One could judge the strength of these with a few empirical tests: such as for (2), comparing industries where it is clear that the skills learned in college (or in a particular major) are particularly relevant vs. industries where it is not as clear, and comparing the number of college grads w/ the relevant skill-signals vs. college grads w/o the relevant skill-signals vs. non-college grads; and for (3), looking to industries where signals of pre-existing ability in that industry do not conform to being in college and comparing their rate of hiring grads vs. non-grads. (This would presumably be jobs in sectors where some sort of loosely defined intellectual ability is not as important. These jobs are becoming more scarce due to automation, and in First World countries in particular, but the tests should still be possible.) (1) is harder to test, as it is agnostic, but trying to see how these intuitions conform to those in hiring positions could be informative. Other signals, as mentioned in the comments, probably have their own tests which can be run on them.
One could judge the strength of these with a few empirical tests: such as for (2), comparing industries where it is clear that the skills learned in college (or in a particular major) are particularly relevant vs. industries where it is not as clear, and comparing the number of college grads w/ the relevant skill-signals vs. college grads w/o the relevant skill-signals vs. non-college grads; and for (3), looking to industries where signals of pre-existing ability in that industry do not conform to being in college and comparing their rate of hiring grads vs. non-grads. (This would presumably be jobs in sectors where some sort of loosely defined intellectual ability is not as important. These jobs are becoming more scarce due to automation, and in First World countries in particular, but the tests should still be possible.) (1) is harder to test, as it is agnostic, but trying to see how these intuitions conform to those in hiring positions could be informative. Other signals, as mentioned in the comments, probably have their own tests which can be run on them.