Touchdown! UBC study uses novel form of decision-making modeling to calculate football’s winning formula
Researchers at the UBC Sauder School of Business and the University of Toronto analyzed over 160,000 NFL plays to find out what works and what doesn’t — and the method can be applied to other sports.
For decades sports fans, teams and broadcasters have been obsessed with statistics, whether they’re runs batted in, pass completions, shots on goal or serve percentages. But commonly used football statistics, such as yards gained, are missing out on the underlying story because they don’t take into account in what situation yards were gained or lost.
But a new UBC Sauder and University of Toronto study is putting a high-tech spin on sports data using fundamental mathematical theory and a novel form of decision-making modeling — and it has the potential to seriously boost teams’ strategy and performance.
For the study, titled Points Gained in Football: Using Markov Process-Based Value Functions to Assess Team Performance, researchers analyzed play-by-play data from 1,034 regular season NFL football games (2013-2016), and an astonishing 164,299 individual plays.
For each play, they extracted the year, quarter, down, yards to go for first down, yard line, offensive and defensive team and play type, then used “Markov process” methods to quantify the value created or lost on each play.
In other words, with unprecedented precision, they computed what worked and what didn’t and when.
“Sports data is as prolific as financial data — but data by itself, until it's analyzed correctly, doesn't really make sense,” says UBC Sauder Professor Emeritus Martin Puterman (he, him, his), a sports fan who has spent his career crunching numbers. “This takes data and uses it intelligently to figure out what you're good at, what you're not good at, and what areas to emphasize in your practice sessions.”
The key difference, says Professor Puterman, is the calculations don’t focus on end results like touchdowns or goals scored, but rather use what short-term steps were required to achieve a particular end goal.
“It’s a way of looking at how to make a series of decisions in the best way possible. So you might sacrifice something over the short term to gain over the long term,” explains Professor Puterman, who coauthored the study with University of Toronto engineering professor (and UBC grad) Timothy Chan and University of Toronto graduate student Craig Fernandes. It was published in the journal Operations Research.
“If I wanted to run a 10k race, I wouldn't sprint as fast as I can for the first kilometre, because I have to balance my long-term objective of achieving a good total time with the energy I spend in the short term. So a Markov decision process tells you how to do that, and determine the trade-off between the present and the future.”
The authors found that in NFL football, passes generate the most points gained, whereas running plays tended to generate a negative value. Among passing plays, short passes were behind most of the top teams’ success — and the bottom teams’ poor outcomes. They also found the best teams distinguish themselves in the middle of the field, as opposed to at either end.
What’s more, they examined the results for specific NFL teams, and found the top three during this period (New England Patriots, New Orleans Saints and Green Bay Packers) gained more points by passing than average teams, leading the researchers to conclude that outstanding passing leads to outstanding performance.
They also found that the New Orleans Saints needed to work on its field goals — they ranked lowest on points gained on field goal attempts — and that the Dallas Cowboys were best in terms of running, but only slightly above average in terms of passing.
Professor Puterman says the approach is especially suited to sports like football, tennis and golf, where the game is based on discrete plays, as opposed to sports like hockey and soccer where the action is more fluid.
So how might teams apply the findings? Professor Puterman emphasizes that there is no one formula that will take a team from the bottom of the standings to the top. Rather, the method precisely identifies what teams’ strengths and weaknesses are so they can reinforce what they do best and improve the areas that need work.
“You can look objectively at how you compare to other teams based on the value gained for each point, and find out where your strengths are so you can emphasize those skills. In the areas with the greatest losses, you can build up strength on that particular play or stop running the play altogether,” says Professor Puterman, who hopes to see points gained or lost as a regular feature of sports stats in football and other pro sports news coverage.
“But then other teams will figure the same things out, and they’ll develop their strengths. So it should go back and forth.”
The approach isn’t limited to helping teams or players improve their game, however: it can also be used to analyze any number of systems that require complex series of steps to achieve particular outcomes, from manufacturing to health care.
“The idea is to take the old-fashioned measure that everybody used before and to convert them into measures that relate to the objective of the game,” he says. “And those points can be in business generating revenue, or in healthcare saving lives.”
Interview language: English