The earnings advantage of college students has been substantial, but graduation rates have barely moved. Why? Bryan offers evidence that these earnings are highly conditional on graduation, and that since expected graduation rates vary many students with low expected graduation rates are not going to be enticed by the premium.
I think this is a great example of differentiating between marginal behavior properly understood and naive assumptions about what the outcomes of marginal behavior should be.
We have really great evidence that students respond to price signals when it comes to making educational investments. This is true broadly, but also for specific majors (like STEM majors). Whenever I raise this point, though, one thing that people often respond with is asking why STEM majors still earn such a high wage premium if people are responsive to prices. Doesn't that indicate a shortage?
Well, no. The law of one price is great, but if you really expect to see an elimination of all human capital premiums you need to assume a homogeneous population, and that is unrealistic. If all you needed to do to succeed in the STEM labor market was walk into and out of a brick building covered with ivy and filled with microscopes for four consecutive years, then there would be no problem. Wage differentials would compensate for the cost of spending four years doing that, but after brushing aside credit constraints and things like that you have relatively easy substitution between, for example, day laborers and scientists.
But that's not the world we live in. There are more detailed ways to talk about this (for example, in the paper that Bryan cites), but the simplest way of putting it is that different occupational labor markets face labor supply curves of varying elasticities. It is harder to bring a STEM worker online than it is to bring a cashier online. So you observe two things in the data:
1. Occupational labor markets that are responsive to price signals, and
2. Potentially big wage differentials, particularly after demand shocks.
We can of course change the shape of these supply curves by changing institutions that produce skill. A big part of why the STEM labor supply curve is inelastic has to do with K-12 education in the United States. Another reason for wage premiums is that we don't do particularly well at producing "middle skill" workers (which has consequences for the college labor market insofar as middle skill workers are substitutes for college educated workers).