Using random forests to estimate win probability before each play of an NFL game

Abstract

Before any play of a National Football League (NFL) game, the probability that a given team will win depends on many situational variables (such as time remaining, yards to go for a first down, field position and current score) as well as the relative quality of the two teams as quantified by the Las Vegas point spread. We use a random forest method to combine pre-play variables to estimate Win Probability (WP) before any play of an NFL game. When a subset of NFL play-by-play data for the 12 seasons from 2001 to 2012 is used as a training dataset, our method provides WP estimates that resemble true win probability and accurately predict game outcomes, especially in the later stages of games. In addition to being intrinsically interesting in real time to observers of an NFL football game, our WP estimates can provide useful evaluations of plays and, in some cases, coaching decisions.

nflWAR: a reproducible method for offensive player evaluation in football

“…Similar to the evaluation of the EP model, we again use LOSO CV to select the above model, which yields the best calibration results. Figure 6 shows the calibration plots by quarter, mimicking the approach of Lopez (2017) and Yam and Lopez (2018), who evaluate both our W P model and that of Lock and Nettleton (2014). The observed proportion of wins closely matches the expected proportion of wins within each bin for each quarter, indicating that the model is well-calibrated across all quarters of play and across the spectrum of possible win probabilities.…”

“…With the notable exception of Lock and Nettleton (2014), researchers typically only vaguely discuss the methodology used for modeling expected points and/or win probability. Additionally, prior researchers in this area typically do not provide their specific expected points and win probability estimates publicly for other researchers to use and explore.…”

“…Play-by-play data, or information specifying the conditions, features, and results of an individual play, serves as the basis for most modern sports analysis in the public sphere (Kubatko et al, 2007, Macdonald (2011, Lock and Nettleton (2014), Thomas and Ventura (2017)). Outside of the public sphere, many professional sports teams and leagues have access to data at even finer levels of granularity, e.g.…”

Player evaluation using win probabilities in sports competitions

“…One disadvantage of the empirical approach in computing win probabilities is that one has to subjectively decide how to bin or partition the variables (time remaining, field position, and score differential) to get reasonable estimates of the win probabilities. Lock and Nettleton illustrate the use of random forests methodology which uses the data to adaptively choose the partitioning of the variables to minimize prediction error. In addition, their methodology uses pre‐game point spreads to adjust the probabilities of the teams winning at the beginning of the game.…”

“…Lock and Nettleton provide several applications of the win probability methodology. Changes in win probabilities can be used to judge the most influential plays in a given football game.…”