Does Glicko Rating System Work Well With CC?

But, the Glicko System is assuming you have started and finished and played in only one game in that week and therefore that your chess skils are getting rusty.  But, while playing CC chess, you have more games going simultaneously and the result is that you are using your chess skills continuously even though, perhaps, only one game concluded in that week.  Do you think both of these scenarios are equal?  The Glicko System as is applied on chess.com is saying that they are!

Yes, I do think it is a good system for CC.

Like you, I don't play too many games simultaneously and have a fairly high RD. My rating is very inflated currently and I have started playing people of a similar rating, so I fully expect it to take a serious dip in the next few months. That would be fair. If I win half and lose half of the games against the similarly rated players then my rating will oscillate for a while (possibly quite dramatically), but will end up around the same as I am now, which would also be fair.

I play a lot of lower rated players in tournaments. If I lose to one of these guys my rating will dip by 60 points. Fair enough. If I lose to these guys then my rating ought to take a serious beating.

In summary, my rating is not reliable because I don't play enough games, so it will possibly change quite dramatically. This is appropriate. If someone plays a lot then their rating is more reliable, so it shouldn't change by much. If one chooses to play few games so as to concentrate more on individual games then their rating might be better than someone who plays a lot of games simultaneously, but is still statistically less reliable, so I think it is the fairest possible system.

Fair engouh.  I can respect your opinion.  But, I don't think one should have to play 30-50 simultaneous games in order to prevent wild fluctations in rating.  I mean, I play chess every single day.  No, I don't complete games every single day.  But, I am quite sure that my chess skills are not getting worse just because I do not complete a game every single day.  The Glicko system as is applied here on chess.com is penalizing me for not completing games every single day!  It is asserting that I am, in fact, getting worse at chess even though I'm playing the game every day!

 Key Point: Your rating is not your actual rating.  What do I mean by that?  Well, Glicko (and other ratings) are mathematical representations of what your true rating is.  No one really knows what their true rating is, since 1. it changes over time, and 2. it requires a large number of games to accurately determine.  The best we can do is approximate your actual rating by using a formula.  The main reason that Glicko is used is because it actually suffers less from errors than other systems.

Now, on Chess.com, does 1500, 1700, etc. mean you are a bad player?  NO, nothing could be farther from the truth!  Ratings are a product of the population they exist in, and they mean very little outside that population.  To say that, "correspondance Chess shouldn't use Glicko", is a very meaningless viewpoint, since the number doesn't translate at all towards what that person's USCF or FIDE number is.  The best you'll be able to say is that of two players, rated 1500 and 1700 on this site, the 1700 is most likely to have a better game.  Anything other than that is ignorance.

The system here works correctly, and if it doesn't "jibe well with ELO on other sites" (by the way, Elo is a name, not an acronym), then the problem is either that the competition is better/worse on the other site, or their system is flawed (typical if it is based on rewards rather than performance).

 You didn't touch on the main idea of this thread.  You just talked about how a rating is never a true rating and that the ratings here are valid only within our population.  Both are true statements.  But, should the Glicko System for CC as applied to chess.com take into account the fact that not everyone can play 30-50 simultaneous games which provide enough finished games per day to get an RD closer to 30 than the much more common 100+ ?

My advice to the creator of this topic is play people in incriments.  Play people 100 pts. plus or minus from your rating.  When you consistently beat this group, your rating will improve.  My current rating is close to 1700, with my average opponent a shade over 1400.  Also, moderate your opponents ability.  Play six games around your same rating and play a few 200 pts below and above.  This will help alleviate the fluctuations of the ratings system.  Always plan on a few you can win, and others you expect to challenge, mixed with some in your range.  It has really helped me improve. 25% below, 50% in proxi to your rating, 25% above.   Hope this helps. (It may not make sense, but I tried)

 Thanks for that info.  I am not, though, concerned with my rating, since I am overrated!  I just want chess.com to use the very best rating system available and I believe a modified Glicko would be more appropriate for CC.  Others disagree with me and they may very well be right, but so far I think it could be improved!

I think you are misunderstanding how it works.  It doesn't think you are getting "rusty" or "worse at chess" from not playing.  Quite the opposite, in fact: it makes it easier for you to gain rating points if you have improved at chess.

 Basically, what I think he's saying is... take a break. =)

RoundTower wrote:

 It does indeed.  Each day that goes by without a finished game increases your RD.  This by definition means that the Glicko system thinks your rating has become more uncertain. 

you are contradicting yourself.  No, it does not think you are getting "rusty" or "worse at chess".  It thinks that it is now less certain of what your rating should be.  Which makes sense.

It doesn't judge that you've got worse (or better).  It merely acknowledges that there is a decent chance you could have got better (or worse!) since last it checked.

I think you should read carefully my posts and the webpages where the details of glicko are explained.  It may well be possible that it could be improved, but it's not very productive for you to suggest ways to change it when you don't really understand how it works.

Good point.  I understand it just fine, just mistyped.  It certainly can, and I believe should be improved, though.

First, I would say that glicko is the correct system. I've done a fair bit of looking around at rating systems, and I understand the statistics involved, and glicko is the best I've seen so far. Now, that defines "best" in terms of most accurate reported rating. That may not be what you're looking for in a rating. It sounds like you want a more precise rating. Not the same thing.

Second, the problem may not be in the system, but in the constants. Some have suggested a switch to elo, but elo is actually glicko, just with different values for the constants than are normally used for glicko. I have not seen anywhere on this site exactly which values are used for the constants here.

Third, I'm not seeing the problem. I'm not playing very many games, and my past few ratings adjustments have been a handful of points each. I'm looking at a potentially significant shift in the near future, but I'm about to beat someone rated 300 points higher than me. Maybe it's a perspective issue more than a system issue.

Being somewhat of a math dude, even though I am not familiar with the algorithm

of the Glicko and not even altogether familiar with the Elo, it can be said that an

algorithm is fair and accurate if it does not discriminate between players and the players are familiar with the rules of the rating game itself. Ratings are only methods of matching players, that is all that they are for. A rating does not mean anything except as a comparison with a particular player with an average over a group of players and of games.The truth abut players is about the matches actually played and who won and lost. But math does not know to whom a particular number or game result belongs to. That is the beauty of math, as Bertrand Russell said.  Math is only a comparison of data so that a potzer does not unwillingly or unknowingly play a grand master and vice versa.

The constants are set up here such that most players have an RD near or above 100.  If that is how chess.com wants it to be, then there is no problem.

How do you "want" a particular value of a constant - it's not like a flavour of ice cream.

What, surely, we all want is the value of constants that gives the best rating system. If RDs go too high, you get too much fluctuation around your true strength. If they go too low, there is not enough movement to reflect a genuine improvment. So there is a delicate balance to strike.

My impression is that RDs are too high and there is too much fluctuation. But it ought to be possible to do some reasonably advanced diagnostics on how well the ratings predict the results of games, to fine tune the constants.

Yes, and since the average RD on this site 130 (see the thread about inflated ratings for the link to chess.com page that shows the average players' stats) and achieving a low RD requires that you play hundreds of simulataneous games, high fluctuation is something we must all deal with on this site.  The admins of this site have so far been unwilling to have a debate on this subject.  In the thread about inflated ratings a staff member commented once or twice and then abruptly dropped out of the conversation as soon as my link showing the average RD of chess.com players appeared.

Strange. To go to all the effort of using a Rolls Royce system like Glicko, and not tweaking the constants to get the best results. It's like using a Ferrari to pull a caravan. 

 Is it ridiculous? In order to get that big a shift, he would have to beat someone over 400 points higher than he is (and someone with a low RD). Isn't that evidence that in the past year he has learned enough to improve his rating by that much? If he has learned that much, he deserves the rating boost. If he hasn't learned that much, he'll lose the ratings boost when he loses the next couple games against similarly rated opponents.

 Your RD is 109..

The lowest i'v seen on this site is 46 so far..

I agree though.. Ouachita is 150 RD.. thats ridiculous.. no wonder hes so high in rating.. (but then he can drop REALLY low... even a draw insures him to drop over 70 rating)

They should lower the RD for every1.. so new comers RD's a bit lower then it should be and every1 elses is lower then it should be etc..