I think the sheepskin effect is strong evidence in favor of the signaling model of education.  Modeled Behavior’s Adam Ozimek’s not so sure:

[W]hat does it tell you about someone when they have invested a lot of
money into college, come very close to collecting the payoff, but then
failed to do so 75% through their senior year? Is the only difference
between this person and someone who finishes college really just the
small amount of extra education and the sheepskin effect? It seems
extremely likely to me that someone who fails to collect their
sheepskin effect is going to be systematically very different from
someone who has, and most of the plausible differences I can imagine
will tend to decrease future wages for the dropout.

In other words:

[P]eople who drop out at 96% have more wrong with them than just having 4%
less education than graduates. If u could cause random drop outs at 96%
completion you would observe a much smaller sheepskin effect.

My reply:

1. If Adam is right, he’s still making a massive concession to education skeptics.  According to the David Card consensus, standard OLS estimates of the return to education are, if anything, underestimates; ability bias is basically a myth.  Adam’s story directly contradicts this consensus.  He’s saying, contrary to most people in this literature, that pushing people with 15 years of school to wrap up their B.A. would have an unusually small payoff for them.

In short, Adam and I both interpret the sheepskin effect as evidence against the orthodox human capital view.  We’re just disagreeing about why the orthodox human capital view is wrong.

2. OK, so why is the orthodox human capital view wrong: ability bias, signaling, or a mix of the two?  Both stories predict that people who fail to finish degrees will, on average, be a lot less productive than people who finish.  The key controversy: In a pure ability story, employers can cheaply and accurately detect a “diamond in the rough” – a student who failed to finish his degree, but nevertheless had all the Right Stuff to do so.  In the pure signaling story, employers must utterly lack this diamond-detection ability; the only way to convince them you have the Right Stuff is to actually finish your degree.

Adam grants that signaling plays some role in the skeepskin effect.  I’ll grant that ability bias plays a role.  But all things considered, signaling seems a lot more important.  Detecting diamonds in the rough isn’t impossible.  But it’s very difficult – especially when employers use credentials to decide whether to interview people in the first place. 

But maybe I’m wrong.  So tell us, dear readers: Where would you be today if external forces prevented you from finishing your degree?