Eliot T. Jacobson Nov 3 1992, 1:53 pm show options Newsgroups: rec.games.chess From: jacob...@oucsace.cs.ohiou.edu (Eliot T. Jacobson ) - Find messages by this author Date: 3 Nov 92 19:55:52 GMT Local: Tues, Nov 3 1992 11:55 am Subject: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse This is an open letter in complaint of the current ICS rating system. When I first registered with ICS over a year ago, my rating was in the low 1600's. Now it at 1475, and is in steady decline. Is this because I have grown worse as a chess player? NO! I noticed the gradual decline of several players that I surveyed (cron, but, aries, etc.), and I determined to find out why. First, I felt the need to confirm my crude feelings, and for this I collected data from the ICS statistics over a month long period. DATE Average ICS Standard Total number rating Deviation registered/rated ____________________________________________________________________ 10/5 1287 314 ***/*** 10/6 1284 317 1101/*** 10/7 1282 319 1107/*** 10/8 1280 321 1106/829 10/9 1281 320 1118/831 10/12 1277 320 1133/845 10/14 1277 320 1142/849 10/15 1276 321 1147/850 10/16 1276 322 1155/856 10/18 1276 321 1177/865 10/19 1275 321 1182/869 10/20 1273 323 1186/873 10/21 1272 321 1191/880 10/22 1271 321 1197/883 10/25 1269 320 1211/895 10/26 1270 318 1219/899 10/28 1269 319 1225/906 10/29 1267 320 1234/910 10/30 1265 322 1244/917 11/2 1264 322 1254/929 11/3 1263 323 1267/937 ===================================================================== Thus in the period from 10/5 to 11/3 the average ICS rating has dropped from 1287 to 1263, a total of 24 points. Over a 12 month period (about how long I have been on ICS) this amounts to an average drop of about 288 points... of course, ICS has been growing rather fast of late, and the rating drop is increasing in proportion. But, as I recall when I joined, the average rating was in the upper 1300's. But of course, you say, the reason is obvious! There are more weak players joining, pushing down the overall average. I say no! In fact, there are more stronger players joining, and the fact that their initial provisional rating is 1000 is killing the established players who play these people, and this point drain is filtering through. Can I back this up? Well, yes. I wrote a computer program to model a new player joining ICS. For this program I established the following parameters. Person A: Joe (ICS) Average. He is an established (20+ games) ICS player, with an ICS rating of 1300. His rating is calcualted by the USCF rating formula, well, based on ICS ratings. Person B: Mary Variable. She is a player whose theoretical strength is known prior to joining ICS, who begins the process of getting an ICS rating with the provisional ICS rating of 1000. Her rating is a performance rating. For the purpose of this model is was convenient to have Joe play against Mary for a 24 game match. The result of the match is listed as (points for Joe)-(points for Mary). The "Net Loss to rating pool" is the sum of both players point loss from the match. Mary's Theo- Result of Joe's Final Mary's First Net Loss to retical Strength 24 games ICS rating ICS rating rating pool ___________________________________________________________________________ 1000 21-3 1279 1000 21 points 1100 18-6 1269 1084 47 points 1200 15-9 1217 1162 121 points 1300 12-12 1254 1258 88 points 1400 9-15 1189 1329 182 points 1500 6-18 1218 1426 156 points 1600 3-21 1200 1509 191 points ____________________________________________________________________________ A lot has to do with the specific sequence of wins and losses, as Mary's rating is quite variable to begin with, but the pattern is clear. The Higher Mary's Theoretical Strength, the more drain on the ICS rating pool she places by entering ICS and becomming rated. So over the course of the month I studied this, 100 such Mary's entered the pool, and as the number of registered players grew from 829 to 937, the average rating dropped about 20 points. If I was better at math, I could regress to compute what the average rating of these last 100 newly registered players really is, but I don't really care. The point is, EVERYONE'S RATING WILL CONTINUE TO HAVE DOWNWARD PRESSURE, WILL CONTINUE TO FALL, IF THE CURRENT SYSTEM IS NOT CHANGED. And the easiest FIX is: RAISE THE INITIAL PROVISIONAL RATING FROM 1000 TO SOMETHING MORE REALISTIC. And along the way, why don't you also: INCREASE EVERYONES RATING BY ABOUT 500 POINTS, TO MAKE THE RATINGS FEEL MORE REALISTIC. Mika Kukkonen Nov 3 1992, 8:42 pm show options Newsgroups: rec.games.chess From: mkukk...@klaava.Helsinki.FI (Mika Kukkonen) - Find messages by this author Date: 4 Nov 92 10:11:19 GMT Local: Wed, Nov 4 1992 2:11 am Subject: Re: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse In <1992Nov3.195552.13...@oucsace.cs.ohiou.edu> jacob...@oucsace.cs.ohiou.edu (Eliot T. Jacobson ) writes: [some interesting statistics deleted] > But of course, you say, the reason is obvious! There are more > weak players joining, pushing down the overall average. I say > no! In fact, there are more stronger players joining, and the I also have found (hard way :) ) that there are more and more good players (= master level or above) joining ICS. On the other hand the amount of poor players has been quite constant, only increasing as the number of people on ICS at the SAME time has been increasing. In first days (yes, I was there :) ) there were about 10 players at most, now I have seen 46 and more... > fact that their initial provisional rating is 1000 is killing Was not this 1500 in first days? I can not see any point in lowering it to 1000 ... > the established players who play these people, and this point > drain is filtering through. Can I back this up? Well, yes. The biggest annoyment as high rated player is, that if you play against low rated player, you have 1 point to win and 32 points to lose... [interesting simulation deleted] > EVERYONE'S RATING WILL CONTINUE TO HAVE DOWNWARD PRESSURE, WILL CONTINUE > TO FALL, IF THE CURRENT SYSTEM IS NOT CHANGED. I agree with this, except in one point. Computers like Scrathcy will probably have a high rating, just because they are not humans, and well, humans make blunders, something I get constantly reminded in ICS. :) > And the easiest FIX is: > RAISE THE INITIAL PROVISIONAL RATING FROM 1000 TO SOMETHING MORE REALISTIC. Well, this looks like an emergency measure, not a fix. Something like pumping more air into tyre with a hole in it. But 1500 is naturally much more realistic, *IF* we want ICS ratings compare to USCF or FIDE. > And along the way, why don't you also: > INCREASE EVERYONES RATING BY ABOUT 500 POINTS, TO MAKE THE RATINGS FEEL > MORE REALISTIC. I thought polls showed that current ratings were about 300 pts lower than peoples USCF. First I would like to say something about current system. Following information can be had from valkyries by command "help ratings". >Topics : Ratings on ICS (USCF hybrid system) ------------------------------------- > Ratings are computed as follows : >-- If you have played less than 20 games : I think this 20 game limit is ok. >o When you win, you add 400 to your opponents rating, and average in. >o When you lose, subtract 400 from your opponents rating, and average in. But these two are totally ridiculous. Do you really do it that way in USCF? How did you get this magic number "400"?? >o When you tie, you average your opponents rating in with yours. >-- After 20 games : >o If rated < 2100, you can fluctuate a max of 32 points. >o If rated < 2400, you can fluctuate a max of 24 points. >o If rated >= 2400, you can fluctuate a max of 16 points. This is clearly a design flaw, because there currently is only 3 players (= scratchy, RFischer, aries) with normal game rating of 2100 or over, and also 3 in blitz ratings (= scratchy, RFischer, Woodpusher), of which two are computers (I don't remember if Woodpusher was, as he has not been around for a LONG while), and aries is a strong IM, if my memory servers me right. >o For win, take diff. in ratings, divide by 25, and add to the max > fluctuation / 2. If this is less than 1, it is set to one. >o For loss, take the diff. in ratings, divide by 25, and add to the > negative max fluctuation / 2. If this is > -1, it is set to -1. >o For tie, if equal rating, no change, otherwise higher rated + max > fluctuation/8, lower rated -max fluctuation/8. At least they had the decency to put 1 as minimum limit, in first days it was 0... >o You also have a rating floor, which, after 20 games, is set to your > maximum rating rounded down to the nearest 100 - 100. Your rating will > never drop below that. This was eliminated, as it gave an easy way for two people to boost each others ratings to the sky. So, as a "better" alternative I would like to give some pointers how rating calculations are done in Finland. This system gives ratings that are almost on the same level than FIDE ratings. (Although my experiences show that ELO is getting inflated.) 1) there is also a "beginner" rating, set to 1525. You play 15 rated games with that rating, and then you get a real one. This real one is calculated in some simpler way than the real rating. (It is based on your oppponents average ratings. I can't remember it now, as my beginners days were LONG time ago ...) 2) The real ratings are calculated following way: When you play a rated opponent, you first count the difference of your ratings. Then there is a table that gives your estimated win percentage. You substract this percentage from the result of game (1 pt from win, 0.5 from draw and 0 from loss) and multiply it with your level multiplier, and add the result to your rating to get a new one (Note that the result can be negative). Looks simple, right. The problem of course is the percentage table. The table is ordered so, that while the difference gets bigger, the percentage grows (or get smaller) slower. So at the beginning the "fork" is three points (0-3 difference gives percentage of 0.5) and at over 200 pts the "fork" is over ten points). There is also a rule, that stronger players percentage never drops below 0.05, to avoid situations where you got 0 points to gain. An example: I (rated 2061) am going to play against a top Finnish team next saturday, where my opponent will be rated about 2200. Difference is 149 pts. My win percentage (I am taking this from hat) will be about 0.25. My level multiplier is the smallest one, 20. Lets say I get lucky, and nail that sucker. :) So my new rating is: 2061 + (1 - 0.25) * 20 = 2076 You may notice, that this makes slow coming from 1525 to 2000. This is compensated by level multiplier, which is 45 on ratings under 1650, and drops 5 pts every 100 from that on, until reaching minimum 20 at over 2050. So, what do you think? Does this system sound more reasonable (at least it seems to work in real life :) )? If there is enough intrest, I can dig up that rating table and other stuff I don't remember now. Mika -- Mika Kukkonen = Mika.Kukko...@Helsinki.FI Trying to learn to live in Helsinki University of Computer Science. "I am studying, so I am, or am I (studying)?" Onar Aam Nov 4 1992, 8:12 am show options Newsgroups: rec.games.chess From: o...@hsr.no (Onar Aam) - Find messages by this author Date: Wed, 4 Nov 1992 14:29:45 GMT Local: Wed, Nov 4 1992 6:29 am Subject: Re: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse >EVERYONE'S RATING WILL CONTINUE TO HAVE DOWNWARD PRESSURE, WILL CONTINUE > TO FALL, IF THE CURRENT SYSTEM IS NOT CHANGED. until it reaches equilibrium. >> And the easiest FIX is: >> RAISE THE INITIAL PROVISIONAL RATING FROM 1000 TO SOMETHING MORE REALISTIC. >Well, this looks like an emergency measure, not a fix. Something like ^^^^^^^^^^^^^^^^ >pumping more air into tyre with a hole in it. But 1500 is naturally much >more realistic, *IF* we want ICS ratings compare to USCF or FIDE. This is not true. If I understand the system correctly then 1000 points are added to the pool each time a new player enters. And this makes the 1000 rating an attractor for the whole system. The average is converging towards 1000. By setting the provisional rating to the average of the international rating the system as a whole will slowly recover. I agree that a provisional rating of 1500 will fix the problems of ICS. >>o When you win, you add 400 to your opponents rating, and average in. >>o When you lose, subtract 400 from your opponents rating, and average in. >But these two are totally ridiculous. Do you really do it that way in USCF? >How did you get this magic number "400"?? Yup, that's how it's done, and it works. The reason for doing this is to prevent deflation in the system. For a player to get a certain rating he has to "steal" points from others. When the player retires or dies he literally takes the points with him into the grave and they are lost for ever. The initial 20 games are a way of adding new points into the system. >>o When you tie, you average your opponents rating in with yours. >>-- After 20 games : >>o If rated < 2100, you can fluctuate a max of 32 points. >>o If rated < 2400, you can fluctuate a max of 24 points. >>o If rated >= 2400, you can fluctuate a max of 16 points. >This is clearly a design flaw, because there currently is only 3 players >(= scratchy, RFischer, aries) with normal game rating of 2100 or over, >and also 3 in blitz ratings (= scratchy, RFischer, Woodpusher), of which >two are computers (I don't remember if Woodpusher was, as he has not been >around for a LONG while), and aries is a strong IM, if my memory servers me >right. This is to keep linearity in the system. In theory this means that a player at rating X will win 1/3 of the matches against a player with rating X + 120. Onar. Paul A. Lane Nov 4 1992, 8:17 am show options Newsgroups: rec.games.chess From: pal...@iastate.edu (Paul A. Lane) - Find messages by this author Date: 4 Nov 92 15:20:36 GMT Local: Wed, Nov 4 1992 7:20 am Subject: Re: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse I'm surprised that the provisional rating is 1000. Why not adopt a system similar to the USCF where your rating is the average opponent's rating for the first n games +400*(W-L)/n? andrew m. boardman Nov 4 1992, 9:31 am show options Newsgroups: rec.games.chess From: a...@cs.columbia.edu (andrew m. boardman) - Find messages by this author Date: Wed, 4 Nov 1992 18:13:36 GMT Local: Wed, Nov 4 1992 10:13 am Subject: Re: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse The "why have the artificial rating" question is a really good one -- but is anyone from ICS reading to answer? >and aries is a strong IM, if my memory servers me right. Ginsberg. >What? Did you not read what I said? Why in earth is there is a limit of >2400, if nobody ever reaches it?? Current system is, that almost >everybody has to live with max fluctuation of 32 pts, which is really >annoying. They took these numbers from a system which they didn't take the rest of. Inconsistency... andrew Alan Baljeu Nov 4 1992, 9:36 am show options Newsgroups: rec.games.chess From: balje...@uther.calvin.edu (Alan Baljeu) - Find messages by this author Date: Wed, 4 Nov 1992 18:51:56 GMT Local: Wed, Nov 4 1992 10:51 am Subject: Re: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse pal...@iastate.edu (Paul A. Lane) writes: >I'm surprised that the provisional rating is 1000. Why not adopt a system >similar to the USCF where your rating is the average opponent's rating for >the first n games +400*(W-L)/n? I think it is that way. The problem is that in your first game, your opponent is playing a player rated 1000, because that rating is assumed before any games occur. Mika Kukkonen Nov 4 1992, 9:37 am show options Newsgroups: rec.games.chess From: mkukk...@klaava.Helsinki.FI (Mika Kukkonen) - Find messages by this author Date: 4 Nov 92 16:38:38 GMT Local: Wed, Nov 4 1992 8:38 am Subject: Re: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse In <1992Nov4.142945....@hsr.no> o...@hsr.no (Onar Aam) writes: >>EVERYONE'S RATING WILL CONTINUE TO HAVE DOWNWARD PRESSURE, WILL CONTINUE >> TO FALL, IF THE CURRENT SYSTEM IS NOT CHANGED. >until it reaches equilibrium. I think that the first poster proved his point much better, unless by "equilibrium" you mean a position where everybody's rating is 0. :) >This is not true. If I understand the system correctly then 1000 points are added >to the pool each time a new player enters. And this makes the 1000 rating an >attractor for the whole system. The average is converging towards 1000. By >setting the provisional rating to the average of the international rating the >system as a whole will slowly recover. I agree that a provisional rating of 1500 >will fix the problems of ICS. It may fix the problem of ICS ratings being 300 points behind USCF, but I doubt it, because that was not the case even when it was 1500 ... >Yup, that's how it's done, and it works. The reason for doing this is to prevent >deflation in the system. For a player to get a certain rating he has to "steal" >points from others. When the player retires or dies he literally takes the points >with him into the grave and they are lost for ever. The initial 20 games are a >way of adding new points into the system. Yes, I read the discussion earlier about thi, and now that I am at home I looked the way it is done in our rating system: UnratedPlayersRating = OpponentRatingsAverage + 4*(percentage result -50) - Hide quoted text - - Show quoted text - >>>o When you tie, you average your opponents rating in with yours. >>>-- After 20 games : >>>o If rated < 2100, you can fluctuate a max of 32 points. >>>o If rated < 2400, you can fluctuate a max of 24 points. >>>o If rated >= 2400, you can fluctuate a max of 16 points. >>This is clearly a design flaw, because there currently is only 3 players >>(= scratchy, RFischer, aries) with normal game rating of 2100 or over, >>and also 3 in blitz ratings (= scratchy, RFischer, Woodpusher), of which >>two are computers (I don't remember if Woodpusher was, as he has not been >>around for a LONG while), and aries is a strong IM, if my memory servers me >>right. >This is to keep linearity in the system. In theory this means that a player at >rating X will win 1/3 of the matches against a player with rating X + 120. What? Did you not read what I said? Why in earth is there is a limit of 2400, if nobody ever reaches it?? Current system is, that almost everybody has to live with max fluctuation of 32 pts, which is really annoying . Mika -- Mika Kukkonen = Mika.Kukko...@Helsinki.FI Trying to learn to live in Helsinki University of Computer Science. "I am studying, so I am, or am I (studying )?" Richard V. Nash Nov 4 1992, 10:36 am show options Newsgroups: rec.games.chess From: n...@visus.com (Richard V. Nash) - Find messages by this author Date: 4 Nov 92 18:08:46 GMT Local: Wed, Nov 4 1992 10:08 am Subject: Re: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse In article < 1992Nov3.195552.13...@oucsace.cs.ohiou.edu> jacob...@oucsace.cs.ohiou.edu (Eliot T. Jacobson ) writes : > RAISE THE INITIAL PROVISIONAL RATING FROM 1000 TO SOMETHING MORE REALISTIC. Unless the current administrators have changed something, there is _no_ initial provisional rating. You only acquire a rating after you play your first rated person. If you win you are their rating + 400, if you loose you are their rating - 400, or if you draw you get their rating. I do agree that ICS ratings are way below USCF or FIDE ratings. I also agree that the average rating seems to be dropping. I think it is because more weak players (like me) have been joining. Anyway, the code that does the ratings calculation is simple and I will provide it to anyone who asks. -- +---------------------------------------------------------------+ | Richard V. Nash | Visual Understanding Systems, Inc. | | n...@visus.com | Tel. (412)-488-3600 Fax. (412)-488-3611 | +---------------------------------------------------------------+ Tony Chamberlain Nov 4 1992, 12:11 pm show options Newsgroups: rec.games.chess From: acham...@cs.weber.edu (Tony Chamberlain) - Find messages by this author Date: 4 Nov 92 19:26:06 GMT Local: Wed, Nov 4 1992 11:26 am Subject: Re: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse |> This is not true. If I understand the system correctly then 1000 points are a dded |> to the pool each time a new player enters. And this makes the 1000 rating an |> attractor for the whole system. The average is converging towards 1000. By |> setting the provisional rating to the average of the international rating the |> system as a whole will slowly recover. I agree that a provisional rating of 1 500 |> will fix the problems of ICS. It seems to me that with the +/- 400 thing, 1000 is not an attractor, and 1000 is not what people start with, thus ratings arent' deflated taht much. for example during the first 20 games if i play someone 400 points lower than me and win they lose one and i don't gain any. This seems to be the only case where points aren't put into the system. If i play someone my level, i gain 400 points( to be averaged) while they only lose 16. if i play someone 400 points higher than me they lose 32 and 800 points get averaged into my score. Unless i play really bad i should start out with a score between 1100-1200 (granted this is well below the 1500 people have stated are good starting places. Of course my math isn't strong, and i might not be fully comprehending waht's going on with the rating sytem , but does this seem reasonable? annan Nov 4 1992, 3:04 pm show options Newsgroups: rec.games.chess From: a...@vax.oxford.ac.uk - Find messages by this author Date: 4 Nov 92 09:20:28 GMT Local: Wed, Nov 4 1992 1:20 am Subject: Re: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse In article <1992Nov3.195552.13...@oucsace.cs.ohiou.edu>, jacob...@oucsace.cs.ohiou.edu (Eliot T. Jacobson ) writes : > This is an open letter in complaint of the current ICS rating system. > When I first registered with ICS over a year ago, my rating was in > the low 1600's. Now it at 1475, and is in steady decline. [much deletia] - Hide quoted text - - Show quoted text - > Person A: Joe (ICS) Average. He is an established (20+ games) > ICS player, with an ICS rating of 1300. His rating > is calcualted by the USCF rating formula, well, based > on ICS ratings. > Person B: Mary Variable. She is a player whose theoretical > strength is known prior to joining ICS, who begins > the process of getting an ICS rating with the provisional > ICS rating of 1000. Her rating is a performance rating. > Mary's Theo- Result of Joe's Final Mary's First Net Loss to > retical Strength 24 games ICS rating ICS rating rating pool > ___________________________________________________________________________ > 1000 21-3 1279 1000 21 points .... > And the easiest FIX is: > RAISE THE INITIAL PROVISIONAL RATING FROM 1000 TO SOMETHING MORE REALISTIC. Hmmmmmm. If the rating system really is such that one player loses points and the opponent doesn't gain any when they perform as expected, then I don't think this fix is really much of a solution. How about first fixing the algorithm? I do however agree with the gist of the article. How about this for a proposal: a previously unrated player's first rating is based on the results of their first n games, for your favourite natural number n. Seems like a good chance to settle the arguments about rating inflation/deflation in the real world! > And along the way, why don't you also: > INCREASE EVERYONES RATING BY ABOUT 500 POINTS, TO MAKE THE RATINGS FEEL > MORE REALISTIC. And also decrease everyone's USCF rating by about 150 (?) points to make those ratings feel more realistic :-) -- James Annan. [--------------------------------------] [ "The only good Tory is a lavatory" ] [--------------------------------------] Kenneth Sloan Nov 4 1992, 3:17 pm show options Newsgroups: rec.games.chess From: s...@cis.uab.edu (Kenneth Sloan) - Find messages by this author Date: Thu, 5 Nov 1992 03:28:21 GMT Local: Wed, Nov 4 1992 7:28 pm Subject: Re: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse In article <1992Nov4.092028.9...@vax.oxford.ac.uk> a...@vax.oxford.ac.uk writes: >... >Seems like a good chance to settle the arguments about rating >inflation/deflation in the real world! Not unless the ICS rating system actually behaves in the same way as those "in the real world". My only knowledge is the ICS system comes from the (conflicting) reports posted here in the last few days - but I'll guess at what's happenning anyway. It sounds as if new, or provisionally rated players use an artificial (often wrong) rating during their first several games - after which their performance rating in those games becomes their established rating. This, if true, is a very bad idea. Idealy, established players should, as a group, neither gain or nor lose rating points to "provisional" players. In the USDF system, individual established players can lose points when they lose to a provisional player, but unless the provisional player never loses a game, these points are returned to the pool via the players who defeat the provisional player. In general, a provisional player *by definition* performs precisely as predicted by his first established rating, and neither takes, nor donates rating poits to the pool. If the ICS system doesn't do this properly, the small size of the pool, and the relatively large number of new players, will virtually guarantee large shifts in the "average rating". What are the *goals* of the ICS system? Is the intent to match USCF ratings? Player-by-player (for those with ratings in each system)? Statistically? How detailed? Mean and std dev? How are those goals supposed to be met? By magic? Why should anyone care if their ICS rating matches their USCF, or FIDE rating? Unless there are specific mechanisms to keep the two (three) systems correlated, I would think it a positive virtue to have a completely different scale, to avoid confusion. Perhaps the ICS administrators ought to find some of the proponents of "superior" ratings systems, built upon superior statistics of pair-wise comparisons, and let them use the ICS community as a testbed? -- Kenneth Sloan Computer and Information Sciences s...@cis.uab.edu University of Alabama at Birmingham (205) 934-2213 115A Campbell Hall, UAB Station (205) 934-5473 FAX Birmingham, AL 35294-1170 Steven L Harrington Nov 4 1992, 5:19 pm show options Newsgroups: rec.games.chess From: sharr...@cs.tamu.edu (Steven L Harrington) - Find messages by this author Date: 5 Nov 92 06:14:00 GMT Subject: Re: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse Who cares whether the ICS ratings are lower than uscf/fide? What difference does it make? I could care less myself. Just map them over. It seems that 1700 ~= master, 2000~= senior master, etc... What's the big deal? Just like everyone else, my rating on ICS is much lower than my uscf, but I could care less. Chess is chess and the stronger players rise to the top whether the top is 2100 or 2600. A rose by any other name would still play like a rose... --steve harrington --ICS:cowboy Bill Vaughn Nov 6 1992, 9:26 am show options Newsgroups: rec.games.chess From: b...@myope.cvs.rochester.edu (Bill Vaughn) - Find messages by this author Date: Thu, 5 Nov 92 16:51:41 GMT Local: Thurs, Nov 5 1992 8:51 am Subject: Re: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse In article pal...@iastate.edu (Paul A. Lane) writes: >I'm surprised that the provisional rating is 1000. Why not adopt a system >similar to the USCF where your rating is the average opponent's rating for >the first n games +400*(W-L)/n? Why not adopt a system where the provisonal rating is the current average rating. Onar Aam Nov 6 1992, 6:57 pm show options Newsgroups: rec.games.chess From: o...@hsr.no (Onar Aam) - Find messages by this author Date: Thu, 5 Nov 1992 06:47:11 GMT Local: Wed, Nov 4 1992 10:47 pm Subject: The ICS rating system (was Re: Why I Hate the ICS rating system.) Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse - Hide quoted text - - Show quoted text - >|> This is not true. If I understand the system correctly then 1000 points are a >dded >|> to the pool each time a new player enters. And this makes the 1000 rating an >|> attractor for the whole system. The average is converging towards 1000. By >|> setting the provisional rating to the average of the international rating the >|> system as a whole will slowly recover. I agree that a provisional rating of 1 >500 >|> will fix the problems of ICS. >It seems to me that with the +/- 400 thing, 1000 is not an attractor, and >1000 is not what people start with, thus ratings arent' deflated taht much. >for example during the first 20 games if i play someone 400 points lower than >me and win they lose one and i don't gain any. This seems to be the only case >where points aren't put into the system. >If i play someone my level, i gain 400 points( to be averaged) while they only >lose 16. if i play someone 400 points higher than me they lose 32 and 800 points >get averaged into my score. Unless i play really bad i should start out with a >score between 1100-1200 (granted this is well below the 1500 people have stated >are good starting places. >Of course my math isn't strong, and i might not be fully comprehending waht's >going on with the rating sytem , but does this seem reasonable? It sounds reasonable enough but you missed a little point. In your very first game the server treats you as if you are rated to 1000. This does not affect your own rating since that follows the "+/- 400 thing" BUT your opponent is affected by it. Imagine that you are a 2000 player and in your first game you play someone rated at 1500. It is most likely that your opponent will lose to you since you in real life are a lot better than him. AND , the crucial point, he loses 31 points because the server assumes that you are a 1000 player. These 31 points are lost forever. And this means that the total number of points in the pool on average will decrease until 1000 equals the average rating of all the players. Lets for a sec. imagine that the provisional rating is 2000. What would happen then? 2000 is above average and therefore the average would rise until it reached 2000. Ex. A player rated 1000 plays his first game against a 1500 player. the 1500 player is most likely to win BUT the server thinks that the new player is rated 2000 and therefore the 1500 player gets 31 points out of nowhere. (He would normally only get one point) And the average would slowly rise till it reached 2000. >In article <1992Nov3.195552.13...@oucsace.cs.ohiou.edu> >jacob...@oucsace.cs.ohiou.edu (Eliot T. Jacobson ) writes: >> RAISE THE INITIAL PROVISIONAL RATING FROM 1000 TO SOMETHING MORE REALISTIC. >Unless the current administrators have changed something, there is _no_ initial >provisional rating. You only acquire a rating after you play your first rated >person. If you win you are their rating + 400, if you loose you are their >rating - 400, or if you draw you get their rating. This is an indirect rating. The server "pretends" you have an initial rating of 1000. As I said this does not affect your own rating but rather your first opponent. >I do agree that ICS ratings are way below USCF or FIDE ratings. I also agree >that the average rating seems to be dropping. I think it is because more weak >players (like me) have been joining. In fact, if an abnormally large number of below-average people entered the net they would in fact skew the net back up again! The average would still be sinking, but each player's rating would get closer to their FIDE rating. Someone mentioned that there was a skew problem even when the provisional rating was 1500. The reason for this is unclear. I cannot imagine that the rating system is intrinsically skewish since it is directly based on the ELO rating system. Therefore there must be other reasons. But putting the provisional rating back up to 1500 will definitely help. That I can guarantee. Onar. Kenneth Sloan Nov 7 1992, 11:15 pm show options Newsgroups: rec.games.chess From: s...@cis.uab.edu (Kenneth Sloan) - Find messages by this author Date: Fri, 6 Nov 1992 18:56:45 GMT Local: Fri, Nov 6 1992 10:56 am Subject: Re: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse In article balje...@uther.calvin.edu (Alan Baljeu) writes: >s...@cis.uab.edu (Kenneth Sloan) writes: >>What are the *goals* of the ICS system? Is the intent to match USCF >>ratings? Player-by-player (for those with ratings in each system)? >>Statistically? How detailed? Mean and std dev? >If you ask me, I think a good goal is to have a player of constant strength >have a constant rating. Then a person can tell if (s)he is improving, >staying the same, or going senile. Other than that, it doesn't really matter >to me. Good. That's a reasonable goal. Now - do you have any ideas on how to achieve that goal? I don't. -- - Hide quoted text - - Show quoted text - Kenneth Sloan Computer and Information Sciences s...@cis.uab.edu University of Alabama at Birmingham (205) 934-2213 115A Campbell Hall, UAB Station (205) 934-5473 FAX Birmingham, AL 35294-1170 Alan Baljeu Nov 7 1992, 11:15 pm show options Newsgroups: rec.games.chess From: balje...@uther.calvin.edu (Alan Baljeu) - Find messages by this author Date: Fri, 6 Nov 1992 17:41:31 GMT Local: Fri, Nov 6 1992 9:41 am Subject: Re: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse s...@cis.uab.edu (Kenneth Sloan) writes: >What are the *goals* of the ICS system? Is the intent to match USCF >ratings? Player-by-player (for those with ratings in each system)? >Statistically? How detailed? Mean and std dev? If you ask me, I think a good goal is to have a player of constant strength have a constant rating. Then a person can tell if (s)he is improving, staying the same, or going senile. Other than that, it doesn't really matter to me . Richard Scott Nov 8 1992, 4:35 am show options Newsgroups: rec.games.chess From: l...@yoyo.cc.monash.edu.au (Richard Scott) - Find messages by this author Date: Sat, 7 Nov 1992 14:58:21 GMT Local: Sat, Nov 7 1992 6:58 am Subject: Re: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse Who cares? Just lower everyone's ratings :) It is not that big a deal, is it? I suppose people are deadly serious about this, but it should just be fun. I, for one, do not enjoy playing people who don't care about anything but winning and their rating. Part of the reason I haven't been playing much. However, why not just use the ELO system? Wouldn't that make more sense? I am not very good (only about 1120) , which I am sure is too high!! Richard -- AussieVamp AussieMan, Neophyte Vampire and DarkWolf (Eric Clapton is God!!!) [ and there is Bobby Fischer... ;-) ] (Walter Payton, Mike Singletary and Neal Anderson form the Holy Trinity!) David Goldfarb Nov 8 1992, 11:04 pm show options Newsgroups: rec.games.chess From: goldf...@ocf.berkeley.edu (David Goldfarb) - Find messages by this author Date: 9 Nov 1992 07:47:45 GMT Local: Sun, Nov 8 1992 11:47 pm Subject: Re: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse In article <1992Nov4.180846.14...@visus.com> n...@visus.com writes: ) In article <1992Nov3.195552.13...@oucsace.cs.ohiou.edu> )jacob...@oucsace.cs.ohiou.edu (Eliot T. Jacobson ) writes : )> RAISE THE INITIAL PROVISIONAL RATING FROM 1000 TO SOMETHING MORE REALISTIC. ) )Unless the current administrators have changed something, there is _no_ initial )provisional rating. You only acquire a rating after you play your first rated )person. If you win you are their rating + 400, if you loose you are their )rating - 400, or if you draw you get their rating . When someone plays their first rated game, their opponent's rating changes as a result. As things are, that opponent's rating changes as though the unrated player were rated 1000. In actual USCF play, this is less of a problem because people usually play several games in their first tournament. So you can calculate a rating based on several games and rate all the opponents against that. On the ICS, however, all games are rated as soon as they end. (Side note: anyone ever wonder what the USCF does when two players both play their first rated game against each other and then don't play any more games in that tournament? This actually came up recently at the Berkeley Chess Club. The winner got a rating of 1001/1 and the loser 1000/1.) -- David Goldfarb | "Bagels can be an enormous force for good or for goldf...@ocf.berkeley.edu | evil. It is up to us to decide how we will use goldf...@UCBOCF.BITNET | them." -- Daniel M. Pinkwater Sorvari Jarmo Nov 8 1992, 11:40 pm show options Newsgroups: rec.games.chess From: j...@cs.tut.fi (Sorvari Jarmo) - Find messages by this author Date: 9 Nov 92 13:12:20 Local: Mon, Nov 9 1992 1:12 pm Subject: Re: The ICS rating system (was Re: Why I Hate the ICS rating system.) Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse In article <1992Nov5.064711.6...@hsr.no> o...@hsr.no (Onar Aam) writes: It sounds reasonable enough but you missed a little point. In your very first game the server treats you as if you are rated to 1000. This does not affect your own rating since that follows the "+/- 400 thing" BUT your opponent is affected by it . How about if the opponents of novices would not be affected at all? So that after 20 games, new players would enter the pool with a more realistic approximation (based on the +/- 400 rule) of their strength than just bringing in 1000 rating points automatically (even master class players). I don't think new players would find it difficult to find opponent willing to play them. -- ----------------------------------------------------------------------------- ! Jarmo Sorvari Control Engineering Laboratory ! ! j...@ae.tut.fi Tampere University of Technology ! --------------------------------------- BOX 527, 33101 Tampere, Finland ----- Onar Aam Nov 9 1992, 6:13 am show options Newsgroups: rec.games.chess From: o...@hsr.no (Onar Aam) - Find messages by this author Date: Mon, 9 Nov 1992 16:48:47 GMT Local: Mon, Nov 9 1992 8:48 am Subject: Re: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse - Hide quoted text - - Show quoted text - >In article balje...@uther.calvin.edu (Alan Baljeu) writes: >>s...@cis.uab.edu (Kenneth Sloan) writes: >>>What are the *goals* of the ICS system? Is the intent to match USCF >>>ratings? Player-by-player (for those with ratings in each system)? >>>Statistically? How detailed? Mean and std dev? >>If you ask me, I think a good goal is to have a player of constant strength >>have a constant rating. Then a person can tell if (s)he is improving, >>staying the same, or going senile. Other than that, it doesn't really matter >>to me. >Good. That's a reasonable goal. Now - do you have any ideas on how to >achieve that goal? I don't. >-- >Kenneth Sloan Computer and Information Sciences >s...@cis.uab.edu University of Alabama at Birmingham Set the provisional rating to the average! That solves the problem. Then this rating will become an attractor for the average. (Or we could just wait until the average reaches the current provisional rating, 1000) Onar. Kenneth Sloan Nov 9 1992, 6:18 am show options Newsgroups: rec.games.chess From: s...@cis.uab.edu (Kenneth Sloan) - Find messages by this author Date: Mon, 9 Nov 1992 19:23:51 GMT Local: Mon, Nov 9 1992 11:23 am Subject: Re: The ICS rating system (was Re: Why I Hate the ICS rating system.) Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse In article j...@cs.tut.fi (Sorvari Jarmo) writes: >... >How about if the opponents of novices would not be affected at all? >So that after 20 games, new players would enter the pool with a more >realistic approximation (based on the +/- 400 rule) of their strength >... Under slightly idealistic assumptions, this is what happens under the USCF system - to the extablished players *as a group*. Surely, a player's rating ought to drop when he loses a game, and ought to rise when he wins. The problem, with unrated players, is how much. Assume: an unrated player's strength does not change appreciably during the provisional period. (N.B. - you may not want to make this assumption0 Then - the USCF system requires that you first rate the UNRATED players in a tournament, and then use their post-event ratings as their pre-event ratings, for purposes of rating everyone else. So, in the extreme case where an unrated plays only one game against a rated player, the rated player wins/loses the absolute minimum number of rating points from that game. Essentially, he wins/loses 0. [assume the unrated player beats a 1600 established player. His provisional rating is 2000, and the 1600 player loses the same number of points he would ordinarily lose by being defeated by a 2000-rated player - for all intents and purposes, 0]. The unrated player gets a provisional rating of 1600. If he continues to play at 1600 strength, then the rating points won/lost against other established players will be correct. Minor variations in performance result in rating points being taken from some established players, and then re-distributed to other established players. Some people are only convinced of this after actually crunching through the numbers - so I'll make an offer. If you will construct a "history" [up to 21 games - broken up into tournaments of at least 3 rounds each] for a new player, I will be happy to crunch the numbers and demonstrate the effect of that history on both the unrated player, and all of his opponents. Feel free to construct as strange a history as you like - but I reserve the right to comment on the likelihood of the history. I am confident that it will be difficult to construct a history which has any appreciable "incorrect" effect on either an individual established player, or on the established player pool as a whole. >I don't think new players would find it difficult to find opponent >willing to play them. A potential problem with your proposes scheme is that many players do not want to play, or do not take the games "seriously" if rating points are not at stake. The established player would literally have "nothing to lose" and this might affect their strength. But, this is probably a minor nit. For a system with the apparent constraints of ICS (single game matches, rated game-by-game), I would agree that it would be acceptable for individual opponents of unrated players to see no rating change whatsoever. Changing established ratings based on an "assigned" provisional rating is simply wrong, unless the goal is to attract the `average rating' to that value. In that case, there isn't much point in even distinguishing between `provisional' and `established' ratings - except possibly to select an appropriate K, as FIDE does. Notice that FIDE is different that ICS in that there is a certain minimum competence required of new FIDE-rated players, and a certain assurance that a new FIDE-rated player is reasonably close to that minimum competence level. Players of 900 strength do not enter FIDE-rated tournaments, and players of 2900 strength who have never competed in a FIDE tournament simply don't exist (well, no humans, anyway...) -- - Hide quoted text - - Show quoted text - Kenneth Sloan Computer and Information Sciences s...@cis.uab.edu University of Alabama at Birmingham (205) 934-2213 115A Campbell Hall, UAB Station (205) 934-5473 FAX Birmingham, AL 35294-1170 Kenneth Sloan Nov 9 1992, 8:05 am show options Newsgroups: rec.games.chess From: s...@cis.uab.edu (Kenneth Sloan) - Find messages by this author Date: Mon, 9 Nov 1992 21:10:11 GMT Local: Mon, Nov 9 1992 1:10 pm Subject: Re: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse - Hide quoted text - - Show quoted text - In article <1992Nov9.164847.4...@hsr.no> o...@hsr.no (Onar Aam) writes: >>In article balje...@uther.calvin.edu (Alan Baljeu) writes: >>>s...@cis.uab.edu (Kenneth Sloan) writes: >>>>What are the *goals* of the ICS system? Is the intent to match USCF >>>>ratings? Player-by-player (for those with ratings in each system)? >>>>Statistically? How detailed? Mean and std dev? >>>If you ask me, I think a good goal is to have a player of constant strength >>>have a constant rating. Then a person can tell if (s)he is improving, >>>staying the same, or going senile. Other than that, it doesn't really matter >>>to me. >>Good. That's a reasonable goal. Now - do you have any ideas on how to >>achieve that goal? I don't. >>-- >>Kenneth Sloan Computer and Information Sciences >>s...@cis.uab.edu University of Alabama at Birmingham >Set the provisional rating to the average! That solves the problem. Then this >rating will become an attractor for the average. (Or we could just wait until the >average reaches the current provisional rating, 1000) Nice try. Let's see how that works under the following assumptions: 0) you enter the pool, with your average strength and average ability. Suppose there are 10,000 players in this pool, all with perfect ratings, and your strength entitles you to the average rating. 1) the system becomes popular, and 20,000 new players enter. Alas, chess players just aren't as good as they used to be, so the average strength of the 20,000 new players is well below the average strength of the existing pool. So - what happens to *your* rating, assuming that your strength does not change? Calculate it. Your goal was to have your rating remain the same, because your strength remains the same. Does your proposal meet that goal? Read Elo! If not for enlightenment, then as a tribute. -- Kenneth Sloan Computer - Hide quoted text - - Show quoted text - and Information Sciences s...@cis.uab.edu University of Alabama at Birmingham (205) 934-2213 115A Campbell Hall, UAB Station (205) 934-5473 FAX Birmingham, AL 35294-1170 Michael Schmahl [Black-Robe Mage] Nov 9 1992, 2:06 pm show options Newsgroups: rec.games.chess From: f...@camelot.acf-lab.alaska.edu (Michael Schmahl [Black-Robe Mage]) - Find messages by this author Date: 10 Nov 92 02:01:00 GMT Subject: Possible fixes for ICS (Was: Why I Hate the ICS rating system.) Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse Richard V. Nash - Hide quoted text - - Show quoted text - writes > In article <1992Nov3.195552.13...@oucsace.cs.ohiou.edu> > jacob...@oucsace.cs.ohiou.edu (Eliot T. Jacobson ) writes: > > RAISE THE INITIAL PROVISIONAL RATING FROM 1000 TO SOMETHING MORE REALISTIC. > Unless the current administrators have changed something, there is _no_ initial > provisional rating. You only acquire a rating after you play your first rated > person. If you win you are their rating + 400, if you loose you are their > rating - 400, or if you draw you get their rating. > I do agree that ICS ratings are way below USCF or FIDE ratings. I also agree > that the average rating seems to be dropping. I think it is because more weak > players (like me) have been joining. > Anyway, the code that does the ratings calculation is simple and I will provide > it to anyone who asks. I have an idea which may help.... This should work like USCF's quick-rating and correspondence system. During the registration process, you should ask for a player's USCF or FIDE rating, and if they have one, use this rating b/10. Also, have the first 10 or 15 games count toward both the regular and blitz ratings. Also I have noticed that there is a base rating for when you assess a new player. I get a negative draw value. I think either a) calculate the new player first, or b) use a 1500 base value... Hope this helps. -- The secretary says he's on another line. Can I hold for a long, long time? -TMBG Michael Schmahl a.k.a. Black-Robe Mage Michael Schmahl [Black-Robe Mage] Nov 9 1992, 2:14 pm show options Newsgroups: rec.games.chess From: f...@camelot.acf-lab.alaska.edu (Michael Schmahl [Black-Robe Mage]) - Find messages by this author Date: 10 Nov 92 02:16:51 GMT Local: Mon, Nov 9 1992 6:16 pm Subject: Re: The ICS rating system (was Re: Why I Hate the ICS rating system.) Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse Onar Aam writes > Someone mentioned that there was a skew problem even when the provisional rating > was 1500. The reason for this is unclear. I cannot imagine that the rating system > is intrinsically skewish since it is directly based on the ELO rating system. > Therefore there must be other reasons. But putting the provisional rating back up > to 1500 will definitely help. That I can guarantee. The ELO scale is a sliding scale which means that only the differences in rating matter, not the actual ratings themselves. So the only thing you can glean from the numbers is how you rate with respect to the other players. However, I agree that it would be a Good Thing [tm] if ICS ratings approximated USCF ratings. Perhaps just add about 200-300 points to the current ratings, or use my earlier suggestion. -- - Hide quoted text - - Show quoted text - The secretary says he's on another line. Can I hold for a long, long time? -TMBG Michael Schmahl a.k.a. Black-Robe Mage Onar Aam Nov 9 1992, 5:49 pm show options Newsgroups: rec.games.chess From: o...@hsr.no (Onar Aam) - Find messages by this author Date: Tue, 10 Nov 1992 06:34:23 GMT Local: Mon, Nov 9 1992 10:34 pm Subject: Re: Why I Hate the ICS rating system. Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse - Hide quoted text - - Show quoted text - >>Set the provisional rating to the average! That solves the problem. Then this >>rating will become an attractor for the average. (Or we could just wait until the >>average reaches the current provisional rating, 1000) >Nice try. Let's see how that works under the following assumptions: >0) you enter the pool, with your average strength and average ability. > Suppose there are 10,000 players in this pool, all with perfect > ratings, and your strength entitles you to the average rating. >1) the system becomes popular, and 20,000 new players enter. Alas, > chess players just aren't as good as they used to be, so the average > strength of the 20,000 new players is well below the average > strength of the existing pool. So - what happens to *your* rating, > assuming that your strength does not change? Calculate it. >Your goal was to have your rating remain the same, because your strength >remains the same. Does your proposal meet that goal? >Read Elo! If not for enlightenment, then as a tribute. >-- >Kenneth Sloan Computer and Information Sciences >s...@cis.uab.edu University of Alabama at Birmingham >(205) 934-2213 115A Campbell Hall, UAB Station >(205) 934-5473 FAX Birmingham, AL 35294-1170 I don't need to calculate it, I know what would happen. My rating would rise. Onar.