Here is a blog post discussing data. After graduation, the sheepskin effect is about 50% of the value of a degree. (Over time, as a worker builds up a track record, this goes down.)

http://econlog.econlib.org/archives/2a012/01/the_present_val...

Unless you believe human capital jumps massively between 116 credits and 120 credits, this is strong evidence in favor of the signalling model.

Another big chunk of the measured returns to education are merely ability bias.

http://econlog.econlib.org/archives/2012/01/correcting_for.h...