Junior Grades

[Paul Cooksey]

If the expectation is indeed reality, then if a junior has a higher k than an adult, then while yes, his rating will increase more quickly, you will also have the side-effect of creating an inflation within the system.

I think rating inflation is usually defined as a higher rating for a given playing strength, not an increase in overall mean rating. Almost all players enter the pool below the mean in any case.

[Alex Holowczak]

If the expectation is indeed reality, then if a junior has a higher k than an adult, then while yes, his rating will increase more quickly, you will also have the side-effect of creating an inflation within the system.

I think rating inflation is usually defined as a higher rating for a given playing strength, not an increase in overall mean rating. Almost all players enter the pool below the mean in any case.

Right. But as those improving juniors become adults, and then they play the people at the top of the list, this inflation will be carried through the system. So the people at the top will be higher than would otherwise be the case.

For example. Yang-Fan Zhou had these results in one rating period last year: http://ratings.fide.com/individual_calc ... 2011-05-01

With k = 15, he gained (as near as makes no difference) 100 points. If k = 30, which is what it is now for players in their first 30 games, he'd have gained 200. So all of a sudden, his rating would be 2522. Should he then go up against another 2522-rated player, the expectation of his opponent would go down from 0.64 to 0.5. This is a pretty dramatic change, and is bound to have a knock-on effect to England's other top players. For example, if Short/Adams etc. beat him a few months later at Sheffield, their rating would go up by a lot less by winning against him than would be the case if Yang-Fan's k was only 15 in the previous time period. This is deflationary, but according to the original hypothesis - that Yang-Fan is more likely to win against Adams/Short than vice versa, because he's an improving junior - then again Yang-Fan's rating will rise by more than Adams/Short go down, and thus, you have yet more inflation in the system. That's the theory, anyway.

Is there any reason why k shouldn't always be the same? To hark back to my Test cricket experiment, I kept k = 25 throughout, and it seemed to cause no problem. If you're worried about rapidly improving people when players first get on the list, why not make it harder for them to get on the list in the first place? That way, you can ensure they have a realistic rating when they get there, rather than letting them onto the list and then fudging the system so that it makes itself look realistic.

[Paul Cooksey]

I'll steal a joke from chessbase when they were reporting on attempts to create new rating systems. They joked that any rating system that could successfully predict Shirov's record against Kasparov, was probably flawed in many other ways. I'm going to say the same about that burst from Yang-Fan, and observe that his rating continued to go up, and that a high k corrects overshooting faster than a low k.

A high starting level does mitigate this issue. But with the FIDE rating floor getting lower, the problem of lag is likely to get worse.

[Alex Holowczak]

But with the FIDE rating floor getting lower, the problem of lag is likely to get worse.

No disagreement on that one!

I tried various levels of k in my experiment, and found 20 was too slow, and 25 was about right. (The WI clung on to their #1 status for far longer in the 1990s than they reasonably should have done at k=20.) It's all subjective really. If Aronian was 2810 and Kramnik 2790, and then two months later they were the other way around, would that necessarily say anything about their ability relative to one another? I would say that it wouldn't.

[IM Jack Rudd]

It could be argued that, if a junior is improving, (1) is more likely than (2), and (3a) is more likely than (3b). I've no statistics that show that is the case, but that's what I would expect. If the expectation is indeed reality, then if a junior has a higher k than an adult, then while yes, his rating will increase more quickly, you will also have the side-effect of creating an inflation within the system. This isn't a disaster - the differences are the key, not the magnitude of the ratings - but I would imagine that it's non-desirable.

I would argue that it is desirable; what you are likely to have is a situation in which the juniors are improving while the adults are staying much the same. In such a scenario, you actually want inflation to happen because that's what accurately describes the reality of the system - the players under consideration are, as a group, improving.

[Roger de Coverly]

I would argue that it is desirable; what you are likely to have is a situation in which the juniors are improving while the adults are staying much the same. In such a scenario, you actually want inflation to happen because that's what accurately describes the reality of the system - the players under consideration are, as a group, improving.

One of the desirable features of a rating system is that it facilitates comparisons over time. So if you have more "strong" players, you want the rating system to be able to reflect that. If you believe in the theory that Keith is as good as Capablanca, the rating system may have failed to revalue enough. Anyone want to claim that a contemporary GM below the top 100 is as good or better than the 1970 to 1972 Fischer?

[E Michael White]

There is a special case where a pocket of ungraded players - they need not be juniors - have played entirely among themselves. But that means real ungraded players. No grades at all, not even calculated "start-of-season-grades". Since these players have no contact with the outside graded world, grading their games is impossible and not attempted. Even so, they don’t get grades of 0. They just don’t get grades.

There is a bigger issue when two or more groups of ungraded players play within the group and groups have one or more members, who play different graded player(s) from the outside world but otherwise the groups are unconnected.

This would be much more dificult to spot and more likely, in fact probably quite common (for group read junior chess club). As a result the groups would likely become out of line, shades of NCCU v SCCU in the 1960s. There would be two sources of error:-

1. the ungraded in one group perform better against the graded than their compatriots in the other group
2. the grades of the graded opponents for the two groups are temporarily out of line.