- Blind mode tutorial
lichess.org
Donate

Glicko2 has downsides

In order to improve, lower-ranked players must play against those who are better than them of course - there is no better way to improve than by examining why one lost.

The Glicko2 has a flaw in which the comparatively higher-ranted are discouraged from playing against those rated lower than them.

Over the course of 17 games of 0+1 and 1/2+0, I have a 14-3 record against Snafu, an interesting player whose approximate rating of 1450 mystifies me. Our games are usually tight, but he has beaten me twice very well and once when my cat sat on the computer and disconnected the internet :O. Despite a win ratio of nearly 5:1, my rating has gone down by 4 against this player.

Against another player, Bereikua, I have gained a total of 59 points with a 34-12 record - a ratio of about 3:1 against a player approximately 90pts below me in ratings. That's a gain of around 2.5 per game.

Is it not the case that Glicko2 is fundamentally flawed for fast bullet games?

In order to improve, lower-ranked players must play against those who are better than them of course - there is no better way to improve than by examining why one lost. The Glicko2 has a flaw in which the comparatively higher-ranted are discouraged from playing against those rated lower than them. Over the course of 17 games of 0+1 and 1/2+0, I have a 14-3 record against Snafu, an interesting player whose approximate rating of 1450 mystifies me. Our games are usually tight, but he has beaten me twice very well and once when my cat sat on the computer and disconnected the internet :O. Despite a win ratio of nearly 5:1, my rating has gone down by 4 against this player. Against another player, Bereikua, I have gained a total of 59 points with a 34-12 record - a ratio of about 3:1 against a player approximately 90pts below me in ratings. That's a gain of around 2.5 per game. Is it not the case that Glicko2 is fundamentally flawed for fast bullet games?

I have many questions about the glicko-2 system, and also some reservations. However, I have to say, it's a lot better than some systems out there. English Chess Federation grading works on averages or something, and churns out some quite ridiculous grades. Suffice to say, there are problems with every system. Your examples are good though. In the end, I don't really know what to think.

I have many questions about the glicko-2 system, and also some reservations. However, I have to say, it's a lot better than some systems out there. English Chess Federation grading works on averages or something, and churns out some quite ridiculous grades. Suffice to say, there are problems with every system. Your examples are good though. In the end, I don't really know what to think.

I agree it is better than some systems.

I think the only way to be truly accurate would be to do away with individual scores and instead rate people through a points system where points are gained through competitive play - in other words in tournaments.

For example, supposing the system in place scored the top 100 people in a tournament but only the best 100 results that a player has in a specific timescale defines their overall rating...

That would keep their score 'current' and stop score inflation from the number of games someone plays.

I'm not sure how one would go about testing that except by taking a sample of current players of different levels and examining their rating under a new system... and I'm not sure I'm clever enough to do that :D

Glicko 2 works, of course, very well for longer games, where the differential in level between players rated say 200 points apart is so much greater in reality.

I'm going to have to go away and think about this :)

I agree it is better than some systems. I think the only way to be truly accurate would be to do away with individual scores and instead rate people through a points system where points are gained through competitive play - in other words in tournaments. For example, supposing the system in place scored the top 100 people in a tournament but only the best 100 results that a player has in a specific timescale defines their overall rating... That would keep their score 'current' and stop score inflation from the number of games someone plays. I'm not sure how one would go about testing that except by taking a sample of current players of different levels and examining their rating under a new system... and I'm not sure I'm clever enough to do that :D Glicko 2 works, of course, very well for longer games, where the differential in level between players rated say 200 points apart is so much greater in reality. I'm going to have to go away and think about this :)

This topic has been archived and can no longer be replied to.