Statistical categories being rated are listed below. Starred categories are considered absolute measuring sticks and are therefore used in final rankings.
*Scoring Offense
*Scoring Defense
*Total Offense
*Yards/Play
1st Downs
*Percentage of Plays that are 1st Downs
3rd Downs
3rd Down Success
*3rd Down Conversion
*Passing Offense
Completions
Attempts
Yards/Pass
*Efficiency Rating Offense
*Rushing Offense
Attempts
*Yards/Rush
*Total Defense
*Yards/Play
1st Downs
*Percentage of Plays that are 1st Downs
3rd Downs
3rd Down Success
*3rd Down Conversion
*Passing Defense
Completions
Attempts
Yards/Pass
*Efficiency Rating Defense
*Rushing Defense
Attempts
*Yards/Rush
*Penalties
*Penalty Yards
*Net Turnover
*Turnovers Committed
*Turnovers Recovered
In each category, a percentage score is calculated per game. As an example, I will present the calculation of Alabama's Total Defense Score from the 2010 BCS Championship Game.
Texas Offense Stats: Alabama Defense Stats:
Season Avg: 421.2 ypg Season Avg: 245.4 ypg
Season Std Dev: 144.5 ypg Season Std Dev: 73.7 ypg
In the BCS game, Texas gained 276 yards of total offense. Using the data from Texas' overall season statistics, this is equivalent to a 0.157 score (15.7 percentile on the season).
Alabama therefore gave up 276 yards of total defense. Using data from Alabama's overall season statistics, this is equivalent to a 0.339 score (33.9 percentile on the season).
I combine these two scores using a gradient formula based on arctangent graphs. Because even though Alabama gave up more yards than its season average, it played much better than Texas' Offense (comparing the scores). Therefore, after combining the two percentiles using my formula, I come up with a total defense score for Alabama of 0.708 for the game.
This method is used for each statistic, for each game. The "absolute measuring sticks" (starred above) are averaged to give a "game statistic score." The opponent's winning percentage is given a percentile score (percentile out of all 120 FBS schools). If the team lost the game, this winning percentile is divided in half. The winning percentile score is averaged with the game statistic score and weighted by where it occurs in the season (more recent games are weighted higher, to adjust for team improvement or degradation). Each game is averaged together, then averaged with the team's winning percentage to achieve a final overall score. This overall team score is ranked highest to lowest (1.000 being the maximum possible score and 0.000 the minimum).
This is the ranking system for Lightsheaber's College Football Rankings.
Because there are so many averages happening, it is hard to break down what is important to the system, but in experiments, I have discovered some areas the computer values:
1. Stability - week-in-week-out uniformity is important, because the computer uses a geometric mean instead of an arithmetic mean. Therefore, achieving scores of 0.550 every week is preferable to receiving 0.810 then 0.129 then 0.754, etc.
2. Winning Percentage - it ends up being over 50% of the score, all things considered
3. Strength of Schedule - because each game score is based on the winning percentage of the team being played, Alabama's undefeated season in 2009-10 gave it a score almost 5% higher than Boise State, who also went undefeated (BSU ended up #3 in my rankings, behind Texas).
Though these three areas are also what AP voters tend to take into consideration, I believe my system DOES give schools like Boise State or TCU the credit they deserve because rolling over cupcakes, though not as helpful as beating an SEC team on the road, does provide momentum, provided the thrashing occurs week-in and week-out. (#1 above)
