So bowl season is upon us and at Empirical Sports we are excited about the upcoming games. This time of year the talkingheads decide to promote the best conference based upon which conference wins the most bowl games. This, of course, ignores the matchups and the quality of opponents that a given conference faces. For example, if the PAC-10 is favored in 7 of 8 games but only wins 6, the talkingheads given them kudos for going 6 and 2 when really they underperformed. Similarly, if the ACC is only favored in 1 of 7 and they win 3, they overperformed.
So the better question is how does each conference do compared to what is expected. (A very typical statistician-type thought, what did we get and what should we have expected beforehand.) To know what we should expect, we’ll use Sagarin’s PREDICTOR rating (we used the ratings from December 10th) which is a least squares approach with home field advantage and we’ll use the Vegas odds for a given bowl games as of that same date. We’ll assume that the standard deviation of the difference between predicted and actual is about 12 points and following Stern (1991), we’ll assume these differences are roughly a normal distribution (the usual bell shaped curve) with mean 0(zero). Our choice of 12 comes from approximating the root mean squared error at Prediction Tracker. That values seems high to us but as always, data wins. We do the same with the Vegas line with a similar standard deviation again with information from Prediction Tracker (a very useful and informative site by the way). We will only consider the performance of the six BCS conferences. The table below gives a summary.
Expected Conference Performance from Sagarin’s PREDICTOR and the Vegas Line
So the table above has the total number of bowl games that each of these conferences is playing in the first column. The next two columns contain the expected number of wins (from the sum of the probabilities of winning each game) and the total number of games in which each conference is favored (total number for which probability of winning is more than 0.5) based upon the Sagarin rating. We do the same for the Vegas line. The probability of a team winning a game is found by taking the line (for Vegas) or the difference in the ratings (Sagarin) and getting a z-score from a mean of 0 and a standard deviation of 12. For example, the Pitt-UNC games has a line of Pitt -3 and Sagarin says Pitt should be -3.86. For Vegas, that means that the probability that Pitt wins is 0.5987 and for Sagarin the probability that Pitt wins is 0.6261. We do this for all of the bowl games and total these probabilities for each conference.
First, the discrepancy in the number of games favored (Big 12 and Pac 10) is due to the Arizona -Nebraska matchup in the Holiday Bowl. Sagarin’s PREDICTOR likes Nebraska slightly more than Vegas does, enough to make Nebraska a favorite while Vegas has Arizona (-1). Second, as of 12/14 there is not a line on the Temple-UCLA matchup or the Navy-Missouri matchup. I’ve used the Sagarin ratings for each to get the Vegas odds. A couple of notes on other specific games. PREDICTOR likes Virginia Tech (-10) over Tennessee while Vegas has Virginia Tech (-4.5). I don’t know much about either team to suggest why this might be. Vegas seems to also like Northwestern (+7.5) more than PREDICTOR (+12.1) in their matchup with Auburn. This I would suspect is due to the return of Mike Kafka, the Northwestern QB.
Our current assessment of the conferences this year (and we emphasize this year since we believe that conference quality fluctuates from year to year — witness the Big 12 from 2008 to 2009) is that the SEC and the Pac-10 are clearly the best but there’s not much light between them. The Big 12, the ACC and the Big East are next as a group and the Big Ten lags behind them all. We based this on looking at Sagarin’s Central Mean though we prefer to consider a central mean of the PREDICTOR ratings, we have not calculated those explicitly.
So what should we expect from the bowl season.
We should expect each conference to win the number of games in the expected wins columns plus or minus. While it is clear that the Pac-10 and the SEC are the two best conferences this year, they’ve got really favorable matchups in their bowls. I would not be surprised to see them win at least 60% and more likely 70% of their bowls. Less than that many wins is underperforming by these conferences. Similarly, the Big Ten should only win 2 games and the Big 12 only 3 or 4 of their 8. We ought to be assessing how a conference performs relative to what we expected them to do. No one should be surprised by the Big Ten only winning 2 or 3 games and the Big 12 only winning 3 or 4, but it should not be used as evidence that they are worse than we currently think. If the Big Ten wins no bowl games, then they are worse than we think but 2 wins and they are what we think they are.
My hope for the bowl season: TCU crushes Boise State and Alabama crushes Texas and there is more BCS chaos. I don’t like to say this since I love to watch Boise State play, but BCS chaos is the goal.
You can see the assessment of previous years bowl performance relative to expected performance post-bowls here.