In the design phase of an intervention study, the first issue to be decided is how to assign people to different experimental groups. There are two groups in studies such as the model example above: the intervention group and the control group. One option is to assign whole schools to a particular group, so that all pupils from the same school experience the same conditions. Another option is to assign the pupils from each school to different groups so that each group is represented at each school.
The sample size needs to be determined in the next phase. This should be designed in such a fashion that any difference between the groups has the greatest possible chance of being identified. In the model, the question is whether many schools should be included and a small sample taken from each, or just a few schools with a large sample taken from each. From Moerbeek's research it can be concluded that a trial using many small schools is preferable, if the costs at school level are relatively low and the degree of mutual influence between pupils within the same school is high.
This research was conducted because a few years ago little was known about the number of people in trials using nested data. Using too few people usually means that the effect of an intervention cannot be demonstrated on the basis of statistical data analyses, whereas the inclusion of too many people means unnecessarily high costs. The results of this research will contribute to the improved design of trials using nested data, which will enable trials to produce good results against reasonable costs.
Nested populations
In the social sciences, experiments are often carried out in nested populations. A nested population means that people are nested in groups. It can also mean that repeated tests are carried out among people. The example of an intervention to prevent or reverse unhealthy lifestyles in young people is often carried out in schools. Here, the phrase "nested data structure" is used: the pupils are nested within schools. The results from nested data depend on the interaction between people within the same group. In the abovementioned example, results such as the smoking and drinking behaviour of pupils at the same school depend on mutual influence and communication, the school rules and the behaviour of teachers.