You would need to measure the individual effects in order to compensate them. Most of them are not measurable in practice, however.

Personally, I believe that some sort of Whole History Rating would be a much better model than anything ELO-derived.

_________________
A good system naturally covers all corner cases without further effort.

Imagine an isolated group of 1 kyus, somewhere in the middle of nowhere, all at rating 2000. They don't visit tournaments, but every week they have a tournament amongst themselves and send the result to the EGD. Since they form a closed system, no rating points from outside enter the system. The epsilon parameter, meant to compensate for deflation, does however enter points into their system. Over time, their average rating will drift upwards. (This is not unreasonable, as presumably they get stronger from playing those round robins).

Now imagine the same group of 1 kyus, but instead of a weekly round robin, they just play over the internet (doesn't count for the rating). Once per year they play a real over the board tournament together and send the result to the EGD. With only one entry into the database, they will hardly drift at all.

From this example, we can conclude that playing more tournaments results in more rating movement upwards.

Of course real life is not that simple, and there are other factors at work, such as rating resets, new players entering the system, old players leaving the system, etc.

In general, we might expect:

• Players from countries with fewer tournaments will tend to be underrated.
• Players from countries with many new players entering will tend to be underrated.
• Players from countries which do not allow rating resets will tend to be underrated.

And, of course, the opposite will also hold:

• Players from countries with many tournaments will tend to be overrated.
• Players from countries with no new players entering will tend to be overrated.
• Players from countries which use rating resets often will tend to be overrated.

Since each country has different combinations of these factors, it is very hard to pinpoint which countries are under or overrated by how much.

To compensate for this behavior, you might use a per person factor determining the reliability of their rating. That's what the AGA ratings do with their Sigma, and also what a rating system like Glicko does.

Thanks HermanHiddema for the clarification.

/Mats

_________________
mohsart - games & books
http://spel.mohsart.se/

WHL, Glicko, Glicko2, Maximum Likelihood would all, IMHO, be a really nice improvement to a system that has to rely on data as sparse as 6 games per year for accurately rating people.

Organise a twice yearly 3 day 8 round residential Chatteris tournament with cheap booze or a camp site week long summer event by a pub on the Norfolk broads with the option to frolic in the water.

Then you can get an accurate GoR

Sheesh, that sounds really fun actually

Since it was conceived, GoR has not been updated. It will take more than arguments pointing out that a better system can be designed to change that.
I added some data to the bottom of this page that are vaguely relevant to this thread
http://senseis.xmp.net/?EGFRatingPerRank

I thought epsilon was changed, as Javaness noted some time ago...

However, I would like to see the dutch system (for dan players) explained here. Maybe, it is a viable alternative to rating magic.

There is a short explanation at SL's EuropeanRanks already, but I'll give an answer here as well:

Promotions for dan grades are regulated by the Dutch Go Association via a "Promotion Committee". This committee consists of three people, who award promotions based on a player's tournament results.

To get promoted to X dan, the most important criterion is:

• Score at least about 50% against other X dan in serious even games, over a sufficient number of games.

The "sufficient" here depends on the grade you're trying to get: higher grades require more games (something like 8+4*X games to get X dan, I think).

If this criterion is met, then there are two additional criteria that come into play:

• Score about 67% against players 1 rank weaker than X dan.
• Score about 33% against players 1 ranks stronger than X dan (at least 1 win required).

The above three criteria can, additionally, compensate a bit for each other.
For example: If you have ca. 60% against X dan, that could compensate for getting only ca. 25% against X+1 dan.

Very strong results can also lower the "sufficient" number of games in the first criterion. Personal example: I was promoted to 4d with a result of 8 wins, 3 losses against other 4 dan. Normally something like 12-12 is required.

For edge cases, there are several additional fuzzy factors:

• How did the player do against players 2 or more ranks higher than X?
• How old is the player, and how long has he been his current rank?
• Did the results include foreign players? Are they from a country known to be weak/strong? How is their rating?
• Do we like the player's hair style? What is his shoe size? Is today a thursday?

This can be a pretty fuzzy system, and in the end it is the decision of the committee, and we trust their expertise. There are rarely complaints.

I don't think it's really a viable alternative for a European wide system though

Nice explanation Herman! Did you ever ask if they're looking for a new third member now that Ad is inactive? You seem to be the perfect candidate

_________________
My name is Gijs, from Utrecht, NL.

When in doubt, play the most aggressive move

Sounds like a very good system to me. Thanks very much Herman!

I've corrected the links for the old BGA promotions committee there - it is http://www.britgo.org/history/gradcomm.html. The points system was created by Andrew Daly and was in operation for about 25-30 years.

Interesting system, I like it. What about demotion?

In practice, demotion only happens if a player asks for that himself. The nice thing about this system is that a dan grade becomes somehow a nice "title" (promotion norms are relatively strict). The downside is, of course, that there are a few (usually older) players that have been underperforming for their grade for many years. I think this is not so bad though, because many players apply for demotion anyway after a sufficient number of bad results.

_________________
My name is Gijs, from Utrecht, NL.

When in doubt, play the most aggressive move

Sankyu Herman. Nice system, but probably suitable for limited populations only. (Most European countries are small enough, however.)

I've also read about the finnish system... it seems to be awfully complicated and I wonder how many players are really promoted that way? I.e. why do they wait so long for tournaments to get promoted, when basically the local club can do it by itself irrespective of tournament results and GoR?

Actually I guess the first change was the introduction of tournament classes, after that initial change everything stayed the same. If you suggested a change, you were told that it would involve changing something. Then, yes, after 10 years along came a small change to one of the parameters. Since that changed the average rating of a shodan has been crawling upwards instead of downwards..

Isn't that fine, in about 30 years everything will be aligned to rating math again.

In the meantime we can comfort ourselves with the knowledge that self ranking is better in predicting results than rating anyway (acc. to andre engels case study for dutch kyu players). In countries with more emphasis on "earned" ranks, we will probably see that their ranks are not very predictive for the results at all.