Tyler Cowen raises a good argument about the signaling model.
Education is good for more than getting a good first job offer right off the bat.
Presumably, relying on education as a signal causes firms to make mistakes when they first hire. Over time, as they correct these mistakes, wage differentials should shrink as workers gain experience. This effect may be difficult to measure in practice. But my guess is that you will not find such an effect. Like Tyler, I am skeptical about the signaling story. I prefer a credentialism story. See point 2 below.
Since as a toddler you can't possibly pay to educate yourself, anyone who does end up footing the bill for your education will probably end up not seeing a positive return on their investment. So if education is purely private, we will probably end up with a lot less human capital investment than is optimal (or that your grown-up self will wish you'd received as a toddler).
1. This may be a good argument for government subsidization of education. However, it does not explain why government should provide education. In the United States, government provision has produced monopolization and, in particular, politically powerful teachers' unions. This makes it less than clear whether spending more on education produces benefits that are not captured by private interests. In our area, we are seeing cutbacks in school programs as counties struggle with under-funded teacher retirement plans. I would like to see the public-goods rationale for that phenomenon.
2. The relationship between education and earnings is not entirely market driven. Within the government sector, pay grade is affected by education levels. Government also has educational credentials that affect many professions, including teaching, health care, and law. That is why I do not look to signaling as the explanation for the returns to schooling. For signals of ability, there are alternatives available. But strict credential requirements leave no alternative.
3. Students are heterogenous. Incomes earned by the average college graduate are not the same as the returns on the margin for a given individual. I will repeat myself here, but I do not see research on education as addressing the issue effectively. Observational studies only tell you about averages. Randomized controlled trials are needed to study marginal effects.
4. Very little of the research on the returns to education is conducted, reviewed, and published within an institutional setting that is scientifically objective. Does the Department of Education wish to publicize negative results? Universities? Left-leaning think tanks? My guess is that the ratio of funding received by education-promoting researchers to that received by education skeptics is more than 100 to 1.