1. When classrooms or schools are the unit of randomization, the “effective sample size” is substantially lower than would be if students were the unit of randomization. This is because of the high within-class or within-school correlation (Kish 1965).
2. This is an issue that was of sufficient concern during the study design that an external review of the study design was commissioned to ensure that there was strong professional support for the random assignment design adopted for the study.
3. A methodological experiment was conducted to assess whether using personal data-recording devices increased reporting of sensitive behaviors. It did not have any such effect for the evaluation sample.