Students Say They Can Guess NFL Play Calling 75 Percent of the Time

0812_Russell_Wilson_Interception
Seattle Seahawks head coach Pete Carroll and quarterback Russell Wilson (3) talk after Wilson's game-deciding interception in the fourth quarter of Super Bowl XLIX. The play call, a pass instead of a run, led to discussion about methods of play calling in the NFL. Brian Snyder/REUTERS

On the last play of Super Bowl XLIX, the Seattle Seahawks were lined up one yard outside of the New England Patriots’ end zone. Down four points with 24 seconds left, the stakes were clear: score and win the game, or lose. Since they possessed one of the NFL’s best running backs in Marshawn Lynch, most fans watching on TV assumed that the Seahawks were going to call a simple inside running play and pound the ball over the line. Then, this happened:

Disasters on critical plays often spark conversation about play-calling. Often, statistics can be more reliable in managing game situations than instincts.

Consider: A new statistical model created by students at North Carolina State University correctly predicted play types at an eye-popping rate when tested against data from real NFL games. Put into practice, the tool could become indispensable to play-calling coaches. Presenting the model’s findings at the Joint Statistical Meetings in Seattle, William Burton, an engineering student at NCSU, explained that "if the offensive play type can be predicted—say a pass—the defensive coordinator can call a blitz or coverage play to gain an advantage."

According to a press release on the study, the model averaged about 75 percent accuracy in predicting calls that were made in 20 NFL games during the 2014 season. The best result—95 percent accuracy for a game between the Dallas Cowboys and Jacksonville Jaguars—correctly predicted 109 out of 119 plays. If you watched the Cowboys last year, you might not think much of those results; after all, they were a pretty run-heavy team and fairly predictable. But the model also correctly called about 80 percent of the plays from the Super Bowl, which featured the much less predictable New England Patriots.

Football coaches and aficionados possess many methods to predict what type of play a team will call. Various computer models for play calling have been around for almost a decade, and the use of analytics is on the rise in the NFL. ZEUS, a play-calling program developed in 2005, was designed to help coaches call their own plays on a laptop from the sideline. But the new model developed by Burton and his collaborator, statistics student Michael Dickey, might be more useful, since it is designed to be used as a tool for anticipating the opposing team’s play. Burton and Dickey’s interactive visualization of the model, the press release says, will make the statistical probabilities of running and passing plays interpretable at a glance.

Statistical sophistication can be the difference between winners and losers in professional football. According to a 2006 study from the University of California, Berkeley, “the behavior of teams…on fourth downs [during the 2005 season] departs from the behavior that would maximize their chances of winning in a way that is highly systematic, clear-cut, and statistically significant.” The old school image of the gutsy coach with a rolled up program under his arms giving the “game of inches” speech is going the way of the dinosaur.

But there’s still a place for pure instinct in anticipating plays. When New England cornerback Malcolm Butler made the game-winning interception in Super Bowl XLIX, he jumped the route on a hunch. After the game, Butler told reporters: "I don't know how I knew. I just knew. I just beat him to the point and caught the ball.”

Correction: A previous version of this article incorrectly referred to Malcolm Butler as a cornerback playing for the Seattle Seahawks in Super Bowl XLIX. Butler played for the New England Patriots.