[jlmorton]

> There are techniques available to control for correlations like that.

If we knew all the correlations, we might stand a pretty good chance of controlling for them, but we don't know them. We might hypothesize that drinking 1% milk is associated with higher rates of exercise, and we might go out and measure that against a control group. But what if it's also associated with drinking less alcohol? Or not smoking? Or anything else.

We don't know all of the correlations, because human beings are complex, and we can't hold all other variables static, which is why nutrition science has been so abysmal. Only if we first hypothesize the correlation, or collect the data necessary to make the connection, can we even attempt to control for it.

[pdpi]

> Only if we first hypothesize the correlation, or collect the data necessary to make the connection, can we even attempt to control for it.

If only we had some way to know which confounding factors the original authors accounted for. Oh wait, that's what RTFAing achieves.

[jlmorton]

That's not really the point. They attempted to control for demographics, and for dietary covariates like protein, total dietary fat, fiber, and saturated fat. But that is a tiny sliver of the possibly correlated factors.

Did they control for the possibility that 1% milk drinkers learned to drink 1% milk at a young age, and that population happens to live in areas further away from sources of pollution, even in comparison with other households in the same income quintile? Did they control for blood lead levels in childhood?

The point is that we don't know what the covariates are. We have no way to even approach the problem.

[lukego]

I'm with GP. I'd read the study if they did an experiment - randomly assigned milk to people and watched the impact - but there's too little value in these purely observational studies from nutritionists. The field hasn't grown up yet.