2010 Strength of Schedule approximation
by gtne91 on Jul 2, 2010 2:23 PM EDT
Email
PrintI was mentioning this in another thread and figured I would post details. I used my favorite computer ranking program ( I have a few), and "predicted" each game for this year based on order of finish last year. Then, for each of the 120 FBS teams, I had the prediction modified so they went 12-0 with others game held the same and ran the program, producing a rating number for each team, IF they went undefeated. Since there difference between ratings for two undefeated teams is their strength of schedule (this program doesnt consider score), the rating if they go undefeated is a proxy for the SOS. Some numbers for context:
Last year, at 12-0 (pre conf championship game) Texas was #1 and Bama #2 by this system, with a rating of 3.961 and 3.919. After getting to 13-0, Alabama moved ahead of Texas. So, all of the top #36 have a schedule strong enough to be ahead of Texas and Bama after week 12 last year. For those wondering, Cincy was the worst of the 12-0 teams at 3.703 (scheduling SE MO St killed them). At the bottom, Western Michigan's 3.069 would have been ranked 27th after getting to 12-0. Thats a bad schedule.
1. UCLA 4.371
2. Washington St 4.356
3. North Carolina 4.345
4. Oregon St 4.301
5. Washington 4.270
6. Miami FL 4.270
7. Virginia Tech 4.229
8. Duke 4.228
9. South Carolina 4.183
10. Oklahoma 4.165
11. LSU 4.150
12. Vanderbilt 4.119
13. Iowa St 4.119
14. Southern Cal 4.111
15. Georgia Tech 4.107
16. Texas A&M 4.073
17. Michigan 4.057
18. Oklahoma St 4.054
19. Mississippi 4.050
20. Stanford 4.044
21. Illinois 4.033
22. North Carolina St 4.021
23. California 4.020
24. Minnesota 4.017
25. Florida 4.012
26. Auburn 4.007
27. Arizona 4.002
28. Florida St 4.001
29. New Mexico 3.999
30. Colorado 3.996
31. Arkansas 3.994
32. Notre Dame 3.992
33. Tennessee 3.986
34. UNLV 3.973
35. Arizona St 3.970
36. Penn State 3.968
37. Texas 3.937
38. Texas Tech 3.932
39. Pittsburgh 3.930
40. Wyoming 3.921
41. Iowa 3.893
42. Alabama 3.892
43. Mississippi St 3.890
44. West Virginia 3.880
45. Georgia 3.874
46. Baylor 3.871
47. San Jose St 3.850
48. Utah 3.848
49. Virginia 3.837
50. South Florida 3.833
51. Oregon 3.830
52. Ohio State 3.827
53. Connecticut 3.812
54. Fresno St 3.808
55. Boston College 3.791
56. Wake Forest 3.786
57. Louisiana Tech 3.786
58. SMU 3.775
59. Rutgers 3.773
60. Kansas 3.771
61. Syracuse 3.768
62. Nevada 3.759
63. Boise St 3.754
64. Maryland 3.752
65. Kansas St 3.752
66. TCU 3.750
67. Nebraska 3.743
68. Brigham Young 3.741
69. Utah St 3.729
70. Wisconsin 3.724
71. Missouri 3.724
72. Louisville 3.721
73. Colorado St 3.718
74. Cincinnati 3.708
75. Northwestern 3.707
76. Hawaii 3.705
77. Air Force 3.696
78. New Mexico St 3.691
79. Marshall 3.682
80. East Carolina 3.674
81. Toledo 3.647
82. Miami OH 3.647
83. Michigan St 3.630
84. Memphis 3.627
85. Rice 3.615
86. Temple 3.612
87. Idaho 3.609
88. Clemson 3.560
89. Kentucky 3.549
90. Indiana 3.533
91. Tulane 3.526
92. San Diego St 3.508
93. Navy 3.507
94. Louisiana-Lafayette 3.497
95. Eastern Michigan 3.485
96. Tulsa 3.433
97. Houston 3.419
98. Purdue 3.416
99. Bowling Green 3.410
100. Southern Miss 3.385
101. Louisiana-Monroe 3.361
102. Alabama-Birmingham 3.358
103. Central Michigan 3.323
104. Arkansas St 3.318
105. Florida Atlantic 3.295
106. Florida Intl 3.274
107. Ohio U. 3.263
108. Northern Illinois 3.257
109. Western Kentucky 3.233
110. Troy 3.214
111. UTEP 3.194
112. Kent St 3.160
113. Ball St 3.149
114. Middle Tennessee St 3.145
115. Central Florida 3.145
116. Army 3.115
117. North Texas 3.114
118. Buffalo 3.106
119. Akron 3.086
120. Western Michigan 3.069
FanPosts are user-submitted, and thus NOT representative of EDSBS editorial or any of our opinions unless posted by us ourselves. Please refrain from posting blatant spam or self-promotion, because this makes us hate you. Thanks!
36 comments
|
0 recs |
Do you like this story?
Comments
I like the method, even if I don't care for the results
I still think Penn State has the most difficult road to undefeated of any team: they play three 2009 BCS bowl game winners on the road: (Alabama, Iowa, Ohio State), three other 2009 bowl teams (Minnesota, Michigan State, Temple), and the winningest team in college football history in Michigan. I also think if PSU had UCLA’s schedule, you could pencil them in for at least 10-2 (Texas and USC would be the only speedbumps).
That being said, I do like the method.
Idle talk and hollow promises; cheating Judases; doubting Thomases
absolutely agree
But the eyeball test heartily disagrees. For instance with Miss. State (and these are just the bowl teams):
At Auburn, Arkansas, MTSU, Kentucky
and Home v. LSU, Ga. Tech, Houston, Florida, Alabama, Ole Miss.
I don’t see a whole lot of teams clamoring for that “43rd most difficult” schedule.
"Hush now, let it go now. I know it's time to go. Time to let this fall from my hands" VNV Nation, "From My Hands"
by Stuck in the Plains on Jul 2, 2010 8:18 PM EDT up reply actions
Miss St
plays Houston, UAB, Memphis, Alcorn St OOC in 2010
Looking at wrong year. Thanks.
"Hush now, let it go now. I know it's time to go. Time to let this fall from my hands" VNV Nation, "From My Hands"
by Stuck in the Plains on Jul 3, 2010 11:37 AM EDT up reply actions
I hate when a method I like gives me results I dont
Fortunately, not the case this time around for me. But Ive had it happen many times.
The P10 is rewarded for depth. 9 conference games does that for them. Washington St also gets the bonus of not haviing to play Washington St. :) That kind of thing hurts Penn St, as they dont get to bump up their strength by playing Penn St. In a true round robin, the last place team will always have the toughest SoS.
Which of the following games...
….dragged Purdue to the lowest ranking among BCS schools?
a) Western Illinois
b) Ball State
c) Toledo
d) Notre Dame
hint: this private independent school also torched the schedules of Western Michigan (120 of 120,) Army (116,) Tulsa (96) and Navy (93.)
That would be Ball State...
weren’t they winless?
"Hush now, let it go now. I know it's time to go. Time to let this fall from my hands" VNV Nation, "From My Hands"
by Stuck in the Plains on Jul 3, 2010 11:37 AM EDT up reply actions
Western Illinois primarily
They were 1-10 last year.
Ball St hurt too, they were ranked last year below a significant number of FCS teams. Western Illinois was below a signifcant number of D2 teams.
Playing a very bad FCS team gets you crushed
Bama’s schedule must be otherwise tough to survive Ga St as well as they did.
It's not a walk in the park
There are lightweights (Ga. St., San Jose St., probably at Duke)
Home v. Penn St., Florida, Ole Miss, Miss. St., Auburn
Roadies at So. Car., Arkansas, Tennessee, LSU
Final Top 25 teams will likely be PSU, LSU, Auburn, So. Car., Arkansas, and Florida; probably 8-9 bowl teams in all.
"Hush now, let it go now. I know it's time to go. Time to let this fall from my hands" VNV Nation, "From My Hands"
by Stuck in the Plains on Jul 3, 2010 2:27 PM EDT up reply actions
Big deal is
Florida rotating on from the East, Duke didnt completely suck last year, and Penn St. Missing Vandy from the East also prevents that from pulling them down.
agreed.
and, this is the year that Duke gets to the scoreboard…jsut have a feeling.
"Hush now, let it go now. I know it's time to go. Time to let this fall from my hands" VNV Nation, "From My Hands"
by Stuck in the Plains on Jul 3, 2010 3:36 PM EDT up reply actions
Got a huge issue...
with any system that would reward a team for playing 11 games rather than playing those 11 and a 12th game vs. a bad team. Or one that could have a game versus Ball State hurt more than one versus Florida could help.
Would really like an additive system rather than one for averages.
Not only that, but
a system that says 12 games over a 14-15 week span is just as hard as 12 games over a 12-13 week span.
This is one area where the Big Ten has really gotten short changed in recent years.
Idle talk and hollow promises; cheating Judases; doubting Thomases
That is there own fault. They have the same scheduling possibilities as anyone else.
Wisconsin, for example, tends to take advantage of it, scheduling a late OOC game after Big10 play.
Actually, for teams with a championship game
if is 12 over 13(14) weeks or 13 over 14(15) weeks. But once again, that is their own damn fault, you dont have to have a championship game, they chose to.
Even before expansion...
wasnt the BigInt discussing moving the final week to thanksgving weekend? Why they ignored the last two weeks of the season has always baffled me.
Sorta discussing
It’s something JoePa wanted, which generally means it will happen in about 5-10 years, but there was a lot of resistance to it. The main reasons were tradition and academics.
Regardless of the reasoning, it still increases the real strength of schedule, and the teams never got credit for it. Saying “that’s their own fault” is no different than saying “it’s their own fault for scheduling hard teams.” Both are true, but both increase the difficulty of the schedule.
Idle talk and hollow promises; cheating Judases; doubting Thomases
large reply accidently deleted
short version, you are sorta right, but mostly wrong, because you havent seen the algorithm. :)
medium length version
Scheduling a moderately weak team (a team whose rating is 1-2 pts below yours) will NEVER hurt you. It is only scheduling a team that is ludicrously below your level (2+ pts) that hurts your average.
This is set up this way for two reasons – 1, because I agree with you about additive v average (to a point). 2, because below a certain point, teams have no reasonable chance to lose, and should be punished for refusing to compete. To pick on Bama again, they could turn the ball over 7 times vs GSU and still win. That probably wouldnt be true vs Jax St, for example.
The way my system works, the further below that cutoff point a team is, the worse is hurts you for playing them. Scheduling a team 2.1 rating pts below you hurts negligibly – this kind of thing happens if you have a Vandy or Duke in your conference. Scheduling a team that is 3 pts below you (if you are competing for a national title, the dregs of the Sun Belt/MAC and mediocre FCS teams) hurts worse. Scheduling a team 4+ pts below you (Cincy v SE Mo St) crushes you, and rightly so. Notice that this means Fla Intl could schedule SE MO St with little affect on their rating, and rightly so, they arent that good anyway.
"punished for refusing to compete"
Doesn’t the fact that your rating goes up only minimally while others see theirs rising substantially count as sufficient punishment?
This is one thing I absolutely hate about the RPI and at least one of the BCS computers (Colley). If you look only at wins and losses, not margin, then a win should never hurt your rating. (It can in points-based systems, because they expect you to win by a certain margin and if you win by less it’s a data point that suggests you aren’t as good as they thought.) It need not help much, but if you’re telling me that Michigan is a worse team because they beat Delaware State by 60 than they would be without having played them, something is out of whack.
It’s also something the Bradley-Terry method solves quite nicely, in that your rating adjusts to make your predicted wins against your schedule match your actual record. Playing a team you’re 99% likely to beat adds 1 to your actual win total and 0.99 to your expected win total; your rating doesn’t need to go up much to account for the remaining 0.01. but it does have to go up. This also puts much less emphasis on just how bad your cupcakes are; the difference between playing a team you should beat 99% of the time and one you should beat 99.99% of the time isn’t a whole lot (nor should it be; for all practical purposes they’re both nothing but body-bag games).
In this system, your rating works out to the number of wins you should have against an average (rating 1) team for every loss; a 4-0 team with four wins over perfectly average teams would have a rating of 9 (that is, you’re estimated at a 90% chance of beating an average team; to avoid unpleasant divide-by-zero operations each team is assumed to have tied one game against an average team in addition to their actual results, so you’re effectively 4.5-0.5 against five average teams). Beat someone with a rating of 10 (say, someone 4-0 against a slightly tougher schedule) for your fifth win, and you jump all the way up to 23.65. Add a fifth win over an average team, and that gets you to 11. But if that fifth win is over a cupcake (rating 0.1, someone a Motor City Bowl team would be a 10:1 favorite against) you only move up to 9.22 – and if it’s over a real patsy (0.01, or someone a Motor City Bowl team would be a 99% favorite against) your rating climbs all the way to 9.02. (All of this neglects the effects your win would have on the other teams’ ratings, just to simplify the math.)
Even Krach can cause you to drop with a bad win
If Team A schedules Delaware St and Team B schedules a real opponent, then A will drop, even with a win, relative to B.
There isnt a real option to not schedule anyone, teams are going to play 12.
BTW, played arount with a B-T/Krach the other day, and saw no real differences between it and mine (sure, teams jumbled a bit, but that is true between and 2 methods). The top 26 teams in 2009 were exactly the same, with two teams being more than 3 places different – Wisconsin dropped 5 and Arkansas went up 5.
My issue with it is I think it really fails at the high and low end, B-T is a MUCH better method for a sport like baseball where records dont approach perfection on either end.
Relative, yes
But your absolute rating won’t go down due to a win. (That is, beating a godawful team is better, if only minimally, than having a bye.) It may still drop if your strength of schedule suffers enough from prior opponents losing, but it won’t be because you won.
One thing it does, admittedly, is put a lot of emphasis on record at the top or bottom end. Your SOS has to be about 3x as good with one loss to get the same rating as an unbeaten team. I do think that’s fair, though – a team whose ceiling hasn’t been shown (by losing to someone) should be ahead of one whose has unless there’s a major difference in SOS. It also runs into trouble when you try to directly compare teams with minimal connection (for instance, if you have 1-A, 1-AA, II, III, and NAIA teams all in one group); for this reason I’d be tempted to group all 1-AA teams into one entity for ranking purposes (this gives you better connection but doesn’t account for discrepancies between good 1-AAs and bad 1-AAs, and might overstate how bad a typical 1-AA is by piling up a 2-50 record or something crazy like that).
Part of the reason for my solution
Is that I WANT to be able to group FBS to NAIA all together. I found I had to break the “all wins are positive” axiom to make that possible with reasonable results.
I was able to keep “any win against a team with a skill level reasonable close to yours is at very least not-negative”.
The scale basically runs from 4 to -4 with 0 being the average team (In 2009: Alabama finished slightly above 4, Trinity Bible didnt break -4). No win over a team within 2 pts of you could ever cause a negative result. That gives a lot of scheduling leeway. If you are at 4 pts, that gives you ~97 teams to schedule, and some of those are FCS. (98 teams finished 2 or better) If you arent competing for a national title but are just around the edge of the top 25, with a 3.100, that gives you to 1.1, which is 220 teams. You have to be scheduling the dregs of FCS to hit below that.
So, really, the only teams that have to be concerned about their schedule hurting them are those going for the national title. And I stand by the principle that scheduling Ga State should almost disqualify Bama from consideration in 2010.
There is a piece of advice for your school that comes out of my system...
If you want to compete for a MNC, schedule from the top half of the MAC, not the bottom half. CMU not EMU.
I wonder ...
… there might be a way to do it by fudging the starting point (make the fictional tie that gets rid of the divide-by-zero possibilities against a team equal to your SOS instead of an average team). Unfortunately, I don’t think that converges. An average of the other teams in your conference or 1-A teams (assuming you’re in 1-A), maybe? Still might not converge, although it’s more likely to.
I’m not sure there is a useful way to do an NAIA-to-1A ranking when there’s so little interconnection; it’s hard enough getting a coherent ranking of 1-A when you play 10% of the other teams (assuming you don’t play a 1-AA).
It doesnt converge
Well, it does, but not on the right point. That was my first attempt at B-T, seemed having the tie against a team equal to yourself made the most sense, it does, but has convergence problems.
Well, I think my method does a pretty good job on 1A to NAIA. It still has a bit of overweight effect on 15-0 teams form D3, but not as bad as many systems Ive seen. It and B-T give very similar results, B-T overweights them a bit more.
The answer, of course, is their is no one RIGHT way to do it. If I could redesign the BCS, I would eliminate the coaches poll (for obvious reasons), count the other poll as 1/3 of the total instead of 2/3 and count computers as 2/3 without throwing out high and low. I would chuck billingsley and any non-open systems, and select a set of computer systems with different ideologies:
For example:
Colley, B-T, mine for non-score based
pure power for the other extreme,
A few sagarin, massey type mix of pts and results type systems – but open systems. Maybe 3 of those type.
Then finish it off with 3 that look at internal statistics, yardage and etc.
That would be 10 different systems. Nice round number.
I'd chuck the BCS entirely in favor of a playoff (another debate entirely)
But if you’re going to work within the BCS instead of blowing it up, that’s probably the way to go.
My system originally worked similarly
Originally, all wins had positive value (absolute rating, not ranking). However, it had issues with teams with 1 very, very good win and then an otherwise bad schedule. Boise St, in particular, last year, due to beating a damn good Oregon team. The adjustment I made solved two problems (and added one):
1. It handled that problem above, although Cincy got hurt worse by it than Boise.
2. It handled a general level issue that sometimes comes up with undefeated teams from lower levels — that 15-0 NAIA team shouldnt be ranked too highly, obviously.
The problem it added is that it is possible for a win to be a negative compared to not playing the game at all. However, its not a matter of playing a below average team, the level before this happens is much lower, and depends upon your teams skill level. In some ways it says that you are what you schedule. If you 11 FCS teams, you are FCS level. If you schedule 12 SEC teams, you are SEC level. There is only so low Vandy can drop with their schedule.
a win should never hurt your rating
Im not sure I accept that as an axiom.
If you look only at wins or losses, why should a win hurt you (directly)?
Certainly beating an awful team shouldn’t help much, but if you let it hurt a team’s rating then you are stating that there’s no result that would not make them look worse than you thought prior to the game.
In points-based systems, a win by a sufficiently narrow margin (how narrow depends on the gap between teams) can and should hurt – the system expected you to win by 40 and you only won by 30, so that’s a data point that suggests you aren’t as good as previously thought. But if you only care about a win or loss, a win (even over the worst team in existence) does not give any reason to believe you’re worse than the system thought. The penalty for playing such an awful opponent is not that you look any worse in a vacuum, but that others improved their appearance and you did not.
Im saying if you play Ga State while contending for the national title
There is no result that would not make you look worse than prior to the game.
This, of course, is subject to GSU being as bad as I expect them to be.
Comments For This Post Are Closed
Secondary Sidebar
User Tools
FanPosts
Community blog posts and discussion.
