the answer to this question is when two variables are correlated, we cannot be sure what is causing the correlation.
For example, let's there is a study that found an increase in consumption in tofu lead to an increase in breast cancer.
Even if it's true that those two really correlated (let's just assume it), we wouldn't be able to know why it is correlated without further researches.