[unimpressive]

>The fact that college graduates get paid more seems adequately explained by the signalling model he presents. How else do you explain the big pay differences for college graduates vs people who dropped out early? They are bigger than you'd expect from just skill/knowledge gain.

Well, one thing to consider is that you can structure things so that you avoid all the hard classes until the end, and then leave without taking them. I found myself in this situation and had to 'boss rush' all the hard STEM classes at once, which wasn't fun and probably usually results in someone just dropping out.

Confirming this isn't what's happening might be one explanation for the sheepskin effect. (Of course, just because it might be doesn't mean it is, I think Occam's Razor favors Caplan here).