For example, "X is a college graduate, and is black" doesn't tell you all that much more than "X is a college graduate"

I'm not sure that what you have in mind here is screening, at least in the causal diagrams sense. If I'm not mistaken, learning that someone is a college graduate screens off race for the purpose of predicting the causal effects of college graduation, but it doesn't screen off race for the purpose of predicting causes of college graduation (such as intelligence) and their effects. You're right, though, that even in the latter case learning that someone is a college graduate decreases the size of the update from learning their race. (At least given realistic assumptions. If 99% of cyan people have IQ 80 and 1% have IQ 140, and 99% of magenta people have IQ 79 and 1% have IQ 240, learning that someone is a college graduate suddenly makes it much more informative to learn their race. But that's not the world we live in; it's just to illustrate the statistics.)