2010 Elite Selection Playoff: AV Ranking Bowl Predictions
Editor’s note: Special thanks to ESP and AVR co-creator Vincenzo Siciliano for his effort in the work outlined below. He did the analysis, I just made a few observations and wrote up this summary.
Disclaimer #1: The following is not intended for betting purposes.
Disclaimer #2: The following analysis does not include the Army—Navy game last weekend. It is not expected that excluding this single contest appreciably skews the results.
Making the Predictions
Similar to last year and the year before, the final AV Ranking (AVR) point values will be used to predict the bowl game winners against the spread. For historical and reference purposes the weekly ESP and AVR results (including the top 10 AWP, SOS, TPR, MOV and QWL rankings) are linked here for weeks 14, 13, 12, 11, 10, 9, 8, 7 and 6.
Last season’s predictions were especially strong going 4-1 in the five BCS games and only missing on the Cincinnati—Florida Sugar Bowl contest which had so many extenuating and motivating factors—Urban Meyer’s semi-retirement, Tim Tebow’s last college football game, Brian Kelly leaving Cincinnati for Notre Dame and not coaching in the game, etc.—that any computer prediction of its outcome was mostly useless. Unlike last season, however, this year the AVR predictions have been extended to include all the bowl game matchups.
The Bowl Season Lineup
It’s difficult to believe that there are 35 bowls featuring 70 teams. It seems that the reward of post-season play is diminished when over half of FBS programs receive a bowl bid. The table below lists all 35 bowls starting with the latest and working backwards.
2010 Bowl Season Lineup
[table id=575 /]
Despite the dilution of reward due to the number of games, there are several intriguing matchups.
The BCS National Championship game features two teams in Auburn and Oregon that have bucked the 10-year national championship blueprint. While the Tigers and Ducks feature run-heavy, high-scoring offensive units neither defense is exceptional (although Oregon does have the 10th ranked defensive TPR).
Ohio State and Arkansas face off in a Sugar Bowl that will feature speed on power. The Razorbacks offense features gunslinger Ryan Mallett throwing to a host of fast, talented receivers while the Buckeyes have the traditional Big 10 build of a power football team on both sides of the ball.
Michigan State heads to Orlando to take on the Crimson Tide with hopes of proving the doubters wrong. The Spartans posted the same record as Ohio State and Wisconsin and even beat the latter head-to-head, yet the Badgers and Buckeyes both received BCS game bids while head coach Mark Dantonio’s troops were left out of the mix.
And, finally, the Irish and Hurricanes meet in the Sun Bowl in what is a preview of a three-game series over the next few years. The game is being built as a renewal of the fierce rivalry from the days when Lou Holtz and Jimmy Johnson roamed the sidelines.
Predicting the Winners and Point Deltas
Using the process described below, the final, week 14 AVR computer model rankings were used to predict the winner and point delta of each bowl contest listed in the table above.
A regression was performed with the final AVR results using only the “bowl equivalent” contests from the regular season. These bowl equivalent games are a subset of the regular season matchups that include only those contests where two criteria are met:
- The two participants had a final AVR point value at or above the minimum final AVR point value of all 70 bowl game participants (Middle Tennessee State at 0.365)
- The AVR point delta of the two participants was less than or equal to the maximum final AVR point delta of all 35 bowl game matchups (Connecticut and Oklahoma at 0.275)
The resulting data set constituted 277 regular season games.
The table below lists the results of these predictions for every bowl game. The table includes the Vegas favorite and associated line, the AVR point delta (home minus away), the AVR favorite and associated line, and two confidence columns.
The first confidence column (AVR win confidence) represents the percent chance the AVR will correctly predict the winner of the game based on the statistical variation of the 277 data points described above. The second confidence column (AVR Vegas line confidence) represents the percent chance the AVR predicts the favorite will cover the Vegas spread using the same variation.
2010 AVR Bowl Predictions*
[table id=576 /]
* The Vegas line is based on a snapshot value.
A few noteworthy items:
- As mentioned above, the largest AVR point delta of any bowl game is the Fiesta Bowl matchup between Oklahoma and Connecticut. This at least partially explains some of the outrage of the Huskies receiving an automatic BCS bid while teams like Michigan State and Boise State have been relegated to lower tier bowls. This also happens to be one of the games where the Vegas and AVR predicted lines most closely agree—both give the Sooners (roughly) a 17-point advantage.
- Other games where Vegas and the AVR lines closely match include the Orange Bowl between Stanford and Virginia Tech, the Rose Bowl between the Badgers and Horned Frogs, the Chick-fil-A Bowl between South Carolina and Florida State, the Pinstripe Bowl between Kansas State and Syracuse, the Texas Bowl between Illinois and Baylor, and the Champs Sports Bowl between the West Virginia and North Carolina State. Each of these games had less than a one-point difference in predicted lines.
- Vegas and the AVR agree on 25 of the 33 (75.8 percent) predicted winners. The average line difference in the eight games where there was disagreement was 10 points compared to only 1.5 points when both Vegas and the AVR installed the same favorite. In other words, when the two disagree on the favored squad, they do so by a substantial margin. When they pick the same favorite, however, the disagreement between predicted margin of victory is much more modest.
- Excluding the Fiesta Bowl the average Vegas line in the other four BCS games is 2.9 points while the average AVR line is 3.1 points. Vegas and the AVR agree on two of these games—both predict Auburn to win the national championship and Stanford to best the Hokies. The AVR gives a slight nod to TCU and Ohio State, but the AVR Vegas line confidence is very low for both. Expressed differently, other than the Fiesta Bowl, the BCS games project to be very evenly matched contests.
So Where Did Vegas (or the AVR) Get It Wrong?
Eight of the 35 AVR predicted favorites have a win confidence in excess of 75 percent, i.e. these AVR installed favorites are a lock to win. These games feature a minimum line of 11.2 points and include the following winners (Vegas/AVR line in parentheses):
- Nevada over Boston College (9.5/16.5)
- Miami (OH) over Middle Tennessee State (2/13.8)
- Oklahoma over Connecticut (17/16.9)
- Nebraska over Washington (13/15.6)
- Oklahoma State over Arizona (6/14.9)
- Missouri over Iowa (1/12)
- Air Force over Georgia Tech (2.5/11.7)
- Boise State over Utah (16.5/11.2)
Additionally, five games feature AVR Vegas line confidences in excess of 75 percent, i.e. these AVR installed favorites are a lock to cover the spread (nota bene, in two of these contests there is disagreement between Vegas and the AVR as to which team is the favorite). These five matchups include (Vegas/AVR line in parentheses):
- Miami (OH) over Middle Tennessee State (2/13.8)
- Michigan State over Alabama (-11/4.4)
- Central Florida over Georgia (-7/8.6)
- Missouri over Iowa (1/12)
- Southern Mississippi over Louisville (3.5/7.9)
Vegas believes Bama will win by two scores while the AVR favors Michigan State by just over a field goal. Similarly, Vegas has installed Georgia as a touchdown favorite over Central Florida while the AVR has them reversed at about the same margin of victory. These two games represent the biggest line disagreement among all 35 bowl contests, i.e. something about how Vegas and the AVR have measured these four teams is very different.
The intersection of the two data sets above is where a) Vegas and the AVR agree on the predicted winner and b) the AVR predicts a larger line than Vegas does. This includes Miami (OH) over Middle Tennessee State and Missouri over Iowa. In these contests there is no disagreement about who will win, only by what margin.