Okay, so here’s my new plan. Poke all the holes in it you want…
What determines the "best" football team? By nature, this is a subjective determination. However, in the interest of making the same subjective decision over and over again, many people turn to a computerized ranking system based on certain criteria. Obviously, their own biases go into the algorithm that orders the teams, but at the very least, this algorithm is objective given the input data. That is to say that given the same set of data, it will always make the same decision, without regard to conference or team favoritism or what games it could or couldn’t watch that day. That said, how does one determine an algorithm?
I decided to go back to the root of what makes a team the best team in football. It may seem obvious, but it has to the be the best at playing the game of football. To achieve that, a team must be, overall, the best team on all sides of the ball. It must win games; and it must beat good teams and do it convincingly. The best football team will have a strong offense, a strong defense, and to a lesser extent, (with apologies to Frank Beamer), good special teams.
A strong offense will be indicated by points scored, yards gained, and not committing turnovers. A strong defense will be indicated by not allowing points, not allowing yards, and forcing turnovers. Strong special teams will be indicated by scoring, and a negative turnover margin. These are easily quantifiable things, making this system easy to keep track up. It also directly relates game performance to rank, without a lot of intermediate ranking.
This system rewards teams who win, who out gain and outscore opponents, and hold onto the ball. The inherent problem is that a good team may elect to play poor teams to boost its points and yards margins. This must be accounted for with a strength of schedule component. I haven’t decided exactly how to do this yet, or how much of the equation it should be. I’m thinking the way they did it a few years back in the BCS was pretty good: your opponents’ records and your opponents’ opponents’ records. But I think a record alone does not indicate enough about an opponent. This is something that requires more thought.
Here is the preliminary method for determining ranking:
.5*(Wins/Games Played) + .35*(.37*(Offensive Yards Gained/Defensive Yards Surrendered) + .38*(Points Scored/Points Surrendered) + .25*(Turnovers Forced/Turnovers Committed))/Games Played + .15*(.667*Opponents’ Strength + .333*Opponents’ Opponents’ Strength)
The beauty of this method is that it is totally dependent on season performance and is not influenced by any pre-season mumbo jumbo. Every team starts out the season being equal. After week one, roughly half the teams will be equal because of the 50% weighting for winning a game, but because of the game stats component, the team that played the best will be team #1. Over the course of an entire season, I think this will all work out pretty nicely. It can break down to an individual week or which team was the best for that given week, or which team is the best up to a given point, depending what data is put in. I’m making it my goal for next weekend to code this up and see what happens.
What determines the "best" football team? By nature, this is a subjective determination. However, in the interest of making the same subjective decision over and over again, many people turn to a computerized ranking system based on certain criteria. Obviously, their own biases go into the algorithm that orders the teams, but at the very least, this algorithm is objective given the input data. That is to say that given the same set of data, it will always make the same decision, without regard to conference or team favoritism or what games it could or couldn’t watch that day. That said, how does one determine an algorithm?
I decided to go back to the root of what makes a team the best team in football. It may seem obvious, but it has to the be the best at playing the game of football. To achieve that, a team must be, overall, the best team on all sides of the ball. It must win games; and it must beat good teams and do it convincingly. The best football team will have a strong offense, a strong defense, and to a lesser extent, (with apologies to Frank Beamer), good special teams.
A strong offense will be indicated by points scored, yards gained, and not committing turnovers. A strong defense will be indicated by not allowing points, not allowing yards, and forcing turnovers. Strong special teams will be indicated by scoring, and a negative turnover margin. These are easily quantifiable things, making this system easy to keep track up. It also directly relates game performance to rank, without a lot of intermediate ranking.
This system rewards teams who win, who out gain and outscore opponents, and hold onto the ball. The inherent problem is that a good team may elect to play poor teams to boost its points and yards margins. This must be accounted for with a strength of schedule component. I haven’t decided exactly how to do this yet, or how much of the equation it should be. I’m thinking the way they did it a few years back in the BCS was pretty good: your opponents’ records and your opponents’ opponents’ records. But I think a record alone does not indicate enough about an opponent. This is something that requires more thought.
Here is the preliminary method for determining ranking:
.5*(Wins/Games Played) + .35*(.37*(Offensive Yards Gained/Defensive Yards Surrendered) + .38*(Points Scored/Points Surrendered) + .25*(Turnovers Forced/Turnovers Committed))/Games Played + .15*(.667*Opponents’ Strength + .333*Opponents’ Opponents’ Strength)
The beauty of this method is that it is totally dependent on season performance and is not influenced by any pre-season mumbo jumbo. Every team starts out the season being equal. After week one, roughly half the teams will be equal because of the 50% weighting for winning a game, but because of the game stats component, the team that played the best will be team #1. Over the course of an entire season, I think this will all work out pretty nicely. It can break down to an individual week or which team was the best for that given week, or which team is the best up to a given point, depending what data is put in. I’m making it my goal for next weekend to code this up and see what happens.
