Sunday, May 20, 2012

Game Theory for Football (Grid Iron)

Came across this research paper by Kenneth Kovash, Professionals Do Not Play Minimax: Evidence from Major League Baseball and the National Football League. It was done in Sept. 2009, where Kovash found that professional team strategies deviated clearly from minimax play, and were exploitable enough to have left money/points on the table.
With the now publicized "Moneyball" story, perhaps professional baseball management has begun to apply hardcore Game Theoretical analysis. As I have more interest in football, let's focus on its portion of the research effort.

Kovash had data mined over 125,000 plays of NFL (National Football League) games for the years 2001 - 2005. By studying expected payoffs across different pure strategies, it was found that average offensive plays do systematically better by passing the ball than running. Empirically, NFL teams have been more likely to run if the previous play was a pass, especially if the previous play was unsuccessful; and vice versa. This negative serial correlation between passing, running is exploitable by an observant opposing team.

(Right-click, and pick "View Image" to see entire chart)

Below are the NFL statistics between runs and passes of offensive plays. The value "Success Metric" is Kovash's estimate of a given play's contribution to the offensive team's score relative to the average play from this down, distance, and yards to goal.
*note that the near 0 P-Values suggest that there are indeed significant differences between results off Runs and Passes.

(Right-click, and pick "View Image" to see entire chart)

According to Kovash, the average NFL team sacrifices 1 point per game on offense as a consequence of these mistakes. It doesn't sound like much, until we estimate the mistake in dollar value on the NFL franchise, about $5Million a year.

0 Reflections: