A team that has played five league games and a team that has played thirty can have the same rating, but you should not trust the two numbers equally. Glicko-2 is a rating system that says how much to trust each one. We run it next to our ELO ratings for every club in Europe's top five leagues.
It was created by the statistician Mark Glickman as an extension of ELO. On top of the strength rating it keeps two more numbers for each team: Rating Deviation, how sure the system is about the rating, and Volatility, how consistent the team has been.
How Glicko-2 differs from ELO
In ELO the winner takes points from the loser, and the amount depends on how surprising the result was. Every rating is treated as equally trustworthy. Take two teams both rated 1600: one after 30 steady matches, the other after 5 matches of mixed results. ELO sees two identical teams. Glicko-2 knows the second rating is much less certain, lets it move more after the next result, and gives less weight to results against teams it is unsure about.
Rating Deviation (RD)
What RD means
RD is the margin of error around a team's rating. A team rated 1700 with an RD of 30 is almost certainly between 1640 and 1760. A team rated 1700 with an RD of 90 could be anywhere from 1520 to 1880. The lower the RD, the more you can trust the rating.
Every team starts the season at 1500 with an RD of 350, the most uncertain value. Each result narrows it. That is why RDs are high in August and settle as the season goes on, and why a team with erratic results keeps a higher RD for longer.
Volatility
What volatility means
Volatility measures how much a team's results surprise the system. A team with low volatility performs about as expected, win or lose. A team with high volatility beats a top side one week and loses to a relegation candidate the next. High volatility keeps the RD from shrinking, because the team's strength is harder to pin down.
The ratings we keep for each team
The public Power Rankings table shows the overall rating and its RD. Behind it we keep a few more numbers that go into our prediction models:
Overall rating
The team's general strength, updated after every match from the result (win, draw or loss), the opponent's rating and the uncertainty of both. The home team gets a head start when working out the expected result.
Attack and defence
Separate ratings for scoring and for keeping goals out, each with its own RD and volatility. Scoring more than a team's attack and the opponent's defence suggested lifts the attack rating, conceding less lifts the defence rating.
Home and away
Simple home-only and away-only ratings, updated like ELO, so the models can see teams that are much stronger at home than on the road.
What this changes in a prediction
Glicko-2 also turns the ratings into win, draw and loss probabilities. When either team's RD is high, those probabilities are pulled toward an even contest, because the system is less sure which team is stronger. A 65% chance built on two well-known teams and a 65% chance built on two uncertain ones mean different things, and our models get both the probabilities and the RDs behind them.
Glicko-2 vs. ELO
What each system captures
Classic ELO
- One rating per team
- No measure of how reliable it is
- No measure of consistency
- Every rating treated as equally certain
Our Glicko-2
- Rating, Rating Deviation and Volatility
- A confidence measure for every rating
- Tracks how consistent a team is
- Uncertain ratings move more after each result
- Separate attack, defence, home and away ratings for our models
How to read the Power Rankings table
Each league's Power Rankings page shows ELO and Glicko-2 side by side. These are the columns:
| Column | What it shows |
|---|---|
| ELO | The team's overall ELO rating. The table is sorted by it. |
| G2 | The team's overall Glicko-2 rating. Higher is stronger, and every team starts the season at 1500. |
| RD | How uncertain the Glicko-2 rating is. Below 80 it is reliable, above 120 it is still unsettled, usually early in the season or after mixed results. |
| ELO± | How much the ELO rating changed on the latest matchday. |
| G2± | The same for Glicko-2. Because uncertain ratings move more, G2± can be larger or smaller than ELO± for the same result, and occasionally go the other way. |
| Rank Δ | Glicko-2 position minus ELO position. Positive means Glicko-2 ranks the team lower, negative means higher. It is the quickest way to spot where the two systems disagree. |
Note: Volatility and the attack, defence, home and away ratings are model inputs and are not shown in the public table. Glicko-2 is one input among many. Football stays unpredictable, and no rating removes that.
See Glicko-2 for Europe's top leagues
The Power Rankings pages show the Glicko-2 rating and RD next to ELO for every club, updated after every matchday.
In short
ELO tells you how strong a team is. Glicko-2 tells you that too, and adds how sure it is and how consistent the team has been. Where the two systems disagree, the Rank Δ column shows you straight away.