Dullard's College Football W12 - dramatically new optimizations by request

[dullard]

Hey I do listen to your posts.

There have been a lot more posts complaining about #wins and #losses this year than in the previous years I've posted this. So I thought about how I feel would be the best way to highlight this aspect of the game and to minimize the margin of victory. So I took the same data as in my other program (so yes I'm still using scores) and reoptimized. The other rating that I have been posting optimized to get the most accurate prediction of next weeks scores. Basically the idea is that the best possible prediction of the score is the best possible prediction of how teams should be ranked - plus it gives the added bonus of a slight advantage if you choose to bet on football games. This of course had the disadvantage of placing teams with 1 loss often far behind teams with 2 losses (and on down the line). So two years ago I increased the weight of losses and that is what I've been posting here (it had only a slight negative effect on the score prediction).

Well with even more complaints this year, I thought about it. What if I didn't optimize for score. There is one other thing that can be optimized for: the best prediction of next week's winners. This gives an advantage to teams that play really horribly against bad teams, but still eek out a win in the end. Or an advantage to teams that consistantly go into overtime but win when the opponent's field goal hits the goal post and bounces out. I personally feel that this is a bad way to judge teams and thus I went with optimization for score. But many people disagree - and who cares that you won by luck, you still won.

Differences:
1) The old program optimized for the best score prediction, with a few % decline in the prediction of the winner. The new program optimizes for game winner, with about a 0.7 point greater error on average in the score prediction.
2) The old program put the score cap in the mid 30's: winning by more than the score cap does nothing to your ranking. The new program puts the score cap in the lower teens - a win by 15 is the same as a win by 77, but is better than a win by 10. Lowering the cap further gives worse predictions of the winner AND worse predictions of the score. Raising the cap gives slightly better predictions of the score but slightly worse predictions of the winner.
3) The new program slightly reduces the weight of the score (the old program multiplied the score difference by ~1.15, the new program multiplies the score difference by ~1.05).
4) The new program increases the weight of a win by ~10 times. So now # of wins is significant.
5) The new program increases the weight of a loss by ~4 times. Still it is slightly more imporant than # of wins, but the difference is much less.

With those optimization changes (nothing else changed), here is the new rating system.

Place Rating Team name
01 132.5 Oklahoma
02 129.3 Southern California
03 124.2 Ohio St.
04 123.7 LSU
05 123.3 Georgia
06 122.5 Purdue
07 122.1 Washington St.
08 122.1 Tennessee
09 122.0 Michigan
10 121.6 Texas
11 121.3 Florida St.
12 121.1 Miami Ohio
13 120.6 Virginia Tech
14 120.2 Nebraska
15 120.1 Miami Florida
16 119.2 Pittsburgh
17 119.0 Minnesota
18 119.0 Boise St.
19 119.0 Kansas St.
20 118.9 Utah
21 118.9 Arkansas
22 118.2 Iowa
23 117.7 TCU
24 117.6 Mississippi
25 117.3 Michigan St.
26 117.3 Florida
27 115.6 Oklahoma St.
28 115.3 Texas Tech
29 115.3 Air Force
30 115.1 Northern Illinois
31 114.9 North Carolina St.
32 114.0 Oregon St.
33 114.0 Maryland
34 113.9 Bowling Green
35 113.6 Wisconsin
36 113.4 Auburn
37 113.3 Louisville
38 112.8 Missouri
39 112.2 Clemson
40 111.6 West Virginia
41 111.6 UCLA
42 111.3 Oregon
43 111.3 Southern Miss
44 111.1 Georgia Tech
45 110.8 Virginia
46 110.7 Syracuse
47 110.7 Colorado St.
48 110.6 Alabama
49 110.5 New Mexico
50 110.2 South Carolina
51 109.9 Marshall
52 109.9 Stanford
53 109.8 Northwestern
54 109.8 California
55 109.5 Wake Forest
56 109.4 Connecticut
57 109.2 North Texas
58 108.8 Hawaii
59 108.5 Washington
60 108.1 Kansas
61 107.9 Notre Dame
62 107.9 Boston College
63 107.5 Memphis
64 107.5 Fresno St.
65 106.8 UNLV
66 106.3 Texas A&M
67 106.0 Colorado
68 105.8 Wyoming
69 105.8 Brigham Young
70 105.4 Toledo
71 105.3 Tulsa
72 105.3 San Diego St.
73 105.1 Navy
74 104.7 Kentucky
75 104.6 Louisiana Tech
76 104.2 Arizona St.
77 104.0 South Florida
78 103.9 Houston
79 103.7 Rutgers
80 102.0 Cincinnati
81 102.0 UAB
82 100.8 Akron
83 100.7 Penn St.
84 100.3 Iowa St.
85 100.2 Nevada
86 099.7 Ball St.
87 098.5 Duke
88 098.5 Tulane
89 098.3 Arizona
90 098.3 Western Michigan
91 098.1 Troy St.
92 097.3 North Carolina
93 096.7 Kent St.
94 096.4 San Jose St.
95 096.2 Baylor
96 096.0 Mississippi St.
97 095.5 Utah St.
98 095.5 Middle Tennessee
99 094.3 Rice
100 94.1 Indiana
101 93.7 Illinois
102 93.0 Arkansas St.
103 91.8 Vanderbilt
104 90.9 Louisiana-Lafayette
105 90.5 Ohio
106 90.0 Temple
107 90.0 New Mexico St.
108 89.6 UCF
109 89.1 Central Michigan
110 86.8 Idaho
111 86.6 East Carolina
112 86.2 Louisiana Monroe
113 85.8 UTEP
114 84.8 Buffalo
115 84.3 SMU
116 84.0 Eastern Michigan
117 82.5 Army

 

[mpitts]

Originally posted by: dullard
06 122.5 Purdue
09 122.0 Michigan

I am assuming that there is nothing in your formula that accounts for head-to-head, considering Michigan and Purdue both have 2 losses and Michigan destroyed Purdue 31-3.

 

[dullard]

Originally posted by: mpitts

Originally posted by: dullard
06 122.5 Purdue
09 122.0 Michigan

I am assuming that there is nothing in your formula that accounts for head-to-head, considering Michigan and Purdue both have 2 losses and Michigan destroyed Purdue 31-3.

There is nothing that I added to manipulate the results when two teams are closely rated, but one team beat the other (since doing so would quite possibly give worse score predictions AND worse winner predictions - and since I cannot come up with a fair formula to do it). As you can see the program that gives margin of victory a greater weight, had Michigan ahead of Purdue. But now that Michigan doesn't get credit for that full 28 point win, in this optimization Michigan is slightly behind Purdue. Like I've said before, this will always happen no matter how you rank teams - you will be ranked below someone you beat.

 

[NeoV]

much better!

 

[Tom]

I would think your new system would be better at predicting, which I think is your stated goal. But since you have them both it will be interesting to see.

But I hope you feel that there is a difference between making predictions about likely scores, and deciding who gets to play who in the Championship game, which I feel is and should be based on at least a partly subjective opinion about what teams deserve to play for the Championship the most, I think the coaches and sportswriter's polls add this to the overall BCS ranking.

 

[dullard]

Originally posted by: Dead Parrot Sketch
But I hope you feel that there is a difference between making predictions about likely scores, and deciding who gets to play who in the Championship game, which I feel is and should be based on at least a partly subjective opinion about what teams deserve to play for the Championship the most, I think the coaches and sportswriter's polls add this to the overall BCS ranking.

I don't think there should be anything subjective. That is where we got into trouble in the past. Teams that were undefeated won't play each other, and one gets a championship and another gets nothing (1993 Penn State - if I got the year correct). Or you have teams that have a football bounce off of someone's foot and get caught to win the game and they get a championship solely since their coach is leaving that year (1997 Nebraska). Other more informed people could list many other flaws.

No I think the only fair way would be a playoff. Here is my playoff ideal:
1) 16 team playoff.
2) All major conferences play a conference championship. Lets suppose there are 6 major conferences. Minor conferences can play one if they want, it might help get them an at large bid for the playoff.
3) If there are 6 big conferences that gives you 12 of those 16 teams in the playoff (and their conference championship is the first game of the playoff). Using the BCS formula (or a similar one) we find the top four teams that didn't make it to those 6 major conference championship games. The four slots could go to anyone: someone not in a major conference, or to someone who was just outside their conference championship game. These four teams are paired off (in any mannor I don't care) and play one game (in one of the minor bowl locations).
4) Now we have 8 teams left. Do a normal playoff with those teams - again using the minor bowls for this purpose.
5) Now we have a fair method of choosing the national championship game - nothing subjective. The season doesn't go past January 1st. The bowls still exist and all get important games. Most teams play the same number of games. A few play 1 or 2 games more than they would have otherwise (only the two national championship teams play 2 or 3 games more than normal). The teams just outside the 16 chosen can go to the remaining minor bowls if chosen.

 

[sciencewhiz]

Considering that almost all teams play 12 game schedules, and only a few teams play 13 (for a conference championship) Isn't the # wins almost the same as # losses?

 

[dullard]

Originally posted by: sciencewhiz
Considering that almost all teams play 12 game schedules, and only a few teams play 13 (for a conference championship) Isn't the # wins almost the same as # losses?

Yes they are almost the same now. But in prior years, things weren't the same. Nebraska recently played 14 games by adding two preseason games. Other teams might have a game cancelled (for a variety of reasons) and thus have only 11. So is a 13-1 team as good as a 10-1 team? Its tough to say. I find that giving teams points for # of wins helps the predictions slightly.

Also during the season the # of games played can vary dramatically. Stanford has played 8, so has Syracuse. Many other teams have played 11. How Does Syracuse's 5/3 record compare to Kansas States 8/3 record? In these middle weeks, # of wins in my opinion has an important meaning.

 

[Tom]

I don't disagree with you about a playoff system being more objective, but even then it's subjective because you got to pick who gets in the playoffs. Like the basketball playoffs always have to leave a few teams out of the 64.

My point about the good aspect of the subjective part of the BCS is that any attempt to be objective is going to be flawed and so as long as there isn't any objective way to do it, might as well be upfront about the subjectivity.

 

[Mookow]

I like the playoff idea, except I think that the first game of the playoff shouldnt have to be a conference championship game. I would say that a random drawing for who faces who would be nice... although, maybe take the top 8 ranked teams, give them home field advantage, and then randomly draw.

 

[Alkaline5]

Re-read dullard's idea again. By making the first round of the playoffs conference championship games you remove all subjectivity for those 12 teams involved. They've already proven that they're the best in their respective divisions. Conferences like the Big East or ACC would have to create divisions to make this work. Choosing the remaining 4 wildcards would still be based mostly on opinion, but the weaker teams would be weeded out quickly. Most importantly, all the big boys would finally get a shot at each other w/o being at the mercy of sports writers or computer rankings.

 

[dullard]

Originally posted by: Alkaline5
Re-read dullard's idea again. By making the first round of the playoffs conference championship games you remove all subjectivity for those 12 teams involved. They've already proven that they're the best in their respective divisions. Conferences like the Big East or ACC would have to create divisions to make this work. Choosing the remaining 4 wildcards would still be based mostly on opinion, but the weaker teams would be weeded out quickly. Most importantly, all the big boys would finally get a shot at each other w/o being at the mercy of sports writers or computer rankings.

What I stated isn't entirely objective. As I posted it, I let the conferences choose who are their best 2 teams. Sometimes that is subjective, but if those teams choose that method, then let that conference be subjective. Yes there is a 3rd best team that really is left out (which is why there is still 4 open wildcard slots). Hopefully though, the BCS or some other mostly objective method will be used for the selection of those wildcards.

I choose the conference championship to be the first playoff game because the conferences are generally already seperated (there are teams that have no chance, teams in the middle of the pack, and usually 1-3 good teams at the top). It makes choosing the teams easy and fair (since most will have played each other anyways). Plus the conference championship is already set up in several of the conferences, why ruin what is already in place?

Sure there will need to be some changes. Some conferences will need to add teams, and hopefully in doing so the conferences will stay relatively balanced in toughness. Things don't need to be perfectly balanced since there are 4 wildcard slots. I suppose in a strange year, 6 teams could all come from the same conference (4 wildcards + the top 2 in that conference championship game).

This is just my ideal. I haven't heard anything better - but my ears are open to suggestions.

 

[sciencewhiz]

One thing that I find interesting is that in the new rankings extremely narrows the gap between Oklahoma and USC. That seems nonobvious, since Oklahoma had more wins, less losses, and has beaten the MOV cap all but 2 games. On the other hand, USC has beaten the MOV cap in all but the game they lost.

With the new rankings, is home field advantage factored the same?

 

[Tom]

"Re-read dullard's idea again. By making the first round of the playoffs conference championship games you remove all subjectivity for those 12 teams involved. "

What if the six of the twelve best teams are all in the same conference ? The subjective decision to give twelve spots to 6 conferences is still subjective, even if it doesn't seem that way on the surface.

 

[MaxDepth]

 

[Alkaline5]

Originally posted by: Dead Parrot Sketch
What if the six of the twelve best teams are all in the same conference ? The subjective decision to give twelve spots to 6 conferences is still subjective, even if it doesn't seem that way on the surface.

I did forget about the minor conferences. I suppose if you wanted to be completely fair, then you could assign the remaining 4 spots to the WAC, MAC, C-USA, and MW champions. But for now it's quite obvious that (except for a couple of teams like TCU) those conferences aren't on the same level of play... that's a fact. Maybe a periodic reevaluation would be neccessary, but as it stands right now, I'd call that as objective a system as is possible.

 

[sciencewhiz]

For a playoff system, I think that maximizing the number of at large picks is important.

Both of my scenarios are based on next year's realignment.

For an 8 team playoff, Give the 5 major conferences (Pac 10, Big 10, Big 12, ACC, SEC) one automatic bid, that is forfeited if their top team is not in the top 16. Give one automatic bid to the highest ranking non major conference team. Have two at large picks.

For a 16 team playoff, give the 5 major conferences one automatic bid, that is not forfeited. Give 1 automatic bid to the highest ranked team between the Big East and MWC and 1 automatic bid to the highest ranked team (or independent) in the other conferences. Then use the rest for the at large bids. Do not have a conference championship game. Right now, with a 12 game regular season, conference championship, and bowl game, teams are finding room for 14 games. With a normal 11 game regular season, and maximum of 4 playoff games, you play a maximum of 15 games.

 

[Tom]

I had this idea about a way to include margin of victory, based on a concept that it should be considered margin of loss rather than margin of victory. What I mean is that rather than rewarding the winning team for the point spread instead you punish the loser.

One way to do it.

give a team 100 points for a victory.
give a team -100 for a loss, plus -(point spread.)

so Oklahoma beats Texas A&M 77-0

Oklahoma gets 100, Texas A&M gets -177.

Then somehow or other these figures are used to determine some SOS value that is recalculated each week, as your current system does, and this is factored in somehow.

Just a thought. I was thinking that point spread may tell more about how unequal a match up is than how good the winner is, so using it to determine mostly SOS factors might be more useful.