It's a well-known fact that college graduates make more money than high school graduates who do not go to college, but the reason is not clear. Bryan Caplan offers a typology where he splits the gap between the income of college graduates and of high school graduates into three parts: human capital, ability bias, and signaling. Caplan also warns economists of education against conflating ability bias with signaling. The (human capital + signaling) total gives the "return on education" for a college degree -- the part that is causally due to getting a college degree. The (human capital) part alone gives the return on education through the channel of improved productivity, whereas the (signaling) part gives the return on education through the channel of being able to convince potential employers that they have higher productivity. The (ability bias) part can be thought of as selection bias for pre-existing ability: people who go on to graduate from college differ from people who do not go to college in terms of their pre-existing abilities (here "pre-existing" does not mean "innate" -- but rather it means what they had before starting college, or what they would have acquired through natural maturing even if they hadn't gone to college) and these people would likely have earned more (compared to the ones who didn't go to college) even if they had chosen not to go to college.

I want to look at the signaling channel more closely. A college degree does send a signal of some sort to potential employers, and stories of college dropouts who achieve great success notwithstanding, many employers, particularly for high-skilled occupations, prefer workers with college degrees even if the degree is not directly related to the work on the job. But why, precisely, does the employer value the degree more? I can categorize the possible explanations into three categories:

1. Causally agnostic signaling: Employers are relying on either their own experience or on the conventional wisdom that people with college degrees are more productive employees than people without. They don't actually care whether this is due to pre-existing productivity or due to human capital acquired in college.
2. Signaling of human capital acquisition: Employers believe that college students acquire valuable human capital from their college experience (whether it's their coursework or other aspects of the college experience). The college degree is a signal of this human capital acquisition. Note that this story has a weak hole, namely, first, that the human capital acquired in colleges is irrelevant for many jobs, and second, that if all one has to signal is human capital, it's easy to signal this by taking a relevant test or battery of tests, as this Wall Street Journal article noted. Note that, in so far as colleges do not provide valuable human capital to students, employers are treating the college degree as an important signal based on a mistaken belief, and at least in principle, the market should punish such beliefs.
3. Signaling of pre-existing ability: Employers believe that people whoget through college have greater ability (intelligence, conscientiousness, conformism, etc.) than people who cannot or choose not to. Note that some of these abilities (such as intelligence and conscientiousness) can be signaled by taking tests. Others, such as conformism, are harder to signal directly, because to signal conformism, one has to signal it for a sufficiently lengthy period of time in the "usual" setting (namely, college), and therefore, status quo bias makes it hard to move away from the equilibrium of requiring college degrees. Note also that some of these "pre-existing abilities" may be acquired and honed in college (for instance, college may teach people better work discipline and make them more conformist). The reason I use "pre-existing ability" is that it's not the primary purpose of college to teach these skills, and it's unclear that colleges in general teach these skills to the majority of their students.

What do you think is the breakdown of signaling between (1), (2), and (3)? Any other thoughts about whether the question is well-conceived, and about alternative formulations of the question?

UPDATE: Lauren Rivera's article on how elite firms hire, which was discussed by Bryan Caplan in an EconLog blog post, is relevant.