Writers of Pro Football Prospectus 2008

Most Recent FO Features

SandersEmm10.jpg

» Scramble for the Ball: Quarter Pole Projections

Mike and Tom weigh the chances of this year's class of receivers, running backs and tight ends who are on pace to break the magical 1,000-yard mark for the first time.

21 Aug 2013

FEI Preseason Primer Part I

by Brian Fremeau

We’re not going to get the projections for every team right.

I have to remind myself of that every offseason. But I don’t fret over teams that surprise us with unexpected runs to BCS bowl games, nor teams that crash and burn despite what the model suggested. That’s college football, and it’s awesome.

I get caught up tweaking and testing the FEI projection model not in a pursuit of perfection, but in a pursuit for more meaningful data. How important are data points like returning starters, quarterback reliance, and recruiting rankings? How important are data points from last season as compared with data collected over several seasons? Is the goal to project wins and losses or overall team strength? Do those two goals always go hand in hand?

I value the end-of-year FEI ratings, a measure of opponent-adjusted possession efficiency, so I build my projections with end-of-year FEI ratings as the target. Strength of schedule and mean wins are a function of the projections. That approach has not changed this year. What has changed is a bit of the input data.

For as long as I have been producing preseason projections, I have been using Program FEI as the baseline measure from which all other adjustments are made. The Program FEI ratings are calculated the same way I calculate single-season FEI, but I include five years of drive efficiency data instead of just one. Program FEI helps illustrate the trajectory of a given program (see the 25-year Program FEI charts for Alabama and its opponents below), but it also correlates very well with next-year FEI (.746), better than any other single measure I’ve tested.

To calculate these "Alternate FEI" ratings, I turned to the Game Splits data I have began processing over the last two years. Game Splits represent the scoring margin components of a game: the value contributed to the non-garbage time margin of victory or defeat.

Let's take last year's BCS national championship game as an example. Alabama defeated Notre Dame 42-14, but according to my methodology, garbage time began after the Crimson Tide took a 35-0 lead in the third quarter. Which units contributed to that 35-point margin for Alabama? By breaking down the value of field position at the start of each drive, the units responsible for generating that field position, and the value generated by the offenses at the conclusion of each drive, my method attributes 26.6 points of scoring margin value to Alabama's offense, 4.7 points to its defense, and 1.7 points to its special teams. An additional 1.9 points was generated by none of these units, attributed to the fact that the Crimson Tide also had one more offensive possession than the Irish at the time garbage time began. Add them all up (26.6 + 4.7 + 1.7 + 1.9 = 34.9; rounding accounts for the 0.1 discrepancy) and you get the 35-point scoring margin of the game.

Game Rewind: Relive every NFL moment…subscribe to Game Rewind.

The Game Splits for all 2012 games can be found here, along with links to 2007-2011 Game Splits. Two other columns in these tables indicate the turnover value and the field position value of each game. In the BCS championship game, Alabama had a 2.3 point advantage on turnovers, and a 2.7 point advantage on field position.

Readers familiar with FEI columns from last season know that I used the turnover, special teams, and field position data to recalculate "Revisionist Box Scores," the alternative outcome of games if those factors had been neutralized. Last season, turnovers were a deciding factor in 16.4 percent of FBS games, special teams were a deciding factor in 7.9 percent of games, and field position was a deciding factor in 10.1 percent of games. That is, the value generated by those factors exceeded the non-garbage scoring margin of the game.

This year, I'm taking the Game Splits analysis a step further. Instead of simply publishing a revisionist box score, I recalculated FEI for each neutralized environment. In the turnover-neutral environment, Alabama defeated Notre Dame by 32.7 points instead of 35 points (35 - 2.3 = 32.7). In the special teams-neutral environment, Alabama defeated Notre Dame by 33.3 points. In the field position neutral environment, Alabama defeated Notre Dame by 32.3 points.

In blowout games such as the title game, the impact is pretty insignificant. But in other games, the neutralized environment flips the outcome. Against Texas A&M, the Crimson Tide lost 27-22, a five point margin of victory. But Texas A&M had a turnover value advantage of 12.2 points in that game. In the turnover-neutral environment, Alabama "won" by 7.2 points.

To produce Alternate FEI ratings, I simply recalculated game efficiency according to the neutral scoring margins and made the same opponent-adjustments as with FEI. The complete set of Alternate FEI ratings for 2012 can be found here. For many teams, the neutral rankings aren't dramatically different, but there are some notable exceptions. Florida State ranked No. 13 overall in FEI, but jumped up to No. 6 overall in turnover-neutral FEI. Stanford ranked No. 7 overall in FEI, but fell down to No. 13 overall in special teams-neutral FEI.

This wasn't just an interesting exercise: I found that the Alternate FEI ratings were another factor that would help with the annual projection adjustments.

I also investigated the way I calculate the start of non-garbage time, and which data is discarded in the FEI and Program FEI ratings. I didn’t find anything substantive enough to alter the way in which I determine garbage time versus non-garbage time, but I did unexpectedly find that previous year garbage time data has some value as it relates to next-year non-garbage time value. I can only speculate at this point as to why it has value; my colleague Bill Connelly offered a plausible suggestion that garbage time value relates in some way to team depth and the strength of second stringers who won’t make their impact until the following season.

In the end, this year’s FEI projections are still rooted in Program FEI, but are now adjusted a bit by the garbage time factor, turnover-neutral FEI, special teams-neutral FEI, and field position-neutral FEI. I have also retained factors in the projection model such as number of returning starters, five-year weighted recruiting rankings, and quarterback reliance (the percentage of total offense run through the team’s quarterback). Since 2004, the correlation of projected FEI to actual FEI is .788.

The final FEI 2013 preseason projections are published below. Readers that have been following my preseason content at ESPN Insider and ESPN the Magazine will note some changes in this latest round of projections. FEI is a little higher on Louisville and Michigan now than it was a month ago, both mostly due to the inclusion of the new Alternate FEI data and garbage time factors in the projection model. In addition, these projections are based on FEI only, not both FEI and Bill Connelly's S&P+, and therefore will be different from the projections in FOA 2013.

In addition to everything discussed so far, I have new Field Position data tables published here at Football Outsiders which can be found using the drop-down statistics menus above. Here is the 2012 Field Position table, breaking down FPA into elements such as each team's actual average starting field position and the percentage of drives from short and long field position distances. The field position tables are available for all seasons since 2007. On my site, I recently published points per drive data dating back to 2007 as well.

This is only Part I of the FEI preseason primer. Next week, we'll dig into individual game projections with all this new data at our disposal.

FEI 2013 Preseason Projections

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, not play-by-play 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. Strength of Schedule (SOS) is calculated as the likelihood that an "elite team" (two standard deviations above average) would win every game on the given team's schedule. SOS listed here includes future games scheduled.

Mean Wins (FBS MW) represent the average total games a team with the given FEI rating should expect to win against its complete schedule of FBS opponents.

Preseason projections are a function of Program FEI ratings, previous-year FEI and garbage time data, previous-year turnover-neutral, special teams-neutral, and field position-neutral FEI, returning starters, recruiting success, and quarterback reliance. As the season progresses and actual 2013 data is collected, the weight given to projected data will be reduced each week until Week 7, at which point it will be eliminated from the rankings entirely. Offensive and defensive FEI ratings will also debut in Week 7.

Rank Team FBS
W-L
FEI SOS SOS
Rk
FBS
MW
1 Alabama 0-0 .283 .335 58 11.0
2 Oregon 0-0 .278 .296 48 10.9
3 Stanford 0-0 .253 .187 24 10.1
4 LSU 0-0 .216 .096 5 9.1
5 Florida 0-0 .209 .164 16 9.3
6 Texas 0-0 .208 .221 33 9.4
7 Texas A&M 0-0 .188 .209 30 9.4
8 TCU 0-0 .185 .158 15 8.7
9 Ohio State 0-0 .182 .422 73 10.0
10 Notre Dame 0-0 .182 .179 21 8.9
11 Florida State 0-0 .175 .310 52 9.7
12 Oklahoma State 0-0 .173 .265 38 9.1
13 Georgia 0-0 .172 .199 27 8.6
14 South Carolina 0-0 .169 .269 40 8.9
15 Oklahoma 0-0 .162 .155 14 8.2
Rank Team FBS
W-L
FEI SOS SOS
Rk
FBS
MW
16 Oregon State 0-0 .157 .168 17 8.8
17 Louisville 0-0 .155 .573 101 10.2
18 Wisconsin 0-0 .152 .333 57 9.2
19 USC 0-0 .149 .213 31 9.4
20 Michigan State 0-0 .147 .286 47 9.2
21 Clemson 0-0 .144 .371 66 9.1
22 Cincinnati 0-0 .136 .643 106 10.2
23 Boise State 0-0 .125 .546 97 9.9
24 Nebraska 0-0 .123 .416 71 9.0
25 Penn State 0-0 .121 .311 53 8.5
26 Northwestern 0-0 .115 .279 44 8.0
27 Michigan 0-0 .104 .195 25 7.0
28 Kansas State 0-0 .101 .223 34 7.8
29 Missouri 0-0 .098 .231 35 7.2
30 Arizona State 0-0 .096 .177 20 7.1
Rank Team FBS
W-L
FEI SOS SOS
Rk
FBS
MW
31 Virginia Tech 0-0 .087 .305 51 8.1
32 Mississippi 0-0 .087 .111 6 6.9
33 Pittsburgh 0-0 .078 .371 65 7.6
34 Baylor 0-0 .077 .204 29 7.1
35 Iowa 0-0 .074 .275 42 6.7
36 Miami 0-0 .072 .317 54 7.5
37 Northern Illinois 0-0 .068 .716 116 9.8
38 Georgia Tech 0-0 .067 .362 61 7.1
39 BYU 0-0 .064 .250 36 6.6
40 North Carolina 0-0 .063 .362 60 7.6
41 Arizona 0-0 .060 .280 45 7.6
42 Rutgers 0-0 .057 .426 75 7.9
43 Washington 0-0 .052 .134 11 6.5
44 Utah State 0-0 .050 .480 86 8.4
45 Vanderbilt 0-0 .048 .173 18 6.6
Rank Team FBS
W-L
FEI SOS SOS
Rk
FBS
MW
46 Central Florida 0-0 .046 .383 68 7.3
47 Utah 0-0 .045 .121 7 5.8
48 UCLA 0-0 .044 .085 3 5.7
49 Syracuse 0-0 .043 .297 50 6.7
50 West Virginia 0-0 .039 .186 23 6.2
51 Texas Tech 0-0 .038 .203 28 6.1
52 Mississippi State 0-0 .026 .128 9 5.5
53 Fresno State 0-0 .023 .697 114 8.6
54 Connecticut 0-0 .020 .455 78 6.9
55 Tennessee 0-0 .015 .060 1 5.2
56 Indiana 0-0 .003 .215 32 5.8
57 San Jose State 0-0 .002 .426 76 7.4
58 Tulsa 0-0 -.001 .579 102 7.8
59 Auburn 0-0 -.006 .153 13 5.4
60 Louisiana Monroe 0-0 -.008 .518 92 7.6
Rank Team FBS
W-L
FEI SOS SOS
Rk
FBS
MW
61 San Diego State 0-0 -.009 .425 74 7.0
62 Iowa State 0-0 -.011 .180 22 4.2
63 Boston College 0-0 -.013 .296 49 5.9
64 Arkansas State 0-0 -.015 .656 108 8.0
65 Toledo 0-0 -.020 .412 70 7.0
66 Bowling Green 0-0 -.021 .748 121 7.9
67 Rice 0-0 -.025 .529 94 7.9
68 Arkansas 0-0 -.027 .072 2 4.2
69 Nevada 0-0 -.027 .363 62 6.0
70 SMU 0-0 -.032 .266 39 5.0
71 Minnesota 0-0 -.032 .274 41 4.7
72 Ohio 0-0 -.034 .600 104 8.2
73 Houston 0-0 -.038 .420 72 5.5
74 Ball State 0-0 -.039 .738 120 7.9
75 California 0-0 -.040 .093 4 4.1
Rank Team FBS
W-L
FEI SOS SOS
Rk
FBS
MW
76 Virginia 0-0 -.040 .276 43 4.5
77 South Florida 0-0 -.040 .353 59 5.0
78 North Carolina State 0-0 -.041 .478 85 5.9
79 Louisiana Tech 0-0 -.043 .829 125 7.9
80 Purdue 0-0 -.046 .152 12 3.5
81 Temple 0-0 -.047 .322 55 5.5
82 Navy 0-0 -.053 .491 87 6.1
83 Louisiana Lafayette 0-0 -.054 .668 111 7.0
84 Western Kentucky 0-0 -.057 .772 123 6.9
85 East Carolina 0-0 -.059 .709 115 6.7
86 Kent State 0-0 -.059 .388 69 5.9
87 Kentucky 0-0 -.060 .132 10 3.7
88 Maryland 0-0 -.067 .382 67 4.5
89 Wake Forest 0-0 -.067 .365 63 4.6
90 Troy 0-0 -.071 .649 107 6.1
Rank Team FBS
W-L
FEI SOS SOS
Rk
FBS
MW
91 Middle Tennessee 0-0 -.071 .659 109 6.9
92 Duke 0-0 -.073 .566 99 4.9
93 Marshall 0-0 -.084 .683 112 6.5
94 Central Michigan 0-0 -.087 .660 110 5.9
95 Buffalo 0-0 -.088 .475 83 5.7
96 North Texas 0-0 -.097 .547 98 5.5
97 Southern Mississippi 0-0 -.103 .504 90 4.7
98 Wyoming 0-0 -.105 .445 77 5.8
99 Western Michigan 0-0 -.106 .370 64 4.7
100 South Alabama 0-0 -.108 .796 124 5.5
101 Florida International 0-0 -.116 .603 105 5.1
102 UNLV 0-0 -.121 .737 119 4.8
103 UAB 0-0 -.123 .465 81 4.7
104 Air Force 0-0 -.123 .528 93 4.9
105 Illinois 0-0 -.128 .256 37 2.7
Rank Team FBS
W-L
FEI SOS SOS
Rk
FBS
MW
106 Texas State 0-0 -.132 .755 122 5.3
107 UTSA 0-0 -.132 .594 103 4.3
108 Colorado State 0-0 -.135 .284 46 5.0
109 UTEP 0-0 -.138 .543 96 4.8
110 Army 0-0 -.143 .573 100 4.3
111 Hawaii 0-0 -.145 .462 80 3.0
112 Florida Atlantic 0-0 -.146 .689 113 4.1
113 Memphis 0-0 -.150 .499 88 3.2
114 Colorado 0-0 -.159 .198 26 2.3
115 Washington State 0-0 -.161 .121 8 2.5
116 Kansas 0-0 -.173 .175 19 2.0
117 Eastern Michigan 0-0 -.178 .505 91 3.1
118 Tulane 0-0 -.195 .732 117 3.4
119 Akron 0-0 -.199 .533 95 3.3
120 Miami (OH) 0-0 -.202 .735 118 2.8
Rank Team FBS
W-L
FEI SOS SOS
Rk
FBS
MW
121 New Mexico State 0-0 -.219 .476 84 3.6
122 New Mexico 0-0 -.223 .499 89 2.5
123 Georgia State 0-0 -.223 .327 56 2.3
124 Massachusetts 0-0 -.252 .471 82 2.5
125 Idaho 0-0 -.265 .462 79 2.2

Posted by: Brian Fremeau on 21 Aug 2013

7 comments, Last at 19 Nov 2013, 6:38am by empower network

Comments

1
by Hang50 :: Wed, 08/21/2013 - 3:49pm

As the season progresses and actual 2013 data is collected, the weight given to projected data will be reduced each week until Week 7, at which point it will be eliminated from the rankings entirely.

My reading of that statement is that "Week 7" is a hard date (e.g. the seventh Saturday of the NCAA football season) that applies to every team at the same time.

If my reading is accurate, have you considered a per-team alternative to the effect of "after the team has played seven games against FBS competitors"? My suspicion is that it would keep some jitter out of mid-season rankings for teams with early-season FCS opponents or byes.

4
by Brian Fremeau :: Sun, 08/25/2013 - 8:14pm

Good suggestion of something to look into. I've tried several variations on how to reduce the weight of preseason data in the formula over the early weeks of the season, but there are always others to consider.

2
by JPW (not verified) :: Sat, 08/24/2013 - 11:15pm

Georgia State opponent graph is only showing Alabama's FEI.
Also, the SOS seems intuitively backwards. A higher FEI rating should make the SOS harder, which seems should be a 'higher' ranking (lower number, 1=highest). As is, when I read a team has a SOS rank in the 100s, I think "oh, that's an easy schedule."

Just my opinion. Great work. Glad ESPN is giving you recognition.

3
by Brian Fremeau :: Sun, 08/25/2013 - 8:13pm

Georgia State is new to FBS, so they don't have a PFEI history.

When a team has an SOS rank in the 100s, it is an easy schedule. I think the confusion might be in misinterpreting what the SOS rating means. It represents the likelihood that an elite team would go undefeated against the entire schedule, so a low SOS rating is a tough schedule.

5
by Thomas Hubbard (not verified) :: Tue, 08/27/2013 - 3:37pm

Just thought I would mention -- the infographics in this article are wonderful. I have sent them to a couple of fellow graphics geeks as examples of excellent work.

6
by Brian Fremeau :: Tue, 08/27/2013 - 5:45pm

Thanks very much. I post other graphics I've done here: http://www.bcftoys.com/graphics

7
by empower network (not verified) :: Tue, 11/19/2013 - 6:38am

This is a good tip particularly to those fresh to the blogosphere.
Simple but very accurate info… Thank you for sharing this one.
A must read article!