Written by Chad Langager at Sporting Charts
The Elo rating system is best known in the world of chess to compare the relative skill levels of players based on their historical performance in ranked matches. Players gain points when they win a match and drop points when they lose a match – the number of points won or loss will depend on the ranking of their opponent. You will gain far less points by beating a lower ranked player than by beating the top ranked player. You can read more about Elo here.
We decided to use this Elo rating system as the backend of our NHL Power Rankings. Our goal is to provide a guide on the relative skill level of each team based on their historical performance against other NHL teams.
These rankings can be used to calculate the probability of one team beating another. A 100 point difference between two teams is equal to a 64% win probability for the higher ranked team while a 200 point difference is equal to a 75.9% win probability.
We won’t go too deep into the calculations behind the Elo rating as it uses a significant amount of code but here are some key considerations:
- Our Power Rankings look at games going back to the 2005-06 season
- Each team started out with a 1500 ranking at the start of the 2005-06 season
- To account for home-ice advantage, home teams receive a 45 point boost for each home game
- The impact of each game on the overall score depends on two factors: Elo score difference and goal differential.
- The greater the Elo difference between the two teams the greater the impact on the Elo rating after the game.
- The greater the goal differential in the game the greater the impact on the Elo rating after the game.
- Elo rating is zero-sum: where one team gains another loses the exact same amount
- After each season, Elo ratings are regressed back to the mean (1500) to account for off-season changes
In future posts, we will add some more information on our methodology but lets get to the rankings. As of December 15, 2015 here are our NHL Power Rankings:
To continue reading this article, click here
