FEI Final Ratings

FEI Final Ratings
FEI Final Ratings
Photo: USA Today Sports Images

by Brian Fremeau

For the last decade, Alabama has held steady as one of the most consistently dominant programs in college football history. It isn't just that they claimed five national championships in that span, or that they played for the championship on two other occasions, or that they have finished the season ranked in the top five, if not top two, nearly every single year under Nick Saban. Their status as the top program in college football from 2009 to present has been supported by what they had never done as much as it has been supported by what they had done. They never suffered a dominant loss. Until Monday night.

Alabama lost by 28 points to Clemson in the College Football Playoff national championship game. They had previously never suffered a loss by more than 14 points under Nick Saban. From the start of the 2015 season through their semifinal victory over Oklahoma on December 29, a span of 58 games, they lost only three times by a combined total of only 21 points. Alabama could be beaten, but they couldn't be beaten down. Or so we thought.

The Clemson Tigers' 44-16 victory over Alabama to wrap up an undefeated championship season was extraordinary on many levels. They were physically dominant in the trenches, though it didn't show up in the box score, as both teams racked up similar stat lines on a total and per-play level. They didn't create more scoring opportunities than the Crimson Tide, they just capitalized on them far more frequently. They played their best when it mattered most, and looked spectacularly impressive doing it. Quarterback Trevor Lawrence made every difficult throw look easy and his receivers made every play that counted, including several one-handed grabs. Clemson's third-down conversion rate was otherwordly, considering their opponent had only allowed only a 31.9 percent third-down conversion rate coming into the game, and had allowed a conversion rate over 60 percent only twice previously in the last ten seasons -- against Johnny Manziel and Texas A&M in 2012 (61.1 percent) and in 2013 (62.5 percent). Clemson went 10-for-15 (66.7 percent) on third downs against the Tide, while facing an average distance to gain of 7.7 yards, and success on those plays turned what very well could have been a competitive down-to-the-wire classic into a blowout.

My numbers don't scrutinize performances at the play level, so a 62-yard completion on third-and-14 in the first quarter that led to Clemson's first offensive touchdown of the game isn't distinguished as an unlikely event in and of itself. Clemson scored a touchdown on a drive that started at its own 25-yard line, and FEI doesn't care whether they achieved that feat methodically, explosively, or otherwise. FEI does take note that Clemson's first touchdown on a 44-yard pick-six interception return, as well as its 46-yard interception return in the second quarter that set up a short field touchdown for the Tigers, are unique plays (specifically, the returns themselves) that don't necessarily represent sustainable possession efficiency. Call both plays opportunistic, and both were certainly great defensive stops against a previously near-flawless Alabama passing attack, but those two plays also inflated the final scoring margin a bit. My game splits calculated Clemson's combined offensive and defensive margin of victory to be 19.6 points -- better than any other performance against the Crimson Tide under Saban, but more along the lines of a three-score victory than a four-score one.

That distinction doesn't matter much, especially against what could and should still be considered an exceptionally formidable opponent. According to my single-game, opponent-adjusted GFEI ratings, Clemson's victory is clearly distinguished as the best of the season, and by a solid margin.

Top 10 GFEI Single Game Performances in 2018
Team Opponent G Final NG Final GFEI
Clemson Alabama W 44-16 W 44-16 1.08
Alabama LSU W 29-0 W 22-0 0.92
Auburn Purdue W 63-14 W 56-7 0.88
Clemson Notre Dame W 30-3 W 30-3 0.84
Georgia Georgia Tech W 45-21 W 45-7 0.78
Missouri Florida W 38-17 W 38-17 0.75
Alabama Oklahoma W 45-34 W 45-27 0.71
Alabama Georgia W 35-28 W 35-28 0.69
LSU Georgia W 36-16 W 36-16 0.68
Florida Michigan W 41-15 W 41-15 0.68
G Final is final score including garbage time; NG Final is final score in non-garbage time

Clemson also produced the fourth-best opponent-adjusted single game performance of the season nine days earlier in its semifinal victory over Notre Dame. That's what peaking when it matters most looks like, and it underscores how Clemson claimed the national championship in such impressive fashion. Still, there's something else that jumps off this list. Aside from Clemson, the best performances of the year were put forward by SEC teams, led first and foremost by Alabama. As far as FEI is concerned, the SEC stood out in its best moments, and as you might suspect, that means the SEC is looking awfully good in the final FEI ratings of the season.

For the third straight season, the national champion isn't ranked No. 1 in the final FEI ratings of the year, but rather No. 2. Alabama remains ahead of Clemson because they were so far ahead of the Tigers on the season to date; the 28-point loss head-to-head didn't fully overcome that difference. FEI doesn't adjust for recency, and Alabama's collective performances of the year, including a record of 7-1 against FEI top-20 teams, still compares favorably to Clemson's overall season (3-0 against top-20 opponents), even after the Crimson Tide laid an egg in the finale.

There are eight SEC teams ranked in the final FEI top 20, and three in the final FEI top 4. That, justifiably, will draw some perplexed reactions (including my own) to the merit of these ratings. Were both Georgia and LSU really as good as FEI suggests? How can five-loss Mississippi State, dominant on one side of the ball but not very special all around, really rank ahead of one-loss Notre Dame? Did the 2018 season "break" the FEI ratings? I'm not sure about any of these questions. The ratings are designed to measure something specific -- possession efficiency -- which usually parallels the eye test for most teams while still challenging our gut instincts about who is better than who. But this final set of ratings, more than any other I've produced previously, appears to have more opponent-adjustment peculiarities than usual. Perhaps that has something to do with this year's lack of connectivity. Perhaps I need to adjust or fix something in the formula that had been previously obscured. I'm going to have to take some time this offseason to examine these questions.

I was already planning on an offseason review of FEI in totality. Earlier this year I proposed a new format for the ratings output that is more tangible, and I think at a minimum I will make that change. I'm also scrutinizing FEI's predictive power. Projections this year were unremarkable, but I identified a few opportunities to isolate games in which FEI may have more predictive power than others. It will take a good amount of work, but I'm looking forward to a thorough review of FEI projections at both the game and season level so that I can be more transparent next year on what the system is and is not capable of in terms of predictive reliability.

I'm also interested in producing a companion set of ratings that better articulate achievement versus schedule rather than efficiency versus schedule. The most rabid debates in college football this year -- check that, every year -- revolve around competing perspectives on "best" versus "most-deserving," and I think there are better ways to frame those debates. Personally, I think that both perspectives are important. We don't crown the national champion based on its roster strength, but rather on what it has accomplished. Often, what it has accomplished meshes well with team strength. But sometimes, it doesn't, and the national champion earns the title by playing its best when it matters most rather than playing best more often over the course of the year.

I don't think it's taking anything away from Clemson to say that they were definitively the national champion in 2018, but they weren't definitively head and shoulders above Alabama, except on Monday night. And it's very healthy to remind ourselves that each game should be taken as a single outcome among many other results, and not necessarily the sole data point that distinguishes the two teams involved. Ohio State lost to Purdue by 29 points this year. Ohio State was not 29 points worse than Purdue this year. Both of those statements are true, and both are important.

In terms of merit, I introduced Elite Win Ratings this year as an output of achievement and not efficiency, and a better comparison with the College Football Playoff selection priorities. I adjusted the output somewhat to close out this year, now representing the adjusted wins (considering the strength of opponents faced) each team achieved. It is based on FEI, but it is a better representation of how accomplished each team was in 2018 rather than how efficient each team was in its games.

Top 10 EWR Teams
Team Rec EWR
Clemson 14-0 15.60
Alabama 13-1 14.63
Ohio State 13-1 13.29
Notre Dame 12-1 12.33
Oklahoma 12-2 11.81
Central Florida 11-1 10.61
Fresno State 11-2 9.82
Georgia 10-3 9.37
Washington State 10-2 8.81
Cincinnati 10-2 8.66

This output needs some offseason evaluation as well, both for its calculation and utility. But it feels more representative of what we prioritize most in our national champions, that they face elite challenges and win those that they face.

Both Clemson and Alabama distinguished themselves as the best teams this year, and every output I produce should reflect that. They also aren't going anywhere, and I fully expect both teams to be at or near the pinnacle of the sport next year and beyond. Clemson hasn't yet sustained a decade of dominance like Alabama, but they appear to have pulled even with, if not ahead of, the Crimson Tide in terms of current program power. It's not certain we'll get another playoff rematch next season, and I suppose some feeling of inevitability will again color our impressions of the season if they do dominate once again. But it's good to have two worthy adversaries rather than just one program head-and-shoulders above the rest. Cheers to Clemson, and cheers to college football rising to their challenge.

FEI Final Ratings

The Fremeau Efficiency Index (FEI) is a college football rating system based on opponent-adjusted possession efficiency. Adjusted Possession Advantage (APA) ratings represent the per-possession scoring advantage a team would be expected to have on a neutral field against an average opponent, calculated as a function of current FEI overall, offense, defense, and special teams ratings.

Strength of Schedule ratings (SOS) represent the average number of losses an elite team (two standard deviations better than average) would have against the team's schedule. Offensive FEI (OFEI) is scoring value generated per drive adjusted for starting field position and opponent defenses faced. Defensive FEI (DFEI) is scoring value generated per opponent drive adjusted for starting field position and opponent offenses faced. Special Teams FEI (SFEI) is scoring value generated per possession by a team's non-offensive and non-defensive units adjusted for opponent special teams units faced. The team's record to date against opponents ranked in the FEI top 10 (v10), top 20 (v20), top 30 (v30), top 40 (v40), and top 50 (v50) are also provided.

Ratings and supporting data are calculated from the results of non-garbage possessions in FBS vs. FBS games.

Click here for ratings for all 130 FBS teams.

Rk Team Rec FEI APA Rk SOS Rk OFEI Rk DFEI Rk SFEI Rk v10 v20 v30 v40 v50
1 Alabama 13-1 .326 2.91 1 2.63 1 3.95 2 1.19 4 .241 3 4-1 7-1 7-1 7-1 7-1
2 Clemson 14-0 .311 2.76 2 1.60 21 3.49 5 .90 1 .092 21 2-0 3-0 3-0 6-0 8-0
3 Georgia 10-3 .262 2.33 3 2.37 4 3.62 3 1.43 11 .180 7 1-2 3-3 4-3 4-3 5-3
4 LSU 9-3 .219 1.62 7 2.37 5 2.55 38 1.37 8 .288 2 2-2 4-3 4-3 5-3 5-3
5 Oklahoma 12-2 .218 1.84 4 1.81 13 4.26 1 2.64 100 .157 11 0-1 1-2 4-2 5-2 5-2
6 Ohio State 13-1 .206 1.63 6 1.29 41 3.42 7 2.10 52 .170 8 1-0 3-0 4-0 5-0 7-1
7 Washington 9-4 .200 1.53 9 1.45 28 2.59 35 1.42 9 .009 56 0-1 0-2 3-2 5-2 5-3
8 Mississippi State 7-5 .182 1.70 5 2.22 7 2.68 28 1.00 2 .118 15 0-3 2-4 2-5 2-5 2-5
9 Notre Dame 12-1 .177 1.48 10 1.33 37 2.74 25 1.45 12 .078 24 0-1 1-1 1-1 4-1 5-1
10 Florida 8-3 .177 1.34 13 1.74 15 2.74 26 1.72 26 .203 5 2-1 3-2 3-3 3-3 4-3
11 Central Florida 11-1 .175 1.61 8 .61 100 3.53 4 1.96 39 -.010 64 0-1 0-1 1-1 1-1 3-1
12 Iowa 8-4 .165 1.24 16 1.00 62 2.41 48 1.49 13 .127 14 1-0 1-1 2-1 2-2 4-4
13 Michigan 10-3 .164 1.39 11 1.60 20 2.80 20 1.57 17 .184 6 0-3 1-3 2-3 3-3 5-3
14 Fresno State 11-2 .163 1.38 12 .82 74 2.65 30 1.42 10 -.022 71 0-0 1-1 1-1 2-1 2-2
15 Missouri 7-5 .152 1.31 14 1.94 12 2.90 13 1.72 24 .018 50 1-2 1-2 1-3 1-4 2-5
Rk Team Rec FEI APA Rk SOS Rk OFEI Rk DFEI Rk SFEI Rk v10 v20 v30 v40 v50
16 Texas 10-4 .149 1.08 23 1.68 18 2.71 27 1.95 38 .024 47 2-1 2-1 3-2 3-3 3-3
17 Penn State 9-4 .145 1.17 19 1.29 42 2.48 44 1.52 14 .149 12 0-1 1-2 1-4 2-4 4-4
18 Auburn 7-5 .145 1.22 17 2.54 2 2.67 29 1.59 18 .213 4 1-4 2-4 2-4 2-4 3-4
19 Boise State 10-3 .142 1.16 20 .84 73 2.79 22 1.81 30 .017 51 0-0 1-1 2-1 2-2 2-2
20 Texas A&M 8-4 .141 1.08 24 2.51 3 2.83 17 2.00 43 .291 1 1-3 1-4 2-4 3-4 4-4
21 Kentucky 9-3 .139 .96 28 1.71 17 2.26 64 1.66 19 .169 9 2-1 4-2 4-2 4-2 5-2
22 Washington State 10-2 .131 1.05 25 .81 76 3.30 8 2.45 82 .061 33 0-1 0-1 2-1 3-1 4-1
23 Utah State 10-2 .130 1.25 15 .62 98 3.04 10 1.76 27 .002 58 0-0 0-1 0-2 0-2 1-2
24 Army 9-2 .129 1.21 18 .76 82 3.18 9 1.97 41 -.139 107 0-1 0-1 0-1 0-2 1-2
25 Utah 8-5 .125 .89 29 1.35 34 2.27 63 1.68 22 .167 10 0-2 0-2 0-3 1-5 2-5
26 West Virginia 7-4 .125 .96 27 1.07 57 2.89 15 2.16 57 .070 28 0-1 1-1 1-2 1-4 1-4
27 Cincinnati 10-2 .117 1.09 22 .66 94 2.45 47 1.37 7 -.108 98 0-0 0-1 1-1 1-1 1-2
28 Ohio 8-4 .115 1.13 21 .56 105 3.43 6 2.26 69 .070 26 0-0 0-0 0-1 0-2 1-2
29 Michigan State 7-6 .114 .81 30 1.25 45 1.63 110 1.09 3 .004 57 0-1 1-2 2-2 2-4 3-6
30 Iowa State 7-5 .103 .79 31 1.36 31 2.29 59 1.68 21 .042 40 0-1 0-3 1-4 2-4 2-4


22 comments, Last at 10 Jan 2019, 5:16pm

1 Re: FEI Final Ratings

Aside from Clemson, the best performances of the year were put forward by SEC teams, led first and foremost by Alabama. It's fair to point out that the SEC also put forth the most poor performances, with Alabama, LSU, and Georgia owning a spot on both sides of the ledger.

Looking at the year-end tallies, Washington and Iowa also appear very over-rated by FEI. In Washington's case, it's probably a really good team that's badly unreliable at QB.

2 Re: FEI Final Ratings

It's fair to point out that the SEC also put forth the most poor performances.

Not necessarily. Because of opponent adjustments, the best Game FEI ratings (like the best single-game DVOA ratings) are not necessarily in the same games as the worst Game FEI ratings. These Game FEI ratings are high *because* you get extra credit for beating SEC teams that are so strong otherwise.

3 Re: FEI Final Ratings

How do you stop the feedback effect? Where initial states keep getting fed back into the amplifier until the mechanism fails?

Basically, how do you keep the initial seedings from determining the final rankings?

Is Washington high just because they started high and got lucky that FEI also liked two of the teams that beat them?

Is Mississippi State just the beneficiary of a ranking that thought highly of an SEC west that didn’t have much after Alabama and LSU? Although that doesn’t explain Iowa.

It’s a more minor point, but in the playoff era, a lot of non playoff teams seem to acquire the givenoshits. Georgia and Michigan seemed to suffer from that.

4 Re: FEI Final Ratings

One problem with college football in general is that feedback is a real worry. The only real way to dampen feedback is to have a lot of schedule mixing. And when the length of the schedule is compared to the number of teams involved, college football is the worst. The college basketball season is more than twice as long, and the NFL has considerably fewer teams.
Non-conference schedules are (a) a very small part of the college football season, and (b) often designed so the top teams don't risk losses. This combination can often result in teams from power conferences being slightly overrated. I would argue that this "power conference" phenomenon is why Alabama is still seen as superior by FEI than Clemson, even after the opposite was clearly demonstrated on the field. Clemson would have to have three tough non-conference opponents to have a hope of getting a strength of schedule comparable to any SEC schedule, and having games vs. Furman and Ga. Southern didn't really help on that front.

Ultimately any measurement of the relative strengths of the various conferences depends entirely on the small number of interconference games. And that leaves us wondering, every year before the bowl games start, whether the reputations of the top conferences (esp. the SEC) is entirely justified. The SEC went 6-6 in bowl games (including both of Alabama's results), suggesting they were rated about right.

8 Re: FEI Final Ratings

Yeah, college football is the ultimate in Small Sample Size Theater, given 100-plus teams playing 13 games. I will say that if you don't weight latter games more heavily, it's not crazy to still say Alabama played better this season. Clemson's qb is quite likely a much better player today, compared to 120 days ago. On the other hand, weighting latter games more heavily likely makes more sense in a league with a majority of players younger than 22.

11 Re: FEI Final Ratings

He was on the roster last year, and appeared in 8 games. Relative to high school preparation, I'd guess working hard within the structure of a program like Alabama's for a year has a lot of value for 19 year old qb.

12 Re: FEI Final Ratings

Clemson also switched to Trevor Lawrence after they struggled in a couple of games (notably Texas A&M) with Kelly Bryant as QB. Then, Lawrence got knocked out of the Syracuse game, and Clemson struggled to win that one by only 4 points. Bryant was also gone as an option by then. Note that Alabama didn't have any close games until they played Georgia in the SEC finals but there's two right there for Clemson. And, yes, I do wonder if you only included Clemson games that Lawrence played in full, if they would be #1. Based on regular season dominance, I suspect it makes sense for a computer to pick Alabama as #1 over Clemson.

14 Re: FEI Final Ratings

Clemson was alternating Bryant and Lawrence in the Texas A&M game (basically two series each). They scored 21 with Bryant, with a failed 4th-goal, and 7 with Lawrence. They stuck with Bryant in the 4th quarter.

9 Re: FEI Final Ratings

One way of handling this is to account for scheduling Sisters of the Blind teams.

FEI, I think, just drops all sub-FBS games entirely from its calculations. That could be replaced by treating them all as the 131st-ranked team (There are 130 FBS teams).

This is vastly under-rating teams like Eastern Washington, North Dakota, Northern Iowa, Youngstown, and Villanova (and prior examples like Marshal, Appalachian State, and Georgia Southern) who have actual rivalries with FBS schools and are much better than the 130th-ranked FBS team, but still over-rates bottom-half FCS teams that get scheduled by the likes of Alabama (Citadel was 5-6 as an FCS school).

At the moment, there's no penalty in the current scoring to account for actually losing one of these games.

13 Re: FEI Final Ratings

The feedback loop is a challenge to tackle. Alabama is great in part because it dominated Texas A&M, who is pretty good in part because they gave Clemson a scare, and the Tigers just whipped Alabama...

But to clarify, if it already isn't apparent, there is no influence on these ratings based on preseason "initial seedings" or "starting high". All of the preseason projection stuff is eliminated from the FEI model by mid-season. These final ratings are based exclusively on 2018 season data.

15 Re: FEI Final Ratings

Because there's no crossover, though, do those pre-season projections ever go away? Or are SEC teams considered strong at season end because they were considered strong at mid-season, but they were strong at mid-season because they were strong at pre-season?

Basically, what would happen if you considered the MAC to be world-beaters in the pre-season and the SEC to be the MAC? Would they recover?

Or would the Big Ten get a huge boost for beating a bunch of MAC teams that weren't actually worth anything? Would the SEC never recover because they play tons of FCS teams whose games don't count towards FEI? Alabama's schedule is illustrative. They played the 86th, 129th, and <130th FEI teams this year, non-con. Next year is almost as bad, but at least Duke is top-half.

Texas A&M played Clemson at home and Alabama on the road. Louisville also played both, both on the road. I'm not sure how helpful it is to know they lost to Alabama by 37 and Clemson by 61, but that does predict the Clemson-Alabama neutral delta. (They also lost to Kentucky at home by 46; Louisville was terrible)

16 Re: FEI Final Ratings

There is no calculation in this system that leans on a team's previous ranking, early season, mid-season, or otherwise. The ratings (since Week 7) have no prior inputs. FEI is calculated each week (since Week 7) as if an entirely unique season of x-number of games and y-number of possessions were inputted at once, and each week's data is published anew.

17 Re: FEI Final Ratings

Maybe I'm not getting something.

I was under the impression that each week's FEI took opponent quality into account. That happens, right?

19 Re: FEI Final Ratings

I'm under the impression that you are suggesting that opponent adjustments are calculated based on the previous week(s) ratings of opponent quality. I'm trying to make clear that the previous week ratings aren't influencing anything, and that opponent adjustments are entirely re-calculated each week on the new, complete set of possession efficiency data. That data, of course, includes all 2018 results to date -- raw results, not the previously calculated opponent-adjusted results. Maybe we are actually talking about the same thing in different terms, but your references to a "starting point" for the SEC vs MAC, for instance, suggest that you think I'm using a team's pre-existing opponent-adjusted data and simply adding in more data each week.

21 Re: FEI Final Ratings

That's more or less what I thought was going on.

Fully fleshed, what I thought was going on is this:

Team gets possession efficiency score for a game (A). This is filtered through opponent quality score (B) which for simplicity sake we'll assume is a multiplication. Team game score (C) is A*B. Season score is a moving average of game score.
This is an over-simplification of the original presentation, but at least captures some of my thought process.

Mostly I'm confused about when the strength adjustment plays into the calculation, and how. Is a team's strength newly calculated during every weekly run, and then back-applied to prior weeks? Or does a team enter a week with a given strength, that gets applied to that week's new game scores, and then it gets adjusted thereafter to be applied to the next week?

Also: If all that matters is week-7 on (which is intra-conference for everyone but independents), why does intersectionality matter at all and how on earth could you ever parse interconference strength?

22 Re: FEI Final Ratings

That reminds me (again) that I need to publish an entirely new "introduction to FEI" post to reference, since the way I produce the ratings has changed often since that original post in 2006. I'll try to make it simple here, at the risk of oversimplifying.

Every FBS vs FBS game produces a "game efficiency" number (how well a team played over the course of non-garbage time, as a function of maximizing success on its own possessions and minimizing success of its opponents on their possessions). Let's call those game efficiency numbers "A". Over the course of a season, the team has 10 to 14 "As" on its ledger. Its raw season efficiency is, basically, the average of those As. We'll call that X. Compare each team's A data against it's opponent X data to calculate how the team performed in the given game relative to the overall raw strength of its opponent. Average those adjusted As to create a "first wash" adjusted X. Do it again, comparing the A data against each team's first wash adjusted X data to produce a second wash adjusted X. The second wash adjusted X is FEI. (Why two washes and not one, or five, or ten? I tested each of these in the development of the formula and found a curve, peaking with the second wash having the highest correlation with projected future results).

There are no "previous week" numbers used in the formula. And the strength of schedule data I produce is an output after FEI is calculated, not prior to its calculation. To clarify on the Week 7 thing, I have a different formula for weeks 1 through 6 that DOES include preseason projections because there aren't enough data points for the adjustment washes described above to produce anything other than extremely volatile results. But early season game efficiency data, including of course data prior to Week 7, is included in the formula just as late season data is included. I have no "recency" weighting in the formula (though I am considering if and how to do that as well).

5 Re: FEI Final Ratings

One team was better in the Reg. Season but another team was better in the Post-Season. Happens so many times, especially in the NFL. It is always interesting to me to see which teams, which QB's will step up their game under the brightest lights and biggest pressure games. In this case I agree with you 100%, we (Clemson) really outplayed them overall and Trevor really came through for us.

6 Re: FEI Final Ratings

Can't help but think about the NFL in '07. Who amongst us would ever have dreamed that the Giants would end up being the most successful team or that Eli Manning would end being the most successful QB for '07.

18 Re: FEI Final Ratings

I'd say the gap between Clemson and 'Bama was much smaller over the course of the season and larger in the championship game.
That the Giants made the Super Bowl was a huge shock. (That they won was an even bigger shock.) By contrast, it wasn't really a huge shock to see Clemson in the title game. On the contrary, they were expected. And their victory over 'Bama, while surprising, was hardly of the same historical importance.

7 Re: FEI Final Ratings

Keep in mind that being successful means the achievement of desired visions and planned goals.