Insights & Articles

Using Player Ratings to Predict Game Outcomes

Understanding Elo and Beyond

Have you ever wondered how football analysts are able to predict the outcomes of games with such accuracy? The answer lies – at least in part – in the Elo rating system, which assigns numerical ratings to teams based on their past performance. But what if we told you that we could take this a step further and predict game outcomes based on individual player ratings?

Traditionally, player performance has been evaluated based on their statistics, but another method for player ratings merges this with the Elo rating system. This method classifies raw statistics into a win or loss per game, using a draw line or win/loss threshold for each evaluated statistic. If a player’s performance exceeds the statistical draw line, it is considered a win, and if not, it is classified as a loss.

To calculate a player’s Elo rating, we need the current Elo ratings of both the player and his opponent. But who is the opponent? For example, if we're evaluating a running back, the opposing team is too general to be considered the opponent, as the team has defense, offense, and special team components. Instead, we leverage the Elo rating model and apply it to players pitted against an opponent’s ability to oppose that player’s position (e.g., running back vs. rush defense, receiver vs. pass defense, etc.). In the case of a running back, for each game, the player is evaluated against the opposing team’s rush defense.

Player Elo ratings are computed for each statistic independently and then aggregated into a composite player Elo rating for each position, currently evenly weighting the individual statistics to generate an overall metric. Draw lines for a selected set of team and individual offensive-oriented ratings are based on historic team game statistics from previous performances. Other offensive and defensive-oriented draw lines were developed to establish a comprehensive position-by-position win-versus-loss assessment.

A Proven Methodology

Using team and player game statistics from the 2000-2019 college football seasons from Sports Reference, this approach was found to predict game outcomes more accurately than traditional methods. While the data covers the NCAA Division I Football Bowl Subdivision (FBS) teams and players, the results are missing players from the Football Championship Subdivision (FCS). However, the methodology could be adapted to evaluate these players if the same team and player statistics at a game summary level were available.

This method of using player ratings to predict game outcomes can provide valuable insights for football analysts, enthusiasts and bettors alike. By combining statistics and the Elo rating system, we can predict not only the performance of teams, but also that of individual players. By leveraging both statistics and the Elo rating system, this methodology offers a more nuanced and accurate way to evaluate player performance and predict game outcomes.