Every year, countless predictions are made about the upcoming college football season. And every year, despite looking like very plausible, reasonable projections, many of them fall flat.
Here we used a preseason model that assigned a numerical rating to every FBS team and used those ratings to project the outcome of every college football game this season before it started. The insights resulting from this model and those projections were plentiful but certainly not 100% accurate. Below are some of the preseason model’s predictions that didn’t quite come to fruition:
Georgia was projected by the preseason model to win 9.6 games and take home the SEC East crown. The Bulldogs were No. 8 in the model’s rankings with a 31.42 rating but, after a disappointing 7-5 regular season, currently check in at No. 44 with a 20.99 rating. They were 4-4 in conference play, finishing in a three-way tie for second in the SEC East with Tennessee and Kentucky, with Florida winning the division for the second straight season.
Georgia was projected in the preseason to be favored in all 12 of its regular season games, although five by single digits — against Florida (projected to win by 4.3 points), North Carolina (projected to win by 6.0), Ole Miss (6.9), Tennessee (8.8) and South Carolina (9.0). The Bulldogs went 2-3 in those games, to go along with one-point defeats to Vanderbilt and Georgia Tech.
The Fighting Irish were No. 5 in the preseason model’s rankings with a 32.65 rating and were projected to win 9.7 games. The model gave them a 28.5% chance to win at least 11 regular season games and projected them to be favored in all 12. Instead, Notre Dame went 4-8 and were No. 64 in the model’s rankings at the end of the regular season with a 16.28 — almost exactly half what their rating was going into the year.
In an article looking at the Irish’s upcoming season, I wrote their “game against USC could decide their Playoff fate. Notre Dame is currently projected to win that game by 4.4 points.” What actually ended up happening was the Irish took a 4-7 record into the Coliseum and lost, 45-27.
The preseason projections had both divisions in the Pac-12 being two-horse races — the North between Washington and Washington State and the South between USC and UCLA. The Huskies beat the Cougars in the Apple Cup to hand Washington State its only loss in conference play this year and punch their ticket to the Pac-12 title game. And USC gave Colorado its only loss in Pac-12 play this season but the Buffaloes finished a game ahead of the Trojans to advance to the conference title game.
The Bruins were No. 23 in the preseason model’s rankings with a 24.16 rating, just two spots behind No. 21 USC. Four months later, UCLA is No. 122 in the model’s rankings with a 2.75 rating and the Trojans are No. 19 with a 30.14 rating. The Bruins, who had the 14th-toughest non-conference schedule in the nation, lost to Texas A&M in overtime and beat BYU by three before going 2-7 in conference play. UCLA was outscored by 2.6 points per game this year and went 1-7 in games decided by single digits.
The Sooners had a 33.81 rating in the preseason model, good for No. 4 in the preseason model’s rankings — ironically, exactly where they currently sit now with a 40.69 rating. Oklahoma was given a 47.9% chance to win the Big 12, which they did, and a 20.7% chance to win at least 11 regular season games, which they fell a victory shy of doing.
The Sooners were projected to beat TCU in their Big 12 opener by 1.2 points and lost to Ohio State at home by 0.4 points, their two toughest games on their schedule, and went on to beat the Horned Frogs, 52-46, but fall to the Buckeyes by 21 points in Norman this September. At 1-2, their Playoff hopes seemed to be dead but they got themselves back in the mix by running the table in conference play. Oklahoma was originally projected to win their last eight regular season games by 17.8 points per game and ended up winning them by 20.4 points per game, finishing No. 7 in the final CFP rankings.
After finishing third in the Heisman voting last year behind winner Derrick Henry and runner-up Christian McCaffrey, the highest finish ever by a Clemson player, I predicted Watson would win the Heisman this year. In 2015, he became the first player to ever pass for 4,000 yards and run for 1,000 more in the same season and had every significant playmaker coming back this year, along with three starting offensive linemen from a year ago. Clemson was the model’s preseason No. 1 team and was projected to win 11.0 games, which it did.
He entered the season with 9-2 odds to win the Heisman, according to Bovada, the second-best odds, behind only Leonard Fournette (4-1). Watson’s 117.48 adjusted pass EPA (Expected Points Added) and 120.49 adjusted total EPA were the 14th-most and 9th-most in the country, respectively, last year. He topped 100 pass and total EPA again this year, with 108.62 pass EPA and 106.85 total EPA in 2016, as he finished second in the Heisman voting behind Lamar Jackson.
Christian Corona is a contributing writer to 247Sports focusing on analytics-oriented college sports content. He is a data analytics consultant based in New York whose college football model went 54.1% picking every FBS game against the spread over the last two months of the 2015 regular season and better than 53.5% six of the last ten weeks of this season. You can reach him at @ChristianC0rona on Twitter.
Source : https://247sports.com/Bolt/What-the-preseason-model-got-wrong-in-2016-50016072