On 2014 Dec 18, DAVID ALLISON commented:

Regression to the mean is a concept that does not seem to be known by or intuitive to many scientists. This paper by Love-Osborne et al. provides an example of that. Specifically, the authors state “For the power calculation, we estimated that 50% of the participants in the IG [intervention group] would either decrease or maintain their BMI z-score and that only 25% of the participants in the CG [control group] would.” Yet, given that the subjects studied were “adolescents with BMI ≥ 85%” (i.e., above the 85 percentile) and who had baseline BMI z-scores nearly two standard deviations above the mean, by regression to the mean alone, we would expect subjects’ BMI z-scores to decrease on average and that more than 50% of subjects would have decreased BMI z-scores even without any intervention. Despite this, the randomization makes their study internally valid, but by not considering regression to the mean, the authors’ power analysis makes little sense.

On 2014 Dec 18, DAVID ALLISON commented:

Regression to the mean is a concept that does not seem to be known by or intuitive to many scientists. This paper by Love-Osborne et al. provides an example of that. Specifically, the authors state “For the power calculation, we estimated that 50% of the participants in the IG [intervention group] would either decrease or maintain their BMI z-score and that only 25% of the participants in the CG [control group] would.” Yet, given that the subjects studied were “adolescents with BMI ≥ 85%” (i.e., above the 85 percentile) and who had baseline BMI z-scores nearly two standard deviations above the mean, by regression to the mean alone, we would expect subjects’ BMI z-scores to decrease on average and that more than 50% of subjects would have decreased BMI z-scores even without any intervention. Despite this, the randomization makes their study internally valid, but by not considering regression to the mean, the authors’ power analysis makes little sense.