Band 16, Heft 2 - Special Issue on Player Tracking Data in the National Football League (NFL) / Guest Editor: Michael Lopez

Most historical National Football League (NFL) analysis, both mainstream and academic, has relied on public, play-level data to generate team and player comparisons. Given the number of oft omitted variables that impact on-field results, such as play call, game situation, and opponent strength, findings tend to be more anecdotal than actionable. With the release of player tracking data, however, analysts can better ask and answer questions to isolate skill and strategy. In this article, we highlight the limitations of traditional analyses, and use a decades-old punching bag for analysts, fourth-down strategy, as a microcosm for why tracking data is needed. Specifically, we assert that, in absence of using the precise yardage needed for a first down, past findings supporting an aggressive fourth down strategy may have been overstated. Next, we synthesize recent work that comprises this special Journal of Quantitative Analysis in Sports issue into player tracking data in football. Finally, we conclude with some best practices and limitations regarding usage of this data. The release of player tracking data marks a transition for the league and its’ analysts, and we hope this issue helps guide innovation in football analytics for years to come.

Using high-resolution player tracking data made available by the National Football League (NFL) for their 2019 Big Data Bowl competition, we introduce the Expected Hypothetical Completion Probability (EHCP), a objective framework for evaluating plays. At the heart of EHCP is the question “on a given passing play, did the quarterback throw the pass to the receiver who was most likely to catch it?” To answer this question, we first built a Bayesian non-parametric catch probability model that automatically accounts for complex interactions between inputs like the receiver’s speed and distances to the ball and nearest defender. While building such a model is, in principle, straightforward, using it to reason about a hypothetical pass is challenging because many of the model inputs corresponding to a hypothetical are necessarily unobserved. To wit, it is impossible to observe how close an un-targeted receiver would be to his nearest defender had the pass been thrown to him instead of the receiver who was actually targeted. To overcome this fundamental difficulty, we propose imputing the unobservable inputs and averaging our model predictions across these imputations to derive EHCP. In this way, EHCP can track how the completion probability evolves for each receiver over the course of a play in a way that accounts for the uncertainty about missing inputs.

The NFL collects detailed tracking data capturing the location of all players and the ball during each play. Although the raw form of this data is not publicly available, the NFL releases a set of aggregated statistics via their Next Gen Stats (NGS) platform. They also provide charts showing the locations of pass attempts and outcomes for individual quarterbacks. Our work aims to partially close the gap between what data is available privately (to NFL teams) and publicly, and our contribution is two-fold. First, we introduce an image processing tool designed specifically for extracting the raw data from the NGS pass charts. We extract the pass outcome, coordinates, and other metadata. Second, we analyze the resulting dataset, examining the spatial tendencies and performances of individual quarterbacks and defenses. We use a generalized additive model for completion percentages by field location. We introduce a naive Bayes approach for estimating the 2-D completion percentage surfaces of individual teams and quarterbacks, and we provide a one-number summary, completion percentage above expectation (CPAE), for evaluating quarterbacks and team defenses. We find that our pass location data closely matches the NFL’s tracking data, and that our CPAE metric closely matches the NFL’s proprietary CPAE metric.

Tracking data in the National Football League (NFL) is a sequence of spatial-temporal measurements that varies in length depending on the duration of the play. In this paper, we demonstrate how model-based curve clustering of observed player trajectories can be used to identify the routes run by eligible receivers on offensive passing plays. We use a Bernstein polynomial basis function to represent cluster centers, and the Expectation Maximization algorithm to learn the route labels for each of the 33,967 routes run on the 6963 passing plays in the data set. With few assumptions and no pre-existing labels, we are able to closely recreate the standard route tree from our algorithm. We go on to suggest ideas for new potential receiver metrics that account for receiver deployment and movement common throughout the league. The resulting route labels can also be paired with film to enable streamlined queries of game film.

This paper details a route classification method for American football using a template matching scheme that is quick and does not require manual labeling. Pre-defined routes from a standard receiver route tree are aligned closely with game routes in order to determine the closest match. Based on a test game with manually labeled routes, the method achieves moderate success with an overall accuracy of 72% of the 232 routes labeled correctly.

Statistical analysis of defensive players in football has lagged behind that of offensive players, special teams, and coaching decisions, largely because data on individual defensive players has historically been lacking. With the introduction of player tracking data from the NFL, researchers can now tackle these problems. However, event and strategy annotations in the NFL’s tracking data are limited, especially when it comes to describing what defensive players do on each play. Moreover, methods for creating these annotations typically require extensive human labeling, which is difficult and expensive. Because of the importance of the passing game and the limited prior research on the defensive side of passing, we provide annotations for the pass coverage types of cornerbacks using unsupervised learning techniques, which require no training data. We define a set of features from the tracking data that distinguish between “man” and “zone” coverage. We use mixture models to create clusters corresponding to each group, allowing us to provide probabilistic assignments to each coverage type (or cluster). Additionally, we quantify each feature’s influence in distinguishing defensive pass coverage types. Our work makes possible several potential avenues of future NFL research into defensive backs and pass coverage strategies.

Continuous-time assessments of game outcomes in sports have become increasingly common in the last decade. In American football, only discrete-time estimates of play value were possible, since the most advanced public football datasets were recorded at the play-by-play level. While measures such as expected points and win probability are useful for evaluating football plays and game situations, there has been no research into how these values change throughout the course of a play. In this work, we make two main contributions: First, we introduce a general framework for continuous-time within-play valuation in the National Football League using player-tracking data. Our modular framework incorporates several modular sub-models, to easily incorporate recent work involving player tracking data in football. Second, we use a long short-term memory recurrent neural network to construct a ball-carrier model to estimate how many yards the ball-carrier is expected to gain from their current position, conditional on the locations and trajectories of the ball-carrier, their teammates and opponents. Additionally, we demonstrate an extension with conditional density estimation so that the expectation of any measure of play value can be calculated in continuous-time, which was never before possible at such a granular level.