My (or my opponent's) rating changes very dramatically after a game and the numbers don't seem to add up. How are ratings calculated?

Free online Chess server. Play Chess now in a clean interface. No registration, no ads, no plugin required. Play Chess with the computer, friends or random opponents.

Ask a question

Popular questions

What is "Average centipawn loss"?
All About Cheating: Questions and Answers
What is the LM title?
Could we have a Bell graph of all player rating ?
What does the 'li' in 'lichess' stand for?
What's the diference between IM, NM, CM, LM, GM and FM?
Training mode in mobile app
Way of Berserk trophy?
How does the rating system work on here?
Request: hide ratings in rated games
What to do about players that time down?
How do you get a Zug Miracle trophy?
Were the Correspondence trophy's and top ten players removed from Lichess?
What does a question mark (?) next to my rating mean?
Is ban in the chat/forum forever?
Can I contribute to lichess? I am a developer.
How do you pronounce Lichess?
Feedback regarding "This player uses chess computer assistance" of another player
Is it possible to favorite a chess puzzle for future references?
How can I send a challenge request (correspondence) to my friend without having to sit and wait for my friend to accept?

4 Answers

Ratings are calculated using the Glicko-2 rating method developed by Mark Glickman. This is a very popular rating method, and is used by a significant amount of chess organisations (FIDE being a notable counter-example, as they still use the dated Elo rating system).

Fundamentally, Glicko ratings use "confidence intervals" when calculating and representing your rating. When you first start using the site, your rating starts at 1500 ± 700. The 1500 represents your rating, and the 700 represents the confidence interval.

Basically, the system is 90% sure that your rating is somewhere between 800 and 2200. It is incredibly uncertain. Because of this, when a player is just starting out, their rating will change very dramatically, potentially several hundred points at a time. But after several games, the confidence interval will narrow, and the amount of points gained/lost after each game will lessen.

Another thing to note is that the confidence interval changes over time. If you win or lose many games (or rather "points") in a row, the confidence interval will increase allowing you to gain/lose points points more rapidly. This is because a winning/losing streak means that you are incorrectly rated/seeded and the rating system should compensate for that.

"Why use 1500, and why use Glicko in the first place? I much preferred it when the ratings started at 1200 and Elo was used; it was far more accurate with FIDE ratings."

1500 is used because it is the figure specified by the Glicko-2 documentation that can be accessed here: www.glicko.net/glicko/glicko2.pdf

The change from Elo to Glicko was a conscious decision. The main issue with Elo was that ratings moved too rapidly. Against a player of equal strength, it was possible to gain or lose 100 points in less than a handful of games. There were other issues as well, such as most players (of vastly different strengths) being bulked together near 1200. This was incredibly problematic.

After changing to Glicko-2, players have been far more evenly spread-out within the rating ranges, making it far easier to find an opponent of appropriate strength. It is also far easier to maintain a rating.

"But ratings are far higher compared to other sites and organisations such as FIDE, USCF, the ICC, etc... What's up with that?"

It is best not to think of ratings as absolute numbers, or compare them against other organisations. Different organisations have different levels of players, different rating systems (Elo, Glicko, Glicko-2, or a modified version of the aforementioned). These factors can drastically affect the absolute numbers (ratings).

It's best to think of ratings as "relative" figures (as opposed to "absolute" figures). By this I mean: within a pool of players, their relative differences in ratings will help you estimate who will win/draw/loss, and how often. Saying "I have X rating" means nothing unless there are other players to compare that rating to.

Fundamentally, Glicko ratings use "confidence intervals" when calculating and representing your rating. When you first start using the site, your rating starts at 1500 ± 700. The 1500 represents your rating, and the 700 represents the confidence interval.

Basically, the system is 90% sure that your rating is somewhere between 800 and 2200. It is incredibly uncertain. Because of this, when a player is just starting out, their rating will change very dramatically, potentially several hundred points at a time. But after several games, the confidence interval will narrow, and the amount of points gained/lost after each game will lessen.

Another thing to note is that the confidence interval changes over time. If you win or lose many games (or rather "points") in a row, the confidence interval will increase allowing you to gain/lose points points more rapidly. This is because a winning/losing streak means that you are incorrectly rated/seeded and the rating system should compensate for that.

"Why use 1500, and why use Glicko in the first place? I much preferred it when the ratings started at 1200 and Elo was used; it was far more accurate with FIDE ratings."

1500 is used because it is the figure specified by the Glicko-2 documentation that can be accessed here: www.glicko.net/glicko/glicko2.pdf

The change from Elo to Glicko was a conscious decision. The main issue with Elo was that ratings moved too rapidly. Against a player of equal strength, it was possible to gain or lose 100 points in less than a handful of games. There were other issues as well, such as most players (of vastly different strengths) being bulked together near 1200. This was incredibly problematic.

After changing to Glicko-2, players have been far more evenly spread-out within the rating ranges, making it far easier to find an opponent of appropriate strength. It is also far easier to maintain a rating.

"But ratings are far higher compared to other sites and organisations such as FIDE, USCF, the ICC, etc... What's up with that?"

It is best not to think of ratings as absolute numbers, or compare them against other organisations. Different organisations have different levels of players, different rating systems (Elo, Glicko, Glicko-2, or a modified version of the aforementioned). These factors can drastically affect the absolute numbers (ratings).

It's best to think of ratings as "relative" figures (as opposed to "absolute" figures). By this I mean: within a pool of players, their relative differences in ratings will help you estimate who will win/draw/loss, and how often. Saying "I have X rating" means nothing unless there are other players to compare that rating to.

If you are familiar with bell curves, the average here is normalised at 1500. That doesn't mean that the average person plays at 1500. It means the average of everybody playing has been set at the arbitrary figure of 1500. Somebody from math want to jump in and explain it better or more better or even accurately? :D Is there an absolute minimum and maximum rating -

http://www.englishchess.org.uk/wp-content/uploads/2012/04/The_Glicko_system_for_beginners1.pdf

For beginners?!

For beginners?!

That paper I just linked to describes the way the formula works precisely but I think that's the jist of it....

but i want to understand what is the temporary rating with the sign '?' and what is the reason of usng it?

if someone can explain

Absolutely incredible answer, thank you so much!

Great answer and great approach, but this explanation of confidence intervals is actually wrong and would be more suited for bayesian credible intervals. From wikipedia:

> The confidence interval can be expressed in terms of a single sample: "There is a 90% probability that the calculated confidence interval from some future experiment encompasses the true value of the population parameter." Note this is a probability statement about the confidence interval, not the population parameter.

how is posible that player has 2000 elo and play versus 1300-1400 players all time?@/ENGINEHOUDINI7

Only registered members with one week of lichess activity can contribute to the Q&A.