Loading...
Wins and losses tell you what happened. Elo ratings tell you how strong a team actually is — and how much each result should shift your expectations.
TL;DR
Elo is a rating system that updates after every game based on the result and the strength of the opponent. Win against a strong team = big rating boost. Lose to a weak team = big drop. Over time, ratings converge on each team's true strength.
The concept is straightforward. Every team carries a numerical rating that goes up when they win and down when they lose. The key insight is how much the rating changes after each game.
Every team starts at a baseline rating, typically 1500. This represents a perfectly average NHL team.
After each game, the winner gains rating points and the loser loses the same number. It's a zero-sum exchange.
The expected result is calculated from the rating difference. A 100-point gap means the higher-rated team wins roughly 64% of the time.
Beating a team you were supposed to beat earns a small gain. An upset win earns a large gain. The K-factor controls this sensitivity.
Over a full season, ratings naturally sort themselves. Teams that beat good opponents climb. Teams that lose to bad opponents fall. No human judgment required.
Not every sport suits Elo equally. Hockey happens to be a strong fit for several reasons.
An 82-game season gives Elo enough data to stabilize. By mid-season, ratings reliably separate contenders from pretenders.
Strength of schedule is baked in. A team that beats three top-10 opponents gains more than one that beats three bottom-feeders. No manual adjustment needed.
A team on a 10-game win streak against quality opponents will be rated higher than their record alone would suggest. Elo captures momentum before the standings do.
Elo is purely result-based. It doesn't directly account for injuries, trades, goalie changes, or shot quality. For that, you need a more complete model.
Elo isn't the only way to rank teams. Here's how it compares to the alternatives.
Simple and transparent. Uses only game outcomes and opponent strength. No underlying stats, no subjectivity. Easy to calculate, easy to explain.
Often produced by writers or polls. Can incorporate context (injuries, eye test) but are inherently subjective. Two rankers can produce very different lists from the same data.
Models like PuckCast use 150+ features including Elo-like strength ratings, expected goals, goaltending metrics, and schedule factors. More complex, but more accurate for game-level predictions.
PuckCast doesn't use a pure Elo system. The prediction model builds on the same core idea — adjusting team strength after every game based on result and opponent quality — but layers in far more context.
Recent form is weighted more heavily than early-season results. Strength of schedule feeds directly into power rankings and playoff projections. The result is a team strength signal that's more responsive and more accurate than standalone Elo.
Elo in the model
Most teams hover around the baseline of 1500. A team rated above 1550 is performing well above average, and anything over 1600 signals a truly elite team. At the other end, teams below 1450 are struggling significantly. The best teams in a given season rarely exceed 1650.
Elo ratings update after every single game. That means on a busy NHL night with 15 games, 30 teams get rating adjustments. This constant updating is one of the reasons Elo captures momentum and form changes faster than standings or weekly power rankings.
Elo ratings are a useful baseline for playoff predictions, but they work best as one input among many. Playoffs introduce small sample sizes, goalie matchups, and coaching adjustments that Elo alone can't capture. PuckCast combines Elo-like strength ratings with 150+ other features for more accurate playoff projections.
The Elo system was created by Hungarian-American physicist Arpad Elo in the 1960s to rate chess players. It was adopted by FIDE (the international chess federation) and later adapted for team sports including football, basketball, and hockey. The core math — expected score based on rating difference — translates cleanly to any head-to-head competition.