Pasteur and parachutes: when statistical process control is better than a randomized controlled trial