ABSTRACT
The previous chapters presented formulae for the purpose of calculating sample sizes to achieve a desired power level in a trial with nested data structures. In almost all cases the power was shown to depend, among other factors, on the size of the intraclass correlation coefficient. Unfortunately, the value of this model parameter is generally unknown in the design stage of a trial. This causes a vicious circle: a trial is designed and implemented to gain insight in the size of some model parameters, in particular the treatment effect, but to efficiently design a trial, the size of another model parameter, namely the intraclass correlation coefficient, needs to be known beforehand.
