[tomrod]

It can't be controlled for if there is no variation in the population. At that point it becomes a fixed effect--a population-specific intercept, as it were--and the only way to net that out would be through a difference-in-difference type study.

[coldtea]

>It can't be controlled for if there is no variation in the population.

You think there are no Tibetans who don't exercise much and don't eat much vegetables either?

[tomrod]

Depends. The external validity may be questionable due to the height of the Tibetan plateau, epigenetic expression that is potentially common in the nation, or any number of other factors. In other words, conditional randomization is not true randomization required.

However, the grandparent comment I understood to be relevant to dietary factors.