FIDE Rating Consultation

[Paul McKeown]

See: https://www.fide.com/news/2538

See also: https://www.fide.com/docs/presentations ... System.pdf

[Wadih Khoury]

Fantastic read for those interested in these things.

[Mick Norris]

Will be interesting what the ECF decide to do

Hopefully if the ECF rating system undergoes a similar change it would take place in summer after the end of the league and county season

[Roger de Coverly]

Hopefully if the ECF rating system undergoes a similar change it would take place in summer after the end of the league and county season

I'm not sure the ECF system will need to be changed. In a manner of speaking it has already implemented the changes now proposed by FIDE, certainly new players are introduced with a dummy game. Also it's very noticeable that players with both ECF and FIDE ratings have higher ECF ratings than FIDE ones. The difference is getting wider. If you play in a tournament where all opponents are rated in both systems, you may find that around a 50% score results in a gain of ECF points and a loss of FIDE ones. Another difference is that the ECF system attempts to rate players of a lower standard than the FIDE minimum. The old ECF grading system gave juniors generously high ratings by treating them as new players every season and had done this for around ten years. These higher grades would have been converted into the new British Elo system.

One might also recall that fifteen years ago, the ECF were persuaded that the players towards the bottom of the grading list were under graded. Part of the alleged evidence for this was a comparison with the FIDE ratings, which at the time were usually above converted ECF grades for players in the 1800 to 2200 band.

[Brian Valentine]

The ECF is aware of the FIDE change and active in looking at the implications.

Two days before FIDE published I circulated draft recommendations for updating advice on ECF/FIDE conversion. That's now torn up!

The FIDE proposals address weaknesses in their approach for "recreational players". These are in areas that ECF addressed in developing its monthly system. With the covid effect it is very difficult to see how well the ECF system is working, but there is no evidence to date that things are deficient.

There are no plans at this stage to re-calibrate ECF ratings but the conversion will clearly change, if these proposals go through. As a veteran of the 2009 recalibration debate at ECF, I'm not going to change player's long term measures of improvement without good reason.

Drafted before Roger posted above.

[Mick Norris]

Thanks Brian, I think we'd all be happy if the ECF didn't have to change the rating system (although P ratings were given a good kicking at the recent MCF AGM)

Roger's comments about the current balance between ECF and FIDE ratings are interesting

[Pete Morriss]

There is a strange omission in this document. Its main claim is that we can see that the rating system is not working properly because lower-rated players have been doing better against higher-rated players than would be expected, and that the cause of this is that the initial ratings of new players are too low. This effect, the document shows, has become far more pronounced since the period 2008-12.

What is not mentioned is that after 2008 FIDE relaxed its requirements for getting an initial rating from having played nine games against rated opponents to have played only five. (FIDE also now allows tournaments in which a player lost all their games to be included in their initial calculation, whilst before they were excluded.) It would seem at least worth considering the possibility that these changes were a significant cause of any problems we now have, and that we could ensure that the initial ratings of new players would be more accurate if they were calculated over more games than at present. That might be better than the drastic, and quite arbitrary, changes proposed.

[Roger de Coverly]

What is not mentioned is that after 2008 FIDE relaxed its requirements for getting an initial rating from having played nine games against rated opponents to have played only five. (FIDE also now allows tournaments in which a player lost all their games to be included in their initial calculation, whilst before they were excluded.) It would seem at least worth considering the possibility that these changes were a significant cause of any problems we now have, and that we could ensure that the initial ratings of new players would be more accurate if they were calculated over more games than at present.

It could be investigated of course, but I would doubt that taking the first nine games as opposed to the first five makes any real difference, The main problem is that without a method of treating much improved players as new players or calculation trickery to achieve a similar result, they will be in the system with ratings that are several hundred points below their current level of performance.

[Ian Thompson]

What is not mentioned is that after 2008 FIDE relaxed its requirements for getting an initial rating from having played nine games against rated opponents to have played only five. (FIDE also now allows tournaments in which a player lost all their games to be included in their initial calculation, whilst before they were excluded.) It would seem at least worth considering the possibility that these changes were a significant cause of any problems we now have, and that we could ensure that the initial ratings of new players would be more accurate if they were calculated over more games than at present.

Wouldn't that change be inflationary, not deflationary? You'd expect the number of players getting an initial rating lower than their true strength to be about the same as the number getting an initial rating higher than their true strength, so it would average out, except that some players getting an initial rating lower than their true strength wouldn't get a rating at all because their performance was below the 1000 points minimum rating meaning that, on average, new players are rated higher than their true strength.

The change that caught my eye, and wasn't justified unless I missed it, was that a new player can't have an initial rating greater than 2200. Why not?

[Pete Morriss]

What is not mentioned is that after 2008 FIDE relaxed its requirements for getting an initial rating from having played nine games against rated opponents to have played only five. (FIDE also now allows tournaments in which a player lost all their games to be included in their initial calculation, whilst before they were excluded.) It would seem at least worth considering the possibility that these changes were a significant cause of any problems we now have, and that we could ensure that the initial ratings of new players would be more accurate if they were calculated over more games than at present.

Wouldn't that change be inflationary, not deflationary? You'd expect the number of players getting an initial rating lower than their true strength to be about the same as the number getting an initial rating higher than their true strength, so it would average out, except that some players getting an initial rating lower than their true strength wouldn't get a rating at all because their performance was below the 1000 points minimum rating meaning that, on average, new players are rated higher than their true strength.

If the new players are improving, as seems to be assumed, the more games necessary to get an initial rating then the higher the initial rating would be: to get nine games against rated players used to require two or three tournaments, and sometimes more, usually spread over at least a year. The real challenge, which the paper doesn't really address, is that this phenomenon has got much more marked since 2008, so what changed then? It mentions that the rating floor was reduced from 1200 to 1000 (surely a minor change) but not the change in awarding an initial rating. There may be other more significant unmentioned changes?

[Roger de Coverly]

It mentions that the rating floor was reduced from 1200 to 1000 (surely a minor change) but not the change in awarding an initial rating. There may be other more significant unmentioned changes?

Going from 1200 to 1000 isn't that minor beacuse it increases the number of improving players coming in at very low ratings. What did the real damage though was moving down from 1800 or so without taking note of the various fudges and tricks national Elo systems had needed to use to deal with the problem of rapidly improving players.
For example
https://www.chessscotland.com/grading/c ... ion-steps/

Juniors and New Adults

Junior players and newly graded adults are assumed to improve their playing strength during the season, whereas established players are assumed to have stable grades. To compensate the opponents of juniors and newly graded adults, a range of bonus points are added to the junior’s grade before expected scores are calculated . For example, if you play a J13 with a published grade of 1200, for your grading calculation the junior is considered to be graded 1200 + 110 = 1310.

200 Up Rule

Fast improving players can quickly find that their published grade is out of date and understates their current strength. Opponents of the improving player are victimised because expected scores are calculated against a published grade which has proved to be inaccurate. To quickly boost improving players to a more appropriate level and avoid their opponents unjustly losing grading credit the 200 up rule was invented.

(Modification July 2001) If a grade goes up by 200 points on the first of the (Nov 2006) final grading runs it will be set to zero and recalculated as if the player was an unrated player . The temporary ratings thus calculated will replace the old published grade. Grades will then be recalculated and opponents will get credit for the temporary rating rather than the old published grade. (The junior addition is deducted from the temporary grade before it is slotted into the 200 up player’s record – since opponents will get the temp rating plus any addition when the grades are recalculated.)

[Mike Gunn]

Under the Richard Clarke BCF system you had to play 10 games before you got any sort of grade/ rating (subsequently reduced to 9 to be in line with FIDE Elo). There is a sound statistical basis for this and my understanding is that the rationale for ECF now publishing ratings for fewer games is that "new players are desperate to get a rating and don't want to wait months to get one". In my opinion the the publication of the P ratings leads to confusion amongst players (and many organisers). If you are going to publish these ratings then (at least) give a likely range or confidence limits so people can appreciate what they really signify.

[Mick Norris]

Yes, it would be good if the P ratings were dropped; then new players would be encouraged to play enough games to get a proper rating, and might play more to do so

[Brian Valentine]

Under the Richard Clarke BCF system you had to play 10 games before you got any sort of grade/ rating (subsequently reduced to 9 to be in line with FIDE Elo). There is a sound statistical basis for this and my understanding is that the rationale for ECF now publishing ratings for fewer games is that "new players are desperate to get a rating and don't want to wait months to get one". In my opinion the the publication of the P ratings leads to confusion amongst players (and many organisers). If you are going to publish these ratings then (at least) give a likely range or confidence limits so people can appreciate what they really signify.

I'm pretty sure the rating team would get a good kicking from a different group if p-ratings were not published.

My rule of thumb is that a new player with just one result (against a K-rated player) has a P-rating +/- 200 (so could believe +/-400), falling to =/-70 after 9 results. For those reaching a stable level it falls to +/-30 (maybe 20 if things are working well). These can be materially higher if the rating is obtained by playing a high proportion of P-rated players.

To Ian's query about the 2200 cut off, I presume that it is to do with titles. Given some qualifications start below 2200, I'm surprised it's set so high.

[Roger de Coverly]

I'm pretty sure the rating team would get a good kicking from a different group if p-ratings were not published.

Congress organisers were those pushing for monthly ratings and presumably also prefer to have something to base seedings and eligibilities on. FIDE's official approach to unrated players is to give them a rating of zero and rank by alphabetic order.

League rules have usually needed estimates for ungraded players, both for "boards in order of strength" and more specific exclusion or handicap rules. Now there's an official estimate in terms of p-ratings.