So the best way to run a controlled trial is to maximize the external noise and confounding variables?