FEI 2016 Primer: Most Average
by Brian Fremeau
It's time to usher in another college football season, and I'm happy to kick things off again at the start of my 10th season with Football Outsiders. I made my first attempt at preseason FEI projections at the start of the 2007 season, and failed spectacularly on a few teams right out of the gate that year. I'd say the 5-2 start I forecasted for Notre Dame in 2007 was just a bit off the mark (the Irish started 1-6), and I didn't boldly predict that every preseason top-10 team would suffer at least one loss over the first seven weeks, even though they most certainly did.
That 2007 season remains one of the wildest in college football history. On an almost weekly basis throughout that year, we were treated to baffling game results. FCS Appalachian State upset Michigan. Stanford upset USC as a 40-point underdog. Boston College (!) and South Florida (!!) were both ranked in the top three of the BCS standings eight weeks into the year. The national championship was claimed by LSU, a team that managed to bounce back from October and November losses to two unranked opponents.
Nothing made sense in 2007. As an analytics guy hoping to continuously to improve the accuracy of his projection model year after year, a small part of me hopes we never see that level of unpredictability again. But the passionate college football fan in me absolutely loves it when chaos takes center stage.
If you're a regular reader of the weekly FEI column here at FO, welcome back. And welcome to new readers as well. My approach here is usually to highlight games from the previous week that had a significant impact on the path to the national championship, and to do so through the lens of my opponent-adjusted possession efficiency data. Virtually everything I do with college football data has its roots in national scoring rates by starting field position. I like to visualize my data whenever possible, so we'll start with a chart.
Click here for larger view and underlying data.
There were nearly 150,000 non-garbage drives in FBS games played over the last nine seasons, and this chart summarizes the results of all of them. It's commonly understood by even the most casual football fans that teams are more likely to score on a drive begun at their opponent's 25-yard line than one started on their own 25-yard line. This chart and the underlying data help us understand how much more likely those scoring rates should be, and to begin to evaluate results based on deviations from those expectations.
The best offenses in the country exceed expected scoring rates, and the worst offenses fall short. Western Kentucky led the nation in points per offensive drive last season (3.62), and Boston College (0.64) finished dead last. On the other side of the ball, the best defenses hold opponents below the expected scoring rates, and the worst let opponent offenses run wild. Alabama led the nation in limiting opponent points per drive (1.02 points per drive), and Kansas allowed opponents to score with ease (3.58).
Points per drive data is easy to digest, but the offensive and defensive efficiency ratings I produce are a bit more nuanced than that. My raw offensive and defensive efficiency ratings take into consideration the value of a team's starting field position and the value of the drive-ending field position that results on scoring and non-scoring drives alike. A successful offensive drive could be one that moves into field goal range, for instance, regardless of the outcome of the kick attempt. A failed chip shot is a failure of the field goal unit, not of the offensive unit. Opponent adjustments are critical as well. Offenses that produce exceptional raw efficiency rates against weak defenses are not as efficient as ones that do so against strong defenses. Ultimately, the chart highlights what an average team against an average opponent is expected to produce. It's a baseline that every team will most likely outperform or underperform against.
Before the rest of the season steers our focus toward the best (and sometimes the worst) teams and units in college football, let's take this moment at the start of the 2016 season to honor the most average. Which teams over the last nine seasons anchor all other teams? When we say, for instance, that an elite team is two standard deviations better than average and an awful team two standard deviations worse, can we be even more precise and name names?
Overall FEI ratings represent how much better or worse than average each team is in managing possession efficiency. Alabama finished on top last year with a .333 rating -- 33.3 percent better than average in maximizing its scoring opportunities and minimizing those of its opponents. Central Florida ranked dead last with a -.266 rating -- 22.6 percent worse than average. In the middle of the pack was Arkansas State with a .000 rating. The Red Wolves went 8-4 against FBS opponents last year but played most of their schedule against below average or worse opponents.
Over the last seven seasons, five teams have finished the year with a .000 FEI rating. Two of the five teams also finished the year with a .500 record against FBS opponents.
|Most Average Teams Since 2007 By Overall FEI Rating|
|Year||Team||FBS Record||FEI Rating|
These are all good candidates for the most average team since 2007, but they weren't all quite as average on offense and defense as they were overall. As with the overall FEI ratings, my Offensive FEI (OFEI) and Defensive FEI (DFEI) ratings represent how much better or worse than average a given unit performs. Instead of representing these ratings in terms of a percentage, the units for OFEI and DFEI are the per-possession scoring values that unit provides.
[ad placeholder 3]
Navy ranked first in OFEI last season, adding 1.51 points per offensive possession better than an average team would have been expected to do against the Midshipmen's opponents and their starting field position. Central Florida had the nation's worst offense according to FEI in 2015, costing its team 1.47 points per offensive possession. Alabama had the nation's best defense last year, generating 1.22 points of scoring value per opponent offensive possession better than average, while Eastern Michigan ranked last, allowing 1.70 points per offensive possession more than average.
From our list of the five most average teams in overall FEI ratings since 2007, only one produced nearly average OFEI and DFEI ratings in the same season. The 2014 Iowa Hawkeyes finished the season with a 6-6 FBS record, a .000 overall FEI rating, a .01 OFEI rating, and a .03 DFEI rating.
Let's unofficially crown the 2014 Iowa Hawkeyes with the retroactive title of most average team of the last decade. Let's also not forget that they turned around the following season and ripped off 12 straight victories and very nearly played their way into a College Football Playoff berth. As we embark on what will surely be another unpredictable college football season, I can't wait to find out which teams will surprise us by exceeding expectations, which ones will fail miserably to meet our projections, and perhaps which teams might unseat 2014 Iowa and claim the title of most average team of the last decade. If you're as excited about the data that defines elite teams, awful teams and average teams alike, you've come to the right place.
FEI 2016 Preseason Projections
The Fremeau Efficiency Index (FEI) is a college football rating system based on opponent-adjusted drive efficiency. Approximately 20,000 possessions are contested annually in FBS vs. FBS games. First-half clock-kills and end-of-game garbage drives and scores are filtered out. Unadjusted game efficiency (GE) is a measure of net success on non-garbage possessions, and opponent adjustments are calculated with special emphasis placed on quality performances against good teams, win or lose.
[ad placeholder 4]
- SOS: Strength of Schedule, measured as the likelihood of an elite team going undefeated against the given team's regular season schedule. Schedule strength data based on FEI ratings and calculated across other dimensions can be explored in this interactive visualization.
- OFEI offense ratings represent value generated per possession adjusted for the strength of opponent defenses faced.
- DFEI defense ratings represent value generated per opponent possession adjusted for the strength of opponent offenses faced.
Preseason projections are a function of program strength and trajectory as well as transition factors including returning production and recruiting success. The weight given to preseason projections is reduced in the FEI formula over the first half of the season until it is eliminated entirely following the results of Week 7.
Note that these projections are somewhat different from the F/+ projections published in Football Outsiders Almanac 2016, which incorporate both FEI and Bill Connelly's S&P+ stats. We'll be publishing the updated F/+ projections on Thursday.
|45||San Diego State||.045||.769||123||-.04||72||.41||24|
|69||North Carolina State||-.003||.197||17||.06||58||-.05||60|
|97||San Jose State||-.092||.630||100||.07||56||-.39||100|
|120||New Mexico State||-.174||.685||110||-.56||114||-.97||127|
4 comments, Last at 01 Sep 2016, 2:53pm
#4 by Brian Fremeau // Sep 01, 2016 - 2:53pm
Something I've been meaning to look at but haven't yet is to examine whether teams within conferences that are clustered together in projections at the beginning of a season tend to decluster as the year goes on.
It isn't just an SEC thing but it could be teams clustered in other ways as well. If FEI projected nearly every ACC team in the 20th-40th ranking positions, for instance, is it more likely that every team will be in that range at season's end, or that those teams will spread out but we just don't know which ones will be better or worse than that yet? I think that's the case, in this example and with the SEC West.