Check out that link, I think someone else above was trying to explain the same thing they talk about in the above link.
Basically the point is that the 1.1v internal reference will be more accurate than your 5v supply. So you measure the more accurate 1.1 with the less accurate 5v. Then you can see the error from 1.1 and assume it's from a difference in the 5v supply, and correct accordingly