A model-based approach to football strategy.
|January 16, 2007|
[Home] [Strategy Review Archive]
Two coaching decisions from the Divisional Round have received so much attention that we will analyze them here, even though they are similar to decisions we have examined in the past. In fact, one decision is included in part because it is eerily similar to a decision faced by the same coach in Super Bowl XXXIX.
With 1:56 left in the 4th quarter of the game between Philadelphia and New Orleans, the Eagles trailed by 3 points, and faced 4th-and-15 at their own 39-yard line. Philadelphia coach Andy Reid had to decide whether to try to pick up the first down, or punt.
Suppose the Eagles try to pick up the first down. If they succeed, they still have two timeouts and about 1:47 in which to get into position for a field goal to send the game to overtime, or score a touchdown to win outright. Even if they fail, they still have a small chance to win if they can force a three-and-out and a pooch punt. Following the pooch punt, the Eagles would get the ball back deep in their own end with about 0:44 left.
If instead of trying to pick up the first down the Eagles kick to the opponents, and force a three-and-out, they will get the ball back at around their own 32-yard line with about 0:44 left.
This decision is remarkably similar to the decision Andy Reid faced at the end of Super Bowl XXXIX. In fact, if in the previous two paragraphs one replaces "pick up the first down" with "recover an onside kick," one obtains almost an exact description of the earlier decision. Our most refined analysis of Super Bowl XXXIX, using our model for the two-minute drill, appeared in a guest article at Football Outsiders, and we won't repeat it here. The main difference is that although the probability of converting on 4th-and-15 is small (around 22%), it's still much larger than the probability of recovering an onside kick. Consequently, although we concluded that in Super Bowl XXXIX kicking deep would have been just as good as kicking onside, in this case punting is substantially worse than trying to pick up the first down. In fact, we estimate that Philadelphia's win probability is about 0.11 if they go for the first down, compared to about 0.05 if they punt. By choosing to punt with 1:56 left against New Orleans, Andy Reid halved his team's likelihood of winning.
With 5:21 remaining in the still-scoreless 1st quarter of the game between New England and San Diego, coach Marty Schottenheimer of the Chargers elected to go for the first down on 4th-and-11 at the New England 30-yard line. The decision to pass up a makeable field goal, with so many yards to go for a first down, is so extraordinary that it deserves formal scrutiny.
We estimate the effect this decision had on San Diego's win probability using the
Dynamic Programming Model.
We assume that, conditional on picking up the first down, San Diego's expected gain on the play is 15 yards. Then according to the Model, if the Chargers go for it, their win probability is either 0.597 or 0.458, according to whether they succeed or fail. The likelihood of picking up the first down is about 30%. Therefore, if the Chargers go for the first down, their win probability is about
If San Diego sends out the field-goal unit, their win probability is 0.547 if the kick is good, but 0.448 if the kick misses and the Patriots take over at the San Diego 38-yard line. NFL place kickers make about 63% of their kicks from 48 yards. This implies that if the Chargers choose to kick, their win probability is about
Copyright © 2007 by William S. Krasker