Why is everyone's strength 800 elo higher on Fridays? • page 5/5 • General Chess Discussion • lichess.org

@CG314 said in #34:

Just to expand on Inventor_1's (correct) point about rating differences (using the standard ELO formula):

If to players of even rating play some games, each can expect to score 50% (Made up of some wins and some draws), for a score of 1:1. As the rating difference grows, this score tilts of course (the numbers are not exact, as I used liberal rounding rules, but are a fair representation, from the point of view of the weaker player):

Diff Expected% Ratio
0 50% 1 to 1
50 43% 1 to 1.3 (or about 3 to4)
100 36% 1 to 1.7 (or about 4 to 7)
200 25% 1 to 3
300 15% 1 to ~6
400 9% 1 to 10
500 ~5% 1 to 18
600 3% 1 to 32
700 1.75% 1 to 56
800 1% 1 to 99

(I get Inventor_1's 1 to 81 at a difference of 765 points, so we are using slightly different versions of the extrapolation, but the formula is badly resolved after some 500 points difference, so that distinction is not noteworthy.)

The main point is that, given this type of formula, and adding reasonable human volatility (we all have days where we are a bit better than others), for a lower rated player to distinguish between playing someone who is 400 points strong, or someone who is 800 points stronger, they would need to play at least 20 games of so (if they lost all 20, the opponent is likely more than 400 points higher, but could still be 500).

Yes. Well done! And I always use the wording “approximately...” for such calculations because I calculate using a simplified formula (it is similar to the data you named, but slightly different) in my head in a few seconds, at most (using a simplified formula, but completely sufficient for any purposes; it gives a result almost equal to reality). For example, approximately 81:1, that 99:1 – in this case, it is the same in meaning (and in all such cases in my practice, it is the same in meaning). Most people absolutely do not understand the order of values. Many of them expect in such cases something like a score of 10:1, and not 99:1. And for even greater differences in rating, the error in the expectations of the majority is even greater. I learned to make precise calculations of chess ratings and did them back in my childhood at 11-13 years old (it was 27-29 years ago). All my teddy bears at home (and many other toys) had chess ratings (regularly recalculated to the current one) for participation in home tournaments! ))))) But I was counting on Elo ratings. And on Lichess, ratings are based on the Glicko-2 algorithm, not Elo. Plus, I now know a simplified formula by which I can easily calculate in my mind an approximate score based on the difference in ratings and vice versa (approximate, but almost like in reality, equal in meaning to the real result). Which I do regularly, without spending much time on it (and in a few seconds, at most).

@CG314 said in #34: > Just to expand on Inventor_1's (correct) point about rating differences (using the standard ELO formula): > > If to players of even rating play some games, each can expect to score 50% (Made up of some wins and some draws), for a score of 1:1. As the rating difference grows, this score tilts of course (the numbers are not exact, as I used liberal rounding rules, but are a fair representation, from the point of view of the weaker player): > > Diff Expected% Ratio > 0 50% 1 to 1 > 50 43% 1 to 1.3 (or about 3 to4) > 100 36% 1 to 1.7 (or about 4 to 7) > 200 25% 1 to 3 > 300 15% 1 to ~6 > 400 9% 1 to 10 > 500 ~5% 1 to 18 > 600 3% 1 to 32 > 700 1.75% 1 to 56 > 800 1% 1 to 99 > > (I get Inventor_1's 1 to 81 at a difference of 765 points, so we are using slightly different versions of the extrapolation, but the formula is badly resolved after some 500 points difference, so that distinction is not noteworthy.) > > The main point is that, given this type of formula, and adding reasonable human volatility (we all have days where we are a bit better than others), for a lower rated player to distinguish between playing someone who is 400 points strong, or someone who is 800 points stronger, they would need to play at least 20 games of so (if they lost all 20, the opponent is likely more than 400 points higher, but could still be 500). Yes. Well done! And I always use the wording “approximately...” for such calculations because I calculate using a simplified formula (it is similar to the data you named, but slightly different) in my head in a few seconds, at most (using a simplified formula, but completely sufficient for any purposes; it gives a result almost equal to reality). For example, approximately 81:1, that 99:1 – in this case, it is the same in meaning (and in all such cases in my practice, it is the same in meaning). Most people absolutely do not understand the order of values. Many of them expect in such cases something like a score of 10:1, and not 99:1. And for even greater differences in rating, the error in the expectations of the majority is even greater. I learned to make precise calculations of chess ratings and did them back in my childhood at 11-13 years old (it was 27-29 years ago). All my teddy bears at home (and many other toys) had chess ratings (regularly recalculated to the current one) for participation in home tournaments! ))))) But I was counting on Elo ratings. And on Lichess, ratings are based on the Glicko-2 algorithm, not Elo. Plus, I now know a simplified formula by which I can easily calculate in my mind an approximate score based on the difference in ratings and vice versa (approximate, but almost like in reality, equal in meaning to the real result). Which I do regularly, without spending much time on it (and in a few seconds, at most).

CG314

#42

And indeed, I also think in Elo, not Glicko, but for the sake of this discussion, there is effectively no difference. In both cases, they used the same base... Assuming someone playing regularly, i.e. no breaks in rating periods, then RD for a player rated 1000 is of order 100, and g(RD) is above 98%... i.e. the correction in winning percentage is less then 2% (and much less for larger rating differences) between the two systems.

The primary difference between Elo and Glicko is how quickly the rating reacts to a performance difference from expectation. In Elo, the reaction is a constant (after the 6th tournament, with different values for kids, and slightly different by rating band), whereas in Glicko-2 it reacts differently for each player depending on how volatile their performance is, how regularly they play, etc. None of this changes the win expectation however.

Same here... For tournament practice, I use +100 = 65%, +200 = 75% + 300 = 85% +400 = 95%... Far from exact, but easy to remember and allows calculating in your head is a second when the pairings are posted (and while not exact compared to the "rea" formula, the difference is less than the human game-to-game variance, so its easily good enough.) And indeed, I also think in Elo, not Glicko, but for the sake of this discussion, there is effectively no difference. In both cases, they used the same base... Assuming someone playing regularly, i.e. no breaks in rating periods, then RD for a player rated 1000 is of order 100, and g(RD) is above 98%... i.e. the correction in winning percentage is less then 2% (and much less for larger rating differences) between the two systems. The primary difference between Elo and Glicko is how quickly the rating reacts to a performance difference from expectation. In Elo, the reaction is a constant (after the 6th tournament, with different values for kids, and slightly different by rating band), whereas in Glicko-2 it reacts differently for each player depending on how volatile their performance is, how regularly they play, etc. None of this changes the win expectation however.

Avyaanagra

#43

maybe the weekend vibe distracted you so you lost but on monday you were bored so you locked in