" | Sex of the child was not assigned at random |
| Is the effect due to the model's sex or to those two particular adults? | No | There was one model of each sex |
| Does the result hold for children in other cultures or social classes? | No | One nursery school in one US university town |
Try this
A researcher wants to know whether a calm model reduces aggression, so she shows 30 children a calm adult and finds that they are not very aggressive afterwards.
Why can't she conclude that the calm model reduced aggression, and which feature of the Bobo design solves this problem?
Show the reasoning
She has nothing to compare her result with. The children might not have been very aggressive anyway. Without knowing how aggressive similar children would be after seeing no model, she cannot tell whether the calm model changed anything.
The Bobo design solves this with a control group: children matched to the others, frustrated in the same way, but shown no model. Only by comparing the calm-model children with the control children can a "reduction" be measured. This is exactly how the study found that the calm male model reduced aggression and the calm female model did not.