[hombre_fatal]

> It is a multivariate situation [...] It could well be that it is not diary or meat but the amount of glucose/fructose one eats which really affects this.

Your missing puzzle piece is the fact that we do multivariate adjustment models to adjust for cofounders.

You might as well ask "How do we know whether it's the cigarettes or the fact that people who don't smoke also tend to exercise and eat vegetables? I guess we can never know!"

[scythe]

It is an epistemic error to say "we did something about X, therefore X is not a problem". There are always characteristics of a population that aren't modeled. This is why randomization is considered so important.