@Aschniedermann in this example, the correlation does not reflect causality, and if we interpret it at face value, and suggest open data decreases reproducibility, we might end up making incorrect policy suggestions. In other examples, the correlation does reflect causality, and we could base sound policy advice on it. To my mind, "just" a correlation does not say much. It could or could not reflect causality, and that is an important distinction.