I ran out of characters to explain my computer poll. It is based off of an ELO rating system with a few additions/adjustments to compensate for the relatively few games per team, limited cross-conference games, and FCS opponents:

• The season is run many times, with ratings being reeled back most of the way to the average rating (1000).

• Performance against the computer’s expected spread is taken into account (to an extent). This means a winning team can still lose rating points, but that’s not very common.

• Conferences get a pseudo-rating based on non-conference games. This is used to adjust expectations for those games.

• All FCS teams are assumed to have a 400 rating. (I’ve tried rating them, but I haven’t figured out how to prevent ratings inflation.)

This algorithm does not know or care about individual players, advanced stats, or historical performance. All it knows about each game is the week, teams playing, and scores. Every team (and conference) starts at 1000 rating.