Amari Cooper, left, and Marcus Mariota helped their teams meet preseason FPI projections.With the regular season complete and bowl matchups set, it’s time to look back at the 2014 college football season. Just as teams reflect on their seasons to evaluate their performance, ESPN Stats & Information will do the same for its Football Power Index.
Below is a breakdown of how FPI performed throughout the year. Which preseason projections were correct? How accurate was FPI at predicting games? This is designed to be unfiltered and informative, so if you have questions, feel free to ask them in the comments section below and we will do our best to answer.
As a quick reminder, ESPN’s Football Power Index is a forward-looking system designed to order each FBS team by which is most likely to beat an average team on a neutral field. FPI is intended to measure team strength, not evaluate a team’s résumé for the playoff, to best predict future performance. Once team strength is captured, FPI can be used to go through each team’s remaining schedule to produce game and season projections (expected W-L, chance to win conference, chance to win out, etc.).
FPI was improved this season to allow for preseason projections. Factored into the preseason ratings were prior years’ efficiencies (offensive, defensive and special teams), recruiting data, coaching tenure and information on returning starters.
Where FPI was correct in the preseason
• Preseason top 3: The top three teams in preseason FPI -- Florida State, Oregon and Alabama -- performed as expected. Ohio State was ranked sixth until days before the season when Braxton Miller injured his shoulder and returning starter information was adjusted. The Buckeyes dropped to 12th.
• Preseason W-L projections: Team strength must be weighed with the difficulty of a team’s schedule to accurately predict win totals. The teams ranked fourth through sixth in preseason FPI -- UCLA, Auburn and Stanford -- had some of the toughest schedules in the nation. Therefore, FPI predicted that they would lose more than three games on average (and they did). The five teams that FPI projected for the highest win totals were the only five teams in the nation with 12 or more wins. Similarly, the top five teams from Power 5 conferences in projected win total finished the season as the top five teams in the final CFP rankings.
Most Projected Wins
In Preseason According to FPI
Along those same lines, FPI projected that Florida State had a 38 percent chance to enter bowls undefeated and no other FBS team had more than a 10 percent chance to win out. The Seminoles are the only undefeated FBS team. The second-most likely team to go unbeaten in the preseason -- Marshall -- had a great chance to accomplish that feat through 11 games.
• Conference projections: Seven of the 10 preseason FPI favorites to win a conference went on to do so, including four of the six that were given the highest chances. The three FPI favorites that didn’t win were in the Big 12 (Oklahoma), Sun Belt (Louisiana-Lafayette) and American (Houston). In the case of the Big 12, Oklahoma (35 percent) and Baylor (33 percent) were very close in the preseason. The Big Ten is an interesting case. FPI favored Ohio State to win the conference without Miller, despite most of the public picking Michigan State.
Where FPI was incorrect in the preseason
Sun Belt champion Georgia Southern was transitioning to the FBS, and FPI underestimated the explosiveness of its offense, which finished 23rd in offensive efficiency. FPI projected that the Eagles would finish about 4-8, but they ended the year with a 9-3 record.
After finishing 4-8 last year, TCU’s improvement in 2014 was a surprise to many. FPI had the Horned Frogs 36th in its preseason rankings, which was higher than most but still significantly below their current fifth-place ranking. FPI was high on TCU’s defense (seventh in the preseason) but did not envision its offensive improvement after it ranked 99th in offensive efficiency in 2013. Overall, FPI projected that the Horned Frogs would enter bowls with around seven wins, and they have won 11 games.
Georgia Tech lost 11 starters, including its quarterback, from last year's seven-win team. In the preseason, FPI projected that the Yellow Jackets would win about six games and had a 5 percent chance to win the ACC Coastal Division. Instead, Georgia Tech reached 10 wins for the first time since 2009, and the Yellow Jackets seek their first Orange Bowl win since 1951.
[+] EnlargeMichael C. Johnson/USA Today
Michigan was returning 15 starters, including quarterback Devin Gardner, to a team that had four of its six losses come by a combined 11 points last season. The Wolverines were ranked 18th in preseason FPI and were projected to have the 16th-best defense in the country. FPI projected that the Wolverines would have about four more wins than they actually did, marking its biggest miss this year.
Texas Tech was supposed to have a top-10 offense this year, but injuries and turnovers stymied the Red Raiders' air attack. And no Power 5 team finished the regular season with a lower defensive efficiency than Texas Tech. It’s safe to say that preseason FPI whiffed on four-win Texas Tech.
South Carolina was ninth in the preseason Associated Press poll, so FPI was not alone in its overestimation of the Gamecocks. Since South Carolina had a favorable divisional schedule, FPI projected that it had the best chance to win the SEC East and the second-best chance to win the SEC. The Gamecocks finished with three conference wins, four fewer than SEC East champion Missouri.
While the preseason ratings served as the basis for FPI, an important part of the system is that it learns from each game during the season and adjusts appropriately as teams play better or worse than expected. This mechanism allows FPI to be fluid as the season goes on, which improves prediction accuracy from week to week.
The team FPI favored won 77 percent of FBS-only games this season, which is better than the win percentage of the Vegas closing-line favorite. There were 50 games in which FPI and the Vegas line differed on the favorite; FPI went 28-22 (56 percent) in those games, including 17-10 in the final eight weeks.
Win Pct Predicting FBS vs. FBS Games
Interestingly, FPI exceeded expectations in games involving teams that finished the season ranked in the CFP Top 25. Most systems would be expected to correctly predict about 66 percent of such games, but the FPI favorite was 34-12 (74 percent), including 18-2 in the final five weeks.
There were certain teams that FPI had a grasp on and others that baffled the system. There were 10 teams, including Michigan State, Clemson, Washington, Florida State and Texas Tech, for which FPI correctly predicted all of their FBS versus FBS games. Add in another 28 teams for which FPI correctly predicted all but one game and the system had a very good understanding of about a third of the FBS.
Texas Tech is a great example of how FPI adjusts as the season progresses. As noted above, FPI was high on the Red Raiders in the preseason but quickly learned of their flaws and adjusted its in-season projections accordingly. FPI correctly identified the favorite in all 11 of Texas Tech’s FBS games.
On the other end, there were five teams -- Virginia Tech, South Carolina, Akron, Utah and Air Force -- where the FPI favorite lost in more than half of the games. Missouri and Texas A&M also were tough to predict with their fluctuating performances.
We have been able to retroactively apply FPI to the past 10 seasons. Since 2005, FPI has correctly predicted 75 percent of FBS games. It is on pace to have its second-best pick percentage in a season but will finish the year far from its 79.5 correct-pick percentage in 2013.
For those looking for a little bowl advice, FPI projects that Marshall (79 percent), Stanford (79 percent) and Georgia (77 percent) are the most likely teams to win their bowl games, and Navy (60 percent) is the Vegas underdog most likely to win.