FIDE Rating Consultation

[Roger de Coverly]

Did you read Sean Coffey’s quote from Elo about the need to resist changes?

He may have said that, but he didn't follow it.

For example he approved the addition of 100 points to all female players except Susan Polgar. That was based on thin evidence and completely ignored the point that in some countries, notably the UK, there were very few female only events. English players of that era mostly earned their ratings in mixed tournaments such as the Lloyds Bank, Hastings Challengers, British Championship and Major Open. Something similar may well have applied to Western European players as well.

Still when the male minimum rating was reduced to 2000, it could be a useful boost to the new FIDE ratings of male 160s or 170s players to have a few opponents with artificially high ratings.

Elo based systems had been extended down to beginner level in national systems long before FIDE did the same. The need to have some measures in place to handle rapidly improving, mostly junior, players was ignored by FIDE apart from the hit or miss k=40 rule.

[Roger de Coverly]

The process may be compared to using a meter stick waving in the wind to measure the position of a cork bobbing on the surface of waving water. The exact position of the cork cannot be stated, but one can give the probable range in which it may be found. The same can be said of ratings.

The problem that FIDE are now rather belatedly trying to address is that whilst a published rating may indicate a strength of between 1000 and 1200, the actual playing strength is between 1600 and 1800. It's not so much the exact position of the cork, but which pond it's in.

[NickFaulks]

Did you read Sean Coffey’s quote from Elo about the need to resist changes?

I believe I read everything he ever published on rating systems, so do not need to be directed to carefully selected snippets.

He was a physicist, so would certainly have recognised that freezing a theory in time, even as a mark of honour to the chap who invented it, is not how science should be done. One point which does resonate, however, is the malign effect of political interference in technical matters. This has been seen in the past and is being seen today.

Elo devised and developed his system to rank the top few thousand players, because that was the extent of the data then available. I cannot believe that he would have opposed extending it down the scale when that became possible although, having no background in statistics, he might not have appreciated the difficulties that would entail.

[Brian Valentine]

Also from Elo:

"Pressures to revise rating processes, to delay declines and to accelerate rises, are natural and never-ending, often arising in high places, often very well intentioned.

Specific proposals abound, to raise ratings through processes other than improved chess play. Unsophisticated proposals such as bribery are rare, but new regulations, say to base ratings on fewer games, or to inflate the bonus points, without regard for probabilistic considerations, can produce undesirable results. Subordination of the rating system to political purposes is ultimately counter productive, leading to vitiation of the integrity of the system and consequently loss of confidence in it." (2nd edition, section 3.12.)

Hmmmm ...

Sean conveniently leaves out the final paragraph in his extract:" Ratings which do not objectively reflect playing abilities inevitably become ineffective for any other purposes as well".

If a system designed for elite players is adapted fo another purpose, rating for recreational chess, then this quote invites modifications.

I think Ian has advocated returning to a Clarke SCALE. The ECF calculations are at least as difficult as FIDE's (speaking as one who has coded both). Also we should not forget that the ECF felt it had to do a compression exercise long before the FIDE investigation.

[Roger de Coverly]

Also we should not forget that the ECF felt it had to do a compression exercise long before the FIDE investigation.

What do we think the ECF problem was? The claim that the BCF/ECF grades had deflated by 1 point a year was nonsence. Had it been so, players who had been 170 BCF and above in the early 1970s would have been 130 or 140s by the 2000s.

I would point the finger at the practice of estimating new players purely by reference to existing players. This came in with centralised grading in the late 1990s and shortly afterwards the first negative grades started to appear. When the initial estimate for a new player was done by hand, the local graders either officially or unofficially would apply a default mimimum. For adults and teenagers I believe it was 100 BCF but lower for those younger.

A Sonas style solution ( adding 400 Elo points to the base equivalent to around 50 ECF) would have been a lot less trouble, So add 50 points to 0 grades and 0 to 175 (or wherever) grades imposing a minimum new grade of 50 Whether treating juniors as new players every year was really necessary particularly for the established higher graded one is perhaps still debatable.

[Ian Jamieson]

Also from Elo:

"Pressures to revise rating processes, to delay declines and to accelerate rises, are natural and never-ending, often arising in high places, often very well intentioned.

Specific proposals abound, to raise ratings through processes other than improved chess play. Unsophisticated proposals such as bribery are rare, but new regulations, say to base ratings on fewer games, or to inflate the bonus points, without regard for probabilistic considerations, can produce undesirable results. Subordination of the rating system to political purposes is ultimately counter productive, leading to vitiation of the integrity of the system and consequently loss of confidence in it." (2nd edition, section 3.12.)

Hmmmm ...

Sean conveniently leaves out the final paragraph in his extract:" Ratings which do not objectively reflect playing abilities inevitably become ineffective for any other purposes as well".

If a system designed for elite players is adapted fo another purpose, rating for recreational chess, then this quote invites modifications.

I think Ian has advocated returning to a Clarke SCALE. The ECF calculations are at least as difficult as FIDE's (speaking as one who has coded both). Also we should not forget that the ECF felt it had to do a compression exercise long before the FIDE investigation.

The SCALE is certainly the main problem. The old 1A, 1B, 2A are superior in my opinion to the BCF’s 3 digit grades which in turn are superior to the ECF / FIDE’s 4 digit ratings. Some of the problems with 4 figure ratings may even disappear under 1A etc because they are second or third order effects.

(Before someone points it out again there would also have to be a change in LOCATION given the strongest players now are stronger than the strongest players when the 1A etc scale was implemented.)

I accept however that the chance of going back to 1A etc is probably nil.

I quite often convert my grade/rating back to 1A etc. and not surprisingly it is far more stable under this scale than 3 figures or 4 figures, perhaps going up a group if I put a lot of work into my chess or going down one or two groups if I am inactive. The only exceptions to this have been when ECF / FIDE have made well-intentioned but perhaps ill considered changes to the grades / ratings.

It still leaves the question however whether you can have one list for the entire range of players or whether you need one list for national / international players and another list for more local players. Under the old 1A etc scheme I understand that the BCF maintained a list down to 3B and the unions maintained separate lists down to 6B

[NickFaulks]

A Sonas style solution ( adding 400 Elo points to the base equivalent to around 50 ECF) would have been a lot less trouble

The problem with that solution is that, unless the root causes are addressed at the same time, after a few years you will be back in the same place.

[Ian Jamieson]

Other scales are also possible.

The old Soviet Union may or may not (different sources disagree) have had a categorisation scheme with each category spanning 200 points e.g. 1800-2000, 2000-2200 etc.

The failure of the rating systems is also demonstrated in my opinion by the number of leagues which permit players whose ratings are within 80 points or similar to play in the opposite order to their ratings because team captains and the leagues recognise that the ratings are not that accurate even if some players and graders/administrators insist they are.

[Roger de Coverly]

I accept however that the chance of going back to 1A etc is probably nil.

History question. The numbers had always been available if not published, being needed for the assignation of 1a, 1b etc. Why was it that the BCF eventually decided to publish the numbers as well as the 1a etc? I might expect it was a perceived need for greater clarity in rankings. Already in the 1960s they had started to experiment with using seedings in Swiss system pairings rather than use random pairings or arbiter's whim. Also if team matches are expected to be ordered it makes it easier to know your rankings.

Nigel Short complained that when FIDE rounded to the nearer 5 Elo points, that organisers would invite by alpahbetic rank and four figures does make ranking easier without arbitrary rules (outside of those inherent to rating systems)

[Roger de Coverly]

The failure of the rating systems is also demonstrated in my opinion by the number of leagues which permit players whose ratings are within 80 points or similar to play in the opposite order to their ratings because team captains and the leagues recognise that the ratings are not that accurate even if some players and graders/administrators insist they are.

A convention of 10 BCF/ECF points of which 80 Elo is more or less equivalent came in with the 4NCL. That was to legalise tactical board order switches, not least to avoid players having two whites or two blacks the same weekend. Older league rules would just say that teams should be in order of strength without being specific about how flexible or otherwise that could be.

[Ian Jamieson]

I accept however that the chance of going back to 1A etc is probably nil.

History question. The numbers had always been available if not published, being needed for the assignation of 1a, 1b etc. Why was it that the BCF eventually decided to publish the numbers as well as the 1a etc? I might expect it was a perceived need for greater clarity in rankings. Already in the 1960s they had started to experiment with using seedings in Swiss system pairings rather than use random pairings or arbiter's whim. Also if team matches are expected to be ordered it makes it easier to know your rankings.

Nigel Short complained that when FIDE rounded to the nearer 5 Elo points, that organisers would invite by alpahbetic rank and four figures does make ranking easier without arbitrary rules (outside of those inherent to rating systems)

I suspect pressure from players (it was before I was born but I have done some reading on the subject and corresponded with one or two people, sadly now dead, around the subject of grades.) and in my opinion the pressure should have been resisted. You can still run Swiss pairings with 1A etc and run team matches.

As far as the fourth figure of ratings are concerned I would argue the 4th figure is at least if not more arbitrary than alphabetic rank or other arbitrary rules.

[NickFaulks]

team captains and the leagues recognise that the ratings are not that accurate even if some players and graders/administrators insist they are.

There are often reasons why captains may not wish to put their players in order of "strength", even if that nebulous concept could be assessed with complete accuracy. Competitions strike a balance between allowing them to do this and stopping them from taking liberties for tactical reasons.

Regarding these graders / administrators you mention, can you give any names? I don't believe I have ever met one. I believe it is generally understood that a rating can at any time only be a best estimate plus a random error.

By the way, the fact that, since data is finite, the random error will always exist is not an argument against making the best estimate as good as possible.

[NickFaulks]

As far as the fourth figure of ratings are concerned I would argue the 4th figure is at least if not more arbitrary than alphabetic rank or other arbitrary rules.

Short's complaint was not that alphabetical order was arbitrary but that it was biased - S coming late in the alphabet!

[Roger de Coverly]

You can still run Swiss pairings with 1A etc and run team matches.

How would you do that, assuming you are using deterministic rules for pairings and board orders which rely on being able to rank players one by one?

[IM Jack Rudd]

I remember encountering an ungraded player whose surname began with W who complained at some tournament that this was the third event he'd played in relatively swift succession where he'd been given the round 1 bye. He did have a point.