An important limitation which is often overlooked is that when you "control for" something by entering it into a regression, as you describe, you are only contr

An important limitation which is often overlooked is that when you "control for" something by entering it into a regression, as you describe, you are only controlling for the linear effect of that thing. It seems to me that this problem is totally fatal for large-scale epidemiological studies with many factors, of which many are sure to have nonlinear effects.