@MaX_sSk said in #48:
> chess.com ratings tend to be lower than lichess ratings. At chess.com, players start out at 1200, whereas on lichess they start out at 1500. That means that, even if both sites had the same pool of players, average players would tend to be about 300 points higher on lichess than on chess.com.(according to Google)
The initial starting point should not matter much in the long run. That's because a stronger player will quickly rise up. The related argument about how lots of new players assigned a high Elo of 1500 has inflated Fide Elo was also shown by Ken Regan, professor of statistics and Fide expert, to be false. In fact there's rating deflation over the last few decades, not rating inflation.
As for what "blunderederxd" says, about the Glicko vs traditional Elo rating formula, that also doesn't matter in the long run since Glicko simply adjusts faster. It's like having a higher "K-factor" in the Elo formula. That's why FIDE had such pushback on their scheme to award junior players a higher K-factor a few years ago, as it made them look like ...Hans Niemann. But in the long run, having a high K-factor, or rising quickly as in the Glicko system, won't make a patzer into a grandmaster.
Edit: in fact, at Chess.com, it is stated: "Chess.com Chess.com uses the Glicko 1 system. New players used to get an initial rating of 1200. Nowadays, new players get to choose their initial rating themselves." So, "nowadays", you can pick your own rating. So let's say you pick your initial rating to be 3000 Elo. After a few days, weeks or months, you should quickly settle to your true rating, and that would be plus or minus some constant from the mean. That mean would be say 1400 Elo on Chess com. At Lichess that mean would be say 1500 Elo.
The initial question still stands: why is Lichess average Elo seemingly about 100 points higher? The only thing I can think of is perhaps LiChess has a different pool of players...but are they stronger or weaker? Does the inflated Elo at LiChess mean they are weaker? Or perhaps stronger? Or perhaps no telling? Probably no telling. After all, a pool of grandmasters, closed to the public, might under the Elo formula have an average rating of 1500 Elo, but that would only be because the pool is closed to the public. Under this scenario GM Magnus Carlsen, who is about 400 points higher than the average GM, would be 1500+400 = 1900 Elo, while GM Lev Aronian, who presently is about 100 points weaker than Carlsen, would be 1800 Elo, and an average GM would be, as I said, 1500 Elo.
But, here's a thought experiment, what if there were two pools, and some grandmasters played in both pools, but the second pool had patzers in it? What would be the rating? So if GMs played in one closed pool, and this same group of GMs were to play in another closed pool, but patzers also played in this second closed pool, would average Elo rise or drop? I'm not sure, but I think average Elo might have a 'bimodal' distribution--two humps--but over time the Elo might be higher in the second pool?!
So under this scenario, LiChess Elo being higher means the LiChess pool of players is stronger than Chess com players!!
In short, for the above thought experiment:
LiChess pool: super GMs, very strong players, and a few patzers --> a relatively high average Elo
Chess dot com pool: some strong players, a few GMs and lots and lots of patzers, with a number of people playing in both pools so you cannot say they are independent of one another --> a relatively lower average Elo
Conclusion: Lichess Elo being higher than Chess com means Lichess has stronger players than Chess com! :)
Seems fair and logical to me...under the assumptions above.
@XsYyLaxa said in #51:
> The initial starting point should not matter much in the long run. That's because a stronger player will quickly rise up. The related argument about how lots of new players assigned a high Elo of 1500 has inflated Fide Elo was also shown by Ken Regan, professor of statistics and Fide expert, to be false. In fact there's rating deflation over the last few decades, not rating inflation.
>
> As for what "blunderederxd" says, about the Glicko vs traditional Elo rating formula, that also doesn't matter in the long run since Glicko simply adjusts faster. It's like having a higher "K-factor" in the Elo formula. That's why FIDE had such pushback on their scheme to award junior players a higher K-factor a few years ago, as it made them look like ...Hans Niemann. But in the long run, having a high K-factor, or rising quickly as in the Glicko system, won't make a patzer into a grandmaster.
>
> Edit: in fact, at Chess.com, it is stated: "Chess.com Chess.com uses the Glicko 1 system. New players used to get an initial rating of 1200. Nowadays, new players get to choose their initial rating themselves." So, "nowadays", you can pick your own rating. So let's say you pick your initial rating to be 3000 Elo. After a few days, weeks or months, you should quickly settle to your true rating, and that would be plus or minus some constant from the mean. That mean would be say 1400 Elo on Chess com. At Lichess that mean would be say 1500 Elo.
>
> The initial question still stands: why is Lichess average Elo seemingly about 100 points higher? The only thing I can think of is perhaps LiChess has a different pool of players...but are they stronger or weaker? Does the inflated Elo at LiChess mean they are weaker? Or perhaps stronger? Or perhaps no telling? Probably no telling. After all, a pool of grandmasters, closed to the public, might under the Elo formula have an average rating of 1500 Elo, but that would only be because the pool is closed to the public. Under this scenario GM Magnus Carlsen, who is about 400 points higher than the average GM, would be 1500+400 = 1900 Elo, while GM Lev Aronian, who presently is about 100 points weaker than Carlsen, would be 1800 Elo, and an average GM would be, as I said, 1500 Elo.
>
> But, here's a thought experiment, what if there were two pools, and some grandmasters played in both pools, but the second pool had patzers in it? What would be the rating? So if GMs played in one closed pool, and this same group of GMs were to play in another closed pool, but patzers also played in this second closed pool, would average Elo rise or drop? I'm not sure, but I think average Elo might have a 'bimodal' distribution--two humps--but over time the Elo might be higher in the second pool?!
>
> So under this scenario, LiChess Elo being higher means the LiChess pool of players is stronger than Chess com players!!
>
> In short, for the above thought experiment:
>
> LiChess pool: super GMs, very strong players, and a few patzers --> a relatively high average Elo
>
> Chess dot com pool: some strong players, a few GMs and lots and lots of patzers, with a number of people playing in both pools so you cannot say they are independent of one another --> a relatively lower average Elo
>
> Conclusion: Lichess Elo being higher than Chess com means Lichess has stronger players than Chess com! :)
>
> Seems fair and logical to me...under the assumptions above.
Chess.com 1200 has to play lichess 1500 and that will tell the rating difference
Lichess uses a different rating system.Glicko-2 and Chess.com uses.elo
@MaX_sSk said in #52:
> Chess.com 1200 has to play lichess 1500 and that will tell the rating difference
No because chess com has changed it. as I said: "in fact, at Chess.com, it is stated: "Chess.com Chess.com uses the Glicko 1 system. New players used to get an initial rating of 1200. Nowadays, new players get to choose their initial rating themselves." So, "nowadays", you can pick your own rating. So let's say you pick your initial rating to be 3000 Elo. After a few days, weeks or months, you should quickly settle to your true rating, and that would be plus or minus some constant from the mean. That mean would be say 1400 Elo on Chess com. At Lichess that mean would be say 1500 Elo."
@Shadow_Warrior_14 said in #53:
> Lichess uses a different rating system.Glicko-2 and Chess.com uses.elo
No that only goes to how quickly your rating changes, like the K-factor in Elo.
@GnocchiPup said in #42:
> You have to look at this from the POV of a server.
>
> 12. So it's useless to compare ratings across sites. As long as someone rated +200 compared to you on any site, within that site, beats you 75% of the time, the ratings are accurate.
I agree except for #12 [edit in brackets: but, if you read the below, I conclude by agreeing with #12, edit to my original reply].
The issue is whether you can correlate, assuming the same people (or similar people) play across both servers, chess com and Lichess, what the distribution of Elos are between the two "camps", two "pools" or two "servers" (they would be represented by a least-squares linear line, y(x) =mx + b, Y axis being one pool and X axis for the other pool) Mathematically there's no reason you cannot. If two Elo pools are relatively open so players can play in two pools, A and B, what would the different average Elos in the two pools mean? [but the different populations having different Elos proves nothing special necessarily about how strong the players are between the different populations, I conclude below]
If you have different average Elos, the question then is why the populations are different. Might lots of 'weak' players in one pool inflate the Elos for those players who are stronger? But then these stronger players might have a lower Elo in the other pool? That appears to be the case between LiChess and Chess com [maybe]. But it's not clear to me which population is weaker [true, it's indeterminable]. I am leaning towards Chess com having lower rated players at levels say below 2200 Elo as it's more popular, and these days popular means weaker (newbies have Elos well below 1000). So perhaps LiChess Elos are higher than Chess com Elos because the players here, below 2200 Elo, are stronger than at Chess com? [maybe, no way to really know] Recall the rule of thumb is that Chess com server Elos are roughly 100 points lower than LiChess server Elos, but, the higher the Elo, the more they converge, and the two pools have Elos cross at about 2200 Elo say some [but, this cross-over proves nothing, other than as with temperatures there's a temperature that's the same both in Farenheit as in Celcius, i.e., -40 degrees is the same on both scales].
But thinking through this, I'm not sure you can say anything at all about the different populations, other than their Elos can be correlated to one another, but this correlation doesn't prove anything as to which population is stronger. [true]
[this next paragraph shows a historical accident it seems, not a truism] It's generally agreed that USCF Elos are 'inflated' by about 100 points, historically, compared to Fide (however at higher Elos the Elo differences of the two camps converge). That means a Fide international player who is rated 2000 in EU/outside the USA would find it "easier" to play in the USA, and after a while the Fide player would have, in the USA, a USCF Elo of about 2100. From Google: "[FIDE] Elo and USCF ratings are roughly equal, with USCF ratings running 100–200 points higher". Or think of what would happen to Elos at the top end of LiChess if the top ten best chess players in the world suddenly were to play at LiChess. I'm pretty sure the average Elo of the top 0.1% of LiChess players would rise. [However, this is just a historical accident, and it's possible the USCF Elo could be inflated *and* stronger than Fide Elo, and though historically that did not happen, it is possible it could happen]
Or imagine you, a 1500 Fide Elo player, playing a bunch of tyros who just learned chess. Suppose the average Elo of these players is say 500 Elo in some rating pool. After a while, your Elo would inflate to a much higher number than 1500 Elo Fide since you to them would be like Nakamura playing any of us. This is a how a prisoner named Claud Bloodgood was able to inflate his USCF Elo (Wikipedia: "Claude Frizzell Bloodgood III (alias: Klaus Frizzel Bluttgutt III; July 14, 1937 – August 4, 2001) was a controversial American chess player. ... Bloodgood organized chess games within Powhatan Correctional Center in Powhatan, Virginia, which were by necessity with fellow inmates.[3] Many of these inmates were taught the game by Bloodgood, and thus began as unrated and inexperienced players. Bloodgood obtained USCF memberships for them. Some accused Bloodgood, with his intimate knowledge of the rating system, of rigging their ratings. The accusation was that he arranged for new prisoners to play rated games against other prisoners, who would deliberately lose, thus giving the new inmate an inflated USCF rating. Bloodgood, it is further alleged, then played rated games against the new highly rated prisoner, and each time he won, gained a few more rating points. This continued for several years, and by 1997 his rating rose to 2759, making the 59-year-old Bloodgood the second highest rated player in the nation, behind only Gata Kamsky. ")
This example clearly shows that within a pool you can have players of different Elo than another pool. Nobody will deny Kamsky was much stronger than Bloodgood. [true, but you don't know about the rest of the players in each pool] So it raises the question: which pool of players was stronger, the USCF pool of players or the Powhatan Correctional Center pool of players? Clearly the former? [Not necessarily] I'm picking Kamsky's Elo was stronger than Bloodgood's Elo [that is nearly certainly true for the top two players, but not necessarily true for the other players]. So therefore you can plot the line of Elos (least squares style) between the Powhatan CC pool and the USCF pool (without the Powhatan prisoners), and come up with a formula for the line, y = mx + b, with m = slope of the line and b being a constant, representing the two pools of players, from the USCF and from the Powhatan Correctional Center. Then you can compare the line. At the top end, we can say that the Powhatan Correctional Center Elo is inflated and the top rated player, Bloodgood, is weaker than an equivalent rated USCF player. [But aside from the top end of the line, there's not much we can say about the Elo strength of the rest of the players, other than come up with a mathematical correlation]
Suppose Bloodgood was say an expert player of 2000 Elo (USCF). Then compared to Kamsky, the Elos are: Powhatan CC + 759 = USCF Elo (2000+759 = 2759 Elo, about the same as Kamsky). But what is the slope of the Powhatan CC Elo (least squares)? For all we know, those prisoners were better than most average USCF players, if they studied chess all day and practiced? [true] And if they deliberately lost to Bloodgood, there's no telling how strong these prisoners were. [true]
This has nothing to do with the odds a player rated about 200 points higher than another player in any given pool can beat another player (it's always about 75%)
In conclusion, I don't think anything can be said about the relative strength of Elos between different pools, so the conclusions at https://chessgoals.com/rating-comparison/#Lichess_Rating_vs_Chesscom are out of date. You will certainly have a least square lines, y = mx + b, between the two pools, but the line can change month to month and does not prove which pool is stronger. The higher rated pool can be stronger or weaker than the lower rated pool. [true]
[true] After all of this I think we're back to square one and that is though you can correlate one rating pool to another, using a table or a least square line, there's no guarantee that the Elo of one pool being lower or higher means the players of that pool are stronger or weaker than the other pool. For all we know, the LiChess Elo being higher by 100 points at various Elos to the Chess com Elo means LiChess players are stronger (given Chess com has lots of newbies). Or it could be the opposite (which is what Chess com fans like to say in their marketing that Chess com is stronger than LiChess). But at the top end the Chess com Elos being so high could mean the players there are stronger (at the 0.1% of players), given that Carlsen and Nakamura play there. Then again, it could mean the opposite on average (aside from MC and Hikaru). There's really no telling.
@XsYyLaxa said in #56:
> I agree except for #12 [edit in brackets: but, if you read the below, I conclude by agreeing with #12, edit to my original reply].
>
> The issue is whether you can correlate, assuming the same people (or similar people) play across both servers, chess com and Lichess, what the distribution of Elos are between the two "camps", two "pools" or two "servers" (they would be represented by a least-squares linear line, y(x) =mx + b, Y axis being one pool and X axis for the other pool) Mathematically there's no reason you cannot. If two Elo pools are relatively open so players can play in two pools, A and B, what would the different average Elos in the two pools mean? [but the different populations having different Elos proves nothing special necessarily about how strong the players are between the different populations, I conclude below]
> [true] After all of this I think we're back to square one and that is though you can correlate one rating pool to another, using a table or a least square line, there's no guarantee that the Elo of one pool being lower or higher means the players of that pool are stronger or weaker than the other pool. For all we know, the LiChess Elo being higher by 100 points at various Elos to the Chess com Elo means LiChess players are stronger (given Chess com has lots of newbies). Or it could be the opposite (which is what Chess com fans like to say in their marketing that Chess com is stronger than LiChess). But at the top end the Chess com Elos being so high could mean the players there are stronger (at the 0.1% of players), given that Carlsen and Nakamura play there. Then again, it could mean the opposite on average (aside from MC and Hikaru). There's really no telling.
Having said all of the above, the more I think about it, the more I feel that if you have enough Fide rated people who play in both pools, LiChess and Chess com, you can figure out which of the two pool are objectively stronger, but you would need a large sample of Fide rated people who play in both pools.
So here's the hypothetical: Pool A = Chess com, Pool B = LiChess. What happens if Pool A, Pool B have exactly the same players? They will have the same Elo distribution. Then add more very weak players to pool A. What happens? Average Elo will drop, with a fatter (more pronounced) left tail in the bell-shaped curve. Now add many more very strong players to pool A. Average Elo will again rise, with a fatter right tail (variance will widen too). Do the same for pool B but add "fewer" very weak players and "much fewer" very strong players. If you now divide the distributions about the mean, the result might be that average Elo for the left side of the mean will be higher for pool B than pool A (since pool B has "fewer" very weak players than pool A), however, on the right side of the curve, pool B will have a lower average Elo than pool A for the right side of the mean (since pool B has "much fewer" very strong players). The Gaussian curves will be flatter and the mean of each curve depends on how many "fewer" and "much fewer" players are added.
This hypothetical would mean LiChess (pool B) has stronger "ordinary players" (hence the Elo is higher than at Chess com) up to the master level (titled players greater than 2200 Elo Fide) but weaker titled players (above 2200 Elo). This is consistent with the idea that Chess com caters to "newbies" more, but, since they have marketing clout, can also afford to lure extremely strong players like Carlsen, Caruana, Nakamura and the like.
But this hypothetical requires statistical analysis of data that are not available to the public. You would need to have a third pool, Fide Elo, and see how these Fide players distributed over both LiChess and Chess.com, and have lots of these players, certainly more than what was done in the simple surveys online.
@XsYyLaxa said in #57:
> Having said all of the above, the more I think about it, the more I feel that if you have enough Fide rated people who play in both pools, LiChess and Chess com, you can figure out which of the two pool are objectively stronger, but you would need a large sample of Fide rated people who play in both pools.
Complicating everything, even if you were to do such a simulation it might turn out, depending on how you distribute the 'strong' and 'weak' players, that average Elos drop even as players get stronger.
I once read that average Elos have dropped over the years ("rating deflation") even as chess players have gotten stronger (the assumption being over time chess players have gotten stronger, since they use engines to check their openings).
So in the end the safest thing to say with confidence is that different rating pools can be correlated but not compared easily. That's to say, you can say that the #1 player of the early 1970s, Bobby Fischer, had an Elo of 2789.7 at their peak and today's #1, Magnus Carlsen, has a peak of 2889.2 Elo, and come up with a formula or table for Elos back then to Elos today, however it's really hard to say who is stronger without further analysis like "average centipawn loss" (where I've seen different things, some studies say Carlsen while others say Fischer was stronger at their peak). Same would be true for comparing whether LiChess or Chess com is stronger.
This topic has been archived and can no longer be replied to.