Why no provisional ratings?
Sort:
shakje
Jun 30, 2008
0
#21
I can understand that it is frustrating, and that this might come across as slightly patronising (that isn't the intention at all), but the solution to people delaying games is to play shorter games. Since this is a correspondence site, when you start a 3 day game, no matter what position they are in they have 3 days to make that move, and there is no reason to complain if they do take those 3 days. Abusing vacation is another issue entirely and I'm pretty sure that site admins will take a very dim view of it. Regardless, I find that the majority of games on here are not like this, and that when they are, the easiest solution is simply to play more games to pass the time (eg enter a big tournament etc.).
One funny thing for me here is, out of your 12 wins against 1600 averages, how many of these were 'blindsided' by the 'unknown' you?
At least 10, maybe 11, -- my 1-10 games, where my rating was lower than my performance and I didn't have losses, were unfair to my opponents' ratings. That's my point. Not having a provisional status causes this problem.
We all have to start somewhere, and just as in any other competitive game, top seeds will be 'upset' by unknowns who eventually become known high-seeds themselves.
We're talking about chess, which has a unique rating system. Upsets can happen, top seeds can be defeated, but none of that has the least bit to do with how to justly break someone into a rating system. I'm not saying new players shouldn't be allowed to defeat established players! ... I'm saying it doesn't make sense for established players to suffer near maximum rating loss when those upsets happen. Rating is supposed to be reflect performance... because the unknown is just that, unknown we can't accurately say whether the established players performance against the unknown was good mediocre or poor. But because there is no provisional status here, the established player's rating is spanked as if we knew his performance was poor.
I do seem to be the only one complaining about this... oh well, I'm fairly certain if there were a hue and cry on chess.com to change the rating system the admins would consider creating a provisional rating status... and I'm even more certain that since there isn't, they won't. C'est la vie.
I don't see the problem with the current system. There is a rating deviation that tells how exact your rating is within a certain range (with an accuracy of p=0.05, which is statistically spoken significant. This means that only 1 of 20 players has a rating range out of the given range). A deviation is effected by N (the number of games played), because how more games you play, how more your rating is becoming the number you really have, and so minimalizing the fluctuations. If N is low, ratings can't be calculated correcty, because there simply is not enough information about how good that player actually is. Each game has therefor a greater impact for a player with a lower N. There is not much to do about it, exept stop whining about it. If your rating is really that important to you, there are 3 possibilities:
1. before you challenge someone, look how much games that player has completed (I recommend a minimum N somewhere around 40, because then a single game hasn't that much influence anymore).
2. Join a chess club if ratings are that important to you, and stop playing online. Ratings aren't that important here I belief.
3. be very good, so that you'll win almost every game. Then it shouldn't matter which rating system you use
shakje wrote: I can understand that it is frustrating, and that this might come across as slightly patronising (that isn't the intention at all), but the solution to people delaying games is to play shorter games. Since this is a correspondence site, when you start a 3 day game, no matter what position they are in they have 3 days to make that move, and there is no reason to complain if they do take those 3 days. Abusing vacation is another issue entirely and I'm pretty sure that site admins will take a very dim view of it. Regardless, I find that the majority of games on here are not like this, and that when they are, the easiest solution is simply to play more games to pass the time (eg enter a big tournament etc.).
Hello Shakje. You are referring to my first post on this thread here; my intention there was not to complain, and I know and realize that every opponent can and may use the full time allowance for every move if they do so wish. My point was that these people ONLY vanish when doom is inevitable and though it is not against the rules, it IS in fact childish and pathetic. And all the more irritating when they try to 'save face' by spouting nonsensical bs about why they can't simply make their next (and final) two moves. Some class. To me, these are people who take the rating too seriously to the point that they lack integrity and gamesmanship. I would argue in these cases that there IS cause for 'complaining', although again that was not my intention anyhow.
My overall point was that new people (such as I am) will wind up having to rely on playing idiots like those to 'clear' their provisionary period, instead of the many great players who are NOT so ignorant nor impudent. I used my own less-than-desirable experience as an example to illustrate - in only 4 games, 2 have gone like that and my current one is looking to go that way with each passing move. These morons are just that, and that is fine. Although it may be known that 3days is the time, we can all plainly see when a player is intentionally avoiding an unwanted fate in YOUR game, but has more than enough time to eagerly play the 3 or 4 that he/she has favourable positions within. Etiquette is what I'm getting at here, which goes well beyond any 'rules'. And if new players are faced with this garbage - to the extent of 50-75% which I would hope to be an anomaly - well, not exactly attractive or welcoming now, is it?
That was my purpose for weighing in on this topic; the thread is about PROV STATUS (or rather the lacking thereof), so I stated my issues with such a thing and some of the reasoning behind my position.
JG27Pyth wrote:
In his example our new RD is 151.something or other, and our new volatility is, get this: 0.05999 -- wow! incredible!... So, uh, our 1500 player is now... uh... yeah, uh... *skritchskritch* uhmmm.... his rating is... well at any rate we do know his volatilty to the 100 thousandth of a... of a... umm... what are the units again? 0.05999 on the Glicko scale?
There seems to be some confusion over the mathematical technicalities here. Basically the point is that there is a sort of "implied provisional period" here. If you haven't played many games, your rating will fluctuate wildly. The more games you play, the more stable your rating becomes. If you have played lots of games, but haven't played many recently, that gets taken into account too.
I'll use myself as an example here. When I first started on this site, I started at 1200 like everyone else. After winning my first game, my rating went up to 1450 (or so). After a few more wins, I was up to 1750. Then a string of losses brought me back to the mid-1500's. Now, after completing 40 games, I'm back up to 1740. Things have stabilized a bit now. For example, I'm currently playing a guy 200 points lower than me. The rating adjustment will be +12 if I win, -37 for a loss, -12 for a draw. That's a fairly small change, really.
The point is that ratings are only meaningful in a long-term sense. It may go up and down by +/- 20 or 30 here and there, but in the long run one game doesn't really affect anything.
For the record, I'm rather indifferent to a "provisional" label. If someone has played less than 20 games their "provisional" status is implied, and it's taken into account by the Glicko system.
For example, I'm currently playing a guy 200 points lower than me. The rating adjustment will be +12 if I win, -37 for a loss, -12 for a draw. That's a fairly small change, really.
I never said you shouldn't lose points for losing to lower rated players. I'm saying you shouldn't lose points to lower rated players who are actually very strong players with ratings that don't reflect their actual performance. If the guy 200 points lower than you is actually a 2200 chess player, playing 2200 chess, then those ratings changes inflicted on you are wrong, and will skew your rating inaccurately*. Is that the end of the world? No of course not. But is it's an easy problem to fix? I recognize that I am in a minority in thinking that provisional ratings are a good idea. I can live with that. But, I honestly don't understand the resistance to provisional ratings.
The major counter-argument here seems to be some version of -- "it's not a big problem; you'll get used to it; ignore it; who cares; play somewhere else if you don't like it (that last bit is rude and not at all intelligent)" -- these are feelings, not reasons, they carry no logical heft. Cyclonus' counter-argument does address the issue. He is logically to the point -- He has a reason for not using provisional ratings: He thinks provisionally rated players would be stigimatized and provisionally rated players would have a hard time finding games. I think, in fact, that is exactly the reasoning of the admins -- possibly based on experience drawn from Yahoo's provisional ratings. I think they're mistaken -- the problem was with Yahoo and their implementation of provisional ratings, not something wrong with the entire provisional rating idea (which is implemented quite successfully by FIDE and the USCF).
If the front door of your house squeaks when you open it -- and someone walks up and says, "hey, a couple drops of oil will fix that" ... it doesn't diminish that argrument to say "aaahhh shaddup, you'll get used to it, we all have." In fact, it supports the argument, acknowledging there is a problem. The germane question is why not oil the door?
Finally, saying -- "I don't care about ratings... if you care about ratings go somewhere else, etc." This is nonsensical! To Knightshade and others who have written these inanities I say: If ratings are so unimportant to you, you can't possible care whether there are provisional ratings or not? Pay the ratings no mind -- Indeed, why even take the trouble to participate in a discussion of these matters? Since the ratings don't matter to you, just pay no attention, and leave the decision to have provisional ratings, or not, to those who do care.
(*37 points is hardly insignificant -- 5 such losses is roughly equal to the loss of a full class).
JG27Pyth wrote:
I never said you shouldn't lose points for losing to lower rated players.
I know you didn't say that. I'm aware that wasn't your point.
(*37 points is hardly insignificant -- 5 such losses is roughly equal to the loss of a full class). True. But if one loses 5 games in a row to players rated 200 points lower, then perhaps one was overrated to begin with?
What I meant was that it is insignificant in the long run. Statistically, I should only lose to players 200 points below me 1 out of every 4 or 5 games - so a one-time drop of 37 points doesn't mean much in the long run.
The major counter-argument here seems to be some version of -- "it's not a big problem; you'll get used to it; ignore it; who cares; play somewhere else if you don't like it (that last bit is rude and not at all intelligent)" -- these are feelings, not reasons, they carry no logical heft.
Ok, but that's not my counter argument. My counter-argument is that you seemed to dismiss the article explaining the mathematical details of the Glicko system, while ignoring the fact that the Glicko system already takes your concerns into account. If a new player joins then yeah, they're rated 1200 (even though they may really be much stronger) - however, they have a very high Glicko RD. That means that you will not lose nearly as many rating points as you would if they were rated 1200 and had finished 200 games.
So essentially, a "provisional period" already exists; the Glicko system already takes this into account, we're just not putting a label on it. Don't only look at a player's rating, also look at how many games they've played. This is not quite as convenient as having a "provisional" label, fine, granted. But is 2 extra mouse clicks really that much of a nuisance?
Statistically, I should only lose to players 200 points below me 1 out of every 4 or 5 games - so a one-time drop of 37 points doesn't mean much in the long run.
First, "you should only lose to players 200 points below you 1 out of every 4 or 5 players" if they are rated correctly in the first place! The whole point of this discussion is that strong new players are rated incorrectly. Not 200 points below rating, either, a brand new 2000 player is rated 800(!) points below his rating. And if a 1600 (400pts above 1200) decides to take him on -- he'll lose 10 out of 10 games and drop to a d class rating. While the news guy's rating climbs to a WRONG rating still well below 2000.
(The drop is more than -37 for the unlucky established fully glicko'ed 1800 who loses to this person, by accident, in a tournament draw...)
Secondly... you've reiterated, for the umpteenth time in this thread, the: "It's not a big problem..." argument. Fine, I'm still standing here, next to the squeaky door, going: But why don't you want it fixed? What the heck is wrong with fixing this problem, big or small?
Bottomline: If you lose to a player rated many hundreds of points below his real level, you will lose rating incorrectly, unfairly, and inaccurately *(Glicko does not fix that!) -- You have given no reason as to why this should not be corrected. You have joined the chorus saying that it doesn't bother you.
Lance, also, this is important:
If a new player joins then yeah, they're rated 1200 (even though they may really be much stronger) - however, they have a very high Glicko RD. That means that you will not lose nearly as many rating points as you would if they were rated 1200 and had finished 200 games.So essentially, a "provisional period" already exists; the Glicko system already takes this into account, we're just not putting a label on it.
No. I thought this too. I think somewhere on the site it is claimed that it works this way. I see no evidence of this being true. IF it did work this way it would help! But from what I can tell your ratings movement is wholly, or mostly, dependent on _your own_ Glicko RD! Your opponent's glicko RD affects how his ratings move! There is no Glicko "provisional" period -- what the Glicko RD does is help shorten the period during which the badly rated player remains badly rated. His ratings move fairly quickly at first. (However, IMHO not nearly quickly enough. BTW.)
I could be wrong. I'm basing this not on a full understanding of the Glicko formula's but on experience while carefully watching how my ratings would be affected by my opponents. If I am wrong, then at a minimum experience shows that whatever effect the Glicko RD has on newly rateds is too small, for newly rateds still affect their opponents ratings far too dramatically -- empirically, they do.
*EDIT* I am wrong, but not too far off... one's own RD is what counts most. Opponents RD does make a difference, but as I've said, not enough difference. And I _still_ don't see what's wrong with a 10-25 game provisional period where the new player has a drastically reduced impact on his established opponent's rating. ICC does it this way.
JG27Pyth wrote:
Secondly... you've reiterated, for the umpteenth time in this thread, the: "It's not a big problem..." argument. Fine, I'm still standing here, next to the squeaky door, going: But why don't you want it fixed? What the heck is wrong with fixing this problem, big or small?
I know from reading your previous posts in other discussions that you're a smart person, and I don't really want to argue. That said, you're missing my point: I don't think the door squeaks at all. As far as I'm concerned, it's not a small problem - there is no problem.
A few things:
1. Someone on another forum once made this analogy, and I'm going to paraphrase it here. Comparing day-to-day changes in ratings is like comparing day-to-day fluctuations in stock prices. The change from yesterday to today doesn't matter; what matters is trends over weeks, months, years.
2. You're only looking at one side of the coin. A player whose real rating is 700 still starts off at 1200 - so you gain more points than you deserve by beating him. It works both ways, and should average out in the long run.
3. TD's can set a minimum # of games required for players to join. If you don't want to play against "Glicko-provisional" players, then don't join tournaments unless the min. # of games is at least 20 - caveat emptor. Having an official "provisional" label for players with less than 20 games complete wouldn't change anything here.
4. Someone with a rating over 1730 is rated higher than 90% of chess.com members; leading one to ask, how often does a 2000+ player join? How many new players are "actually" rated higher than 1730? (Answer: 10%, max). Thus, the scenario you describe is rare. A new 2000+ player will have their rating over 1700 within the first 5 or 6 games they finish.
5. Ratings have more and more meaning the more games you play. After 20-30 games, your rating is "statistically meaningful" but still only a rough approximation. Once you've played hundreds of games your rating starts to become truely accurate. Suppose I play 200 games against players rated 1540; I'm expected to lose 40 of those games. In reality, things don't always work out exactly as statistics predict they should. So in reality, I'd probably lose 30-50 of those games. If one or two of those losses was against a new player who was underrated, so what? My rating would change by:
(150 wins) x (+12 for winning) = +1800 and
(45 losses) x (-37 for losing) = -1665 and
(5 draws) x (-12 for drawing) = -60
Net change: +75.
In total, my rating is hardly affected at all. Note that it doesn't even matter if 2 or 3 of those 45 losses are to underrated "newbs"; these calculations involve numbers which are within reasonable margins of statistical error anyway. (i.e., this could still happen even if all the 1540 players had accurate ratings)
