"cohort analysis" Maybe people exercise because they are healthy and can. We need randomized trails to see if it's real.