Bama Boys are Already Hating

So, it was claimed on Twitter (@rankingsray) that I bashed Bama because of SoS but still had Clemson at 3. I pointed out that there were 2 reasons: Clemson's SoR is higher overall than Bama's, and Clemson had played 1 more game. Plus, I didn't bash anything. The system ranks you where the system ranks you. Don't like it? Get on the horn to your AD and have him/her get you a tougher schedule!

As I have said before, on Twitter and FB and even here, playing extra games is rewarded under my system. But the Twitter comments got me to thinking that really, that only matters at the end of the season when Bama isn't playing for a conference championship while UGA/LSU is winning one. I haven't simulated the results yet, but a few quick peaks makes me think there could be some large movements even at the end of the season if I don't control for number of games played. Which strikes me as odd. Really, as we add more data the variances should settle down as we regress to the mean, so to speak.

Now, the system itself accounts for your opponents' number of games played in the 2nd level points tier. I explained how I screwed that up in my last post. And the 1st level points are an average of your points earned/games played (1pt for P5 win, 0.5pt for G5 win, -1pt for FCS school), so 1st level points are normalized for games played. BUT...2nd level points get skewed if you've played more games than others, assuming you win those extra games.

I'm going to use (3) Clemson and (16) Alabama here for my example, because all of the human polls have them very close together while I have them pretty far apart.

• Clemson has 0.7727 1st level and 2.3874 2nd level for a total score of 3.1601 in 11 games
• Alabama has 0.8000 1st level and 1.2500 2nd level for a total score of 2.0500 in 10 games

If I remove Clemson's latest win over Wake, the numbers look like this:

• Clemson has 0.7500 1st level and 1.8750 2nd level for a total score of 2.6250 in 10 games
• Alabama has 0.8000 1st level and 1.2500 2nd level for a total score of 2.0500 in 10 games

That's a 3% reduction in 1st level points, but a whopping 21.5% reduction in 2nd level points and a 17% reduction in total score!

I'm thinking I need to normalize the 2nd level points for games played until the conference championships. I'm not 100% sure, need to think about it.

That said, let's look at USC vs Alabama, since I'm getting the most pushback from Bama fans (totally expected, and it's fine - I expect to get blasted and roasted this first season).

• Alabama has 8.0 1st level points - 7 points for 7 P5 wins, 1 point for 2 G5 wins (0.5pt/win for G5 teams). They've played 10 games, so 8.0/10=0.8000 1st level score.
• USC has 6.5 1st level points - 6 points for 6 P5 wins, 0.5 point for 1 G5 win. They've played 11 games, so 6.5/11=0.5909 1st level score.
• Losses do nothing other than add a zero to average against.
• Play an FCS team, lose 1 1st level point (see: Clemson v Wofford)

The 1st level score difference is 0.2091 in favor of the Tide.

2nd level scores is where the difference is made up, and then some. 2nd level scores are calculated as such:

• The spreadsheet assumes a 12 game schedule, 10 points per game for a divisor of 120.
• The divisor gets modified based on how many games you've played and how many games the opponents you beat have played (if you've played 3 games and your opponents have played 3 games, the divisor stands at 120; if one of those teams has only played 2 games, subtract 1 from the divisor to make it 119)
• For each team you beat, add up their 1st level points and then divide by the divisor. 
• Then multiply by 10 to move the decimal. 

Alabama has beaten Duke, S.Carolina, Ole Miss, A&M, Tenn, Arkansas, and Miss St in the Power 5, and New Mexico State & Southern Miss in the Group of 5:

Team / 1st level points

Duke  / 1.5

NM State / -1

S.Carolina / 2

S. Miss / 2

Miss / 1.5

A&M / 4

Tenn / 2.5

Ark / -0.5

MissSt / 3

For 15 2nd level points. 

Bama's divisor is 120 because the total number of games played by their defeated opponents is not less than 10/team and Bama has played 10.

(15/120)*10=1.2500

USC has beaten Stanford, Utah (key win here), Zona, Colorado, AZ State, and Cal in P5, and Fresno St in G5:

Team / 1st level points

Fresno / 0.5

Stanford / 4

Utah / 6

Zona / 2

Colorado / 3.5

AZ St / 2.5

Cal / 2.5

For 21 2nd level points (already we see  that USC has beaten better teams than Bama has beaten)

USC's divisor is 112 because they've played 11 games and their opponents have played an average of 9.86 games for a 1.14 difference times 7 defeated opponents. 1.14x7 teams is 8, so we subtract 8 from 120 to get to 112.

(21/112)*10=1.875

2nd level score difference is 0.6250 in favor of the Trojans.

Bama has a 0.8000 1st level score and a 1.2500 2nd level score for a total of 2.0500

USC has a 0.5909 1st level score and a 1.8750 2nd level score for a total of 2.4659

End of story. Maybe. I still need to think about normalizing for extra games. I've got a really smart friend who can maybe sort this out for us. I'll ask him.

Bama is about to be hurt for playing Western Carolina. This is my one nod to a bias, but it is not against the SEC. It is against any FBS team playing any FCS school. But the Tide can certainly redeem much of their season with a win in the Iron Bowl. That will probably depend a lot on Mac Jones and Bo Nix (breathtaking insight there, I know).

USC plays UCLA. The Bruins are probably upset about the beat down last Saturday, but beating USC won't make them bowl eligible so maybe they play a bunch of backups? USC is already bowl eligible, and 8-4 looks a lot better than 7-5. Slovis seems to be blossoming.

Anyway, I hope that explains the way my rankings work. There is literally no personal input. The numbers are arbitrary (and I'm open to input - is the average P5 team twice as good as the average G5 team?).