It is relatively straightforward to control for things like "exercise from other activities" and underlying comorbidities in an attempt to compare "like with li

It is relatively straightforward to control for things like "exercise from other activities" and underlying comorbidities in an attempt to compare "like with like" even for non-randomized studies. One should also keep in mind that while correlation =/= causation, showing a strong correlation is a step along the path toward showing causation.