Page 1 of 2
ELO
Posted: 20 June 2022, 10:13
by sorryimlikethis
I understand how ELO gains and K Factor work, for the most part
What I don't understand is how they come to different results in these two examples.
https://boardgamearena.com/table?table=276747406
K Factor 20
+13 ELO (242 to 255) against opponents with less than 100 ELO
https://boardgamearena.com/table?table=276980349
K Factor 20
+10 ELO (241 to 251) against all opponents at or above 100 ELO
Is it a case of the K Factor being labelled incorrectly? Example 1 was my 21st game of Catan, but only 18th if you don't include timed-out games. Do timed-out games count?
Re: ELO
Posted: 20 June 2022, 10:29
by RicardoRix
Yes you're right. Something is wrong.
If you hover over the new ELO score you can see the calculation.
The 10.3 ELO gain should have been 20.6. So somewhere that has been divided by 2.
Re: ELO
Posted: 21 June 2022, 07:19
by sorryimlikethis
Something is seriously wrong. I couldn't figure out why climbing was so slow when rank #1 reached 300 ELO in 18 games. I have a higher win rate and double the amount of games and I'm still 3-4 wins behind.
My K Factor has been 30 from my very first game.
https://boardgamearena.com/table?table=275225312
K, the 'elasticity factor,' has a value that depends on the number of games you have played: 60 for your first 10 games, 40 between your 11th and 20th games, 20 for your 21st onwards.
Well, *?
https://boardgamearena.com/table?table=277239291
I am meant to gain 10.27 for beating a single player but because there were other people in the game it is less rewarding?
*moderator edit: please remember to use family-friendly language on BGA
Re: ELO
Posted: 21 June 2022, 07:28
by voriki
sorryimlikethis wrote: ↑21 June 2022, 07:19
I am meant to gain 10.27 for beating a single player but because there were other people in the game it is less rewarding?
Because a game is All vs All.
I can't speak for Catan, as I don't play it and don't know all the rules, but other games such as King of Tokyo or Bang!, everyone attacks everyone. When someone is defeated, it's not because you were the only person defeating him, but all had a part in it, all depleted his health, … So the reward for that defeat should also be shared.
Re: ELO
Posted: 21 June 2022, 07:39
by sorryimlikethis
voriki wrote: ↑21 June 2022, 07:28
sorryimlikethis wrote: ↑21 June 2022, 07:19
I am meant to gain 10.27 for beating a single player but because there were other people in the game it is less rewarding?
Because a game is All vs All.
I can't speak for Catan, as I don't play it and don't know all the rules, but other games such as King of Tokyo or Bang!, everyone attacks everyone. When someone is defeated, it's not because you were the only person defeating him, but all had a part in it, all depleted his health, … So the reward for that defeat should also be shared.
Each player has their own reward, it is their own ELO gain or loss from the match. In a free for all 4th place helps ensure 2nd and 3rd place aren't winners by blocking or stealing from them, but they aren't rewarded for that. They suffer an ELO loss against everyone.
Having more opponents does not make a game easier, in a free for all it makes you less likely to finish 1st. It doesn't make sense to give out less ELO for a harder task.
Re: ELO
Posted: 21 June 2022, 07:43
by Romain672
Don't understand the first link neither.
Yes, this is because you can gain thrice the normal elo gain in 4p games.
If you check a game like Saboteur or Secret moon, where there is 8 players, it's game with lots of rng, and so if you win 8 times the normal elo, your elo will just mean nothing.
More there is players, less you have control of what is going on since there is more unknown between your turns.
Catan I suppose is still a game where you can manage it correctly, and playing against someone is hard to do, but in many others games, that's not the case.
sorryimlikethis wrote: ↑21 June 2022, 07:39Each player has their own reward, it is their own ELO gain or loss from the match. In a free for all 4th place helps ensure 2nd and 3rd place aren't winners by blocking or stealing from them, but they aren't rewarded for that. They suffer an ELO loss against everyone.
Having more opponents does not make a game easier, in a free for all it makes you less likely to finish 1st. It doesn't make sense to give out less ELO for a harder task.
Oh yeah it does. If you don't divide your gain, you will gain thrice the amount of elo on a 1st place in 4p than a 1st place in 2p.
Re: ELO
Posted: 21 June 2022, 07:52
by sorryimlikethis
Romain672 wrote: ↑21 June 2022, 07:43
Oh yeah it does. If you don't divide your gain, you will gain thrice the amount of elo on a 1st place in 4p than a 1st place in 2p.
Losses aren't divided. If I get 4th right now I will take the full ELO loss from every other player.
Re: ELO
Posted: 21 June 2022, 08:04
by dschingis27
There are several things to say here.
K, the 'elasticity factor,' has a value that depends on the number of games you have played: 60 for your first 10 games, 40 between your 11th and 20th games, 20 for your 21st onwards.
One rationale here is that players who already know a game in real life are getting a proper Elo much quicker than if the K factor were constant.
I am pretty sure that aborted games do not count towards this counter. So your first example game probably counted as your 18th game with a K factor of 40 under the hood. In your second example, you had a K factor of 20 under the hood. That's why your Elo changes now much slower than in your first games of Catan.
I write "under the hood" because there is also the c factor.
https://boardgamearena.com/forum/viewto ... =9&t=15950
This is something that is not really documented on BGA and I don't know how the c factors are for Catan. It seems that c=0.5 for 4p games of Catan.
Lastly, there are bugs with the Elo calculation display. It seems to me that what is supposed to be displayed is actually k*c which worked out for your first 20 games. It displays 30 for your first 10 games, then it displays 20 for the next 10 games, but then mistakenly continues to display 20, while in truths it is k*c=10 starting from your 21st game of Catan. That means the Elo calculation done was correct in all games according to the system implemented. However, the display was wrong. which is a problem for many years now, I think.
I am not BGA staff and I don't know for sure if any of what I write is true. It is just how all of it makes sense to me. If you are (understandably) bothered by the display issues or the general system, feel free to make a bug report or suggestion and feel free to copy/paste my explanations.
Re: ELO
Posted: 21 June 2022, 08:21
by Romain672
sorryimlikethis wrote: ↑21 June 2022, 07:52
Romain672 wrote: ↑21 June 2022, 07:43
Oh yeah it does. If you don't divide your gain, you will gain thrice the amount of elo on a 1st place in 4p than a 1st place in 2p.
Losses aren't divided. If I get 4th right now I will take the full ELO loss from every other player.
Ignore your first game. Your final gain is divided by 2, so it's like divide by 2 the gain (and loss) against each opponent.
Re: ELO
Posted: 21 June 2022, 08:34
by sorryimlikethis
dschingis27 wrote: ↑21 June 2022, 08:04
There are several things to say here.
K, the 'elasticity factor,' has a value that depends on the number of games you have played: 60 for your first 10 games, 40 between your 11th and 20th games, 20 for your 21st onwards.
One rationale here is that players who already know a game in real life are getting a proper Elo much quicker than if the K factor were constant.
I am pretty sure that aborted games do not count towards this counter. So your first example game probably counted as your 18th game with a K factor of 40 under the hood. In your second example, you had a K factor of 20 under the hood. That's why your Elo changes now much slower than in your first games of Catan.
I write "under the hood" because there is also the c factor.
https://boardgamearena.com/forum/viewto ... =9&t=15950
This is something that is not really documented on BGA and I don't know how the c factors are for Catan. It seems that c=0.5 for 4p games of Catan.
Lastly, there are bugs with the Elo calculation display. It seems to me that what is supposed to be displayed is actually k*c which worked out for your first 20 games. It displays 30 for your first 10 games, then it displays 20 for the next 10 games, but then mistakenly continues to display 20, while in truths it is k*c=10 starting from your 21st game of Catan. That means the Elo calculation done was correct in all games according to the system implemented. However, the display was wrong. which is a problem for many years now, I think.
I am not BGA staff and I don't know for sure if any of what I write is true. It is just how all of it makes sense to me. If you are (understandably) bothered by the display issues or the general system, feel free to make a bug report or suggestion and feel free to copy/paste my explanations.
Thank you, this makes sense to me. They should add the information about the c factor to the FAQ page.
What it does mean is that playing at 4 players is a pointless exercise if trying to climb Catan.
Interestingly on BGA it is recommended as a 3 player game even though on BGG it is voted as best at 4.
Romain672 wrote: ↑21 June 2022, 08:21
sorryimlikethis wrote: ↑21 June 2022, 07:52
Romain672 wrote: ↑21 June 2022, 07:43
Oh yeah it does. If you don't divide your gain, you will gain thrice the amount of elo on a 1st place in 4p than a 1st place in 2p.
Losses aren't divided. If I get 4th right now I will take the full ELO loss from every other player.
Ignore your first game. Your final gain is divided by 2, so it's like divide by 2 the gain (and loss) against each opponent.
If losses are also divided then this is fine, apologies. I was mistakenly looking at one of my earlier games which had a written K factor of 20, but it must have been higher and just labelled incorrectly.