AFAIK the Erdos number was originally defined based on peer reviewed publications and people who maintain lists still do so on that basis. Mathematicians publish lots of collaborative papers. Erdos was a graph theorist and unusually widely collaborated, the idea of making a graph of mathematical collaborations was obvious enough and defining distance from Erdos made a nice joke, because the number was low for almost everyone who'd get the joke.
But it's supposed to be collaborations on research papers. This may be enough to give everyone with a Erdos number a Bacon number (I don't know if Bacon number is usually defined for professional movie and TV appearances or for any common appearance), but it doesn't work in reverse.