Quick ratings, a bold idea

Running Chess Tournaments
1

In this thread I’d like everybody to stop talking about whether, why, or by how much quick ratings for most players are lower than their regular ratings, or about what to call this difference (“deflation” or something else), or about whether this difference is a problem. If you want to debate these points, please do so in another thread, such as Eliminate Dual Rating or A Rating Related Question or Re: Quick Ratings or elsewhere. If I see any posts here that violate this request, I will ask the moderators to remove them as off topic.

This thread is about how to better align quick ratings with regular ratings. Having almost everybody’s quick ratings lower than their regular ratings, often by 100 points or more, creates (at the very least) a serious PR problem. Players often display a sour grapes attitude toward their quick ratings, and organizers of quick events frequently use regular ratings instead of quick ratings for eligibility and prize purposes.

I do not want “everybody’s quick rating to be the same as their regular rating”. I want the average to be about the same. There should be about as many players whose quick ratings exceed their regular ratings as there are vice versa.

This goal is best achieved if we view the quick rating system as a satellite system to the regular rating system. In other words, quick ratings should revolve around regular ratings.

My fundamental idea is that, whenever player X enters a quick tournament, his opponents’ regular ratings, rather than their quick ratings, should be used to calculate player X’s new quick rating from his old quick rating:

  • A. Each player’s post-event quick rating should be calculated from his own pre-event quick rating, his event score, and his opponents’ regular ratings. That way, opponents’ stale ratings would not drag down the system, because they would not be used.
  • B1. A player who is new to quick chess should be handled the same way as one who is new to regular chess. In other words, his initial quick rating should be calculated from his event score and his opponents’ regular ratings.
  • B2. Alternatively, a player new to quick chess could have his pre-event quick rating initialized from his own regular rating (perhaps based on 10 games), and then his post-event quick rating would be calculated as for a rated player, i.e. from his pre-event quick rating, his event score, and his opponents’ regular ratings.

I’m not sure which of the last two would be better, B1 or B2.

But the above may not be enough. It might also be desirable, whenever an event is quick-rated, to take additional direct action regarding each player’s (possibly stale) pre-event quick rating:

  • C1. Define a player’s pre-event quick rating to be completely stale if the player has played 40 or more regular-rated games since his last quick-rated event.
  • C2. Conversely, define a player’s pre-event quick rating to be completely fresh if the player has played 20 or fewer regular-rated games since his last quick-rated event.
  • C3. In between, define a player’s pre-event quick rating to be partially stale if the player has played between 20 and 40 regular-rated games since his last quick-rated event. Define a player’s staleness factor to be 0% at 20 regular-rated games, 100% at 40 regular-rated games, and varying linearly between 0% and 100% as the number of regular-rated games varies between 20 and 40.
  • C4. At the start of the rating process for each quick-rated event, adjust each player’s pre-event quick rating to be pR + qQ, where R and Q are the player’s regular and quick ratings, p is the player’s staleness factor, and q = 100% - p.

Oh, and:

  • D. No more dual ratings. (It’s not clear whether dual ratings have helped or hurt, but with the above plan they should not be necessary.)

TESTING: Go back to January 1, 2017. (It should not be necessary to go all the way back to 2004.) Zero out all quick ratings, and start from scratch at January 1, 2017. Re-rate all quick events played since then.

Look at the results to see if they meet the desiderata stated in my third paragraph at the top of this post. If not, make (at least some of) the following modifications and try again:

  • E1. Make the adjustment described in C4 only if the player’s (unadjusted) pre-event quick rating is lower than his regular rating, i.e. only if Q < R.
  • E2. Replace all instances of “his opponents’ regular ratings” in A, B1, and B2 with “his opponents’ regular or quick ratings, whichever is higher in each case”.
  • E3. Rating floors should apply to quick ratings too, i.e. any player whose quick rating floor is lower than his regular rating floor should have his quick rating floor raised to make it the same as his regular rating floor.
  • E4. Check the formulas involving bonus thresholds, bonus multipliers, K-factors, multiple passes, etc, to see if there are any differences (stated or indirect) that might be holding quick ratings down, and make any necessary corrections.

If any of these modifications seem to be going too far (i.e. if the quick ratings now appear inflated instead of deflated), cut some of them back and try again.

IMPLEMENTATION: Go back to January 1, 2017, the same date as for testing. Zero out old quick ratings, just as in the testing stage. Quick ratings may not be important enough to warrant going back any further:

  • F1. Any players who acquire quick ratings in this rerate will have these new quick ratings listed on MSA.
  • F2. Any players who do not acquire quick ratings in this manner will continue to have their old quick ratings listed on MSA, but these old ratings will be listed as P/0 (provisional based on 0 games). The P/0 listing will have two purposes: (a) to alert the MSA rating calculation programs that these players should be treated as unrated in the quick system; and (b) to alert tournament organizers that they may wish to use these players’ regular ratings, rather than their expired quick ratings, for pairings and prize purposes. (Of course, organizers have the right to use either rating anyway, but the P/0 listing will give them a little extra guidance.)

As always, testing is important. Test, examine the results, adjust the formulas slightly, test again, etc. Results from such empirical methods can be better than those from adhering to any particular person’s opinions of what caused the problem in the first place.

Bill Smythe

2

My quick tournaments are in May, June and July and are 12-14 games if I can get all three in. In the fall and very late winter I have 11-12 regular-rated games normally scheduled. Add two more tournaments and I’d (rarely) trigger a very slight staleness that isn’t really there, but the concept is for a best approximation rather than exactitude in every instance.
Always using my quick rating only for quick events would be to the ratings-benefit of my opponent in a majority of those quick events. If I played in a few more tournaments in August through April then I’d trigger a stale quick rating and my regular rating would be used to rate quick events, to the detriment of my opponents a majority of the time (well the C1 definition would mean that only the first quick tournament would be considered stale while the remainder of the season would be considered fully fresh).

Quick ratings would only change if there was quick play, so truly stale quick ratings (no quick activity for multiple years) would allow for a disconnect between the quick and regular ratings. Validating the concept should probably compare a player’s quick to regular at the time of the most recent quick activity (not simply using the current ratings).

3

Right now your quick rating is slightly lower than your regular, so that wouldn’t be true at this moment. I know, however, that you have previously told us that your quick rating is often higher than your regular, in which case your opponents would indeed be better off to have your quick rating used in their rating calculations. But with most players, their quick ratings are lower than their regular.

In either case, however, the E1 variation (use whichever is higher to calculate opponents’ new ratings) would take care of the problem. I didn’t want to suggest E1 as the main line, though, because I was afraid quick ratings might actually become inflated as a result. Only testing and time would tell.

Side note: Your playing in a lot of quick-only (i.e. not dual) tournaments has probably sharpened your quick skills, so that the quality of your play suffers less from the faster controls than is the case with most of your opponents. In other words, you are “relatively better” at quick than at regular, which explains your (usually higher) quick rating.

Yup. And according to my proposal, those extremely stale ratings would be listed on MSA as P0 (provisional based on zero games), thus performing the dual purpose of (a) notifying the world of the staleness of that rating, and (b) rendering that staleness harmless to others, should that player ever return to active play.

Bill Smythe

4

If there is no reason for individual ratings to be about the same as regular ratings, there is no reason for the average of those ratings to be about the same as regular ratings, either.

5

Disagree.

When the quick system was started, players’ quick ratings were initialized to their regular ratings. Obviously, the intention was that the quick system should have a look and feel similar to that of the regular system. (I guess you’d call this inductive logic rather than deductive logic – trying to figure out why somebody did something in the first place.)

But, somehow, something went wrong. Quick ratings are now generally lower than regular, by 100 points or so. The “similar” look and feel has evaporated. Players no longer have any respect for the quick system, and organizers are often using regular ratings rather than quick ratings for eligibility, pairing, and prize purposes.

It’s time to stop trying to figure out how it happened. Let’s take the bull by the horns and do whatever is necessary to fix an obviously broken system.

Bill Smythe

6

My Quick rating is lower than my Regular rating because I play badly at fast speeds.

7

But that would mean that, whenever you lose to an opponent at quick that you would normally beat at regular, you are feeding points into the quick system, by way of your opponent. That would not reduce the average rating within the quick system. One player’s rating loss is the opponent’s rating gain.

Some players’ playing strength suffers only slightly when they play fast chess. Other players’ playing strength suffers a lot. When players from these two groups get paired against each other, the first group wins and gains rating points, the second group loses and drops rating points. The average rating still should not change. You could say that players in the first group are “relatively better” at quick than at regular, while those in the second group are “relatively worse”.

I know, “relatively better” sounds like a double comparison, sort of like a double negative. But it’s really not the same thing at all.

In other words, it is a fallacy to assume, just because nobody plays quick as well as they play regular, that everybody’s quick ratings should be lower than their regular ratings. Rating systems do everything in a relative way.

So, something else went wrong somewhere along the way. People have various theories as to where things went south. For my part, I’m becoming less and less interested in such theories. I just want to take aim at the problem head-on, by whatever means necessary.

Bill Smythe

8

Player’s Quick ratings were initialized to their regular ratings at the start because they have to be based on something. What else would you have them initialized to? And it is not a sign that something has gone wrong with the system if the two diverge. Some people, particularly us older folks, don’t play as well at fast time controls. Then there is the fact that a lot of us don’t play much at Quick rated time controls, so our ratings tend to get stale.

People pay more attention to the rating system that they use most, and for most of us that’s the Regular ratings. We also pay more attention to the system where the money is, and that’s definitely the Regular ratings. How many CCA tournaments can you think of that use Quick ratings?

9

Analysis shows that the higher rated player generally tends to do a bit better compared to the expected performance formula at dual rated time controls than at slower time controls. I think that was one of the arguments used in favor of lowering K for masters in dual-rated events.

See

Because the higher rated player is doing better than expected, that would also tend to diminish the number of bonus points injected into the quick system, which would lead to overall lower ratings.

At quick-only or blitz time controls, the higher rated player generally did worse than at slower time controls, and there’s a lot more ‘fuzz’ to the data. (However, over 77% of the quick rated games are ones played at dual-ratable time controls.)

These graphs include all players with established ratings, I may try regenerating them with players with a rating of 1200 or higher.

10

No.

11

Which means that the lower rated player tends to do a bit worse. What comes around goes around. One player’s gain is another player’s loss.

OK, so now the K-factor is lower for the group that does better, and higher for the group that does worse. The gainers gain less, the losers lose more. No wonder the quick rating system has been hemorrhaging rating points.

That too.

OK, OK, so multiple factors have caused the quick system to leak rating points.

But, please, let’s get back on topic, which is what do we do about it? Specifically, what about the suggestions in my original post, especially point A (updating a player’s quick rating using opponents’ pre-event regular ratings instead of their quick ratings) and points C1-C4 (dealing with stale ratings)? Would these ideas work? Are there other ideas that might work better? If so, what are these ideas? Be specific.

Bill Smythe

12

In your initial post you stated that you want the average quick rating to be about the same as the average regular rating. This is clearly important to you. The problem is that there is no reason why this should be so, and in fact there are reasons that have been pointed here out why it easily might not be so. Personally I have no problem with the fact that my regular and quick ratings are not close, and I think most people feel the same way. Again, it comes down to what you value. Most people don’t value their quick ratings as much as their regular ratings, but the reasons for that have little if anything to do with the fact that the two are frequently divergent. If finding ways to align the two ratings more closely is a something for which you want to lead the charge, then go for it. But I think you’re going to find that you’re leading a very small army.

13

The lowered K for high rated players in dual-rated events affect ONLY players rated 2200 or higher, so if a master plays a non-master the master has the lowered K but the non-master’s K is not changed.

So assuming the master wins, the master gains fewer points, but the non-master’s loss remains the same.

14

I don’t understand why we continue to spend effort on non-issues.

Should we do something about Quick stale ratings based on change in Regular Ratings? Possibly. Although with the proliferation of ratings at US Chess we must now ask the question if we should do something for all ratings in s similar way.

But rather than go through complex work for no apparent reasons, why don’t we just make a point of telling people that they are separate rating systems played under separate conditions and players tend to perform differently? Seems truthful, correct, and simpler.

Occam wins.

15

I think staleness is an issue that could be addressed independently of other issues, but I’m not sure there’s any consensus as to how to do it or a lot of interest in doing it.

As a former boss of mine was fond of saying, if something is not worth doing, it is not worth doing well.

16

Ditto.

Also, now that we have several rating systems, if we start talking about alignment between systems, why would it be only one system?

Simpler is the communication outlined above. It is true and applies to all systems.

17

IMHO, staleness is only a problem if there is a major change in the underlying skill level of the player, something we see a lot of, especially in younger players. They figure something out, the light goes on. I’ve seen a player’s strength advance at least two classes in a few weeks.

Suppose someone is a bit better at blitz than at slow chess, but not as good at quick chess. Say that person has stable ratings of 1600 blitz, 1550 regular, and 1500 quick.

Now, let’s say that person studies a LOT but only plays regular (slow) chess. If their regular rating goes up to 2000, would it be reasonable to assume their blitz rating should now be 2050 and their quick rating should now be 1950, even if they haven’t played any quick or blitz rated games lately?

That’s probably more accurate than assuming the player is still a 1600 blitz player and a 1500 quick player.

I’m not sure how the mechanics of cross-rating system adjustments should work, but I think we should think about that possibility.

18

I completely agree. Once this idea is implemented for quick ratings, it should be tested and implemented for blitz ratings also.

As for online, now that regular online is just beginning, the same ideas should be considered there. The question is, should online quick and online blitz be tied more closely to online regular, or should all five “other” systems (OTB quick, OTB blitz, and all three online systems) be tied more closely to OTB regular?

Bill Smythe

19

Come to think of it, we should probably see that they align with FIDE and CXR also. :unamused:

20

I agree, although I might change “only” to “primarily”.

Probably, yes. Or at least, as you said:

Absolutely. And my proposal would accomplish exactly the desired effect. The player would have a pre-event quick rating of 1500 on MSA, but it would be listed as P0 (provisional based on 0 games), alerting the TD, who may (or may not) decide to place him in the Open section instead of the U1600 section. Either way, the MSA rating software would use 2000 as his pre-event quick rating, but his actual quick playing skill would probably be just a bit short of that, so he would lose a few points and end up around 1950. Mission accomplished.

Let’s test this idea and find out.

Bill Smythe