This is an archived version. For the current version, please go to

16.3.5  Example of incorporating a cluster-randomized trial

As an example, consider a cluster-randomized trial that randomized 10 school classrooms with 295 children into an intervention group and 11 classrooms with 330 children into a control group. The numbers of successes among the children, ignoring the clustering, are

Intervention: 63/295

Control: 84/330.

Imagine an intracluster correlation coefficient of 0.02 has been obtained from a reliable external source. The average cluster size in the trial is (295+330)/(10+11) = 29.8. The design effect for the trial as a whole is then 1 + (M – 1) ICC = 1 + (29.8 – 1)×0.02 = 1.576. The effective sample size in the intervention group is 295 / 1.576 = 187.2 and for the control group is 330 / 1.576 = 209.4.


Applying the design effects also to the numbers of events produces the following results:

Intervention: 40.0/187.2

Control: 53.3/209.4.

Once trials have been reduced to their effective sample size, the data may be entered into RevMan as, for example, dichotomous outcomes or continuous outcomes. Results from the example trial may be entered as

Intervention: 40/187

Control: 53/209.