|
|
|
|
The Situation:You are watching an NFL Football Game. You know something about the game and what outcomes are probable, but how do you express this knowledge? For example, how likely is it that a team will come back from a 10-point deficit at half-time if they will receive the second half kickoff? What if you could watch hundreds -- or even thousands -- of games starting from that situation? If you could, you'd get an idea of how frequently such a comeback happens. Suppose you watched 1000 games where the team receiving the second-half kickoff trails by 100 points, and in 183 of those 1000 games that team came back to win. Then you could reasonably say that a team in that circumstance has a 18.3% chance of winning. Surrogate Reality:You can't watch 1000 games, of course, because only one of them like this is
being played. But you can see what would have happened if 1000 similar
situations were played out. It's not easy, but it is possible. Build a
mathematical simulator, based on data collected over an entire NFL season. Model
the relationships between down and distance, play calls, etc. Factor in game
situations, things like time remaining and field position to name a few. Then
unify all this mathematically, using the same stochastic behavior seen in real
game situations. Interactive Football Simulation:This applet simulates the conclusion of a football game given the current situation. The internal algorithms are based on all play-by-play data from the 2004 NFL season (over 40,000 plays). Given the current information, the algorithm generates an appropriate strategy (i.e., run, pass, punt, etc.) and simulates the result of the play (net yards, possible penalties, elapsed time, etc.). After updating the current situation, the process is repeated until the game ends.
Try it out here. Or if you'd like to know more, send me a note: David H. Annis, Ph.D. (Dave@SportsQuant.com). Back to Top |
Send mail to
webmaster@sportsquant.com with
questions or comments about this web site.
|
