Removing Bias from College Football Rankings by Using Data
For years, as a college football fan, I've often questioned the decision-making behind college football playoff rankings. While some selections are straightforward, many cases spark debates about whether the right teams were chosen.
The biggest issue influencing these rankings is brand bias. Decision-makers often seem more concerned with selecting teams that attract the most viewers rather than those most deserving of a playoff spot. We saw this last year with teams like Florida State and Alabama, and it’s likely we’ll see it again this year—especially with Alabama.
As an Alabama fan, I would always argue that we are one of the 12 best teams, but our resume might give you an opportunity to really say we don't deserve to be.
This inspired me to explore what a truly unbiased ranking system might look like. With the advent of the NIL system, the playing field should theoretically be more even. Teams like Alabama and Georgia no longer dominate college football as they once did. But is this because they've declined, or because other teams have improved?
My goal is to create a ranking system that is truly based on numbers. The numbers don't care how many national championships you've won, or how much money is in your NIL fund. I am looking at data that exists strictly within this season, with all Power 5 schools on an even playing field. Finding which teams performed the best, and which teams did not.
For those of you here to see the rankings, here they are below. If you want to keep reading, I will explain where the numbers come from.
Normalized Strength
The normalized strength is a value that represents a team's overall performance on a scale of 0 to 100. It is derived from a combination of the team's: offensive performance (points per play, yards per play), defensive performance (opponent points and yards per play allowed), overall win and loss quality.
If every team is on an even playing field, how do we determine what the best teams are outside of their record?
To develop this strength metric, I analyzed game-by-game statistics to identify instances where teams overperformed or underperformed on offense and defense. In order to proportionally value each team and each game, I focused on four key metrics - Points Per Play, Yards Per Play, Turnovers Per Play, and Penalty Yards Per Play.
I used each of these metrics to create a scale that portrays a team's strength, which was then normalized on a scale of 0 to 100. Here are the top 25 strongest teams:
Strength of Schedule
The strength of schedule (SOS) represents the average difficulty of the opponents a team has played by averaging the normalizing strength of those opponents adjusted by tier.
To account for disparities in conference strength and scheduling, I incorporated strength of schedule into the rankings. You can see teams like UNLV are ranked highly in strength, but utilizing a strength of schedule will allow for me to balance their performance out against teams who play in more competitive conferences.
Win Value
The win value is a score calculated on a scale of 0 to 100 for each game won. It reflects the quality and difficulty of the win based on several factors.
How do we truly value a win? Every win has some value, but an SEC team beating a Group of 5 school does not merit much impact, whereas upsetting a top 5 team should be a highly valued win.
Here's my logic:
1.0 for Power 5 conferences or Independents.
0.9 for Group of 5 conferences.
领英推荐
0.8 for all other opponents.
(Opponent Normalized Strength x Opponent Tier) x 10 (This gives a maximum of 10 points for beating a very strong team)
(Opponent Win Percentage x Opponent Tier) + (Opponent Strength of Schedule x Opponent Tier) x 10 (This gives a maximum of 20 points for beating a team with a great record and strength of schedule)
Compare the team’s game performance to its season averages in points per play, yards per play, turnovers per play, penalty yards per play. Add 2.5 points for each metric where the team outperformed its average. Add 2.5 points for each metric where the opponent performed worse than its averages against the team (defense).
Loss Value:
The loss value is a score calculated on a scale of 0 to -100 for each game lost. It reflects the quality of the loss, penalizing poor performances more heavily.
Following the same principles as win value, how should we measure the value of a loss? Every loss has a value, but in some cases, that value is not necessarily very high. My mind goes to Ohio State's loss at Oregon - a game that was lost by 1 point on the road to the only undefeated team in the country. That loss should not reflect as much negative as being upset at home.
If the opponent's adjusted strength is less than the team’s strength then subtract 10 times the difference between the team strength and opponent adjusted strength. This will deduct more value for losing to worse teams.
20 - (Opponent Win Percentage x Opponent Tier) + (Opponent Strength of Schedule x Opponent Tier) x 2 (Because this gives a maximum of 20 points for a win, I am subtracting this from 20 in order to give less deduction for losing to a good team. Losing to team with a win percentage of .9 and a strength of .9 would give a total loss of 4 points, whereas losing to a team with a .5 and .7 respectively would give a total loss of 16 points.)
Compare the team’s game performance to its season averages in points per play, yards per play, turnovers per play, penalty yards per play. Subtract 2.5 points for each metric where the team performed worse than its average. Subtract 2.5 points for each metric where the opponent performed better than their averages.
Final Output
As you can see below, teams with less losses are typically ranked higher. The differences in value between wins and losses are what is accounted for when summing the the win value and loss value calculations.
For those that want to argue if a 3 loss team is better than a 2 loss team because of the conference they play in, there is merit to that argument, but when you look at the data that assesses overall team performance, then that is not necessarily the case.
This analysis shows the bias in the current playoff rankings—but it appears to be directed primarily toward Alabama. Teams like Arizona State, Iowa State, and BYU are unfairly pushed further down the rankings than their performance data indicates, often in favor of SEC schools that underperformed by comparison.
I get the argument that SEC schools play tougher competition, but when it comes to valuing wins and losses, I'd make the statement that a lot of schools are left out of conversations who deserve to be there.
I’d love to hear your feedback, and feel free to ask any questions!