Writers of Pro Football Prospectus 2008

Most Recent FO Features

Harris-Brady.jpg

» Week 12 DVOA Ratings

Denver remains No. 1 in the Football Outsiders DVOA ratings, but New England moves up to No. 2 and has taken over as our Super Bowl favorite.

07 Nov 2007

Fremeau Efficiency Index: Week 10

by Brian Fremeau

The Fremeau Efficiency Index (FEI) principles and methodology can be found here. Like DVOA, FEI rewards playing well against good teams, win or lose, and punishes losing to poor teams more harshly than it rewards defeating poor teams. Unlike DVOA, it 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, a measurement of the success rate of a team scoring and preventing opponent scoring throughout the non-garbage-time possessions of a game. Like DVOA, it represents a team's efficiency value over average.

Only games between FBS (Division 1-A) teams are considered. The Week 10 Ratings represent the results of games played through Sunday, November 4, 2007.


Rank Team W-L FEI Last Wk GE GE Rank
1 LSU (8-1) 0.303 1 0.242 7
2 Oregon (8-1) 0.271 3 0.245 6
3 West Virginia (7-1) 0.229 4 0.357 2
4 Ohio State (9-0) 0.229 7 0.367 1
5 South Florida (5-3) 0.224 2 0.107 24
6 Oklahoma (8-1) 0.220 8 0.315 4
7 Florida (5-3) 0.206 10 0.133 21
8 Auburn (6-3) 0.204 6 0.129 22
9 Arizona State (8-1) 0.198 5 0.216 10
10 Boston College (7-1) 0.192 9 0.172 15
11 Connecticut (7-1) 0.191 11 0.168 16
12 Florida State (6-3) 0.185 17 0.089 29
Rank Team W-L FEI Last Wk GE GE Rank
13 Clemson (6-2) 0.183 16 0.218 9
14 Missouri (7-1) 0.183 21 0.219 8
15 Alabama (5-3) 0.176 12 0.068 39
16 Cincinnati (6-2) 0.173 19 0.167 18
17 Arkansas (5-3) 0.160 26 0.197 13
18 Georgia (6-2) 0.159 13 0.061 41
19 Virginia Tech (6-2) 0.159 27 0.080 33
20 California (6-3) 0.158 15 0.075 35
21 USC (7-2) 0.153 30 0.208 12
22 Kentucky (5-3) 0.146 24 0.034 49
23 Kansas (8-0) 0.144 14 0.332 3
24 BYU (5-2) 0.142 22 0.167 17
25 Michigan (8-1) 0.136 29 0.149 20

Offensive Efficiency (OE) represents the raw per-possession scoring rate for each team, and Defensive Efficiency (DE) represents that of its opponents. Adjusted Offensive Efficiency (AOE) and Adjusted Defensive Efficiency modifies OE and DE for field position and opponent. Field Position Advantage (FPA) represents the difference between a team's average offensive field position and that of its opponents.


FEI Rank Team W-L OE OE Rank AOE AOE Rank DE DE Rank ADE ADE Rank FPA
1 LSU (8-1) 0.412 12 0.439 9 0.177 11 0.169 8 3.4
2 Oregon (8-1) 0.474 4 0.510 2 0.240 36 0.223 23 4.7
3 West Virginia (7-1) 0.477 3 0.462 4 0.174 10 0.194 15 4.2
4 Ohio State (9-0) 0.390 16 0.336 36 0.068 1 0.081 1 10.2
5 South Florida (5-3) 0.289 58 0.322 43 0.199 19 0.176 9 6.3
6 Oklahoma (8-1) 0.462 5 0.444 8 0.167 7 0.157 6 0.3
7 Florida (5-3) 0.498 1 0.548 1 0.367 93 0.324 82 3.5
8 Auburn (6-3) 0.288 60 0.322 42 0.183 13 0.151 4 -0.2
9 Arizona State (8-1) 0.359 23 0.373 21 0.190 18 0.181 11 2.2
10 Boston College (7-1) 0.303 50 0.381 18 0.136 2 0.191 13 0.3
11 Connecticut (7-1) 0.294 55 0.282 61 0.168 8 0.185 12 3.5
12 Florida State (6-3) 0.243 81 0.269 70 0.190 17 0.201 17 5.7
FEI Rank Team W-L OE OE Rank AOE AOE Rank DE DE Rank ADE ADE Rank FPA
13 Clemson (6-2) 0.389 17 0.387 16 0.152 5 0.179 10 5.8
14 Missouri (7-1) 0.438 8 0.497 3 0.253 39 0.227 28 -0.2
15 Alabama (5-3) 0.303 49 0.341 34 0.271 47 0.226 26 1.8
16 Cincinnati (6-2) 0.343 33 0.351 29 0.185 15 0.200 16 5.1
17 Arkansas (5-3) 0.418 10 0.437 10 0.231 31 0.241 39 4.7
18 Georgia (6-2) 0.356 26 0.383 17 0.271 46 0.239 37 1.1
19 Virginia Tech (6-2) 0.232 89 0242 82 0.140 4 0.141 2 5.0
20 California (6-3) 0.346 30 0.366 22 0.282 53 0.267 47 1.3
21 USC (7-2) 0.352 28 0.314 45 0153 6 0.147 3 3.1
22 Kentucky (5-3) 0.410 13 0.424 11 0.332 81 0.284 56 3.3
23 Kansas (8-0) 0.436 9 0.349 30 0.139 3 0.153 5 4.6
24 BYU (5-2) 0.339 36 0.342 32 0.199 20 0.213 20 -1.3
25 Michigan (8-1) 0.339 35 0.310 46 0.171 9 0.168 7 5.0

Click here for rankings of all 119 teams. Click here for Offensive/Defensive Efficiency rankings for all 119 teams.

For the first time in weeks, FEI unveils a new No. 2 team after Oregon stepped up and disposed of previously undefeated Arizona State on Saturday. The Ducks' ranking didn't make a quantum leap, but their FEI rating closed half of the gap between then No. 3 Oregon and No. 1 LSU. The Ducks also separated from the rest of the pack a bit, with the difference between No. 2 and No.3 now about equal to the difference between No. 3 and No. 12.

The Sun Devils found out in a hurry how dangerous Oregon's offense is this season, especially in long fields. Oregon scored touchdowns on its first three drives of the game to take a 21-3 lead, all from starting field position inside their own 30-yard line. The Ducks added an 80-yard drive in the second half to take a commanding and demoralizing lead. Those four touchdowns represent 33 percent of the long-field scoring given up by Arizona State all season. Oregon has scored more points on long-field drives than all but two other teams in football this season, and adjusted for field position and opponent defenses, is the most prolific offense in the nation in long-field situations.

LSU, meanwhile, overcame yet another second-half deficit to close the door on Alabama and virtually lock in a spot in the SEC championship game. Only pushovers Louisiana Tech and Ole Miss remain before a regular season finale against Arkansas. The Tigers and Ducks have now distinguished themselves in both FEI and the eyes of voters as the class of the two best conferences in football, but they won't meet each other at season's end in the BCS unless Ohio State slips up.

For their part, the Buckeyes have moved up into a virtual dead heat with No. 3 West Virginia. As they did last season, Ohio State leads all teams in Field Position Advantage, starting its drives on average 10 yards further downfield than its opponents. Their opponents, in fact, have started only two competitive drives all season on the Ohio State side of the field, and the Buckeyes shut out the opposing offense on both drives. In all of college football, there have been a total of 1,232 competitive offensive drives this season begun at or inside the opponent's 45-yard line. Ohio State has faced just one of those drives. (West Virginia faced the fewest short-field drives in 2006 with only three). OSU closes out the season with Illinois and Michigan, the best two opponents on its schedule, but two it should still shut down if it can continue to dominate the line of scrimmage, play soundly on special teams and avoid costly turnovers.

As for No. 5 ... we'll get to them in a moment.

Breaking Down FEI

Plenty of the commentary upon the release of these ratings each week has revolved around the question of reliability. Does FEI reveal something intrinsic about the teams themselves, predict future results, mesh with the polls, weigh conference strengths properly, breakdown team units to reveal strengths and weaknesses, outperform Sagarin, react appropriately to losses, or determine the best method for crowning a college football champion? Beyond all that, given the proven ability of play-by-play-based DVOA to correlate strongly year to year and VOA to correlate strongly with winning percentage, what can FEI actually be expected to do with more limited data and four times as many teams?

As mentioned in the 2007 Preview article, the correlation of winning percentage from year to year is stronger in college football (0.59) than it is in the NFL (0.24). Even stronger are the year-to-year correlations of college football team Pythagorean Wins (0.67) and FEI (0.76). Consistency and stability in the FEI rankings, though, could be seen as a valuable asset of the system or as a handicap. LSU at the top of the FEI rankings four consecutive weeks? Possibly insightful. South Florida remaining near the top after three losses? Back to the drawing board.

By design, FEI strives for consistency in evaluating every team's resume, crediting playing well against good teams win or lose, and punishing losing to poor teams more harshly than it rewards defeating poor teams. But what defines a good team or a poor team? To determine this, let's step away from FEI for a moment to consider another ranking system altogether, the College Football Ranking Comparison compiled by Kenneth Massey.

The consensus rankings Massey produces on his site constitute an average ranking for each team from more than 100 computer ranking systems. Though all of the individual systems weigh game results differently and disagree somewhat on individual team rankings across the board, the consensus rankings can be understood to be a reasonably reliable and balanced view of the relative position of every team in college football.

I divided the 2005 and 2006 final Massey consensus rankings into the following distribution of types of teams:
Elite (teams ranked 1-5)
Very Good (teams ranked 6-20)
Above Average (teams ranked 21-40)
Average (teams ranked 41-80)
Below Average (teams ranked 81-100)
Very Bad (teams ranked 101-115)
Awful (teams ranked 115-119)


Games between Team Type and Opponent Type, 2005-2006
Team Type Games vs. Elite vs. Very Good vs. Above Avg. vs. Avg. vs. Below Avg vs. Very Bad vs. Awful
Elite 123 (5-5) (15-4) (29-3) (38-1) (18-0) (4-0) (1-0)
Very Good 359 (4-15) (30-30) (57-21) (113-8) (47-2) (25-0) (7-0)
Above Average 486 (3-29) (21-57) (61-61) (117-28) (51-8) (40-0) (10-0)
Average 901 (1-38) (8-113) (28-117) (154-154) (115-25) (92-15) (40-1)
Below Average 450 (0-18) (2-47) (8-51) (25-115) (45-45) (60-20) (12-2)
Very Bad 341 (0-4) (0-25) (0-40) (15-92) (20-60) (34-34) (15-2)
Awful 96 (0-1) (0-7) (0-10) (1-40) (2-12) (2-15) (3-3)

I also broke down Game Efficiency data from every game in 2005 and 2006 into the following distribution of winning and losing performance:

Narrow Win/Loss (games decided by one possession)
Solid Win/Loss (games decided by between one possession and one standard deviation of national GE)
Dominant Win/Loss (games decided by between one and two standard deviations of national GE)
Destroying Win/Loss (games decided by more than two standard deviations of national GE).


Team Type records in Types of Games, 2005-2006
Team Type Destroying W/L Dominant W/L Solid W/L Narrow W/L
Elite (13-0) (38-1) (36-2) (23-10)
Very Good (18-1) (93-16) (87-24) (85-35)
Above Average (21-6) (63-29) (102-60) (117-88)
Average (19-15) (98-89) (167-180) (154-179)
Below Average (0-20) (29-79) (47-113) (76-86)
Very Bad (2-20) (11-83) (29-71) (42-83)
Awful (1-12) (1-36) (1-19) (5-21)

Though the Fremeau Efficiency Index does not use this data directly, many of the principles of FEI were fine-tuned this off-season with these distributions in mind. Elite teams lose infrequently, of course, and when they do, most are narrow losses. Elite and very good teams are rarely destroyed. Almost all destroying wins are performed by Average or better teams. These charts are not particularly earth-shattering, but it is important to use this data to help classify teams when evaluating new data.

Back to 2007 and, once again, the South Florida Bulls. One of the reasons it is difficult to accept USF's high ranking is that its three losses have come in the last three weeks -- even if they were once the No. 2 team in the country, they absolutely must not be anymore. FEI does not weight recent results at the expense of early-season results, so the order of games doesn't make a difference. All three of USF's losses were Narrow, and two of them came against teams currently classified by the Massey Consensus as Very Good and Above Average. Add to that two victories over Very Goods (West Virginia and Auburn), and South Florida finds itself in an awkward position. Is it more appropriate to drop USF behind Connecticut and Cincinnati, or keep them ahead of West Virginia and Auburn? FEI keeps the Bulls behind WVU and ahead of the other three. The Massey Consensus places the Bulls behind WVU, Auburn and Connecticut and ahead of Cincinnati. In both cases, three on-the-field results are violated by the USF ranking, an inevitability for virtually every permutation.

Can a team that loses to Connecticut, Cincinnati and Rutgers possibly be one of the top 10 teams in the nation? It is a more challenging question than may be apparent. USC last season lost to Oregon State and UCLA, both Above Average teams in the Massey Consensus ratings. California (10-3; 2006 Massey Consensus No. 11) lost to an Elite (USC), an Above Average (Tennessee) and an Average (Arizona). South Florida's loss resume currently compares favorably with the 2006 Bears, and USF has already notched two victories better than any Cal tallied last season. The logic of previous seasons may simply not be a perfect fit for the up-and-down 2007 season, and perhaps South Florida has assured its poster-child billing. That said, I do anticipate that simply playing the abysmal Syracuse Orange will negatively impact their rating next week.

It was also suggested that USF's high rating was skewing its opponents' ratings and severely distorting FEI across the board. Just for fun, I ran a set of FEI Ratings Sans USF, removing them from FBS and rating the other 118 teams based on the "un-distorted" results.


FEI Ratings Sans USF
Rank Team W-L *FEI*
1 LSU (8-1) 0.312
2 Oregon (8-1) 0.275
3 West Virginia (7-0) 0.240
4 Ohio State (9-0) 0.231
5 Oklahoma (8-1) 0.227
6 Auburn (6-2) 0.215
7 Florida (5-3) 0.212
8 Arizona State (8-1) 0.202
9 Boston College (7-1) 0.201
10 Florida State (6-3) 0.192
11 Missouri (7-1) 0.187
12 Clemson (6-2) 0.187
Rank Team W-L *FEI*
13 Alabama (5-3) 0.183
14 Virginia Tech (6-2) 0.172
15 Georgia (6-2) 0.166
16 Arkansas (5-3) 0.165
17 California (6-3) 0.163
18 USC (7-2) 0.156
19 Kansas (8-0) 0.148
20 Kentucky (5-3) 0.147
21 BYU (5-2) 0.147
22 Michigan (8-1) 0.138
23 Wake Forest (6-3) 0.135
24 Connecticut (6-1) 0.128
25 South Carolina (5-4) 0.127

Not a lot changes, but if you prefer a college football world that pretends South Florida's big wins and tough losses never happened, feel free to use this handy output.

Aside from South Florida and several other Underrated/Overrated teams previously identified this season, there isn't a great deal of disparity between the poll ratings and FEI this week, with one significant exception.

Overrated

Kansas (8-0; No. 4 BCS, No. 5 AP, No. 19 FEI)
It was a banner day for the Jayhawks and their fans as the team expunged decades of frustration by humiliating Nebraska 76-39 on Saturday. How in the world did that absurdly dominant effort result in a nine-spot drop in their FEI ranking? The problem, as has been the case with most of the teams designated Overrated by FEI this season, is that their opposition as a whole has been entirely underwhelming. Their best opponents to date -- Colorado, Texas A&M, and Kansas State -- all lost badly themselves over the weekend, and as has been well-documented, the Jayhawk non-conference schedule is embarrassingly bad. Only Missouri at the end of the regular season should provide a true test for Kansas before the potential Big 12 championship game.

Posted by: Brian Fremeau on 07 Nov 2007

30 comments, Last at 11 Nov 2007, 12:54am by Sid

Comments

1
by Kal (not verified) :: Wed, 11/07/2007 - 4:40pm

What three other teams have scored more in long-field situations than the Ducks?

2
by PHn (not verified) :: Wed, 11/07/2007 - 4:45pm

FEI predicts a very good Cal-USC contest this week. Even there are no national implications, it should be a tough game.

3
by Kal (not verified) :: Wed, 11/07/2007 - 5:55pm

I actually don't think that's going to happen and also why FEI is a bit poor for rating college in general. Rating the early games the same as the late games in the season removes things like trends and averages out overall performances. Some times that's good, and some times it loses some information.

For instance, FEI can't tell that Cal is on a big losing streak and that their QB is hurt and not playing as well as he was early on. It can't tell that USC was without its first-string QB and that Booty is a better fit.

That would be something I'd like to see; aside from FEI, a weighted FEI (weighted towards the end of the season) would be a good thing and possibly a better predictor than flat FEI.

4
by footballprofessor (not verified) :: Wed, 11/07/2007 - 6:46pm

Maybe I didn't see it, but where is Hawaii in all of this? Aren't they 8-0??? Their schedule may be absolute %#@$, but they've been thrashing their opponents.

5
by Dave (not verified) :: Wed, 11/07/2007 - 7:01pm

#4, err, no, they haven't been...

09/02 N Colorado W 63-6 (I-AA)
09/08 at La Tech W 45-44 (4-5, 5th place in the WAC)
09/15 at UNLV W 49-14 (2-7, 8th place in the MWC)
09/23 C Southern W 66-10 (I-AA)
09/29 at Idaho W 48-20 (1-9, last place in the WAC)
10/07 Utah St W 52-37 (0-9, 8th place in the WAC)
10/12 at San Jose St W 42-35 (3-6, 6th place in the WAC)
10/28 New Mexico St W 50-13 (4-6, 7th place in the WAC)

So the best team Hawaii played was an LA Tech team that's the middle team in their conference -- and they won by one point. In overtime.

6
by Roscoe (not verified) :: Wed, 11/07/2007 - 7:49pm

#3 All of the other rating systems (i.e. AP and USA Today, and by extension the BCS) aren't sure what they are trying to do. In other words, you can't tell if the voters are rewarding teams for what they have accomplished on the field, or for how good the teams are at that precise moment. In fact, I think, these ratings are trying to accomplish both at once, and as a result are really doing neither.

At least this FEI system, with its focus on past data and no skewing for early wins and losses, appears to know what it is trying to do.

7
by Seth Burn (not verified) :: Wed, 11/07/2007 - 8:34pm

Phenomenal article.

8
by pm (not verified) :: Wed, 11/07/2007 - 8:34pm

This system is a joke if they still have South Flordia as a top team. Its hard to respect a very flawed system.

9
by Jesse (not verified) :: Wed, 11/07/2007 - 10:26pm

I'm sorry, but saying that Oklahoma at 8-1, Missouri at 8-1, and Georgia at 6-2 (not to mention a host of other teams) are all worse than South Florida at 5-3 makes it impossible for me take these rankings seriously.

10
by Kal (not verified) :: Wed, 11/07/2007 - 10:49pm

#9: why is it so incredible to think that a 5-3 team is better than an 8-1 team?

I do think that USF's ranking is flawed, but I don't think that this is because they're 5-3 and ahead of other teams.

Another possibility is that FEI is putting too much weight based on the strength of the opponent. If you look at FEI, their ranking is 5 but their GE (the VOA to the DVOA) is 24. I don't think they're the 24th best team in the nation, but they're not the 5th best either. Somewhere in the middle is more likely. Same thing goes for Kansas; while they're probably not the third best team in the nation, they're almost certainly better than 19 and are getting creamed by their poor schedule. It'd be interesting to run various coefficients and test which is better of a predictor for future games based on this year.

11
by The Boilermaster (not verified) :: Wed, 11/07/2007 - 11:41pm

@10:

Excellent analysis. Strength of schedule adjustment seems too large, and as with all parameterizations, the coefficient should be determined a posteori, not a priori (as seems to be the case for the FEI).

12
by The Boilermaster (not verified) :: Wed, 11/07/2007 - 11:56pm

To further the point of the strength of schedule adjustment being too large....check this out.

I took the average of each teams FEI and non-adjusted EI. Assumming (and this is a sketch assumption) that there's a linear relationship of (FEI - EI) as a function strength of schedule coefficient, the table below is nominally the result of halving the impact of strength of schedule on the rankings.

Team (FEI+EI)/2
OSU 0.3
WVU 0.29
LSU 0.27
OU 0.27
Oregon 0.26
Kansas 0.24
ASU 0.21
Mizzou 0.2
Clem 0.2
BC 0.18
USC 0.18
Uconn 0.18
Ark 0.18
Cinc 0.17
Florida 0.17

Seems subjectively better to me.

13
by Brian Fremeau :: Thu, 11/08/2007 - 12:27am

#1:

There are actually only two teams that have scored more in long-field competitive possession situations than Oregon: Texas Tech and Tulsa. The sentence has been fixed.

14
by Jesse (not verified) :: Thu, 11/08/2007 - 12:43am

10- That's probably a fair way of putting it.

15
by Gary (not verified) :: Thu, 11/08/2007 - 2:58am

Re: 12

That seemed like it may have swung a little too far in the other direction. Here it is again with FEI receiving double weight:

1 LSU 0.28297
2 Ohio State 0.27504
3 West Virginia 0.27163
4 Oregon 0.262603333
5 Oklahoma 0.251903333
6 Kansas 0.206536667
7 Arizona State 0.203856667
8 Missouri 0.194553333
9 Clemson 0.19433
10 Boston College 0.185413333
11 South Florida 0.185166667
12 Connecticut 0.18365
13 Florida 0.182106667
14 Auburn 0.178706667
15 Arkansas 0.172193333
16 USC 0.17119
17 Cincinnati 0.171053333
18 Florida State 0.152856667
19 BYU 0.15066
20 Michigan 0.140186667
21 Alabama 0.139866667
22 Virginia Tech 0.132383333
23 California 0.130173333
24 Hawaii 0.127836667
25 Georgia 0.12641

It's hard to argue too much with that list.

16
by Becephalus (not verified) :: Thu, 11/08/2007 - 2:59am

I personally enjoy the FEI posting each week, one of the only things I can bear to read about college ball. I am perfectly fine with USF being three. I really think people drastically overestimate what the data warrants with their individual team.

Oh my team beat 8 teams ranked 35-110 and lost to one team ranked 17. CLEARLY they are better than a team that lost to 12 7 and 21, but beat 11 and 8.

Anyway keep up the god work.

17
by Becephalus (not verified) :: Thu, 11/08/2007 - 3:02am

that is supposed to be "there" not "three".

18
by Brooklyn Bengal (not verified) :: Thu, 11/08/2007 - 12:55pm

Gary, that list is pretty freaking awesome...er, subjectively speaking, of course.

19
by Dennis (not verified) :: Thu, 11/08/2007 - 1:14pm

Re #3: Rating the early games the same as the late games in the season removes things like trends and averages out overall performances. Some times that’s good, and some times it loses some information.

If the justification for not having a playoff is becase it would take away the "every game counts" factor, then all games should count the same in the rankings. If you count recent games more than early games, then you lose "what makes college football special" and you're on the path to a playoff.

20
by Kal (not verified) :: Thu, 11/08/2007 - 4:31pm

#19:

No, that's not what I'm saying at all.

If you want to use FEI as a way of evaluating a team's entire season and how well they did throughout the year, having no weights is a good thing. This would be good for arguing what teams should and shouldn't be in bowl games.

If you want to use FEI as a predictive tool to determine based on prior behavior who is more or less likely to win next week, using weights is likely a better thing. Since Brian uses this each week to predict the next week's scores and evaluates FEI based on that success, I'm saying that it's probably true that it would be even more successful if you weighed recent games more heavily than older games.

I think both are valuable for the same reason that I think weighted DVOA and DVOA are valuable.

21
by The Boilermaster (not verified) :: Thu, 11/08/2007 - 5:24pm

@19...

I don't get it...It seems extremely clear to me that pollsters ALREADY value recent games far far more than early games. Just look at Michigan. Do you really think that if they were a one-loss team coming in to this weekend's (rescheduled) matchup vs. Appalachian State, and LOST, that their ranking would be the same as it is now?

I would go so far as to argue that hardly any voters consider early game performance as more than an abstract boolean value (win=true. loss=false). It's been discussed many times in other threads, but the whole idea that "every game counts" in College Football is just simply a fallacy. Sample criterion for a game counting:

1) You're a big program that loses to a weak opponent.
2) You're a mid-tier program that beats a quality opponent.
3) You're ranked in the top ten at the time of your game.
4) You're a ranked team that blows out a mid-tier or better team.
5) The game is on national TV.

Does anyone really care about more than, say, 2-4 games per week (beyond their hometown teams')?

22
by Eddo (not verified) :: Thu, 11/08/2007 - 5:33pm

21: I'll bite: let's say the Appalachian State game was moved to this weekend, and we say that every game had been bumped up. Michigan's only loss would be to #3 Oregon, so I'd say Michigan could be ranked as high as #5, as low as #10.
So now, if they lost a close game to Appalachian State, they'd probably drop at most 10 places (look what happened to USC when they lost to Stanford). Therefore, Michigan would be somewhere in the 15-20 range.
So yes, I'd say recent games matter (it would cost Michigan ~5 spots in the rankings). However, the same loss this weekend would only drop them ~10 places, as opposed to all the way out of the rankings, like it did in September.

23
by Bayou_Fan (not verified) :: Thu, 11/08/2007 - 5:48pm

While the basis for this list is developed from sound reasoning, I believe that Kal's post hit the nail on the head. Not weighting recent games more than earlier games means that this tool will not be able to pick out short term trends from the noise of the entire season. If you want it to be more predictive of next week's performance, you need to weight it for recent games. This allows it to adjust more for injuries and changing tactics affecting a team's overall efficiency.

Now, if you want to run two seperate result columns where one is weighted and one is unweighted, you can have a tool that produces both predictive output (weighted) and total season body of work output. However, I would suggest that the unweighted column take into account what the rating was for each team at the time they played each other, and not the rating that they currently have. For example, beating USC when their 1st string quarterback was still playing is more impressive than beating them after he was injured. I'd probably tweak the system further by making the first three weeks of games use the week three ratings. The system has seed data to base opponent adjustments on for the first game and the second game is likely also a crap shoot. By the third game, there is enough for rough comparisons to be made.

Then, those saying the body of work should dictate bowl placement can use the unweighted column and those that recent performance is more indicative of true placement can use the first column.

I am personally in favor of a limited playoff for college football. I'm not a "plus one" fan as I think that, while its an improvement, it is still going to be too restrictive for second tier conferences/teams to make it into the hunt. I favor 8 seeded teams, using 4 of the early bowls as playoff games, then, the 4 BCS bowls are rotated through 2 being the second playoff round, one being a consolation game (to establish #3 and #4) and the other to be the championship game.

The scheduling of the bolws themselves doesn't have to change. Early bowls are plenty early enough and the BCS bowls are late enough and spread far enough apart. 8 teams gives a good chance for a non-bcs conference to make the playoff. The payout for all of the playoff bowls would likely increase as greater interest would be on them. Yes, the consolation BCS bowl would likely get a bit of a letdown, but, already every year, at least one of the BCS bowls underperforms enough to be considered a letdown financially. At least, at this point, even the consolation game should be a decent game as the opponents should be relatively decently matched. As for the complaining about there being too many games. Only the top 4 will play two extra games. The rest of the 119 teams will play either their regular schedule, or, their regular schedule plus a bowl. Those two extra games will represent a large chunk of change paid out to those schools. This spreads the wealth even more than before. And, for the purists, it is guaranteed to involve the top 8 teams. No more getting screwed by the BCS bowls by dipping down to #10 or 11 because their fans travel better and leaving out a good #7 or 8.

As for the conference/bowl tie ins, this won't neccessarily break them. Lets be real here, it is unlikely that the conference champion of any BCS conference will be ranked below #10 in any poll. They'll likely all qualify for the playoff. If there's a year where a three loss team makes it out of a BCS conference, and one of those losses is an early year embarrassment against a I-AA team, then, the tie-in deserves to be broken. Aside from the fans of that school, no one else is going to want to see them in a playoff and their presence will hurt the bowl ratings anyway.

So, this playoff will be fair to the teams, lucrative for the bowls and teams, and generate the least amount of controversy. I will only negatively impact a maximum of four teams academically at most (and, the break between the first round of the playoffs and the second round should be about two weeks anyway, allowing them time to take their finals). What's not to like?

24
by Kal (not verified) :: Thu, 11/08/2007 - 10:10pm

Those two extra games will represent a large chunk of change paid out to those schools. This spreads the wealth even more than before.

I don't think that's really true. We'll assume that there is about the same money involved now as there was before. Okay, you've got 4 extra games, and presumably the teams that play in those games will get that money. That has to be split.

It also means that (by comparison to now) 2 teams will be left out in the cold. The 9th and 10th teams won't make it into the playoffs. That leaves only 2 at-large bids, and you'll be getting into the same issue you had before. Who travels better? How do you determine who the 7th and 8th best teams are? It's all subjective.

If you're going to have an 8-team playoff, you need to make sure that the 8 teams get there in some objective way. I liked the prior idea, where you simply make two more conferences, make sure that all conferences have a championship game or have a clear way to identify the champion (the Pac-10 plays everyone in the Pac-10, so it's easily settled who the best is), take two other teams from conferences and move on. Yeah, you might have times where the two best teams are from the same conference, but that's too bad.

I still like the bowl system we have better.

Alternately, you could say you're getting the same money per game that you do now. Except you're keeping it amongst the top 8 teams. The top team will play in three big bowl games; that doesn't spread out the wealth, that concentrates it.

25
by Pat (not verified) :: Thu, 11/08/2007 - 10:41pm

I’m saying that it’s probably true that it would be even more successful if you weighed recent games more heavily than older games.

Possibly, possibly not. Weighting recent games is throwing out data. If teams don't trend that quickly, you're better off using all data, because the reduction in the variation you see game-to-game is more valuable than the week-to-week trend.

Weighted DVOA stretches back about 10 games or so, if memory serves (ballpark number), which implies that weighting probably wouldn't do much until the very last game in the season.

The other problem you have there is that the early games are typically the out-of-conference games. Reducing their importance essentially turns Division IA into a bunch of disconnected islands.

26
by Pat (not verified) :: Thu, 11/08/2007 - 10:55pm

Also, I said this last week, but I really, really don't understand the offensive/defensive efficiency rankings on their own. Normalizing for field position just seems to throw away the performance of the offense/defense on every drive except scoring drives. In a low-scoring game, this could essentially completely remove the performance of the offense/defense.

Plus, the scale doesn't seem right to me. Offensive efficiency, minus defensive efficiency, should give you some sort of ranking equivalent to FEI, yet doing that ends up with Oregon in front of LSU (0.287 vs 0.27), and Oregon had better field position as well, implying that the portion of the offense/defense/special teams that's contributing to field position is also playing better than LSU.

27
by Sid (not verified) :: Fri, 11/09/2007 - 3:57pm

Duke is ahead of Texas A & M! How is that even possible?

28
by Bayou_Fan (not verified) :: Fri, 11/09/2007 - 7:07pm

re: Duke ahead of Texas A&M, umm, the aggies are not impressing anyone on the field this year.

re: spreading the wealth: I meant among the bowls, not the teams. The top two teams will get big payouts. That is a positive benifit of having a winning program. I don't have a problem with that.

Re: recent games being weighted more: Given the shorter college season as compared to the NFL season (and preseason), I would propose weighting be given to the most recent four games in a +40%, +30%, +20%, +10% manner (as a starting point). It doesn't eliminate the effects of the early season games, it just reduces them in importance in late season rankings. I'm trying to pick up short term trends that have immediate relevance with respect to ranking. Given such a limited number of datapoints per school, its tougher to do in the college game, granted. It should still be done, however. I believe some of the other computer models out there do just that.

re: teams 9 and 10 being left out in the cold: I dare say that there will be fewer tears shed over #9 and #10 with respect to the playoffs then over a slighted #3 or #5 in a plus one system. That being said, and I may not have articulated this well (or at all) in the original post, but, I am actually against the BCS conferences getting automatic bids. Its possible for a 3 or even 4 loss team to win a BCS conference. They might not be ranked that well (may just barely clear BCS eligibility in a muddy year) but could still get in ahead of a very good mid-major.

If one wants to keep the conference champions from the 6 BCS conferences with guaranteed bids, then, the rest of the Div 1A teams should be divided into 4 remaining conferences, each without a conference championship game. These represent the mid-majors and minor teams from Div 1A and excludes non-conference teams (Notre Dame should either join a conference or accept never playing in a BCS bowl). These four conferences, labled A,B,C, and D here should be geographical in nature (East Coast, Central East, Central West and West coast+hawaii and Alaska). In a rotating fashion, they should play an interconference championship (first year, A vs B and C vs D, then A vs C and B vs D the next year, etc) for the final two spots. But that would completely eliminate the need for rankings, except to decide final seedings for the playoffs.

I'm not so much in favor of that as I am in saying that the 6 BCS conference champions plus the two highest rated schools outside of the BCS conferences should play an 8 team playoff to decide who is #1-#4, 5 - 8 are judged by bowl seedings or final rankings. Yes, this opens up the possibility that a non-bcs team with a top 8 rating might get shut out of the playoffs, especially if two or three of the BCS conferences have a down year, but, that is less likely and less painful than snubbing a good #3 or #5 in my eyes.

29
by evo34 (not verified) :: Sat, 11/10/2007 - 1:36am

Brian, the biggest issue I see with your score projections is that the total points are usually too low. Not sure how you would adjust for this, but basically your point totals average about 10-20 points below the over/under total for each game.

Also, it appears that favorites are projected to do better than they actually do. That is, I think the largest source of error in your proj. MOV is that you are over-estimating it. Again, not sure how to fix this, but your system is overstating (on average) the margin by which a better team is likely to beat a lesser team. WOuld be interested to see if you have studied this, and have any plans for improvement.

P.S. In response to the "weighting" proponents, generally speaking (throwing out major injury/personnel-change teams), using equally-weighted full-season data is more accurate to predict future performance than weighting recent data more heavily.

30
by Sid (not verified) :: Sun, 11/11/2007 - 12:54am

RE: 28

Of course not. But is Duke better? You totally ignored the point. I didn't say Texas A & M was a good team this year. I'm saying it's pretty crazy to have them behind Duke.