The rise of data and analytics in the NFL
The rise of data and analytics in the NFL
The use of data analytics in the NFL has been evolving for several decades, and has only recently found its way into the mainstream coverage of the sport. As a much less data-rich sport to analyse on the surface, football’s data revolution lagged behind baseball’s and basketball’s for some time, but that is starting to change. What began at the Philadelphia Eagles in the mid-1990s under team president Joe Banner, in an attempt to transfer the “sabermetrics” baseball phenomenon over to American football, has now spread to every team, every website, and every die-hard fans’ vocabulary.
In the secret world of NFL headquarters, as with most industries today, data analytics has slowly pervaded the decision-making processes, in order to make better decisions driven by quantifiable evidence. However, the nature of the sport of American football is that there are so many moving pieces, so many different contexts and situations, that finding the right questions to ask and data to pull is a continuing battle. Whereas baseball is really just pitcher-vs-batter, football has an infinite number of interactions that can take place between players on any given play. While that keeps traditional qualitative player scouting and team analysis at the forefront - e.g watching video and breaking down tactics or performance with a coaches’ eye - almost every NFL team is waking up to the reality that incorporating analytics can optimise their decisions and create that crucial edge in a hyper-competitive league.
There is a wide range of applications for data analytics in the NFL, so we will focus the scope of this article on two broad aspects: team strategy and player evaluation.
To simplify the sport, there’s a general dichotomy on each play for the offense: hand the ball to a running back and run the ball, or have the quarterback keep the ball and throw it. The running game is the more traditional aspect that conservative “old school” coaches like to emphasise, for any number of reasons. Historical game data, however, almost overwhelmingly supports more passing of the football.
Academics and analysts of the game have found that, in most situations, passing the ball generates more yards per play and a greater likelihood of eventually scoring. One of the enlightening metrics in this field is “expected points added” or EPA. It’s a standardised measure of play success based on a variety of inputs such as where the offense is on the field, the time left on the clock, the score, how the game has gone so far, and more. Different teams and analysts have various models for this, but Brian Burke of ESPN was one of the first to bring EPA properly into the public domain in 2014. EPA provides a more accurate measure of play success than traditional stats, and it is through EPA per play, among other metrics, that many have advanced the modern emphasis on the passing game based on the data available.
The passing game’s general superiority to the run game has been clarified and proven by researchers from the likes of MIT and Harvard, but it can be explained simply as well. In the 2019 season, passing plays on 1st-and-10 in neutral situations (close score in the first 3 quarters) netted 7.7 net yards per attempt, while running plays averaged just 4.5 yards per attempt. Despite this clear difference, teams called running plays in 57% of those situations. Generally, the conservative run-first mantra of longtime NFL coaches and figureheads keeps the data in the background of decision making when it comes to team strategy. As analytics becomes further and further ingrained in the sport, those outdated and “gut feel” decisions should be phased out.
Some teams, however, are already listening to the analytics and modernising the game. For example, in 2019 the Green Bay Packers and young coach Matt LaFleur passed on 42% of 2nd-and-short situations - a traditional situation to run the football in - and reaped the rewards of their aggressiveness. They averaged 17.7 net yards per play on 2nd down, nearly three times the league average of 6.6. Granted, their quarterback and passer is future Hall of Famer Aaron Rodgers - but the Packers’ analytics-driven passing game has made him more successful than ever.
Another aspect of team strategy based on data is going for it on 4th down. This is where, instead of the traditional moves of kicking a field goal or punting the ball back to the opposition on 4th down, teams are starting to realise that, in certain situations, the data supports a more aggressive approach. Some data-driven teams such as the Baltimore Ravens or Philadelphia Eagles have been on the frontier of this for several seasons, but more and more are waking up to it every season. In 2019, teams went for it on 4th-and-short about twice as often as in 2008. The trend is coming about due to calculations of the probability of success - i.e making the 1st down marker - which makes for an optimised data-led decision rather than a coach’s mere instinct. It is standard now for many teams to have a quantitative assistant with an ear in the Head Coach’s headset, telling him whether to go-for-it or whether to kick the ball. While team models are proprietary and secret, Twitter accounts like Ben Baldwin’s “4th Down Decision Bot” provide fans with an insight into the melding of analytics and football strategy.
Probably the most important function of analytics in the NFL today is the use of data to optimise player evaluation, and particularly around the NFL Draft between seasons. Every April, all 32 NFL teams take turns selecting from a pool of college football players to fill out their rosters, and potentially find a superstar or a diamond in the rough.
With the top young draftees receiving contracts in the tens of millions of dollars, teams lean on data and science to optimise their selection process, and avoid letting their fans down. In the 21st century, teams can generate and analyse data in a number of ways. Most simply, there’s the NFL Combine, which is a series of physical tests that draft entrants undergo in order to show their speed, strength, agility, and other physical measurements. Each of the 32 teams would have their own complicated model, using those test result inputs and others, in order to find the players that have the best chance of success. Websites such as Football Outsiders have created models like KUBIAK, which are likely similar to what teams have in-house. Using regression and more complicated statistical methods, teams can use past data to find whether a player’s college receiving yards should be more heavily weighted than his hand size, or whether his lack of height should outweigh his blazing speed.
Then, there is the innovation of player tracking. This uses GPS chips on player’s jerseys to give teams instant feedback on speed, movement, and the relation of one player to others during a play. This data comes out as x- and y-coordinates, and has seen sweeping introductions throughout the sports world. In football, NFL.com’s Next Gen Stats provides a publically-accessible look into its applications, showing fans which players reached the fastest speeds, broke the most tackles, or covered the most ground on their way to making a tackle. Beyond that, NFL teams are given exclusive and even more granular data that they can manipulate to evaluate talent and find players with the greatest value for their team.
Published by Will Eddowes