I've been impressed with the way the Glicko rating system works. In particular:

  • more games played recently cause your rating to stabilize
  • fewer games played recently cause your rating to fluctuate until you get several recent games under your belt

USCF modified their system to be similar a few years back (where there is a sliding scale to amplify or dampen the rating change for a given match), and now that Glickman (the inventor of Glicko) is on the rating board for the USCF, I imagine they'll eventually adopt an even more flexible system, like his own.

Is there a reason (other than being resistant to change, or that they just haven't gotten around to it yet) that FIDE doesn't adopt a more accurate rating system?

  • glicko involves rating deviation...is this perhaps more relevant for online people with no particular schedule rather than for actual chess professionals who are expected to play on a more consistent basis? – BCLC Feb 27 at 16:08

It is not proven that Glicko is more accurate. Glicko just solves the problem a different way, with a different emphasis. Consider that all these systems abstract a human mind to a 4 digit number.

There's a lot of politics in ratings; it's a touchy subject. If FIDE were to adopt it, they'd produce a new top-100 list, and a lot of GMs would likely be upset.

The ELO system is very easy to understand; transparency is important. More complex systems have immediate disadvantages.

  • 7
    +1, although I disagree with the first paragraph, the rest of it is true enough. 1) Glicko is proven to be more accurate on average (measured in terms of how often it is able to predict who will win a given match). 2) About abstracting a human mind to a number, this isn't really the case. It is a measure of skill, such that you can calculate probabilities (for win/draw/lose) when comparing skill levels. It's all backed by math and statistics. In Glicko-2 there's even a new factor for volatility, for those players who play inconsistently (allowing their ratings to fluctuate more). – Eve Freeman May 13 '12 at 1:20
  • 1
    Interesting. However, I'd affirm that 'better on average' isn't necessarily 'better'. What if the times when it isn't more accurate causes actual harm? I'll go read up on it, however. – Tony Ennis May 13 '12 at 17:17
  • This is informative. chess.com/article/view/chess-ratings---how-they-work – Tony Ennis May 13 '12 at 17:38
  • glicko involves rating deviation...is this perhaps more relevant for online people with no particular schedule rather than for actual chess professionals who are expected to play on a more consistent basis? – BCLC Feb 27 at 16:08

As far as I understand the Glicko rating system adds a reliability score to the rating number, which may provide a higher fairness regarding the rating score adjustment calculations. Obviously this might lead to the wrong assumption that there is a correlation between playing rated games and resulting playing strength (since not playing decreases the reliability score which in turns de-valuates your rating number somehow). I see a lot of problems connected to the reliability score, although I must admit I am no expert about the subtleties of the system. Just to pick three:

  • A player playing only the rare rated game might get no invitations anymore. Imagine you are a tournament director inviting attractive players. Now you have to choose between 10 players, all of which have the same rating number, but different reliability numbers. Probably you will pick the ones with the best reliability numbers first. Over time the reliability gap between players with many rated games and less ones might widen.
  • Manipulation of rating numbers might become easier. Even the ELO system is not cheat-safe, because some people got ratings or IM/GM norms by playing in rigged tournaments. A good reliability score might make helping players with no reliability score very cheap - the good player gives rating points without losing much himself. As soon as money is involved, being cheat-safe is an important issues. I have the feeling the Glicko rating system is more prone to manipulations.
  • The communication and understanding of playing strength might become more difficult to the masses. When you communicate a ranking list, you run into problems. A player with a rating of 2000-50 has a "real playing strength" somewhere between 1900 and 2100. Another player has got 1950-2000, which translates to a number in the range of 1750 and 2150. So the player with 1950-2000 might be in fact the better player, but will probably be ranked below the other one, only because he played less games or against less reliable opponents. You make explaining your ranking system more difficult to people, without really guaranteeing more fairness. A GM which stops playing might gain "potential strength" and boast about his world champion potential, because his reliability number makes increasing room for deviation (mind it: in both directions, not only downwards!).

To sum it up: You might get more than you bargained for, because the whole package does not only come with improvements. And since there are heavy monetary issues involved when you look at the upper end of the ranking lists, I do not see the merits of changing an existing system.

  • Most ranking lists have a cutoff for the reliability score (RD), rather than trying to use it to rank players. So a 2000 (50) would always be ranked higher than a 1950 (100). According to their current rating, they're higher rated, regardless of the reliability. If the 1950 (100) player would win against the 2000 (50), their rating will increase more significantly than it would if they were 1950 (50). Can you describe how one might cheat the system with Glicko in more detail? Are you talking about sandbagging your rating if you have high RD? – Eve Freeman May 14 '12 at 17:42
  • @West Freeman: I doubt this is the perfect place to prove the relative cheat-proneness of a rating system and I am just taking guesses, because to me it looks cheaper to manipulate the system, since I won't loose that much rating points myself when I lose intentionally against an unrated player. The more important point of my answer is that you should check a system not only for calculation fairness, but to cheat-proneness, communcatability etc., before you throw away a working system. – Ray May 15 '12 at 5:55
  • But how does artificially raising someone's rating help them? Usually, people trying to game the system are trying to lower their rating so they can play in lower sections and win the first prize. This might be easier with Glicko, if you avoid playing for a long period of time, and then lose a few games that you should have won. But combined with rating floors and the fact that this can only be done every once in a while because of the RD taking time to go back up. – Eve Freeman May 15 '12 at 6:01
  • 1
    You might need high ratings to gain a title, get invitations or get paid by clubs to play for them or work as a trainer. I am speaking about professional chess here and where the real money is, because you questioned FIDE's behaviour. – Ray May 15 '12 at 6:30
  • glicko involves rating deviation...is this perhaps more relevant for online people with no particular schedule rather than for actual chess professionals who are expected to play on a more consistent basis? – BCLC Feb 27 at 16:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.