lichess.org
Donate

Why chess ratings don't mean what they used to

thanks for your time and effort, its a great article and i can only assume it is a great book as well.

thanks for your time and effort, its a great article and i can only assume it is a great book as well.

I haven't read the full article but I am pretty sure this has contributed to why Kramnik keeps getting spanked by younger players and keeps blaming cheating because of it.

Just saying.

I haven't read the full article but I am pretty sure this has contributed to why Kramnik keeps getting spanked by younger players and keeps blaming cheating because of it. Just saying.

Ratings are the reason players from Carlsen down to some very young juniors are avoiding playing. In fact for South African players, it has become a thing to go and gain rating overseas, then refrain from playing locally so that you can qualify on rating for closed tournaments from which national teams come. Initially rating was meant to be similar to a golf handicap, indicating a certain level of skill. However, a pool of players playing only each other year after year will remain at the same rating, even though their skill constantly goes up. For example the amount of 2700s are actually less than 10 years ago, while the general level of top chess, influenced by computers, has in fact gone up immensely. Perhaps the rating system can be fixed, else we should look at alternatively structuring chess so that established players will also have something to play for.

Ratings are the reason players from Carlsen down to some very young juniors are avoiding playing. In fact for South African players, it has become a thing to go and gain rating overseas, then refrain from playing locally so that you can qualify on rating for closed tournaments from which national teams come. Initially rating was meant to be similar to a golf handicap, indicating a certain level of skill. However, a pool of players playing only each other year after year will remain at the same rating, even though their skill constantly goes up. For example the amount of 2700s are actually less than 10 years ago, while the general level of top chess, influenced by computers, has in fact gone up immensely. Perhaps the rating system can be fixed, else we should look at alternatively structuring chess so that established players will also have something to play for.

absolutely agreed, though this isn't only youth, sometimes elderly also massively outpeform, last tournament I played against a pretty old guy, who had fairly low rating, but he beat everyone including me, and later I checked what was his chance to show that performance with that rating according to ELO and it was like 1%, clearly his rating was decades old or not seriously active in any case and he had been gaining skill and playing ever since (likely online), he was like 1700 blitz, but played like NM.
I'd say the whole system needs to be scrapped and rebuilt from scratch, not only there is massive issue of rating pools, but the whole system isn't really all that good, and certainly not good for modern dynamics. It ignores that players have massively different levels of motivations, some are more motivated to win when playing significantly lower rated or feel more comfortable with it, while some feel like they have nothing to prove and the rating gain is small so they underperform. Also the rating system can be gamed too much in various ways. And one of biggest issues I see is that it is completely blind to the well known phenomenon of 'kryptonite opponent' and simply someone who's style is very uncomfortable for you to play against while his average results might not be all that much, this is a very common scenario which I think just ends up randomizing ratings in the end, and making them more and more unreliable, an intelligent system would try to do something like mapping players according to styles, I think it could be very easily done by building some sort of opponent tree that would identify styles based on whom you lost against - I think it's not that complicated to do something like that just by linking who's losing against whom, the system does not need to know anything about chess rules, it can just discover types via patterns of losing say there is a player named X, there are 100 players who all have beat him multiple times without losinga single time yet those players should not be beating him according to their average performance, that can identify him as belonging to certain style and them belonging to certain style group. Ratings could be assigned based on every common style group, thus a player would have say 20 different ratings based on how he performs against what type players, perhaps the unified rating then would be average of all, that could capture true strength and true reality 10x better than current system

absolutely agreed, though this isn't only youth, sometimes elderly also massively outpeform, last tournament I played against a pretty old guy, who had fairly low rating, but he beat everyone including me, and later I checked what was his chance to show that performance with that rating according to ELO and it was like 1%, clearly his rating was decades old or not seriously active in any case and he had been gaining skill and playing ever since (likely online), he was like 1700 blitz, but played like NM. I'd say the whole system needs to be scrapped and rebuilt from scratch, not only there is massive issue of rating pools, but the whole system isn't really all that good, and certainly not good for modern dynamics. It ignores that players have massively different levels of motivations, some are more motivated to win when playing significantly lower rated or feel more comfortable with it, while some feel like they have nothing to prove and the rating gain is small so they underperform. Also the rating system can be gamed too much in various ways. And one of biggest issues I see is that it is completely blind to the well known phenomenon of 'kryptonite opponent' and simply someone who's style is very uncomfortable for you to play against while his average results might not be all that much, this is a very common scenario which I think just ends up randomizing ratings in the end, and making them more and more unreliable, an intelligent system would try to do something like mapping players according to styles, I think it could be very easily done by building some sort of opponent tree that would identify styles based on whom you lost against - I think it's not that complicated to do something like that just by linking who's losing against whom, the system does not need to know anything about chess rules, it can just discover types via patterns of losing say there is a player named X, there are 100 players who all have beat him multiple times without losinga single time yet those players should not be beating him according to their average performance, that can identify him as belonging to certain style and them belonging to certain style group. Ratings could be assigned based on every common style group, thus a player would have say 20 different ratings based on how he performs against what type players, perhaps the unified rating then would be average of all, that could capture true strength and true reality 10x better than current system

This blog is totally relatable. Great research and this is another reason that most players even who are Asian travel to Europe most of the time to play tournaments and go on tours even myself has many Asian friends like that... Overall great blob

This blog is totally relatable. Great research and this is another reason that most players even who are Asian travel to Europe most of the time to play tournaments and go on tours even myself has many Asian friends like that... Overall great blob

Let's not forget that players also decline naturally with age. The brain learns best at adolescence, later a mixture of youth and experience helps, and with age comes the decline. Every honest player at the age of 70+ knows what I'm talking about.

Not only 2700+ GM's, also for players around 2000:
Around 50% of them have their rating peak between the age of 36 and 42!
This is not only true for the "new" FIDE development, that was also true decades ago.

If someone finds their rating slowly declimimg around the age of 50, don't solely blame the reasons in the article, but admit your slowly declining skill and work on it harder than before if it's important to you.

In case you had bad luck with opponents from different countries and younger players and therefore lose 100 rating points in a few games:
Congratulations, you are now underrated yourself and win the points back soon!

If you usually play in a country that was a deep blue in the article: Admit that you are overrated compared to the rest of the world ;)

Let's not forget that players also decline naturally with age. The brain learns best at adolescence, later a mixture of youth and experience helps, and with age comes the decline. Every honest player at the age of 70+ knows what I'm talking about. Not only 2700+ GM's, also for players around 2000: Around 50% of them have their rating peak between the age of 36 and 42! This is not only true for the "new" FIDE development, that was also true decades ago. If someone finds their rating slowly declimimg around the age of 50, don't solely blame the reasons in the article, but admit your slowly declining skill and work on it harder than before if it's important to you. In case you had bad luck with opponents from different countries and younger players and therefore lose 100 rating points in a few games: Congratulations, you are now underrated yourself and win the points back soon! If you usually play in a country that was a deep blue in the article: Admit that you are overrated compared to the rest of the world ;)