Oreopoulos and Salvanes’s recent piece in the JEP begins with ample concessions to the signaling model of education:

One issue is that schooling may help develop skills or it may help signal skills that individuals already have. If those with more schooling also have more inherent abilities (perhaps because schooling for them is easier or more enjoyable), employers can use schooling to predict better candidates. This is especially helpful when desirable worker attributes, like perseverance, discipline, and time management, are not easily observed. The distinction matters because, with the signaling story, the private returns to individuals overestimate the total economic gains (perhaps massively so), whereas in the case where schooling develops skills, the private gains are probably a lower bound (due to externalities).

It is very difficult to disentangle the extent to which returns to schooling are driven more by signaling or skill development mechanisms because both theories generate very similar empirical predictions. Our view of literature is that there is evidence of both (for example, Arcidiacono, Bayer, and Hizmo, 2008).

O&P then make the most thought-provoking point against signaling I’ve seen in a long while:

That said, causal evidence of nonpecuniary returns to schooling tends to support the skill development theory more than it does signaling.  According to the signaling theory, exogenous increases to schooling would affect a person’s ranking, which would matter only to employers (or possibly potential spouses), but it should not affect individual decisions such as whether to smoke, vote, spank, or “live for today.” If schooling affects these decisions, it likely plays more than just a signaling role.

This distinction between “causal effects of education on the way other people treat you” and “causal effects of education on how you behave and what you believe” is valuable.  But every reasonable proponent of the signaling model admits that schooling has some effect on skill development.  So when you look at the effect of education on individual behavior, you have to ask yourself, “Is the effect bigger than expected given my estimate of the skill development/signaling breakdown?” 

If the estimated effect of education on individual behavior is bigger than you expected, then O&S are right; it’s time to put less stock in the signaling model.  But if the estimated effect of education on individual behavior is smaller than you expected, the opposite is true.  And if you want to guard yourself against hindsight bias, you should write down your expectations before you read the article.