Why you should care
Because these models have the potential to fix more than the gridiron.
Perhaps there is no time-honored tradition more prevalent than the couch-bound fan doling out advice during Sunday football games. Will Burton coyly admits that he’s just as guilty of trying to predict play calls from his apartment. What makes him different from the rest of us? Most of the time, he’s actually right.
The North Carolina State fifth-year senior built a statistical model this summer that predicts whether an NFL team will run or pass with the kind of accuracy — about 75 percent — that could win Super Bowls (or at least a lot of bragging rights). The model factors in all the stats you might expect, like time remaining, point difference, yards to go and play downs, but Burton, just 23 years old, thinks he has come up with the real trick: combining the right mix of stats and historical data dating back more than 15 years. “We could have made it even more accurate than it is,” Burton tells OZY, if letting teams use it more quickly wasn’t so critical.
It’s cool to predict football plays, but how much does that help?
Could it change the sport as we know it? Of course it could. That is, if the accuracy keeps improving, potentially allowing, say, the Dallas Cowboys to essentially read the minds of opposing quarterbacks. Burton impressed enough stat pros that he presented his algorithms at the 2015 Joint Statistical Meetings in Seattle, the largest annual gathering of stat wonks in North America. NC State published his research online as the first public statistical model to forecast pro football plays — plus, it’s fully interactive!
Sporting a blue polo, cropped hair and a brown NC State cap, Burton is fratty-athletic in a fashion that mimics many of football’s emerging stat geeks. Once relegated to dorky online blogs, those who know how to grapple with Big Data are becoming hot commodities. The Pittsburgh Steelers hired a numbers wrangler in August, the Miami Dolphins added one in March, and there are probably about “10 to 15” franchises with significant stat departments, says Shane Reese, a Brigham Young University statistics professor who works with college, Olympic and pro teams. “We’re going to continue to see people investing serious resources,” he says.
That’s due in part to the explosion of data-gathering tools available. Wearable tech like Fitbit wristbands, Adidas jersey sensors and Catapult Sports helmet devices are able to gather information on everything from speed and agility to dehydration and sleeping habits. Play-by-play data, meanwhile, is crowdsourced from websites like Armchair Analysis, which Burton and his collaborator, Michael Dickey, who declined to be interviewed for this article, used to build their prediction machine. It’s the first project of this nature published openly, although experts like Reese say some NFL teams have almost certainly created their own, similar tools, but are keeping them on the down low. (The league itself and several teams declined to comment on Burton or their own efforts.)
“The best takeaway is that analytics are getting bigger and bigger,” says Burton. The biggest surprise: that early turnovers, more than any other factor, had the greatest effect on play calling. If a team gave the ball away, they ran the ball 90 percent of the time afterward to avoid another costly fumble or interception. As most fans might have already guessed, teams pass more than ever — up 56.7 percent in 2014 from 54.4 percent in 2012. Burton tested his model against a random sample of 20 NFL games, accurately calling three-fourths of plays as runs or passes. The best result was when the model correctly predicted almost 92 percent of the play calls from a 2014 game between the Jacksonville Jaguars and Dallas Cowboys.
In five notebooks, he plots ways to translate the skills he’s learned forecasting football games to disparities in health care coverage.
Yet for all his wonky sports analysis, Burton lights up most when talking about possibilities away from the field. “It’s cool to predict football plays,” he says, “but how much does that help?” Sure, the fact that he wrestled and played tennis while attending Millbrook High School in Raleigh, North Carolina, reinforces the jock look. But the tedium of school wore on Burton, and that feeling followed him to college. That is, until he read two articles about how researchers were using stats to track sex traffickers and evacuate flood victims in Haiti. Burton had found his higher purpose.
But while analytics skills are highly valued by businesses these days, most academics are more likely to thumb their noses at sports-related work. “It’s hard to impress those guys,” Burton admits, recounting how his statistical model wasn’t as highly regarded as, say, an algorithm that isolates cell irregularities in sick people. Distrust of analytics lingers even in pro sports, from coaches who worry their play-calling jobs might become redundant to general managers who don’t think stats paint a complete picture. “I still don’t know that it’s viable as a career,” says NC State’s Justin Post, a professor and co-founder of the university’s sports statistics club. “There aren’t that many sports teams, and not that many jobs to go around.”
That doesn’t bother Burton, though, who has applied to NC State’s graduate analytics program. Today, he scribbles through five notebooks, plotting ways to translate the skills he has learned forecasting football games to disparities in health care coverage and, yes, the educational system that he feels failed him. Between ink blots are schemes to build a model that predicts the number of patients likely to show up at a hospital’s ER, or to use analytics to link environmental factors to a person’s overall health.
Even higher learning doesn’t escape Burton’s calculating: He says he wants to crunch numbers to make sure that college scholarships go to the students “who are most in need,” rather than those who benefited from a privileged background.