You can do correlation studies, and try to control for all the factors that distinguish the people who enter college from the people who don't, to distinguish t

You can do correlation studies, and try to control for all the factors that distinguish the people who enter college from the people who don't, to distinguish that selection effect from the actual benefit of the education. There are studies like that; but since you can never be sure you've controlled for all factors, they're suspect. An artificial experiment would be nice, but the cost, ethics, etc. make it not too practical. The best research uses natural experiments, like when some external factor changes the number of university places available, so that people who would have gone to college in the 2017 cohort don't in 2018 (or vice versa). That gives you the answer for the marginal applicant (i.e., the applicant who was just on the edge of qualifying), though, not the average.