Baylor is (at least according to us) the big story of the college football season as far as power ratings go. After three games it's clear they could be the best team in the nation; they also might be exposed by a team with either a decent offense or a decent defense (we think it will take both).
So far they've won three games by monster margins, but the teams they've defeated aren't anything special. They beat Wofford 69-3; Buffalo, 70-13; and Louisiana-Monroe, 70-7.
It's pretty clear their offense is great, and their defense at least pretty good. Overall, they rank extremely well in our power ratings so far. They're #2 in our modified pre-season ratings, where they're currently credited 50% for this year's performance and 50% for our pre-season ranking of #28. In the early All-Division Strength power rating, which is completely unbiased, they are #1 by a good (and unsustainable) margin.
But evaluating Baylor by their accomplishments—just by who they've beaten—you can't say they've done that much. Wofford is 2-2 in the FCS; Buffalo is 2-2, and ULM is 2-3 now. So using a BCS-style ranking which doesn't take margin of victory into account, the Bears don't fare so well. In our Success power rating they are just #26.
So the schizm we see is: Baylor should be ranked low in BCS-style ratings, and rated high in margin-of-victory-inclusive power ratings.
But here's the rub: Sagarin's ratings have the Bears just the opposite
RATING W L SCHEDL(RANK) | ELO_CHESS | PREDICTOR
3 Baylor A = 89.46 3 0 54.13( 172) | 98.96 2 | 84.50 16
Baylor is #3 in Sagarin's current rankings, with an ELO_CHESS ranking of #2, and PREDICTOR of #16. ELO_CHESS is his BCS-style rating, while PREDICTOR is based on victory margins.
The Predictor ranking we can sort of explain...almost. Baylor's schedule strength is horrible, and while their margins of victory make up for that, Sagarin diminishes the value of wide margins. And it's still early enough in the season that Baylor's pre-season value is being partially used. Even if they rank #1 in pure terms, that could be averaged with, say, a #31 start to end up around #16 (this is obviously not what Sagarin does, but the general idea is the same).
The ELO ranking of #2 is very puzzling however. Baylor hasn't beaten any good teams, yet they have climbed to #2? That might make sense if they had started at #1, but they didn't start in his top ten (they were probably started close to last year's #18 finish).
The way we see it, an ELO ranking of #16 and a Predictor ranking of #2 would make a lot of sense right now, and match what our power ratings are showing for the Bears.
It almost makes me wonder if Sagarin mis-labelled his outputs this year. What's interesting is that it's hard to tell by looking at other teams. Alabama is #1 in our Success ratings, implying that they've played the toughest schedule so far in terms of wins and losses, but they aren't #1 in Strength. Sagarin has them only #3 in ELO (Success) and #1 in Predictor (Strength). UCLA is #5 in ELO and just #18 in Predictor, where we have them about the same ranking in both Success and Strength. There aren't really any obvious teams that have played well but lost close games that should be an obvious case for disparity in the two types of ratings. We have to assume the columns are labeled correctly and that there is some other explanation.
Clearly, Sagarin's use of pre-season foundation numbers, or his technique of combining basis numbers and this year's results has caused some seemingly unusual effects for Baylor. Interestingly the language he used to always include (about the ratings not being WELL CONNECTED, and therefore BAYESIAN) is not included this year (he also seems to have removed the previous years of ratings from his site).
When the teams are all "well connected," and the ratings unbiased, Baylor might have a profile closer to ours, that is, high in Strength and mediocre in Success. Of course by then, they might have lost a game or done something else to change the relationship between the two types of rating systems.
Comments