1. Brand is very important to them, maybe even more so than actual performance on the field. If you're a blue-blood program, or in one of their favorite conferences, you can do no wrong.
2. They don't seem to hold close losses against you, at least not very much. Especially if you're one of the teams in #1 above, and you lose to another one.
3. They do seem to put a lot of weight on blowouts; they don't buy the whole "just win, baby" thing. They seem to see a close loss as more similar to a win, rather than to a blowout loss. And blowout wins seem to carry an inordinate weight.
4. They don't put much weight on, or even take take account of, home vs road. They don't seem to reward teams that play decent teams on the road, or punish those that avoid playing road games.
5. They've stated in the past that they put a lot of weight on winning one's conference, but the situation in the Big 10 this year will force them to prove it. If they leave out the Wisky/Penn St winner but invite tOSU and maybe even Michigan instead, it would signal that winning your conference doesn't really mean much at all.
In other words, this whole process is extremely subjective, and therefore subject to biases. Introducing a more quantitative, computer-based element might help to balance this out. Which is of course exactly what happened to the BCS in the early years.