The Tempo-Free Gridiron is getting a second author to help provide another take on college football. Eddie is a diehard college football junkie with an encyclopedic knowledge of the game. He's the complete opposite of myself in that regard, as I have no allegiance to any particular team or conference. I do this out of pure curiosity for the statistics and predictions with no regard to the names on the jersey (or boxscore). As you'll see, Eddie starts with the exact same data set that I use, but his system attempts to account for more factors out of the gate. This should be an interesting contrast in styles: blank slate versus encyclopedia, and simple versus comprehensive.
In his introductory post Eddie will give you the under-the-hood look at his methodology along with his current rankings and predictions for upcoming games. Over the next week we hope to do per-conference breakdowns and a look at the BCS picture, along with some compare-and-contrast posts that highlight differences in our systems and their results.
Welcome, Eddie.
Showing posts with label introduction. Show all posts
Showing posts with label introduction. Show all posts
Tuesday, November 17, 2009
Thursday, October 22, 2009
Under the Hood
There are three key concepts behind the statistics on this site:
In football the concept of a possession is slightly less clear. Is it a single snap from scrimage? An entire drive? Where does a kickoff fall in this definition? If we were to use a single play from scrimage as our definition of a possession, we find that our symmetric relationship between possessions doesn't hold. One team could easily have 100 possessions in a game, whereas their opponent may only muster 50. In this scenario we would have to account for the amount of offense and defense in each game in addition to the efficiency of each squad. This seems needlessly complicated.
On the flip side, if we were to use a drive as the definition of a possession then the game becomes significantly more symmetric and familiar. One team takes the ball, attempts to score, and then returns the ball to their opponent. Unfortunately this further reduces our already-small sample size. In college basketball there are 340 teams playing a 30-game slate with roughly 70 possessions per game. In football there are 120 teams playing an 10-game schedule with roughly 20 possessions per game. (All numbers approximate) Furthermore, how do we account for the scenario in which a pass is intercepted and returned for a touchdown? A kickoff returned for a touchdown (or the return is dropped and the kicking team returns it for a touchdown)?
I finally settled on the philosophy that every play is a possession. Every snap, kickoff, punt ... it's all a single possession because each team has the potential to score on every play. When one team is on offense and has the ball, they are in essense playing both offense and defense on each snap; the same goes for the "defense". They are also trying to block the other team from scoring and -- if possible -- score themselves.
I may revisit this decision later, but for now it seems to serve me well and allows me to "fit" everything into the other two portions of the model.
This brings us to part two: Pythagorean expectation, a formula developed by Bill James that uses the average number of points scored by and against a team to determine their expected winning percentage. If S is the number of points scored by a team and A is the number of points allowed, then the expected winning percentage of that team is
WinPct = 1 / (1 + (A/S)^E)
where E is an exponent that varies from sport-to-sport. In baseball the most accurate value appears to be 1.81. In college basketball the estimation of the exponent has varied from 14.0 (by Dean Oliver) to 16.5 (John Hollinger) to 11.5 (Pomeroy). After analyzing data from several years of college football, I settled on a value of 3.0 for E, although many values in the range [2.5,3.2] produced reasonable results.
Now that we have a definition of a possession and a value of E for our Pythagorean winning percentage, that brings us to the log5 formula. This formula states that if two teams were to play each other, and one team has a winning percentage of A and the other a winning percentage of B, the odds that the first team would win are
There are two other small factors my system takes into account: home field advantage and the primacy of more recent games. In college football the home team wins approximately 62% of all games. This effect can be seen to varying degrees in other sports as well. It is interesting to note that the magnitude of home field advantage is larger during the early stages of the season but diminishes as the season progresses. From 2006-2009, the home team won 68% of all games played in the first half of September, compared with 57% of all games played in the first half of November. Even taking into account the tendency of Powerhouse U. to schedule early-season games against Sister Mary's School of the Blind, the effect deteriorates as the year progresses. This may be caused by teams becoming more comfortable playing on the road as they mature during the year, but for now it's simply an observation.
There is also the matter of weighting more recent games more heavily. Unlike many other systems, I do not completely discard games from the previous years. I use an exponential decaying factor that causes games from the start of the year to count as roughly 3/5th of a full game, and games from a full year ago as roughly 1/6th of a game. This allows early-season predictions to be "in the ballpark" for most teams. Let's be honest; USC is going to remain USC from year-to-year, and Duke is going to ... well, let's just say Durham becomes a much cheerier place starting in mid-November.
Now that you've had the whirlwind tour of the statistics behind the site, let's get on with the good stuff: rankings.
- offensive and defensive efficiencies;
- pythagorean calculation for expected winning percentage; and
- log5 formula for head-to-head matchups.
In football the concept of a possession is slightly less clear. Is it a single snap from scrimage? An entire drive? Where does a kickoff fall in this definition? If we were to use a single play from scrimage as our definition of a possession, we find that our symmetric relationship between possessions doesn't hold. One team could easily have 100 possessions in a game, whereas their opponent may only muster 50. In this scenario we would have to account for the amount of offense and defense in each game in addition to the efficiency of each squad. This seems needlessly complicated.
On the flip side, if we were to use a drive as the definition of a possession then the game becomes significantly more symmetric and familiar. One team takes the ball, attempts to score, and then returns the ball to their opponent. Unfortunately this further reduces our already-small sample size. In college basketball there are 340 teams playing a 30-game slate with roughly 70 possessions per game. In football there are 120 teams playing an 10-game schedule with roughly 20 possessions per game. (All numbers approximate) Furthermore, how do we account for the scenario in which a pass is intercepted and returned for a touchdown? A kickoff returned for a touchdown (or the return is dropped and the kicking team returns it for a touchdown)?
I finally settled on the philosophy that every play is a possession. Every snap, kickoff, punt ... it's all a single possession because each team has the potential to score on every play. When one team is on offense and has the ball, they are in essense playing both offense and defense on each snap; the same goes for the "defense". They are also trying to block the other team from scoring and -- if possible -- score themselves.
I may revisit this decision later, but for now it seems to serve me well and allows me to "fit" everything into the other two portions of the model.
This brings us to part two: Pythagorean expectation, a formula developed by Bill James that uses the average number of points scored by and against a team to determine their expected winning percentage. If S is the number of points scored by a team and A is the number of points allowed, then the expected winning percentage of that team is
WinPct = 1 / (1 + (A/S)^E)
where E is an exponent that varies from sport-to-sport. In baseball the most accurate value appears to be 1.81. In college basketball the estimation of the exponent has varied from 14.0 (by Dean Oliver) to 16.5 (John Hollinger) to 11.5 (Pomeroy). After analyzing data from several years of college football, I settled on a value of 3.0 for E, although many values in the range [2.5,3.2] produced reasonable results.
Now that we have a definition of a possession and a value of E for our Pythagorean winning percentage, that brings us to the log5 formula. This formula states that if two teams were to play each other, and one team has a winning percentage of A and the other a winning percentage of B, the odds that the first team would win are
.
A - A * B
WPct = -----------------
A + B - 2 * A * B
There are two other small factors my system takes into account: home field advantage and the primacy of more recent games. In college football the home team wins approximately 62% of all games. This effect can be seen to varying degrees in other sports as well. It is interesting to note that the magnitude of home field advantage is larger during the early stages of the season but diminishes as the season progresses. From 2006-2009, the home team won 68% of all games played in the first half of September, compared with 57% of all games played in the first half of November. Even taking into account the tendency of Powerhouse U. to schedule early-season games against Sister Mary's School of the Blind, the effect deteriorates as the year progresses. This may be caused by teams becoming more comfortable playing on the road as they mature during the year, but for now it's simply an observation.
There is also the matter of weighting more recent games more heavily. Unlike many other systems, I do not completely discard games from the previous years. I use an exponential decaying factor that causes games from the start of the year to count as roughly 3/5th of a full game, and games from a full year ago as roughly 1/6th of a game. This allows early-season predictions to be "in the ballpark" for most teams. Let's be honest; USC is going to remain USC from year-to-year, and Duke is going to ... well, let's just say Durham becomes a much cheerier place starting in mid-November.
Now that you've had the whirlwind tour of the statistics behind the site, let's get on with the good stuff: rankings.
What's all this, then?
Welcome to my humble attempt at blatant imitation of real sports statisticians.
I'm a long-time follower of college basketball, and over about the last five to ten years there have been some pretty interesting strides in improving our understanding of how to analyze college ball. One of the more visible people in this arena is Ken Pomeroy, who I've followed since around 2004. His blog and his rating system really sparked my interest in the concept of tempo-free statistics. The basic idea behind tempo-free statistics is that what we currently measure in a sport like basketball -- points per game, number of turnovers, etc -- is in many ways broken. They depend in large part on how quickly a team plays the game; the faster the pace, the more possessions a team and their opponents have, the more opportunities for both teams to score and give up points, rack up assists, and cough up turnovers.
They don't, however, answer the fundamental question "How good is team 'X'?"
For that we turn to tempo-free statistics. Remove the disparity in possessions, normalize all statistics to a common metric such as "points per 100 possessions", and adjust for the quality of the opponent. This allows us to see the fundamental efficiency of a given team. Ken Pomeory has an excellent write-up of how this applies to college basketball that I will not even attempt to duplicate, but simply encourage you to read.
The question this blog examines is "do these concepts apply to football as well as they do to basketball?" To answer that question I plan to lay out how I create my rankings, where I obtain my data, post predictions on upcoming games, and analyze where the system was right and where it went wrong. I also hope to examine some areas in which my system produces vastly different results than either conventional wisdom and/or other computerized rankings such as those used in the BCS.
This is not my first attempt at applying this approach to college football. Last year I participated in the ESPN Winning Formula Challenge, a 12-week-long contest with significant sums of prize money for those who could write a computer program to predict college football results. Unfortunately there was an issue with my code during the second 4-week competition -- remember kids: bounds-check your array accesses because some team somewhere will hang 80 points on their opponents -- but had my code worked throughout the season I would have finished 6th out of approximately 120 competitors. That's a 73% accuracy rate using nothing but the final score and number of possessions in each game.
In my next post I'll go into some of the nuts and bolts of my approach, but for now I encourage you to read Pomeroy's write-up on how to use and understand tempo-free statistics.
I'm a long-time follower of college basketball, and over about the last five to ten years there have been some pretty interesting strides in improving our understanding of how to analyze college ball. One of the more visible people in this arena is Ken Pomeroy, who I've followed since around 2004. His blog and his rating system really sparked my interest in the concept of tempo-free statistics. The basic idea behind tempo-free statistics is that what we currently measure in a sport like basketball -- points per game, number of turnovers, etc -- is in many ways broken. They depend in large part on how quickly a team plays the game; the faster the pace, the more possessions a team and their opponents have, the more opportunities for both teams to score and give up points, rack up assists, and cough up turnovers.
They don't, however, answer the fundamental question "How good is team 'X'?"
For that we turn to tempo-free statistics. Remove the disparity in possessions, normalize all statistics to a common metric such as "points per 100 possessions", and adjust for the quality of the opponent. This allows us to see the fundamental efficiency of a given team. Ken Pomeory has an excellent write-up of how this applies to college basketball that I will not even attempt to duplicate, but simply encourage you to read.
The question this blog examines is "do these concepts apply to football as well as they do to basketball?" To answer that question I plan to lay out how I create my rankings, where I obtain my data, post predictions on upcoming games, and analyze where the system was right and where it went wrong. I also hope to examine some areas in which my system produces vastly different results than either conventional wisdom and/or other computerized rankings such as those used in the BCS.
This is not my first attempt at applying this approach to college football. Last year I participated in the ESPN Winning Formula Challenge, a 12-week-long contest with significant sums of prize money for those who could write a computer program to predict college football results. Unfortunately there was an issue with my code during the second 4-week competition -- remember kids: bounds-check your array accesses because some team somewhere will hang 80 points on their opponents -- but had my code worked throughout the season I would have finished 6th out of approximately 120 competitors. That's a 73% accuracy rate using nothing but the final score and number of possessions in each game.
In my next post I'll go into some of the nuts and bolts of my approach, but for now I encourage you to read Pomeroy's write-up on how to use and understand tempo-free statistics.
Subscribe to:
Posts (Atom)