> The problem with this assumption is you can’t keep everything else identical. Population X and Y don’t grow up live and die in exactly identical areas expect for different radiation levels.

For most other things you would do a placebo controlled trial, but it might be hard to get people to sign up for a trial where you're giving them enough radiation to expect to cause a significant number of them to get cancer.

So controlling for confounders is really the only thing left if you want an answer.

And it works pretty well -- at least better than nothing -- if you actually want the answer. The problem with it is that it unfortunately can also produce whatever answer you want from a political perspective by reverse engineering which confounders to control for based on which ones shift toward the desired outcome, so you can't trust somebody else to do it for you without checking their work at a level of detail that basically amounts to redoing the analysis yourself.

> It turns out skin cancer is a solid proxy

Wait, what are you using it as a proxy for?