7

I noticed that my USCF quick rating is lower than my USCF regular rating. At first I thought maybe it was because I was relatively weaker at faster time controls, but it seemed like the difference was too large to be just that. This got me to looking at the ratings of everyone in my section (It was the open section, there was another section of sub-1200 and unrated) of my most recent dual-rated tournament. On average, the pre-event quick ratings of the 50-ish players in this tournament were 170 points below their regular ratings (and that even includes one player who never played quick before and was therefore given a provisional quick rating equal to his regular rating.)

It occurred to me that maybe I'm just in an area with local weirdness in the ratings, but that doesn't seem to be the case. Wesley So has the highest regular rating in the nation, but his quick rating is 166 points lower than his regular. The 100th best regular rating is 2474, but the 100th best quick rating is 2263, which is 211 points lower.

Although I know the player pools of the two rating systems are not identical, it would seem that the pools have significant overlap - all are USCF members who play in tournaments. Initial quick ratings are literally based on the regular rating, if the player was already regular-rated. And both rating systems appear to use the same algorithm to calculate ratings.

But despite that, quick ratings are significantly lower, apparently by nearly a full class. Why are they so much lower, given the same algorithm and many of the same players? And, assuming it's desirable that the ratings be similar, what could be done to correct this situation?

3
  • This trend is real and is born out by statistics: uschess.org/archive/ratings/ratedist.php. Percentiles for rating x are consistently higher for quick, which indicates a lower average rating. Nov 29, 2020 at 14:50
  • Now that's interesting. There's one place where that flips, which is scholastic ratings for those rated under 1000 - for them, it seems like the quick ratings were higher.
    – D M
    Nov 29, 2020 at 14:54
  • This is interesting (and I wasn't looking at scholastic stuff for my original comment). Perhaps this is because kids play more quick? Maybe There are pockets of scholastic tournaments which only play G25, and few kids get outside that bubble? In any case kids rarely regress to their rating floor, so your answer still holds water. Nov 29, 2020 at 15:02

4 Answers 4

7

This may be partially due to ratings floors.

When a player hits his rating floor, his rating cannot go lower if he loses, but his opponent's rating can still go higher. This adds ratings points into the system.

There is an absolute rating floor of 100 which nobody can go below. (For every event you play in and quarter-point you win over the board, this is raised by 1, to a maximum of 150.) Rating floors exist at every 100 points from 1200 to 2100, and these can be achieved by getting a rating at least 200 points above the floor. There's a rating floor at 2200 which can be earned by getting the Original Life Master title. And finally, if a player earns $4000 or more in an under-2000 section, their rating floor is set high enough so they become ineligible to win the prize again.

Regular and quick ratings both have ratings floors, but they are independent from each other. Most events with very large prizes are probably not quick-rated. And I don't think the Original Life Master title can be earned with a quick rating. This probably means that fewer people are at a floor with their quick rating, and that if they are, the floor is lower than it might otherwise be.

To add rating points to the system, it is insufficient for a player to simply be at the floor - he must actually play games. I would hypothesize that more regular-rated games are played than quick-rated games, especially at the 1200+ level where floors come into play. I would also hypothesize that older players are more likely to be at a floor (because they won a large prize sometime on their life, because they once managed to raise their rating beyond its ordinary level, or because their skills are not what they once were) and that older players in particular play more regular games than quick games.

And if younger people play relatively more quick games, that may be another reason why quick ratings are lower. Younger players are more likely to start with a very low rating but be rapidly improving. When a player's ability improves, his rating lags behind, and his opponents will tend to lose rating points. This tends to have a deflationary effect on ratings.

2

They are not really lower. That depends on the person.

Most lower rated players train the wrong way to be able to play speedier chess well, or at least better than very slow chess games.

GMs can play speed chess quite well and most would not have a lower speed rating than for slow OTB games.

Exacerbating the problem is digital clocks with all their weird add ons delays and similar gimmicks that drastically change the effect of playing faster on different people and their ratings.

OTOH I know a mere NM whose published speed rating is higher than his OTB. So again, it depends a lot more on the person than the ratings.

Also Exacerbating is the lumping of widely different speeds and related ability into one very wide category for the ratings.

With more rating categories the ability vs speed would be better shown although for most players there will be a maximum where they are as good as they can be so slowing down the time control would not raise their ratings more.

2
  • 1
    "GMs can play speed chess quite well and most would not have a lower speed rating than for slow OTB games." - But most do have a lower quick rating than regular. Not all, but most. For some the difference is small (Nakamura has 2836 regular vs 2820 quick) but for many it's larger (Jeffery Xiong has 2787 regular vs 2530 quick.)
    – D M
    Jan 10, 2020 at 0:14
  • Although relative ratings do "depend on the person," this a very real trend; quick ratings are genuinely lower than regular. uschess.org/archive/ratings/ratedist.php Nov 29, 2020 at 14:51
2

The reason is very simple: There are more players playing regular than quick time control.

What the rating algorithm says is that, within a player pool, the relative rating difference between two players should measure the relative strength between them. The weakest player in each pool always has a rating of 100 and every other player's rating represents their strength relative to the weakest player (assuming an ideal world where every player plays a lot). A rating in one player pool is incomparable to the rating in another player pool.

If we assume that each player pool has the same player strength distribution (e.g., a normal distribution), it follows that a player with the same percentile in both player pools gets an overall higher rating in the larger pool. Why? Because the larger the pool is, the more likely that the weakest player in the pool is weaker.

Your argument that regular and rapid pool have significant overlap don't matter. What matters is that they ultimate are very different in size. See the old stats: 65455 regular players vs 37162 quick players.

If you want to know whether you are stronger in regular or rapid, you can look up your percentiles in these two pools and compare them.


Update: To clarify one point: The difference between a player's quick rating and regular rating does not reflect a player's difference in the strength of the two groups of time controls, even though the two ratings are calculated under the same algorithm. This is because each player's rating are always calculated based on the player's strength in that group of time control relative to those of other players in that pool. The algorithm never calibrates the ratings between the two groups of time control.

If this seems hard to understand, try to replace the digits in a quick rating with alphabets (0 -- A, 1 -- B, 2 -- C, etc. e.g., rating 1230 becomes BCDA). Now you run the algorithms to compute a player's regular and quick ratings, you'll see that digits never encounters alphabets. Intuitively, a digital rating isn't comparable to an alphabetical rating. The algorithm never sets a bridge that calibrates the ratings in two different pools.

7
  • Interesting. Do you have the math which would show that a difference of that many players would change the distribution by that much, though?
    – D M
    Feb 23, 2023 at 9:54
  • We don't have the distribution of player strength distribution and can't give an estimate. However, 200 points difference isn't really dramatic. According to the winning rate table, 200 points means the higher-rated player can win the lower-rated player with a probability of roughly 75%. If the weakest quick player also plays regular, a 75% winning rate over the weakest regular player is a reasonable figure.
    – xuhdev
    Feb 24, 2023 at 1:50
  • With the above being said, there are a lot of practical effects: There are in reality many players with 100 rating, even though one can be significantly weaker than another. They together form the lowest rating despite their strength discrepancy. This weakens the assumption of the system, but conceptually the effect of the size of the player's pool should still hold.
    – xuhdev
    Feb 24, 2023 at 2:00
  • 1
    To be clear, improving winning rate against an absolute beginner from 50% to 75% (200 points increase) may be just a matter of 20min careful study of rules. 200 points improvement for a club player is absolutely a lot of work.
    – xuhdev
    Feb 24, 2023 at 7:33
  • I just checked the under 1200 section of that tournament I referenced in my question... the average difference between pre-tournament regular and quick ratings (unrated players excluded, of course) was only 56. And the only two with a difference of over 200 were rated above 1000 regular. If the very lowest players at 100-125 were driving this effect, wouldn't you expect the difference here to be much higher than for the players in the higher-rated section? Or at least similar? I mean, nobody rated 1000+ is going to be getting meaningful rating points for beating a 100 player.
    – D M
    Mar 21, 2023 at 1:33
1

While floors contribute to it, the biggest reason for the difference is the fact that strong players are less likely to play quick rated tournaments as they improve. So their quick rating doesn't increase as fast as their regular. Here's an example scenario:

Tournament 1: Dual Rated. Regular Rating = 1000, Quick Rating = 998
Tournament 2: Regular rated only (longer time control). Regular rating: 1000 => 1100.
Tournament 3: Regular rated only (longer time control). Regular rating: 1100 => 1175.

So at the end of the day, your regular rating is 1175 but your quick remained at 998 . Then when you DO play a dual tournament again, you're quick rating is underrated and brings everyone else down. There are some crazy examples, of GMs like Awonder Liang with only 2100 quick ratings or so.

What USCF should do is automatically boost everyone's quick ratings to minimum 100 points lower than regular, if they want quick ratings to be accurate at all.

1
  • "Then when you DO play a dual tournament again, you're quick rating is underrated and brings everyone else down. " - OK, but if you had played in a bunch of tournaments in the meantime, you would have been bringing them down anyway, just more gradually. Does it matter whether your rating goes from 1000 to 1100 to 1175, or directly from 1000 to 1175? You're stealing 175 points from your opponents either way. In fact, by doing it all at once, you become more likely to get bonus rating points for a performance far above what was expected.
    – D M
    Oct 21, 2022 at 0:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.