FIDE Rating Deflation

[Thomas Rendle]

General pattern seemed to be an increasingly predictably successful raid by English juniors on everyone else's rating points! Personally i think FIDE are going to have to wake up to a potentially serious deflationary problem before very long...

Tell me about it. I drew with Peter Lalic at the weekend - with a FIDE of 1951 and an ECF of 185(A). 1951 roughly converts to 163 and 185 converts to 2130

[IM Jack Rudd]

I can correct the deflation, no problem. All I need is four very strong players, a boatload of unrated players who can hardly play, and one small aubergine.

[Alex Holowczak]

Tell me about it. I drew with Peter Lalic at the weekend - with a FIDE of 1951 and an ECF of 185(A). 1951 roughly converts to 163 and 185 converts to 2130

The vast majority of English juniors seem to have FIDE-ratings which are far too low for their actual ability. I guess there are two potential reasons:
(1) They're entering the rating list too soon, and they're improving quicker than their rating can catch up
(2) They're not playing enough FIDE-rated chess to get an accurate rating

[Alex Holowczak]

I can correct the deflation, no problem. All I need is four very strong players, a boatload of unrated players who can hardly play, and one small aubergine.

At the risk of being guilty of this sort of thing, why the small aubergine?

[IM Jack Rudd]

I can correct the deflation, no problem. All I need is four very strong players, a boatload of unrated players who can hardly play, and one small aubergine.

At the risk of being guilty of this sort of thing, why the small aubergine?

http://en.wikipedia.org/wiki/Goodness_G ... characters

[David Sedgwick]

General pattern seemed to be an increasingly predictably successful raid by English juniors on everyone else's rating points! Personally i think FIDE are going to have to wake up to a potentially serious deflationary problem before very long...

Tell me about it. I drew with Peter Lalic at the weekend - with a FIDE of 1951 and an ECF of 185(A). 1951 roughly converts to 163 and 185 converts to 2130

Thomas, I understand the issue, but it doesn't actually make a lot of difference to you in this particular case. To be precise, one rating point.

As your own rating is 2397, Peter is treated as being 400 points below you and the game cost you 4.2 rating points.

A draw against someone rated 2130 would have cost you 3.2 rating points.

[Thomas Rendle]

General pattern seemed to be an increasingly predictably successful raid by English juniors on everyone else's rating points! Personally i think FIDE are going to have to wake up to a potentially serious deflationary problem before very long...

Tell me about it. I drew with Peter Lalic at the weekend - with a FIDE of 1951 and an ECF of 185(A). 1951 roughly converts to 163 and 185 converts to 2130

Thomas, I understand the issue, but it doesn't actually make a lot of difference to you in this particular case. To be precise, one rating point.

As your own rating is 2397, Peter is treated as being 400 points below you and the game cost you 4.2 rating points.

A draw against someone rated 2130 would have cost you 3.2 rating points.

That's true - I'm not really complaining, a rating point here or there isn't such a big deal, I was just agreeing with Richard really (I know Peter Lalic gained around 30 points so this goes along with the "raid by English juniors")

[Richard Bates]

Tell me about it. I drew with Peter Lalic at the weekend - with a FIDE of 1951 and an ECF of 185(A). 1951 roughly converts to 163 and 185 converts to 2130

Thomas, I understand the issue, but it doesn't actually make a lot of difference to you in this particular case. To be precise, one rating point.

As your own rating is 2397, Peter is treated as being 400 points below you and the game cost you 4.2 rating points.

A draw against someone rated 2130 would have cost you 3.2 rating points.

That's true - I'm not really complaining, a rating point here or there isn't such a big deal, I was just agreeing with Richard really (I know Peter Lalic gained around 30 points so this goes along with the "raid by English juniors")

I actually think that one rating point is quite a lot to be donating every time you play a junior.

None of this is to take away from the performances of the juniors in question - George O'Toole was =1 (and gained i think 45 rating points), Isaac Sanders got 4.5 with several very tenacious performances from dodgy positions against much higher rated players and Ravi Haria also got 4.5 picking up a grading prize (interestingly Southend bases grading prizes on ECF grades) including a draw against me and a win against Dave Ledger.

On the deflationary point, it seems to me that it will impact at far higher levels than the ECF problem - it must be a potential problem for organisers looking to organise FIDE rated weekenders and wanting to attract strong players when they are so rating unfriendly.

[David Sedgwick]

I felt this discussion might be better continued in a new thread.

I actually think that one rating point is quite a lot to be donating every time you play a junior.

A slight exaggeration, surely? And you would spread the pain when you play adults who haven't played as many juniors as you have.

On the deflationary point, it seems to me that it will impact at far higher levels than the ECF problem - it must be a potential problem for organisers looking to organise FIDE rated weekenders and wanting to attract strong players when they are so rating unfriendly.

Why weekenders in particular? Do they contain more juniors?

It's well known that some (not all) English GMs won't play in rated weekenders; they don't want to put their rating on the line in events with fast time limits and when there's no opportunity to prepare for their opponents. (At least I think those are the reasons.)
In contrast such events are popular with players in the 1900 to 2300 range.

There have been anxieties within FIDE for years about rating inflation, so I can't see them worrying if there is now a little deflation. Moreover, I imagine it only affects countries where a relatively small proportion of events are FIDE rated. It's well known that Norwegian juniors are often very underrated. However, the same problem doesn't seem to apply in (say) France or Italy.

If we want to increase the possibilities for English events to be FIDE rated, the change that would do most to facilitate it would be to reduce the minimum length of the session from four hours to three hours. That would, for instance, allow the London League to be rated. To do so wouldn't necessarily be popular, of course - see my comments about GMs above.

This was a change which I tried and failed to get the FIDE Qualification Commission to make at Dresden Congress 2008. Changes will next be made in 2012, so now is the time to make constructive suggestions.

[David Sedgwick]

I've responded to Richard in a new thread - FIDE Rating Deflation.

Carl, please could I suggest moving the last few posts to that thread?

[Roger de Coverly]

Why weekenders in particular? Do they contain more juniors?

Possibly so. You already mentioned on behalf of Hastings that the clash with the London qualifiers is bad for junior attendance at Hastings.

Another point is that if you were to compare an IM playing 9 rounds at Hastings with playing the equivalent in weekend events, that the weekend events would be likely to contain more pairings against under-rated juniors even if there were no more of them in proportion than Hastings. The 2 or more weekend events would contain more "early" rounds where rating mismatches are normal.

At present, the minimum standard for FIDE rating is G/120 or equivalent. Bringing it down to G/105, G/90 or their equivalents would enable many more British events to be internationally rated. It's worth asking, which countries have events which these sorts of move rates facilitate?

On rating deflation, it's still very possible to get an IM or GM title complete with the required rating at a young age. When you look at their rating histories, they usually managed at least 1900 as their starting rating. Are there any examples of current IMs or GMs whose first rating was down in 1600s?

I'm not sure about the theory of pairing a Swiss using rankings if the rankings are wrong. I suppose it reverts back towards the earlier tradition of random pairings where you could get a first round clash of players who might expect to be first and second.

[Alex Holowczak]

Suppose a GM loses to a 2000-rated junior. The GM's loss comes at k = 10, whereas the 2000-rated junior's win will come at k = 15. Since W-We for white will be equal to 1 - (W-We) for black, this means they'll both have numbers of the same magnitude to multiply by k (obviously the winner's will go up; the loser's down). So doesn't this mean that the junior gains more rating points than the GM loses, thus the net effect is inflation, not deflation?

The converse would equally be true; if the GM beats the 2000, he gets the win at k = 10, but the junior loses at k = 15, hence net deflation.

If the junior beats the GM more than vice versa, then the net result is more inflationary results than deflationary ones. As a result, you get more inflation in the rating system.

Is that correct logic?

[Roger de Coverly]

If the junior beats the GM more than vice versa, then the net result is more inflationary results than deflationary ones. As a result, you get more inflation in the rating system.

Is that correct logic?

Variable K's help to some extent.

The problem being referred to is the issue that the 2350 IM is facing not a 1950 player, that they should score almost 100% against, but a 2150 or better player with a 1950 rating. At the very least this distorts the tournament pairings, since the guy on the next board is playing a genuine 1950 adult. The effect over time is probably to dilute (reduce) the ratings of 2350 players.

Elo style systems can respond quickly to changes in strength. Look at http://ratings.fide.com/individual_calc ... 2011-05-01

You have to play a lot of games to get that sort of effect though.

[Alex Holowczak]

If the junior beats the GM more than vice versa, then the net result is more inflationary results than deflationary ones. As a result, you get more inflation in the rating system.

Is that correct logic?

Variable K's help to some extent.

The problem being referred to is the issue that the 2350 IM is facing not a 1950 player, that they should score almost 100% against, but a 2150 or better player with a 1950 rating. At the very least this distorts the tournament pairings, since the guy on the next board is playing a genuine 1950 adult. The effect over time is probably to dilute (reduce) the ratings of 2350 players.

Elo style systems can respond quickly to changes in strength. Look at http://ratings.fide.com/individual_calc ... 2011-05-01

You have to play a lot of games to get that sort of effect though.

I'm just wondering if inflation/deflation is actually caused by the variable ks, on the basis that with varying ks, more points are going to enter or leave the system than if they had the same k.

The other thought I had was that if the ratings of chessplayers are normally distributed (a brave assumption), then by increasing the number of players in the system (the effect of lowering the floor), it makes sense to have more extreme results at the ends; i.e. more 2800+s. It may not be a case of inflation or deflation at all.

[Roger de Coverly]

The other thought I had was that if the ratings of chessplayers are normally distributed (a brave assumption), then by increasing the number of players in the system (the effect of lowering the floor), it makes sense to have more extreme results at the ends; i.e. more 2800+s. It may not be a case of inflation or deflation at all.

If you try plotting the distribution of ECF grades, then the graphical shape is normalish. Published grades are now distorted by adding age based supplements to juniors and the treat juniors as new players rule.

The top of the scale is likely to be Normal, because these sorts of skill related things usually are. It's less obvious why the lower should be, a plausible explanation is that as you are measuring performance in graded events, there's a certain amount of self-selection. So those who aren't any good don't play.

By contrast, without ever having tried to plot it, the international list would be more of a truncated Normal. So the same shape as a national list down to about 2000-2100. This is because you have approaching 100% of players of that standard on the international list. Below that I would expect the coverage to be patchy. So really the international ratings are just the top 5-10% of the underlying distribution of players taking part in competitive chess. This varies by country as well.

So I don't think that increasing the number of rated players creates more 2800s. What does though, is making subtle changes to the rating rules for players of 2000 and above. The biggest change was back in the early nineties when the ratings were extended down to 2000. At a stroke ,a lot more players could aspire to international ratings. Just look at how many 4NCL teams are populated by those in the rating range 2000 - 2250. The extensions to 1800 and lower make it less likely that a player with a 2250 rating just got there by a "lucky" first tournament or two. So I think you can get simultaneous inflation and deflation depending on how the rules interact with the player pools.

Also don't forget that the top players are probably getting better in practical terms at least. The rest of us have to improve as well to maintain our relative distance.