Ok, so if we have 100 players who all have established rating on chess.com, but have never played OTB chess, and as a test we put them to play a tournament against each other OTB. Let's say 25 of the players have 2000+ online rating, while 25 of the players have 1750 online rating, 25 of the players have 1500 and 25 of the players have 1250. Assuming there is no correlation between OTB and online chess, we should expect that approximately 5 players from each rating range score in the top 20, as well as in the bottom 20. Does someone seriously believe that?

Or maybe the majority of 2000+ rated online players will be in top 20, with maybe few occasional lower rated players, who happen to be either playing a surprisingly good tournament (or just be genuinely better in OTB situations), and majority of 1250 rated players will be in bottom 20, with maybe occasional higher rated player who can't handle OTB situation. That wouldn't only prove that there is obvious correlation between OTB and online ratings, BUT also causality between OTB and online playing skills.

I'd go with the latter, but ya'll free to argue.

# whats equivalent to chess.com and uscf rating?

You are mixing up statistical correlation and the possibility to draw individual conclusions from said statistics. The correlation between online, FIDE, USCF and whatever ratings is not very weak, it's actually very strong. Again, this has been proven multiple times, when comparing large enough sample size of peoples ratings in different rating pools.

However, there are always exceptions to every statistic, and indeed a player could do much better in either OTB or online chess, and thus have quite a big difference in those two ratings. But on average majority of players have OTB and online ratings following in tandem (of course adjusted to the average rating system in place). And you are right in sense, that there is no way to make a conversion that if your chess.com blitz rating is 1400, then your OTB rating is around 1500. But that has nothing to do with whether these ratings have correlation or not. I know it's semantics at this point, but you can't claim that the correlation is weak just because it's not possible to make a fool-proof conversion between two ratings, if person has only one of them rated.

There is a lot of incorrect stuff in your post, but I'll focus on the part that is pertinent to the topic at hand.

The correlation is weak because the margin of error is wide. Saying "an online rating of X would mean a OTB rating of X +/- 300 points" saying (literally) "your OTB rating would be within 6 standard deviations around your online rating". That is, you can lump 95% of the entire population into the same grouping! When you graph such a distribution, the error bars cover most of the area!

You see the same kind of distribution for ELO estimators (e.g. elometer.net). If you make the range wide enough, you can never be wrong.

You are mixing up statistical correlation and the possibility to draw individual conclusions from said statistics. The correlation between online, FIDE, USCF and whatever ratings is not very weak, it's actually very strong. Again, this has been proven multiple times, when comparing large enough sample size of peoples ratings in different rating pools.

However, there are always exceptions to every statistic, and indeed a player could do much better in either OTB or online chess, and thus have quite a big difference in those two ratings. But on average majority of players have OTB and online ratings following in tandem (of course adjusted to the average rating system in place). And you are right in sense, that there is no way to make a conversion that if your chess.com blitz rating is 1400, then your OTB rating is around 1500. But that has nothing to do with whether these ratings have correlation or not. I know it's semantics at this point, but you can't claim that the correlation is weak just because it's not possible to make a fool-proof conversion between two ratings, if person has only one of them rated.

There is a lot of incorrect stuff in your post, but I'll focus on the part that is pertinent to the topic at hand.

The correlation is weak because the margin of error is wide. Saying "an online rating of X would mean a OTB rating of X +/- 300 points" saying (literally) "your OTB rating would be within 6 standard deviations around your online rating". That is, you can lump 95% of the entire population into the same grouping! When you graph such a distribution, the error bars cover most of the area!

You see the same kind of distribution for ELO estimators (e.g. elometer.net). If you make the range wide enough, you can never be wrong.

Not sure where you got that +-300 points thing, with chess.com blitz especially it's on average much closer than that. The standard deviations are within +-200. I know 400 points is a lot, but the point never was to claim that one could deviate one rating from another. But you can get estimates, which apply to vast majority of the players, and bigger the sample the smaller the variation is. The whole debate was whether there was correlation between the statistics or not (and it's 0.7+ on 0 to 1 scale, that's quite clear correlation). And in a sense it's also useable to compare the relative strengths of two different rating pools.

How much use is there for an individual interested in what their FIDE or USCF might be? Not all that much, but getting a +-200 points estimation on high probability is better than nothing.

As this user explains quite concisely, the mainstream theory thus far, generally referred to as the "rule of adding 700," is that one's FIDE rating is typically somewhere around one's blitz rating on chess.com + 700. For instance, if someone has a blitz rating of 1800 on chess.com, you can expect that person to have a FIDE rating of around 2500.

macer, macer, macer .... totally serious as always!

well, i will not be challinjin carlsin for the title in the near future with my big rating.

Not sure where you got that +-300 points thing, with chess.com blitz especially it's on average much closer than that. The standard deviations are within +-200. I know 400 points is a lot, but the point never was to claim that one could deviate one rating from another. But you can get estimates, which apply to vast majority of the players, and bigger the sample the smaller the variation is. The whole debate was whether there was correlation between the statistics or not (and it's 0.7+ on 0 to 1 scale, that's quite clear correlation). And in a sense it's also useable to compare the relative strengths of two different rating pools.

How much use is there for an individual interested in what their FIDE or USCF might be? Not all that much, but getting a +-200 points estimation on high probability is better than nothing.

First, the standard deviation of the Glicko/ELO system is 100 points. The post you are referring to is not using "standard deviation" in terms of statistical analysis (it is just the average difference from the chess.com rating and the user's self-reported FIDE rating - which is not always accurate). He is also looking at more people on the high end of the rating ladder (as the mean for each is much higher than it should be - granted, FIDE's mean is higher than it should be naturally since they do not rate below 1000, so even though their system is designed to have a mean of 1200, it will actually be ~1600-1800 simply because they self-skew the data).

The issue I have is with the "estimates" (as I've said several times). If your range is +/- 2-3 standard deviations (i.e. +/- 200-300 points), the estimate is meaningless as you are basically including more than 85-90% of the possible range as their "estimated" range. Even +/- 1 standard deviation covers 68% of the possible range. Put another way: If you go to your doctor and ask "what is the likelihood I have cancer?" and he replies with "well, there is a 20% chance +/- 70%", it answers your question by telling you nothing.

People getting choked up over "well, higher OTB ratings usually means higher online ratings so obviously there is some correlation" are missing the point. The correlation coefficient is closer to .5 than it is to 1.0. So yes, it is somewhat correlated, but the correlation is so weak that any attempt to derive conclusions from it (e.g. estimates going either way) are meaningless; the confidence interval is simply too large.

If you could find a higher correlation coefficient for something like X + Y +/- (.5 * SD) (e.g. 1500 + 100 +/- 50 = [1550, 1650]) it would at least be somewhat useful.

The correlation gets even more distorted when you look at specific sub-sections (e.g. if you look at just the GMs and compare their FIDE rating to their online blitz ratings ... you see some 2500s with 2800+ ratings, with others having sub-2300 ratings - and there are usually reasons for that that go well beyond their playing strength).

Hmmmm

That is your argument? Sometimes (actually every time) you make me laugh.

I have spend many hours in a bar drinking beer. Should I claim it's a good thing?

I have spend a lot of hours doing many other meaningless nonsense(reading magazines, newspapers, just sitting, cooking things that couldn't be eaten, planting things that grow to become some very weird greenish things , etc. ). Should I claim that all these are good things and everybody must waste his time like I did? Or should I claim they are serious business and they are much better than watching a good theatrical play or reading a good book?

Man , you really are funny.

And he is still around... arguing against online chess...in a website where people enjoy playing online...

I am glad to make you laugh my friend.

Not sure where you got that +-300 points thing, with chess.com blitz especially it's on average much closer than that. The standard deviations are within +-200. I know 400 points is a lot, but the point never was to claim that one could deviate one rating from another. But you can get estimates, which apply to vast majority of the players, and bigger the sample the smaller the variation is. The whole debate was whether there was correlation between the statistics or not (and it's 0.7+ on 0 to 1 scale, that's quite clear correlation). And in a sense it's also useable to compare the relative strengths of two different rating pools.

How much use is there for an individual interested in what their FIDE or USCF might be? Not all that much, but getting a +-200 points estimation on high probability is better than nothing.

First, the standard deviation of the Glicko/ELO system is 100 points. The post you are referring to is not using "standard deviation" in terms of statistical analysis (it is just the average difference from the chess.com rating and the user's self-reported FIDE rating - which is not always accurate). He is also looking at more people on the high end of the rating ladder (as the mean for each is much higher than it should be - granted, FIDE's mean is higher than it should be naturally since they do not rate below 1000, so even though their system is designed to have a mean of 1200, it will actually be ~1600-1800 simply because they self-skew the data).

The issue I have is with the "estimates" (as I've said several times). If your range is +/- 2-3 standard deviations (i.e. +/- 200-300 points), the estimate is meaningless as you are basically including more than 85-90% of the possible range as their "estimated" range. Even +/- 1 standard deviation covers 68% of the possible range. Put another way: If you go to your doctor and ask "what is the likelihood I have cancer?" and he replies with "well, there is a 20% chance +/- 70%", it answers your question by telling you nothing.

People getting choked up over "well, higher OTB ratings usually means higher online ratings so obviously there is some correlation" are missing the point. The correlation coefficient is closer to .5 than it is to 1.0. So yes, it is somewhat correlated, but the correlation is so weak that any attempt to derive conclusions from it (e.g. estimates going either way) are meaningless; the confidence interval is simply too large.

If you could find a higher correlation coefficient for something like X + Y +/- (.5 * SD) (e.g. 1500 + 100 +/- 50 = [1550, 1650]) it would at least be somewhat useful.

The correlation gets even more distorted when you look at specific sub-sections (e.g. if you look at just the GMs and compare their FIDE rating to their online blitz ratings ... you see some 2500s with 2800+ ratings, with others having sub-2300 ratings - and there are usually reasons for that that go well beyond their playing strength).

Hey Bobby, as I can see you understand something about the topic.

people here talking about how online chess is nothing like real otb chess is like talking about how spectators watching sports fights and analyzing them without ever experiencing it themselves. most people here probably never played in any otb chess let alone coffee shop chess.