Some of you might be familiar with NYT's 4th Down Bot. It evaluates in real time whether a coach is making the right decision to punt or not. The Bot is based on a model developed at Advanced Football Statistics. The model is based on thousands of plays from the NFL, evaluating how many points a particular scenario is worth to a team.
Long story short: You can go here, enter a particular scenario, and see the statistically sound recommendation. The two data points on the right are "EP" (expected points) and "WP" (win percentage). Those are explained more below.
The calculator has two ways of evaluating a punt/kick/go-for-it scenario.
1. For most of the game, a coach's goal should be to maximize the point differential. So the during the first three quarters, the bot uses the expected point outcome of any scenario. You should do the thing that has the best percentage chance of favorably balancing your points vs. your opponents.
2. With 10 minutes left in the game, the bot switches to "winning percentage outcome." This is based on the likelihood that a team that makes decision X (punt, go for it, kick), will win the game. It switches to this model for two reasons: the later in the game, the more directly a particular decision will impact win outcome, and the easier it is to calculate. It's almost impossible to determine how a first quarter punt affected overall win percentage outcome, so you should instead focus on maximizing the point spread.
NOTE: The models are based on thousands of outcomes from the NFL, not from college. It's possible there could be some statistical differences, but I doubt they would fundamentally change the outcomes.
Long story short: You can go here, enter a particular scenario, and see the statistically sound recommendation. The two data points on the right are "EP" (expected points) and "WP" (win percentage). Those are explained more below.
The calculator has two ways of evaluating a punt/kick/go-for-it scenario.
1. For most of the game, a coach's goal should be to maximize the point differential. So the during the first three quarters, the bot uses the expected point outcome of any scenario. You should do the thing that has the best percentage chance of favorably balancing your points vs. your opponents.
2. With 10 minutes left in the game, the bot switches to "winning percentage outcome." This is based on the likelihood that a team that makes decision X (punt, go for it, kick), will win the game. It switches to this model for two reasons: the later in the game, the more directly a particular decision will impact win outcome, and the easier it is to calculate. It's almost impossible to determine how a first quarter punt affected overall win percentage outcome, so you should instead focus on maximizing the point spread.
NOTE: The models are based on thousands of outcomes from the NFL, not from college. It's possible there could be some statistical differences, but I doubt they would fundamentally change the outcomes.
