As you play more games, your RD will decrease and your rating will change much less. This is the Glicko rating system and probably the most mathematically accurate.
The rating is an estimate of your playing strength. The fact that it jumps a lot when your RD is high is a result of the estimate being poor. As you play more games and the estimate gets better the fluctuations go away.
). I usually make up for it later, but my rating takes a hit for a while. I think that it should be more often than once a month, though. Maybe once every other week would do. It would take a lot of time for players in the 2000 and up range to increase their rating if it was only adjusted once a month.
I said once a month just as example, it can be once a week or once a 3 month, never mind. Point is in system, not calculating rating after every game. Every other week is just fine.
"playing someone earlier in your chess.com rating affects more then playing that same person later."
This is not true. Playing someone when your RD is high affects your rating more than playing that same person when your RD is low. Sinec your RD increases with time and decreases when you play games, you can't say necessarily that your rating is affected more earlier than later. RD is a measure of how accurate your rating estimate is. I think it's a good thing that your rating changes more when the estimate is less accurate.
"I see no reason that if two guys rated 1500 outplay someone rated 2000, for first rating change +70, and for second rating change +5, just because first has played 2 games only and that was his third, and second has played 100 games and that was his 101st."
Then you will be happy to know that that is not how it works. If the rating change is +70 for one player and +5 for the other, it's because the rating of the confidence in the rating of the first player was much less than the confidence in the rating of the second player -- it's determined by number of games played, it's determined by how good is the rating estimate. That the rating estimate gets better when games are played leads to a correlation between number of games played and rating accuracy, but this doesn't mean there is a causal relationship between number of games played and rating change.
"Restarting system every month is much better in my oppinion. Therefore you get rating stable and still not unchangeable after you play 100 games."
You get the appearance of stable ratings. But the ratings are no more accurate.
Every time we get a new piece of information we can improve the estimate of a players rating. So why not give the improved estimate after each game. Waiting for some number of games or some amount of time before recalculating the ratings does not make the rating any better of an estimate of the players strength, it just makes the rating seem to change less.
ok, let's say that I am holding my arguments for a moment. I would like to know why FIDE is using system similar to one I suggest (with even more mechanisms whose intention is to make rating less flucuative), and why not Glicko? You say (if I got you right) that on long term we get to same with both (or nearly same), so why its not widely accepted?
Second question is what piece of information you get if a player 2000 rated outplays someone 1500-? In Fide system his rating wouldn't change at all. (even if he played 7 games with 7 straight wins, because he is supposed to win)
FIDE faces contraints in their rating system that are not faced here. For example, games are sent in to FIDE to be rated from all around the world and FIDE may wish to rate these games in the order they were played, or they may realize that rating them in order is too difficult so they rate them in short batches. In contrast, at chess.com (or any online server) the result of every game is known instantly as it finishes. So the online server has greater freedom to choose a better rating system.
Also important is that FIDE choose their system of rating a long time ago. To make a change in how they rate games would no doubt require some kind of vote among all the member federations. This comes with discussions and debates, etc. etc. It's not easy for a global organization to just up and change how they do things.
Notice that the most popular online chess sites Internet Chess Club (ICC), Playchess.com, FICS, etc. all update ratings as games are finished. The argument that FIDE doesn't do it this way so neither should chess.com is pretty weak.
The information you get when a 2000 rated player wins against a 1500 rated player is in fact very little and the rating of the 2000 player doesn't change at all. This is the same at FIDE and chess.com, so I don't see your point.
Imagine this scenario. Player A has played 3 games and has a rating of 1500. How accurate is this rating? Not very accurate since it is based only on 3 games. If he plays against and beats a 2000 rated player, this tells us a lot of information. We weren't very confident that 1500 was a good estimate and now we have some more information that tells us his rating should be much higher. Now consider Player B who also has a 1500 rating but has played many games recently so that we are very confident in his rating estimate. If he plays and wins against a 2000 rated player this gives us some information, but it does not override all the information we have from the many games he has played recently that gave him a 1500 rating and so his rating increases, but not as much as Player A. How is this unfair?
very interesting debate! the mathematics behind the ratings are solid, and people would panic if they only saw a rating update every month :)
you said that a 2000 rated player should ALWAYS beat a 1500 rated player, but that is not true. they should almost always beat them. which is why their rating goes up very very little when they do.
very interesting debate! the mathematics behind the ratings are solid, and people would panic if they only saw a rating update every month :)
you said that a 2000 rated player should ALWAYS beat a 1500 rated player, but that is not true. they should almost always beat them. which is why their rating goes up very very little when they do.
I don't see why anyone would panic, contidional they know why rating hasn't changed (perhaps you used a word too strong). Second thing is true, I wasn't stating anything different, of course he can lose sometime, but those theoretical odds are so near to zero that they shouldn't be mentioned separatelly.
Loomis, thanks for excellent post. I agree with everything, but example. What if your player A after that win, continues on playing fairly poor chess, winning and losing 1400-1600 players? Why was he awarded those points, just to lose them in next few games?
Cummulating someone's result from entire month to make single rating change looks much more fair. Here is why. Assume that your Player A and your Player B played 5 different games in one month, both with same people and scoring same result. Assume that both were rated 1500 previously. And they both win that unlucky 2000 rated player. Who cares about what happened 2 years ago, or 75 games ago, or how to explain the fact that player A (for SAME, identical result) is awarded few times more rating points that Player B? Player B is discriminated just for playing chess longer.
I agree that we know much about player B and almost nothing about player A, but past results are not and should not be excuse for dicrimination. (in a way, don't take me wrong) If 2 people at work show identical result, they should be awarded with same paying, no matter that one is 25 and other 45 years old, don't you think?
ello!
I'm no stats wiz (read about the rating method from your site) but I think that its important that past results matter due to the widened statistical sample we can draw information from...
taking a month's worth of play or 5 games limits the sample we can obtain information from... a change in streaks between month to month would cause greater fluctuations in the ratings...
more severe would be the difference in ratings between players who play against opponents of differing strengths... a large sample allows us to offset this
as to the point of discriminating against a player who has been playing longer, i think that i would agree that we should treat him differently from a new player
true, his past records prevent him from attaining a rapid raise in ratings, but the same is also true if he plays a weaker player, past records shield him from a sharp drop in ratings = )
anyway, just 2 cents worth... personally, I think that ratings aren't all that important when placed next to the games we play... thought that its interesting to understand the logic behind the system though!
Cheers!
milutinovic, you're still confusing time with accuracy. You continue to say that the two players are descriminated based on how long they have played or how many games they have played, but this is not the case. They are descriminated based on how accurate their rating is. Remember that the rating is just an estimate. So when the newer player with the worse estimate of his rating gains a lot of points for his lucky win, his RD is still high and we know there is a large uncertainty in his rating.
"If 2 people at work show identical result, they should be awarded with same paying, no matter that one is 25 and other 45 years old, don't you think?"
If the guy who is 45 has worked for me for 20 years and he shows me a result that is what out of the norm, I would think "this guy is most likely going to revert to what he has been the last 20 years." This is based on an accurate estimation of his abilities, not on his age or the amount of time he has worked. If the guy who is 25 has just started working for me and shows me a good result I think it is reasonable to say "he deserves a higher respect from me, but so far I still know little of him, so if he has a bad result he will fall back to the norm more quickly."
ok, ok, ok. It seems that I am simply overvoted or catastrophically wrong on this point. I just hate the idea of rating (or estimate rating) being changed after every single game. I don't know any more why.
And btw, I prefer talent to experience, so my line of thoughts would be " This guy showing good results should be rewarded even more than guy working for 20 years. He is already showing success, like he is working for 20 years. I want to motivate him, and keep him in my firm, who knows what kind of a miracle he would do in few (20) years!?"
With this as my last (as I see no support, so I must be wrong), best wishes to you all. See you at the table.
I lot has been posted on subject of ratings, their appearance, and update, so I would like to suggest something I think can help for them to become real playing measurement.
Ratings on this site are very fluent, you can go +-50 points based on a single game, therefore being +-300 points within a week, which I think is not good. It should be like they are showing your overall playing strenght, not just strenght shown in last 3 games.
My suggestion:
1. new people. For them there should be some kind of provisional rating until they complete let's say 10 games. In that period rating should be as jumping as it is now.
2. "old people". For guys already having more than 10 (or whatever) completed games, rating should be updated once a month, based on their rating from previous month, score of games finished in this month, and overall average rating of people they played to in this month.
This system is similar to one FIDE uses, and I find it much better compared to one here. If someone has lost a game to a guy 200 points less rated (which happens all the time), it doesn't mean he should be punished by taking over 50 points (or even more) from him. This way rating would be much stable and changes made to it would more realistically reflect change in someones play and knowledge.