I always try to improve the preseason and early season college football rankings at The Power Rank. The primary rankings on the site still use last season’s games, with this season’s games counted twice. I think they do a good job, but this method reacts slowly to teams that have struggled, such as Texas (31st).
So I developed new model this week. It’s based on the regression model that I used for my preseason predictions, which consider a team’s rating the last 4 years, turnovers and returning starters. Now, the model includes a rating calculated from only games this year.
The visuals shows the top 10 teams in this regression model. While Baylor is mostly likely overrated at 2nd since they have not played anyone, I do like that Alabama has dropped to 3rd and Washington has cracked the top 10.
Let’s look at the predictions this model makes.
How low should Texas be ranked?
Texas checks in at 51st in this regression model. Their moderate success over the past 4 seasons (moderate by Texas standards) and a host of returning starters keep the Longhorns above the average FBS team (125 teams total).
For last night’s game at Iowa State, the regression model predicted a 2.3 win for Iowa State. The rankings that use last year’s games had Texas by 2.8. The regression model has reacted faster to the Longhorn’s struggles, who have lost badly to Mississippi and BYU.
Texas squeaked out a win last night over Iowa State. They needed a hail mary touchdown at the end of the 1st half as well as a no call on a fumble that would have ended Texas’s game winning drive. Further more, Iowa State gained 6.0 yards per play compared to 4.9 for Texas.
Mack Brown is dating Lady Luck.
How good are the predictions of the new model?
I went back and tested how accurately each ranking system predicted game winners. This test considered all games after week 5 from the 2007 to 2012 seasons.
The regression model predicted 69.2% of game winners, while The Power Rank using last year’s games got 68.9% correct. With an error of about 0.8%, both rankings system have the same predictive power.
However, both methods perform better than The Power Rank with only this year’s games. Those rankings predicted 67.5% of game winners, quite a bit less.
Let’s look at the predictions these two models make.
Notre Dame and Arizona State
Notre Dame has disappointed this season. They have already lost twice, and that 7 point win over Purdue looks worse as the Boilermakers continue to lose badly each week.
The rankings with last year’s games predict a 1.3 point loss against Arizona State at a neutral site in Dallas. However, the regression model predicts a 5.5 point loss, the same as the line.
I still don’t know what to think about Notre Dame. Their defense doesn’t tackle well in the secondary. But Oklahoma scored 14 points off of 2 tipped passes against the Fighting Irish last week. Moreover, QB Tommy Rees had a terrible game.
I’d stay away from this game.
Illinois at Nebraska
Illinois has been a pleasant surprise, a rarity in the Big Ten this season. Behind the 9th best offense, the Fighting Illini are 53rd in the regression model, a miracle for a team that finished 115th last season.
They travel to Lincoln to face a Nebraska team that has struggled on defense. The regression model has reacted more quickly to the opposite fortunes of these two teams, picking a 6 point win for Nebraska (the line favors Nebraska by 9).
The rankings with last year’s games have Nebraska by 13.6 points. With the two teams that do not resemble their preseason expectations, it’s safe to ignore this prediction.
This is my upset special for the week. Nebraska’s offense has not lived up to expectations, and QB Taylor Martinez will not play again this week. Illinois gets the win in Lincoln. Next week’s headlines give Mack Brown a week of reprieve and focus on the job security of Bo Pellini.
Kansas State at Oklahoma State
Kansas State lost a host of starters from last season’s stellar team. In addition, the Wildcats had an unsustainable turnover margin in 2012. Hence, my preseason ranking had them at 37th.
The rankings with last season’s games predict a tight game (0.8 points) in favor of Oklahoma State. Again, it’s safe to ignore that given the changes to this Kansas State team.
The regression model predicts a 9 point win for Oklahoma State. This margin is probably to big. Kansas State fumbled the ball 3 times in gifting a win to Texas last week.
The line favors Oklahoma State by 14. This is too much for a team whose offense hasn’t performed at the elite level it did last season.
What do you think?
I’ve copied the rankings from the regression model below. Would you like to see them as the primary rankings?
Let me know in the comments. Thanks for reading.
1. Oregon (4-0), 28.80
2. Baylor (3-0), 26.81
3. Alabama (4-0), 23.22
4. Stanford (4-0), 17.67
5. Georgia (3-1), 15.98
6. Texas A&M (4-1), 15.55
7. LSU (4-1), 15.49
8. Washington (4-0), 14.31
9. Florida State (4-0), 14.14
10. Florida (3-1), 14.07
11. Ohio State (5-0), 13.87
12. Clemson (4-0), 13.46
13. Louisville (4-0), 12.92
14. UCLA (4-0), 12.58
15. Wisconsin (3-2), 11.93
16. South Carolina (3-1), 11.47
17. Miami (FL) (4-0), 10.73
18. TCU (2-2), 10.14
19. Oklahoma (4-0), 10.14
20. Arizona State (3-1), 9.84
21. Texas Tech (4-0), 9.74
22. Arizona (3-1), 9.59
23. Missouri (4-0), 9.01
24. Utah State (3-2), 8.32
25. Mississippi (3-1), 8.17
26. USC (3-2), 7.75
27. Northwestern (4-0), 7.42
28. Oklahoma State (3-1), 7.41
29. Oregon State (4-1), 7.23
30. Northern Illinois (4-0), 5.76
31. Virginia Tech (4-1), 5.24
32. Tennessee (3-2), 4.79
33. Maryland (4-0), 4.72
34. Auburn (3-1), 4.67
35. UCF (3-1), 4.53
36. Penn State (3-1), 4.43
37. Notre Dame (3-2), 4.37
38. Boise State (3-2), 4.28
39. Nebraska (3-1), 4.11
40. Iowa (4-1), 3.92
41. Michigan State (3-1), 3.67
42. Utah (3-2), 3.45
43. Brigham Young (2-2), 3.41
44. Vanderbilt (3-2), 3.05
45. Georgia Tech (3-1), 3.01
46. Michigan (4-0), 2.60
47. Fresno State (4-0), 2.24
48. West Virginia (3-2), 1.88
49. Arkansas (3-2), 1.76
50. Syracuse (2-2), 1.50
51. Texas (3-2), 1.40
52. Kansas State (2-2), 1.21
53. Illinois (3-1), 1.02
54. East Carolina (3-1), 0.94
55. Mississippi State (2-2), 0.80
56. North Carolina State (3-1), 0.73
57. Iowa State (1-3), 0.72
58. Washington State (3-2), 0.50
59. Rutgers (3-1), -0.15
60. Toledo (2-3), -0.17
61. Ball State (4-1), -0.23
62. Cincinnati (3-1), -0.33
63. San Jose State (1-3), -0.38
64. Pittsburgh (3-1), -0.44
65. Houston (4-0), -0.58
66. California (1-3), -0.62
67. North Carolina (1-3), -0.66
68. Kentucky (1-3), -0.75
69. Minnesota (4-1), -1.11
70. Bowling Green (4-1), -1.49
71. Marshall (2-2), -1.65
72. North Texas (2-2), -1.87
73. Indiana (2-2), -1.90
74. Boston College (2-2), -1.92
75. Western Kentucky (4-2), -2.02
76. Buffalo (2-2), -2.33
77. Ohio (3-1), -2.50
78. Rice (2-2), -2.77
79. Navy (2-1), -3.03
80. San Diego State (1-3), -3.31
81. Connecticut (0-4), -3.37
82. Colorado State (2-3), -4.02
83. Virginia (2-2), -4.74
84. Wyoming (3-2), -5.16
85. SMU (1-3), -5.28
86. Arkansas State (2-3), -5.78
87. Louisiana Lafayette (2-2), -6.11
88. Tulsa (1-3), -6.14
89. Nevada (3-2), -6.31
90. Duke (3-2), -6.34
91. Louisiana Monroe (2-4), -6.46
92. Colorado (2-1), -6.49
93. Army (2-3), -7.09
94. Wake Forest (2-3), -8.01
95. Temple (0-4), -8.05
96. Kent State (2-3), -8.11
97. Louisiana Tech (1-4), -8.17
98. Florida Atlantic (1-4), -8.54
99. Middle Tennessee State (3-2), -8.61
100. Purdue (1-4), -9.07
101. Kansas (2-1), -9.54
102. South Florida (0-4), -9.68
103. Tulane (3-2), -10.27
104. Western Michigan (0-5), -10.40
105. Troy (2-3), -11.14
106. UAB (1-3), -11.34
107. Hawaii (0-4), -11.75
108. Air Force (1-4), -11.80
109. UNLV (3-2), -12.39
110. Southern Miss (0-4), -13.19
111. Memphis (1-2), -13.20
112. Akron (1-4), -13.27
113. UTEP (1-3), -14.38
114. Idaho (1-4), -14.94
115. Miami (OH) (0-4), -15.71
116. Florida International (0-4), -15.86
117. Central Michigan (1-4), -16.30
118. Eastern Michigan (1-3), -17.36
119. New Mexico (1-3), -18.10
120. New Mexico State (0-5), -19.98