[justin66]

> It would be an equivalent population who weren't asked to change their diets.

Which is actually an ethical quandary when you think about it. A doctor's ethics are really tested when he has to say to some patients "sure, I think you should change your diet and you'll probably die if you don't, but you're part of the control group!"

One more ethical (and still statistically somewhat interesting) way of administering such a study would be to bring in doctors advocating different sorts of diets. The Cleveland Clinic hasn't done such a thing but they probably could: they've got a group that advocates the rigid vegan diet and another that advocates the Mediterranean diet. Not as good as having a control group that continues to subsist on cheeseburgers and coke but it actually can test a few specific ideas, like the vegan claim.

[ghkbrew]

> Which is actually an ethical quandary when you think about it.

You're right, of course. But whether a study is scientifically valid and whether it is ethical are two orthogonal questions. Similar concerns are why most cancer therapy trials end up looking more like A/B testing than classic experimental/control experiments. The point I was trying to make, though, is that you need at least two groups which differ only by a small set of (ideally one) known variables to draw any real conclusions