well like one example would be if the author's study is predicated on the notion that salaries are optimally allocated, but you suspect that the hidden term rep

well like one example would be if the author's study is predicated on the notion that salaries are optimally allocated, but you suspect that the hidden term representing misallocation of labor is actually quite large. In other words, you have to consider the possibility that degrees are not a perfectly honest signal, and that people are not always paid exactly according to their productivity. You only have to have one flagrantly unprovable assumption to make the whole chain of inference invalid