User:Ruirliu/sandbox

American Football Analytics

Football analytics is the application of using data in the form of metrics and statistical models in the context of American football. As an emerging field in the National Football League (NFL), the major professional league for American football in the United States, it is used to inform coaches and teams on how to game plan against opponents, optimize training, or make decisions during matches, e.g. determining whether to punt the ball or to try to convert on 4^th down. Analytics also plays a role in the world of sports betting and fantasy football, a form of gambling that revolves around players gaining points for the better based on their performance in the games they play.

Early History and Beginnings

Brian Burks, the founder of Advanced Football Analytics, developed win-probability charts on NFL games during his oversea travels. This hobby then contributed to the first wave of modern football analytics in 2008 as the concept of win probability started to influence how games played out in real time. Burks is credited for being among the earliest people who created databases, findings from which resulted in a significant shift in professional football. Innovation arose from reviewing statistics with context rather than taking the numbers for what they are at face-value.

Building on the idea that not every yard gained has equal impact on the outcome of a game, Aaron Schatz, editor-in-chief of Football Outsiders, developed a formula to assign value to plays based on how “successful” they were in the early 2000s. With Schatz recognizing that the down, the field location, and the game clock of any given play are all important in determining the effect the outcome has on the game, the formula takes the context of each play into account, making such a metric more useful when analyzing how well a player or team is performing.

With the advent of faster computers and technology, it became easier to analyze large sets of data, which allowed for concepts that had previously been introduced decades before, e.g. the idea of expected points from Virgil Carter in the 1980s, a founding father of football analytics, to be fully utilized and developed. The realization of these “advanced stats” gradually resulted in the widespread evolution of football analytics heading into the 21^st century.

Use of Analytics by NFL Teams

NFL coaches and teams have more reliable and useful resources to inform their decisions before and during a game. They can better game plan against opponents by analyzing how “successful” certain formations and plays have been, and they can therefore strategize against the more effective ones; with access to contextualized metrics that accredit appropriate value to each play, it is much easier and faster to use analytics rather than just simply watching hours of film tape.

During matches, teams can use analytics to determine be more aggressive or to be more risk-averse. Win probability models provide tangible reasons for making certain decisions as the ultimate goal is to win the game; for example, if it was 4^th down and there were three options: punting the ball, kicking a field goal, or attempting to get a 1^st down, teams that heavily rely on analytics would choose the play with the highest chance of helping them win the game every time as opposed to relying on “gut feeling” (such as coaching experience, sense of momentum, etc.).

Philadelphia Eagles vs. New England Patriots in Super Bowl LII, which the Eagles won 41-33

Ryan Paganetti, a game management coach who formerly worked with the Philadelphia Eagles from 2015 to 2020, contributed to the team’s progressive and aggressive playstyle that helped them win Super Bowl LII in 2017. As the team found success from using analytics to fuel their aggressive decision-making, the number of 4^th down and two point conversion attempts across the NFL went up drastically. The shift from more traditional and conservative game management to more data-based thinking has allowed for NFL teams to make statistically better decisions while also promoting innovation in the utilization of analytics.

Use of Analytics in Fantasy Football and Sports Betting

Fantasy football, a popular form of engagement and betting among the NFL’s audience, has also been affected by the advent of football analytics. The premise of fantasy football is drafting players and setting a line-up every week where the line-up that gets the most points (which are obtained through gaining yards, scoring touchdowns, etc.) wins. With access to more advanced statistical tools and data, bettors can better predict how a player will perform over a given week or season as without just having to look at box scores.

An important point of analysis is the opportunities or usage for a given player on their team; by analyzing both the quantity and quantity of those opportunities (such as targets for a receiver), one can more accurately speculate how well that player would do in a game. Josh Hermsmeyer, a football analyst and writer, developed the “weighted opportunity rating” (WOPR) metric that is heavily used to evaluate the wide receiver position in fantasy football; it takes into account both the receiver’s target share (how many times he is targeting in a game) and his air yard share (the number of yards from the line of scrimmage to the point of the catch).

As more statistical models and made and improved upon by fantasy football sportsbooks like FanDuel, projections of player’s performances become more reliable as they take every bit of context into account. In the world of sports betting, even the smallest adjustments from applying analytical tools could make the biggest differences in outcome.

Notable Advanced Statistics and Metrics

Defense-adjusted Value Over Average (DVOA)

DVOA, a metric created by Football Outsiders, takes into account the down and yards to convert, field position, score, time left, and opponent when evaluating the “success” of any given play. It is measured as a percentage, where a team with a 25% DVOA is 25% better than the league average and a team with a -25% DVOA is 25% worse than the average.

A play deemed “successful” is worth one point, and one deemed “unsuccessful” is worth zero points (fractional points are possible); large plays are worth more points, and losing yards or turnovers (except for interceptions on 4^th down in the last two minutes of a game) are worth negative points. The Value Over Average (VOA) is calculated by adding up the point value of every single play for a team or player and dividing that by certain baselines. DVOA is then calculated by adjusting for the quality of the opposing defense by accounting for the average success in stopping each type of play.

Expected Points Added (EPA)

EPA represents the number of points a given play is expected to contribute to a team, which can be positive or negative (the latter if it would help the opposing team score). It is calculated by using the average of the scoring outcomes of plays in past games that were in the same exact situation (same down, distance, and field position) of a given play in interest; the difference between the point expectancy before the play and that after it is the EPA. For example, if a team gains five yards on a 3^rd and 4, that would yield a positive EPA as it becomes more likely that the team progresses toward their opponent’s endzone and scores.

Completion Percentage Over Expectation (CPOE)

CPOE is a measure of how accurate a quarterback is, accounting for the difficulty of passes thrown. It considers factors such as the air yards the throw covers and the distance of separation of the receiver from the nearest defender. It is calculated by taking the difference between a quarterback’s actual completion percentage (percentage of throws completed) and the expected completion percentage, which is calculated using the factors mentioned before.

Adjusted Net Yards Per Attempt (ANY/A)

ANY/A quantifies a quarterback’s production by accounting for every aspect of his performance: yardage, touchdowns, interceptions, and sacks. In contrast to the yards per attempt (YPA) stat, ANY/A attributes positive value to scoring passes and negative value to sacks and interceptions. It is calculated by using this formula, which represents the total value the quarterback contributes per play:

$ANY/A=(passing\ yards+20\cdot passing\ touchdowns-45\cdot interceptions-sack\ yards)\div (passes+sacks)$

Future of Analytics

As more and more NFL teams adopt data-driven strategizing and game-planning, the field of football analytics is becoming more and more prominent. Schematically speaking, teams are becoming more aggressive and risk tolerant; there has been an approximate 23% increase in the number of 4^th down attempts from 2017 to 2019. [9] Given that passing attempts yield 3.2 more yards than rushing attempts on average, this wave of football analytics might push teams to pass more often on 1^st down as opposed to running the ball, which has been the more traditional route.

Similar to other subfields in data analysis and collection, the progress of football analytics is coupled with the innovation of new or improved technologies and computers. Amazon Web Services has partnered with the NFL to create Next Gen Stats, which provides statistical analysis in a digestible format for teams to use when planning for games or for training. Next Gen Stats takes data from Zebra Technologies, which started working with the NFL by implementing radio frequency identification (RFID) technology in stadiums to track players’ movements and tendencies in real time.

The evolution of data science and machine learning allows for advancements in analyzing film and rendering accurate probability models. As methods for data collection and analysis continue to improve, more dependable predictions and frameworks can be made to help inform decisions made by coaches and watchers alike.