For players, teams and gamblers, millions of dollars ride on this year’s Superbowl. Can math give an edge to predicting the winner? Left: Quarterback Peyton Manning of the Denver Broncos. Photo by Peter G. Aiken/Getty Images. Right: Quarterback Russell Wilson of the Seattle Seahawks Photo by Steve Dykes/Getty Images
This Sunday, millions of football fans will cheer for the Denver Broncos and the Seattle Seahawks in the Super Bowl. More bets are placed on the Super Bowl than on any other single day sporting event, according to the American Gaming Commission. Gamblers in Nevada wagered $98.9 million on last year’s game, and the Nevada Gaming Control Board expects this year’s total to exceed that.
But professional sports gamblers like Rufus Peabody know that when it comes to putting money on football, math is powerful.
“I use zero percent gut instinct,” he said.
Instead, Peabody, who founded Massey-Peabody Analytics, which analyzes professional and college football, looks at statistics. Peabody isn’t betting on a winner in this year’s Super Bowl. He’ll be betting on smaller aspects of the game, where the statistics can give a clearer picture of the outcome. That’s because, even as a professional gambler who relies on the numbers, Peabody says it’s hard to find value against the spread in every game.
“If I’m bidding on football against the points spread, the benchmark is 50 percent,” i.e., from the start, it’s a coin toss. “I’m hoping to win 55 percent of the time,” Peabody said.
We had DEN in top 2 every week of the year. SEA first hit #2 for us after week 3. Neck-and-neck since week 10. pic.twitter.com/dq76HWSmuk
— Massey-Peabody (@MasseyPeabody) January 22, 2014
Using advanced statistics, sports analysts try for better than 50-50 accuracy when it comes to predicting a winner. As of this publication, many analysts are tipping their scales in the Broncos’ favor for Sunday’s game. But there are dozens of different statistical models that analysts use to forecast the outcome of any game.
With so much money riding on one game, how accurate is one prediction? At best, a mathematical model can predict the winner 65-70 percent of the time, said Keith Goldner, chief analyst at numberFire. And even advanced statistics have a limit.
“If you had all the information you needed and you had the perfect mathematical equation to kind of predict a winner, you could only be correct about 75 percent of the time … in theory,” said Brian Burke, co-creator of FourthDownBot for the New York Times and founder of Advanced NFL Stats.
For statisticians, being a sports fan is crucial to knowing which numbers are important. Goldner spends as much time watching the NFL as possible. Sports fans can memorize and recite the stats of games and their favorite players. But in statistical analysis, not all numbers are equal, Goldner explained. Rushing yards, passing yards, turnovers, interceptions — each is given its own weight in a formula.
How important those numbers are depends on what you’re asking your model to do, explained Ben Alamar, a sports analytics consultant. As a result, every analyst has a different model, even though many are using the same techniques. In Alamar’s model, he looks at how each team performs in a certain situation, and weighs each matchup. For example, how does each team perform at first and 10, offensively and defensively? When matched against another team, based on their history, how do they match up at the first and 10?
Goldner, on the other hand, looks at how many points a team can score in any given scenario, and runs the numbers from there. Both Alamar and Goldner are calling the game for the Broncos, by a margin of one to two points.
Burke’s model calls the game for the Seahawks, but only by a point. Burke said his formula is a bit simpler than Alamar or Goldner’s. Using statistics to predict the winner of a football game is a bit like trying to make cupcakes, he explained. Every analyst is using the same ingredients, but they each have their own recipe. While some formulas may be aiming for the perfect cupcake, his looks for what each of those ingredients mean and how important they are in making a good cupcake.
“(The model is) really trying to figure out how football really works and understand what makes teams win, versus what makes teams lose. … It’s not optimized to predict a winner but it’s one of the happy side effects of the model. I stick by it,” he said.
In order to make any formula work, statisticians need good data, and a lot of it. That makes football a more challenging sport to analyze statistically than baseball, Goldner said. Most analysis relies on comparing a team’s performance over time, and an NFL team only plays 16 games in a season. Major League Baseball teams play 162 games in each season. Baseball simply has more data, Goldner said, so playing by the numbers is a bit more clear cut. The book “Moneyball” explains how understanding player statistics made the Oakland A’s more competitive.
With more information, analysts can build better models, and they can tell them a lot about how to improve the game. Goldner points out that statistical analysis is growing in every sport as coaches and managers look to the numbers to recruit the best players and adapt their strategies. Take a look at the Carolina Panthers this year, he said. Based on the odds, their coach tried to “go for it” more often on the fourth down, and the Panthers saw some success as a result.
“In general, the statistical stuff isn’t going to, in one season, completely turn you around…what the stats do is they give you a very marginal increase in your chance of winning games and performing better,” Goldner said. Again, look at the Panthers. Their 76.9% fourth down conversion rate got them to the post-season, but not past the San Francisco 49ers in the NFC Divisional round.
“A lot of the old timers in the sport don’t trust the people coming in touting their numbers and who have never played the game. And similarly, a lot of people who crunch the numbers tout their numbers without giving any credence to the knowledge that the old timers have accumulated,” Goldner said.
And there’s so much random chance in a football game — the way a ball bounces, the wind, a player’s last minute decision — that no formula can account for all of it, Alamar said. When it comes down to it, you just can’t beat the spread at this point, he said, and 70 percent accurate is “not exactly an ‘A’ on an exam.”
“Anyone who tells you they can beat the spread with their statistical model is probably lying,” Alamar said. “Because if you can do that, you don’t tell anybody.”
The post Betting on the Super Bowl? Get math on your side appeared first on PBS NewsHour.