FEI Week 9: Game Factors
by Brian Fremeau
Last week we went under the hood of the FEI ratings to understand why FEI is so smitten with the Pac-12 and Utah in particular. As predicted, Utah’s non-blowout loss to USC over the weekend did little to diminish the Utes’ standing in the FEI ratings. Utah has faced six FEI top-30 opponents to date and has two FEI top-10 victories. Stanford has faced five top-30 opponents. All other FEI top-25 teams have faced an average of only 2.4 top-30 opponents each. Four top-25 teams -– Florida State, Baylor, Ohio State, and Oklahoma State –- have faced zero top-30 opponents.
This week I wanted to go even further under the hood and reveal the opponent-adjusted game-by-game data for every team. The philosophy behind the FEI ratings has been published here and has not changed: teams are rewarded for playing well against good teams, win or lose, and they are punished more severely for playing poorly against bad teams than they are rewarded for playing well against bad teams. How does this philosophy apply to the raw data collected and processed each week?
At the game level, the primary unit of measure is Game Efficiency, a measurement of the success of teams maximizing their own scoring opportunities and limiting the success of their opponent scoring opportunities over the course of the non-garbage possessions in the game. In the FEI formula, those raw individual game measures are adjusted for the strength of the opponent faced to produce GFEI, a single-game measure of the overall performance. I discussed GFEI a bit at the end of last season after Alabama posted the best GFEI rating of the last decade in the 42-14 victory over Notre Dame in the national championship game.
The FEI formula does not take a simple average of GFEI ratings, but also applies a relevance factor in accordance with the philosophical approach described above. Generally speaking, a top team’s games against the best opponents it faces receive more weight in the formula than its games against its weakest opposition. If it loses to weaker competition, those games receive more weight. Similarly, the opponent-adjusted offensive efficiency (OFEI) and opponent-adjusted defensive efficiency (DFEI) ratings are also a function of individual game performances and the relevance of those data points.
Collectively, I’m naming these individual game performances FEI Game Factors. Let’s take a closer look at Alabama’s FEI Game Factor profile to understand where the Crimson Tide’s No. 1 overall FEI rating is coming from.
|FEI Game Factors - Alabama Crimson Tide|
|Alabama||1||Virginia Tech||W 35-10||W 35-10||.655||4||1.922||19||-.413||150||.194|
|Alabama||3||Texas A&M||W 49-42||W 49-42||.570||16||1.710||37||-.220||211||.196|
|Alabama||4||Colorado State||W 31-6||W 24-6||.169||261||.025||545||-.369||161||.113|
|Alabama||5||Mississippi||W 25-0||W 25-0||.643||5||1.118||135||-1.564||5||.188|
|Alabama||6||Georgia State||W 45-3||W 38-3||.244||199||.742||257||.239||390||.044|
|Alabama||7||Kentucky||W 48-7||W 34-7||.360||104||.057||536||-.525||119||.089|
|Alabama||8||Arkansas||W 52-0||W 35-0||.431||61||1.222||105||-.420||149||.074|
|Alabama||9||Tennessee||W 45-10||W 42-3||.591||13||1.534||62||-.525||120||.102|
Alabama has played eight FBS games to date and five of those games rank among the top-20 nationally in GFEI. More than half of their overall game performances rank in the 98th percentile. The relevance distribution highlights how much more important the wins over Virginia Tech and Texas A&M are to Alabama’s rating (nearly 20 percent each), and how little importance is given to the win over Georgia State (only 4.4 percent). That game is the least relevant result of the season for any team according to FEI Game Factors.
Alabama has only allowed 26 points since their shootout victory over Texas A&M, and only 19 of those points were allowed in non-garbage time. That’s dominant, but when we adjust for the strength of the opponent faced in those games, only the shutout of Ole Miss ranks among the top 100 DFEI single-game performances. It seems odd that the best Alabama offensive performance according to OFEI was the opening weekend win over Virginia Tech, a game where the Crimson Tide only scored two offensive touchdowns -- both on a short field. But the Hokies have the nation’s No. 1 defense according to FEI, and success of any kind against that unit rates favorably.
A highly-ranked offensive performance does not necessarily indicate a poor defensive performance. In the same game, Virginia Tech posted the 11th best DFEI of the season, one of five DFEI performances by the Hokies ranked among 35 best of the season (96 percentile).
|FEI Game Factors - Virginia Tech Hokies|
|Virginia Tech||1||Alabama||L 10-35||L 10-35||.380||92||.820||232||-1.181||11||.136|
|Virginia Tech||3||East Carolina||W 15-10||W 15-10||.403||77||.883||202||-.998||29||.153|
|Virginia Tech||4||Marshall||W 29-21||W 29-21||.102||328||-.156||615||-.790||58||.126|
|Virginia Tech||5||Georgia Tech||W 17-10||W 17-10||.505||28||.585||311||-1.170||13||.165|
|Virginia Tech||6||North Carolina||W 27-17||W 27-10||.221||215||.259||448||-.946||35||.124|
|Virginia Tech||7||Pittsburgh||W 19-9||W 19-9||.138||285||.396||377||-.545||113||.122|
|Virginia Tech||9||Duke||L 10-13||L 10-13||.198||231||-.505||729||-1.626||3||.175|
These tables are the guts of the FEI ratings. Why does this team rank so high? Why does that team rank so low? The answers are found in the Game Factors, all of which I’ve published on my site and will update weekly for the remainder of the year.
The weekly updates are of particular importance. Game Splits, the offense, defense, special teams, field position and turnovers splits that are recorded in each game are not opponent-adjusted, and therefore do not change when new data is collected. The Game Factors, however, will change each week. If Texas A&M keeps rising in the FEI ratings, Alabama’s GFEI value and the weight of that result will change.
Florida State’s blowout victory over Clemson ranks as the best GFEI performance of the season to date. And that’s with Clemson ranked all the way down at No. 42 in the FEI ratings. If the Tigers finish strong and climb in the FEI ratings, Florida State’s GFEI performance in that game will look more and more impressive (and the Seminoles overall rating will be impacted as a result).
|FEI Game Factors - Florida State Seminoles|
|Florida State||1||Pittsburgh||W 41-13||W 41-13||.680||3||1.829||25||-.071||259||.211|
|Florida State||3||Nevada||W 62-7||W 45-7||.196||233||.154||498||-.626||91||.113|
|Florida State||5||Boston College||W 48-34||W 48-34||.107||320||.842||220||.698||592||.160|
|Florida State||6||Maryland||W 63-0||W 42-0||.384||89||1.938||18||-.226||207||.140|
|Florida State||8||Clemson||W 51-14||W 41-7||.763||1||1.685||40||-.890||45||.243|
|Florida State||9||North Carolina State||W 49-17||W 42-10||.049||372||.619||300||.070||310||.133|
Even without dramatic changes in the ratings of opponents, the relative weight of each game will shift over time. Florida State’s win over Clemson is the most impressive game of the season, and it carries nearly 25 percent of the weight of the Seminoles FEI rating. Louisville’s loss to Central Florida carries nearly 25 percent of the weight for the Cardinals. Baylor’s win over Kansas State is similarly weighted. Individual Game Factors on average account for about 14 to 15 percent of each team's rating. As each team plays more games, the relative weight of past games will decrease simply due to the influx of more data.
Click here for the FEI Game Factors for all 125 FBS teams
I’m sure there will be many surprises in the FEI Game Factors data tables. There may be more questions raised than answered in the tables as well. Let me know.
FEI Week 9 Top 25
[ad placeholder 3]
The Fremeau Efficiency Index (FEI) rewards playing well against good teams, win or lose, and punishes losing to poor teams more harshly than it rewards defeating poor teams. FEI is drive-based and it is specifically engineered to measure the college game. FEI is the opponent-adjusted value of Game Efficiency (GE), a measurement of the success rate of a team scoring and preventing opponent scoring throughout the non-garbage-time possessions of a game. FEI represents a team's efficiency value over average.
- SOS Pvs: Strength of schedule based on the likelihood of an elite team going undefeated against the given team's schedule to date.
- SOS Fut: Strength of schedule based on the likelihood of an elite team going undefeated against the given team's remaining schedule.
- FBS MW: Mean Wins, the average number of games a team with the given FEI rating would be expected to win against its entire schedule.
- FBS RMW: Remaining Mean Wins, the average number of games a team with the given FEI rating would be expected to win against its remaining schedule.
- OFEI: Offensive FEI, the opponent-adjusted efficiency of the given team's offense.
- DFEI: Defensive FEI, the opponent-adjusted efficiency of the given team's defense.
- STE: Special Teams Efficiency, the composite efficiency of the given team's special teams units - field goals, punt returns, kickoff returns, punts, and kickoffs.
- FPA: Field Position Advantage, the share of the value of total starting field position earned by each team against its opponents.
These FEI ratings are a function of results of games played through October 26th. The ratings for all FBS teams can be found here. Program FEI (five-year weighted) ratings and other supplemental drive-based data can be found here.
6 comments, Last at 05 Nov 2013, 7:22pm
#2 by Kal // Oct 30, 2013 - 6:18pm
Also since I"m thinking about it - it seems like GFEI and GW are a really great way to account for some serious injuries or disruptions in a team in a nice, organic (albeit manual) way. For example, you can reduce the weight of Georgia's good scores against Clemson as an indication that it isn't likely going to happen again due to the injuries. Last season you could deprioritize Stanford's results without Kevin Hogan and reweigh based on more current results.
Have you given any thought to doing these kinds of tweaks? Dealing with injuries is one of the big holy grail of advanced stat analysis, and it's pretty exciting to see a system that may be able to handle it almost effortlessly.
#4 by Brian Fremeau // Oct 30, 2013 - 9:57pm
I have given some thought to modifying the weights of the data in a number of ways, and certainly participation and injury data could be applied here. More recent results could also receive more weight.
#6 by Tai (not verified) // Nov 05, 2013 - 7:22pm
Good snap count data for college (like this site offers for the NFL) would be wonderfully helpful in building a good adjustment model to properly weight games earlier in the season, while ultimately trying to assess the current strength of active rosters. That data could also be a great tool for more accurate preseason ratings in many models that currently look at returning starters; I think you'd get a much better picture if you could look at the percentage of prior-year snaps returning.
None of this has much to do with FEI, but this comment made me think about it. Snap counts are a stat I've been fantasizing about having access to for college analysis since before they were even available for the NFL. I know they probably won't show up any time too soon, data collection would be a massive project if they were going to be comprehensive. Still, it would be so great...