1. ## Planned Rating Changes

Over the past 3-4 mths the ACF Ratings Officers have been experimenting with a number of possible changes to the rating system.

The main aims being :
a) To better handle the impact on players who play rapidly improving players (or for that matter rapidly declining players)
b) To better handle players who are new to the normal list even though they may be active on the rapid list and vice versa.

With regards to a) this is achieved by generating an intermediate rating for all players. For the calculation of the player's rating his rating at the start of the period is still used but the opponents intermediate rating is used as the opponents rating rather than their rating at the start of the rating period.

For example back in 2002 and up to the March 2003 rating period Ronald Yu's rating had been between 1691 in April 2002 and 1776 in March 2003. Now for the June 2003 ratings Ronalds opponents had their results calculated based on his March 2003 rating of 1776 yet he was clearly playing at a much higher level. It therefore seems reasonable for Ronalds opponents to treat him not as 1776 but as a player rated more like 2050. BTW Ronalds performance rating during this period was 2225. Testing shows that it is indeed better to do this as it leads to better predictive accuracy. It should be noted that the USCF uses this intermediate rating idea in their rating calculations.

With regards b) this is achieved by increasing the RD factor of the player when his rapid is used as the seed rating for his first normal games or vice versa in the case of his first rapid games. This increase in RD was already done previously but it was found to improve predictive accuracy if we increased it more. Note increasing it to the point of maximum RD(350) actually reduces predictive accuracy, which we already knew and I had mentioned previously on this board.

Lastly we are investigating whether we should raise the rating floor from is current value of zero to possibly 100.

2. Well done, it's good to see you finally recognise there is a problem! What criterion will you use to determine that a player meets the status of 'rapidly improving'? And how will this work in a pool of players that are generally improving as a cohort, eg groups of juniors that mainly play each other, as might apply at the Gardiner Chess Centre, for example?

3. Originally Posted by David_Richards
Well done, it's good to see you finally recognise there is a problem!
Ha ha ha. Now why doesnt it surprise me that you would be the first to reply.

Originally Posted by David_Richards
What criterion will you use to determine that a player meets the status of 'rapidly improving'?
The system already detects rapidly improving players provided they have a non trivial number of games actually submitted for rating. As to how it works I already explained it in the post.

Originally Posted by David_Richards
And how will this work in a pool of players that are generally improving as a cohort, eg groups of juniors that mainly play each other, as might apply at the Gardiner Chess Centre, for example?
Havent we been through this before.
I suggest you look back at the previous threads.

4. Sounds like a good idea Bill, I know a lot of players have suffered from playing underrated players and hopefully the new technique might better protect against possible future deflation.

Having the rating floor at 100 is a great addition too, I and a couple of others suggested implementing it on a previous thread so thanks for considering it.

I was wondering though, with the way you want to shape the rating system would I be correct in assuming you would want optimally, a figure for each player that represents their true ability closer than simply their past performances?

5. Originally Posted by Bruce Dickinson
Sounds like a good idea Bill, I know a lot of players have suffered from playing underrated players and hopefully the new technique might better protect against possible future deflation.
There isnt any evidence of deflation using the current method and you would want to maintain that with any changes to the system.

Originally Posted by Bruce Dickinson
Having the rating floor at 100 is a great addition too, I and a couple of others suggested implementing it on a previous thread so thanks for considering it.
Yeah I'm still looking at this.

Originally Posted by Bruce Dickinson
I was wondering though, with the way you want to shape the rating system would I be correct in assuming you would want optimally, a figure for each player that represents their true ability closer than simply their past performances?
This is always a tricky question.
E.g make the players new rating their performance rating for the period. Afetr all that must surely represent their true ability.
Bzzzt. Wrong.

What rating systems come down to is merging the players current performance into his previous performance. Hopefully doing it in such a way that their new rating is the best predictor of their future results.
Of course you want everyones rating to meet that criteria.
Easier said than necessarily done.

6. Originally Posted by Bill Gletsos
What rating systems come down to is merging the players current performance into his previous performance. Hopefully doing it in such a way that their new rating is the best predictor of their future results.
Of course you want everyones rating to meet that criteria.
Easier said than necessarily done.
"Hopfully doing it it in such a way ..."

"Easier said than necessarily done."

This reeks of mediocracy. Should it be rewarded with a kiss?

Edit: Lemme say I am not having a go at your hard work, just your system.

7. Originally Posted by Matthew Sweeney
"Hopfully doing it it in such a way ..."

"Easier said than necessarily done."

This reeks of mediocracy. Should it be rewarded with a kiss?
I was simply trying to explain to BD the issues.
Also I never said that was the any particular system.
I was talking in general terms.

Edit: Let me say I am having a go at you.

8. Ha ha ha. Now why doesnt it surprise me that you would be the first to reply.
Because after criticising you continually for not listening, I wanted to be the first to congratulate you for doing so.

The system already detects rapidly improving players provided they have a non trivial number of games actually submitted for rating. As to how it works I already explained it in the post.
No, you've given a single example without any detail of how you arrived at 2050. Is it linked to volatility?

Havent we been through this before.
I suggest you look back at the previous threads.
No, you've never suggested this idea before. It will clearly have a beneficial effect.

9. Originally Posted by Bill Gletsos
Over the past 3-4 mths the ACF Ratings Officers have been experimenting with a number of possible changes to the rating system.

The main aims being :
a) To better handle the impact on players who play rapidly improving players (or for that matter rapidly declining players)
b) To better handle players who are new to the normal list even though they may be active on the rapid list and vice versa.

With regards to a) this is achieved by generating an intermediate rating for all players. For the calculation of the player's rating his rating at the start of the period is still used but the opponents intermediate rating is used as the opponents rating rather than their rating at the start of the rating period.

.
Bill

Another rating thread eh. Will be good for some post counts of certain people.

BTW, Christopher Wallis is a case that bears looking at. He has just had two super tournaments after the last rating calc. and he is going to cause a major distortion to our Interclub team selections (we run 25 teams).

starter

10. Originally Posted by Matthew Sweeney
"Hopfully doing it it in such a way ..."

"Easier said than necessarily done."

This reeks of mediocracy. Should it be rewarded with a kiss?

Edit: Lemme say I am not having a go at your hard work, just your system.
I'd say this would be the mechanism he used to correct the ratings on the Gold Coast, which lead to an average increase of around 65 pts.

11. Originally Posted by David_Richards
Because after criticising you continually for not listening, I wanted to be the first to congratulate you for doing so.
I wasnt sure how to take your orignal comment.

Originally Posted by David_Richards
No, you've given a single example without any detail of how you arrived at 2050. Is it linked to volatility?
You simply do an initial calculation using the standard Glicko2 formula. It gives the 2050.

Originally Posted by David_Richards
No, you've never suggested this idea before. It will clearly have a beneficial effect.
You misunderstand me.
You were obviously referring to supposed isolated rating pools.
It was that I was referring to when I said read the previous threads.

There are lots of ideas we may be looking at.
I just dont bother wasting my time discussing them on the BB.

Once tests have been done and if the results positive then I'll comment on them.

12. Originally Posted by David_Richards
I'd say this would be the mechanism he used to correct the ratings on the Gold Coast
Then you would be wrong.
Of course I had explained at the time what we did.
I'm not going to waste my time repeating it.

13. Originally Posted by Bill Gletsos

There are lots of ideas we may be looking at.
I just dont bother wasting my time discussing them on the BB.

Once tests have been done and if the results positive then I'll comment on them.
I don't think I've been wasting you're time, Bill - I hope I've been providing impetus for change, simply throwing up some ideas!

14. Originally Posted by starter
Bill

Another rating thread eh. Will be good for some post counts of certain people.
I was sorely tempted just to wait until it was implemented before commenting so as not to put up with stupid replys.(DR this is not referring to you so far)

Originally Posted by starter
BTW, Christopher Wallis is a case that bears looking at. He has just had two super tournaments after the last rating calc. and he is going to cause a major distortion to our Interclub team selections (we run 25 teams).
Unfortunately there isnt much that can be done for him so far based on reported results. Even under the new system he would only be about 1656.
Eugene Schon on the other hand would be around 1238.

15. Originally Posted by Bill Gletsos
Then you would be wrong.
Of course I had explained at the time what we did.
I'm not going to waste my time repeating it.

Then maybe you should, run it retrospectively, it might tidy up many anomalies.