FIDE Rating Consultation

[NickFaulks]

.

[NickFaulks]

I accept I’m in a minority but in my opinion the balance is already too far towards largely spurious accuracy and we should be simplifying the system not making it more complicated.

I don't think you are in a minority.

A mathematically inclined GM once pointed out the rather obvious fact that, following some changes to the rating system, the k factor needed to be moved from 10 to ( I think ) 11.6 to maintain consistency - he was appalled that FIDE people such as myself were too stupid to grasp that. He was himself incapable of understanding that players simply didn't want k=11.6.

[NickFaulks]

Here is one suggestion. The rating system is, and always has been, a balance between simplicity and accuracy. It might be worth running a more complicated, but probably more accurate, variation of the system, and comparing results.

This was done in immense depth in the Kaggle contest, an amazing resource which in my opinion left little to be said.

https://www.kaggle.com/c/ChessRatings2

[NickFaulks]

I fear Roger's idea of analysing point changes turns out to be a "red herring".

With no disrespect to Roger that is absolutely right, and a point which I had to make with some ferocity during the FIDE analysis. The past three years has been a period of unprecedented turnover and, in demographic terms, the births have come from a radically different population from the deaths. Any analysis in which this is not a prime consideration will be worthless.

[SeanCoffey]

Here is one suggestion. The rating system is, and always has been, a balance between simplicity and accuracy. It might be worth running a more complicated, but probably more accurate, variation of the system, and comparing results.

This was done in immense depth in the Kaggle contest, an amazing resource which in my opinion left little to be said.

https://www.kaggle.com/c/ChessRatings2

I remember this. All very interesting, but in the end none of the Kaggle contest methods were adopted. I did not see any official explanations why not, but the models were considerably more complicated than basic Elo, and the results did not seem so drastically better as to justify switching.

Somehow, ten years later, changes that are quite radical (for lower-rated players) seem to be on the verge of being introduced. The justifications given seem (to me) to be less substantial than the Kaggle contest work.

[NickFaulks]

I remember this. All very interesting, but in the end none of the Kaggle contest methods were adopted. I did not see any official explanations why not, but the models were considerably more complicated than basic Elo, and the results did not seem so drastically better as to justify switching.

I think we did give a clear explanation, which was essentially as you describe. All the same, I was surprised and impressed by the improvements in predictive power offered by "black box" methods.

Somehow, ten years later, changes that are quite radical (for lower-rated players) seem to be on the verge of being introduced. The justifications given seem (to me) to be less substantial than the Kaggle contest work.

The adjustment is radical, but easy to calculate and to understand. Crucially, this would be a one time change and the underlying process will not be affected.

This is not to say that I agree with the full proposals - reinstating the 400 point scam is politically motivated and quite unjustified.

[Ian Jamieson]

I accept I’m in a minority but in my opinion the balance is already too far towards largely spurious accuracy and we should be simplifying the system not making it more complicated.

I don't think you are in a minority.

A mathematically inclined GM once pointed out the rather obvious fact that, following some changes to the rating system, the k factor needed to be moved from 10 to ( I think ) 11.6 to maintain consistency - he was appalled that FIDE people such as myself were too stupid to grasp that. He was himself incapable of understanding that players simply didn't want k=11.6.

If I’m not in a minority why are there continual changes to the rating system?

I suspect I’d have been with the mathematician but instead of having the changes and a K factor of 11.6, I wouldn’t have bothered with the changes (and kept the k factor at 10)

The only good thing about all the changes is I pay less and less attention to my rating with every change. If I’m playing for a team I leave it to my captain to worry about ratings. I haven’t played many (any?) grading restricted tournaments recently because of Covid but if the changes to the ratings cause too many problems I’ll just stop playing in these tournaments. In my case therefore the changes if anything will reduce the amount of chess I play not increase it.

At some point if there continue to be changes to the ratings there is a real danger of the rating system bringing chess into disrepute. I’m not saying we are there yet but it is a danger if we continue in the direction we are going.

[NickFaulks]

I suspect I’d have been with the mathematician but instead of having the changes and a K factor of 11.6, I wouldn’t have bothered with the changes (and kept the k factor at 10)

The major change in question was increasing the frequency of rating lists. There was undoubted international demand for that.

At some point if there continue to be changes to the ratings there is a real danger of the rating system bringing chess into disrepute. I’m not saying we are there yet but it is a danger if we continue in the direction we are going.

I'm not sure about bringing chess into disrepute, but I don't even want to see the rating system in disrepute.

Organised chess relies upon ratings, not least for titles. Players wish to be confident that if player A has a rating 200 points above player B, then A is very likely to beat B in a ten game match. Following the recent lockdowns, at some parts of the curve, that may not be true, so it is reasonable to consider extraordinary corrective measures. However, you would be right to feel that if in a few years we have to enter another such process to correct the effects of the current proposals, everyone involved will look rather foolish.

[Roger de Coverly]

However, you would be right to feel that if in a few years we have to enter another such process to correct the effects of the current proposals, everyone involved will look rather foolish.

I'm not sure that there's a general acceptance of what the problem is. I would define it as bringing onto the rating list at low ratings players liable to improve considerably without having a reliable mecanism to revalue them towards ratings reflecting their newly acquired strength. I regard the k=40 rule for under 18s as hit or miss given that it penalises poor performances as much as it rewards good ones. Given that this isn't changing, the proposed changes to the international rating system will be relying in the future on bringing new players in at a minimum of 1400 and crediting them with two dummy draws to enable this to be achieved and presumably also to have some players with excessive ratings for others to feed off. That and the one off effect of compressing all players in the 1000 to 2000 range into 1400 to 2000.

In its final form the English Clarke system handled improving junior players by ignoring rating continuity and treating them as new players every season. It seemed to have worked, at least in comparison to the FIDE ratings of the same players.

[NickFaulks]

I'm not sure that there's a general acceptance of what the problem is.

No, we all know what the problem is - the ratings are wrong. There is no agreement on how it happened ( although I think the prime causes are self-evident ) and still less on the most appropriate solution ( if you think it won't sort itself out with enough results reported ).

[SeanCoffey]

...
However, you would be right to feel that if in a few years we have to enter another such process to correct the effects of the current proposals, everyone involved will look rather foolish.

That outcome doesn't seem all that unlikely.

FIDE would probably be better off deferring the changes for a year, while considering the implications further. The proposed changes might be all right. Then again, they might not.

What if the changes result in significant inflation a few years down the line, and predominantly at higher levels? Perhaps there might be general inflation that propagates through the system, mitigated at very low levels by newly arriving players. If we then redo the plot on page 6 in this document, wouldn't we get a similar curve? Once again, high-rated players would underperform in practice against lower-rated players, compared to the Elo model, but this time because the former were over-rated rather than the latter being under-rated.

Are there any plans at all to monitor the performance of the system over time in any rigorous way? This all seems a bit ad hoc, rushed, hope for the best.

[NickFaulks]

Are there any plans at all to monitor the performance of the system over time in any rigorous way?

Yes of course, but in context a year is not very long.

[Ian Jamieson]

I suspect I’d have been with the mathematician but instead of having the changes and a K factor of 11.6, I wouldn’t have bothered with the changes (and kept the k factor at 10)

The major change in question was increasing the frequency of rating lists. There was undoubted international demand for that.

You probably won’t be surprised that I was one of the minority who were not demanding an increase in the frequency of lists.

Just because something is being demanded it does not necessarily make it a good idea.

Also a rating system that is suitable for professional players may not be suitable for amateur players and vice versa. Maybe the solution is to have two separate rating systems.

[Kevin Thurlow]

"Also a rating system that is suitable for professional players may not be suitable for amateur players and vice versa. Maybe the solution is to have two separate rating systems."

Why not? Didn't the Clarke system had a national list which started at roughly 190 (I think) and then separate "Union" lists. So I propose a FIDE list starting at 2200, and then a separate list for amateurs going up to 2200, and if you get there you get promoted. So on the amateur list, you could do as much fiddling of ratings as you liked, maybe use different K etc.

[NickFaulks]

So I propose a FIDE list starting at 2200, and then a separate list for amateurs going up to 2200, and if you get there you get promoted.

When players from the two different lists play each other, will the game count?