Recently, the twin-control design has been used to test causal models (e.g. exercise→happiness, exercise→¬depression/anxiety symptoms, casual sex→depression/suicidal thoughts). The theory is simple. Suppose we do a standard cross-sectional design study and find that X and Y are correlated. Suppose we suspect that X causes Y. Then, if X causes Y, then one would expect to see a relationship within identical twin pairs for X and Y. If the correlation between X and Y is due to shared genetics, then it will not be correlated within identical twin pairs (baring any de novo mutation being responsible for it). If it is found to be correlated within identical twins, then the education model may be true but also some developmental models relying on non-education environmentally caused differences as well as de novo mutation genetic models.

Did anyone do a study like this? I haven’t seen it, but it is quite simple to do. The only thing needed is a dataset with identical twins, educational attainment/years in school and some g proxy. Maybe NLSY? If you know of a dataset, contact me and we will try.