Siggestions for conversion of a rating system

[Dhanish]

In your formula "z=(FIDE rating - 2000)/200", what is "FIDE rating"? EDIT: do you mean the current FIDE rating of a given AICCF player?

Yes.

[Dhanish]

Conversion between different human rating systems is sticky at best.

True, but we have to do something. Thank you for your suggestions, we'll examine all possibilities before taking a final decision. Unfortunately, it is an association and convincing everybody is impossible. By following the ICCF system, perhaps we can sell it better!

[Dhanish]

Thank you very much for your detailed analysis. The ELO system is not accepted by many of our members as they say it is too complicated!

3) FIDE rating system based on Elo system, using Gaussian curve
.
.
mean value: don't know

From an ICCF booklet, I find 2200 used as a median value. Can anybody confirm whether 2200 or 2000 is used as the average by FIDE?

[Sven]

Thank you very much for your detailed analysis. The ELO system is not accepted by many of our members as they say it is too complicated!

3) FIDE rating system based on Elo system, using Gaussian curve
.
.
mean value: don't know

From an ICCF booklet, I find 2200 used as a median value. Can anybody confirm whether 2200 or 2000 is used as the average by FIDE?

The Elo system does not appear to be very complicated, I think all chess players can understand it with almost no mathematical background. There is one relevant table that maps rating differences D to scoring probabilities P(D). This table can be calculated based on the Gaussian curve or the Logistic curve, to name two of the most important models; the values differ slightly but not very much. Everything else is already trivial. For the most common case of evaluating games between players who already have an established rating it is just one formula:

NewRating = OldRating + K * (Score - ExpectedScore)

where ExpectedScore is just P(D) for one game, or the sum of all P(D) in case of more than one game. K can be common for all players or depend on rating, age, activity etc. (e.g. higher K for new players).

For the exceptional case of evaluating games of a player who has no rating yet you can either assign an initial default rating (like AICCS does), but then it will take some time until the new player's rating converges to his actual strength, or calculate the initial rating from the first N games (or >= N) which is the usual way. In the latter case a typical approach to obtain the initial rating is an iterative approach where you start with an arbitrary value for OldRating, calculate NewRating, then set OldRating := NewRating and repeat the same until NewRating == OldRating (or Score == ExpectedScore) with reasonable precision. If all N games are won or all N are lost then you need to wait for more games until that condition breaks, to overcome problems with infinity.

If this is already too complicated for AICCF members then my next question would be, what are their requirements and expectations?

- Accuracy?
- System easy to understand and use for members?
- System easy to implement with a computer program?
- Self-regulating?
- Speed of rating adaptation for new players to their real playing strength?
- Comparability of ratings to other rating systems, e.g. FIDE?
- What else?

[Sven]

Thank you very much for your detailed analysis. The ELO system is not accepted by many of our members as they say it is too complicated!

3) FIDE rating system based on Elo system, using Gaussian curve
.
.
mean value: don't know

From an ICCF booklet, I find 2200 used as a median value. Can anybody confirm whether 2200 or 2000 is used as the average by FIDE?

I think none of both applies. There is no fixed average value, or initial average value, of FIDE ratings. Currently FIDE ratings have a lower bound of 1000. You can calculate the current average of all FIDE ratings with a spreadsheet on your own if you like, visit the FIDE ratings download page. The average ratings can vary due to the appearance of new players, and recently FIDE has included many more weaker players into their rating pool (many years ago the lower bound was 2200, this was changed in several steps).

As I already wrote, I would not look at the average FIDE rating of the whole FIDE pool but at the average FIDE rating of all FIDE-rated players in the AICCF pool. If you take the difference between that value and the current average AICCF rating of the same group of players then you get an idea of the offset you would need for a conversion, provided you want the new AICCF ratings to resemble FIDE ratings for those players who are in both pools.

[Dhanish]

As I already wrote, I would not look at the average FIDE rating of the whole FIDE pool but at the average FIDE rating of all FIDE-rated players in the AICCF pool. If you take the difference between that value and the current average AICCF rating of the same group of players then you get an idea of the offset you would need for a conversion, provided you want the new AICCF ratings to resemble FIDE ratings for those players who are in both pools.

Unfortunately, there are <10 FIDE rated players and since most of them have just joined, there are very few games. A few more FIDE rated players have expressed interest to join AICCF and hence the question.