Tuesday, September 4, 2012

Week 2: Top 25 -- RBA


Rank +/- Team WinPct SoS Adjusted
Off. Def. Pace
1 -- Alabama 1.000 0.552 11 26.9 2 5.4 1 156.7 121
2 -- LSU 0.992 0.549 16 24.1 5 8.1 3 160.0 113
3 -- Oregon 0.984 0.530 50 26.0 4 12.5 28 176.3 3
4 -- Oklahoma St. 0.984 0.551 13 26.9 3 13.9 44 173.3 7
5 +3 Stanford 0.959 0.542 30 23.4 11 12.6 32 166.9 58
6 +1 Oklahoma 0.959 0.543 25 19.7 27 9.5 4 171.7 14
7 -1 Wisconsin 0.959 0.535 38 28.7 1 11.9 23 161.5 109
8 +14 Houston 0.958 0.474 86 22.3 14 14.0 45 175.5 4
9 -4 Boise St. 0.934 0.462 96 23.9 7 10.5 10 168.8 36
10 -1 Arkansas 0.933 0.567 2 23.5 10 11.2 16 165.5 81
11 -- Florida St. 0.932 0.549 15 18.9 32 9.6 6 165.7 77
12 -2 South Carolina 0.925 0.543 27 20.8 21 9.6 5 157.7 120
13 -- USC 0.925 0.533 44 24.0 6 9.8 7 166.8 61
14 -- Georgia 0.900 0.549 17 21.4 18 13.3 41 159.8 114
15 +3 Michigan St. 0.893 0.545 23 20.3 23 10.3 9 167.2 53
16 -- Florida 0.883 0.559 7 16.8 55 10.1 8 161.4 110
17 -2 TCU 0.876 0.464 95 23.5 9 13.3 39 167.8 46
18 NA Nebraska 0.873 0.533 43 21.1 20 12.2 27 166.8 63
19 -- Texas A&M 0.860 0.556 10 22.2 15 12.1 25 172.7 11
20 -- Notre Dame 0.856 0.565 3 21.6 17 10.7 12 164.8 89
21 -4 Michigan 0.851 0.544 24 22.4 13 12.0 24 167.5 50
22 -1 Texas 0.831 0.522 61 19.3 29 11.6 20 168.1 43
23 -- Ohio St. 0.831 0.532 45 20.1 25 11.3 18 159.2 116
24 +1 Missouri 0.824 0.528 56 17.6 40 11.5 19 174.2 5
25 -1 Virginia Tech 0.821 0.527 58 17.5 42 11.1 13 158.2 119
Rankings through games of 2012-09-02

New entries: Nebraska.

Dropped out: Southern Miss.

LOL WUT?  I'm sure you're asking yourself why Houston and Stanford went up in the polls after weak performances against Texas State and San Jose State, respectively.  There's a perfectly rational reason, which is why the BCS computers wait until week 8 before releasing their statistics.  In linear regression, there's a concept called "leverage."  Basically, this means that some outliers have sufficient weight that they throw off the entire curve fit.  When a team has a terrible start to a season, the new point has significant weight.  However, the old data showing them as a successful team remains in the predictor.  Therefore, this creates an odd effect where the team appears to do better against strong competition than weak competition.



The plot above illustrates this effect.  The horizontal axis represents the strength of Houston's opponents.  The vertical axis represents its offensive efficiency in PPH.  The size of the bubble indicates the relative weight of the game.  Since Texas State sits at the far bottom left of the plot and has significant weight relative to other games, it exhibits high leverage.  This creates a plot with a positive slope, rather than a negative slope, which confuses the algorithm.  This will sort itself out, so don't sweat it.  In the meantime, I'm looking into other approaches.

Unsurprisingly, Alabama is the top team in this week's poll.  Drubbing Michigan only proves the point.

Follow us on Twitter @TFGridiron