Another complaint about ELO

[detlefchef11]

In fact, I would argue that, if it is the intention of ELO to accurately evaluate player skill, as has been told to me by those defending the current system. Then number of games played should not be a factor. Because you're not looking at number of games played, you're looking at number of games played here. And, (I think) every game here is an online version of a board game. So someone could have played a game 100s of times before they joined BGA. I know I did. I played a bunch of Wingspan at home on the board and on Steam. So, the only thing that should be evaluated is how well they do and what level of player they're doing it against. Total games played on BGA is basically meaningless to anyone who recognizes that BGA is not the only place one can play these games.

Many videos games does the same thing.
They hide elo during the first 5/10 games, because they make the player move up and down way faster to put them faster to their real elo.

Imagine that you have played 100 games of game X before joining bga and your level is about 200 elo.
Instead of having the actual system: 30elo>60elo>90elo>120elo>150elo>180elo>210elo which make you get your real elo in 7 games.
You would get: 10elo>20elo...>190elo>200elo and make you reach your real elo in 20 games.
So instead of making 7 unlucky player be against you (who got a lower elo than what you should have), you would make 20 unlucky player.

So no, you are wrong there.

But, according to the link I was shown, the K factor eventually ends up at 20, not 10 as you used. Not only that, but the vast majority of players are above 100, so for at least your first games (and likely beyond that), you'll be winning games against people with a higher ELO than you (given you start at 0). So, in all honesty, you'll be at 200 relatively quickly even without exaggerating things through the K factor.

[detlefchef11]

Besides, earlier in this discussion, when I was pointing out the folly of getting less ELO for winning a 4 player game than you do a 1v1 despite the fact that, all things being equal, it's statistically harder to win a 4 player game than it is a 1v1. Thus, one could expect to climb the ranks more quickly playing 1v1.

That was dismissed with "eventually players will normalize at their appropriate ELO regardless of whether they get there playing 4 or 2 player games."

So, which is it? Are you willing to have certain players show an inaccurate ELO for a period of time or not?

[silentProtest]

https://boardgamearena.com/faq?anchor=faq_account_elo

K-factor multiplier

Then it's broken. Because, in a three player game, 1) if the lowest ranked player finishes last, they shouldn't get hammered on ELO and 2) with few exceptions it should be a net zero sum and this one wasn't close. About 30 pts were lost and only 15 gained.

Once again. The game I played that inspired this thread involved an expert level player coming in 4th out of 4 and losing 26 pts. In a game that included two players below 200. So, any metric that results in the lowest ranking player in the game finishing 3rd out of 3 losing MORE is flawed and needs to be revisited.

I think you didnt get it. Use this: https://www.omnicalculator.com/sports/elo
but adjust K-factor. and on bga its: if youre first -> + elo for each under that score (which is everyone). if second -> +elo from everyone -elo from 1st etc. - if needed, repost the link in question and I could paste the elo, to check for a bug.

Also feel free to read up on elo on wiki, or anywhere else. And I think there was a link posted, in this thread, to a discussion about the implementation of elo in >2 player games.

[detlefchef11]

https://boardgamearena.com/faq?anchor=faq_account_elo

K-factor multiplier

Then it's broken. Because, in a three player game, 1) if the lowest ranked player finishes last, they shouldn't get hammered on ELO and 2) with few exceptions it should be a net zero sum and this one wasn't close. About 30 pts were lost and only 15 gained.

Once again. The game I played that inspired this thread involved an expert level player coming in 4th out of 4 and losing 26 pts. In a game that included two players below 200. So, any metric that results in the lowest ranking player in the game finishing 3rd out of 3 losing MORE is flawed and needs to be revisited.

I think you didnt get it. Use this: https://www.omnicalculator.com/sports/elo
but adjust K-factor. and on bga its: if youre first -> + elo for each under that score (which is everyone). if second -> +elo from everyone -elo from 1st etc. - if needed, repost the link in question and I could paste the elo, to check for a bug.

Also feel free to read up on elo on wiki, or anywhere else. And I think there was a link posted, in this thread, to a discussion about the implementation of elo in >2 player games.

This is the original game in question https://boardgamearena.com/table?table=451030890
As you can see, I won a 4 player game that included a player ranked well ahead of me and yet got less points than I did for winning this 1v1 game against someone that was ahead of me by less ELO than the high ranking player in the 1st game https://boardgamearena.com/table?table=448669185

Then, of course, came the strawman arguments using extreme examples where the inclusion of extra players would actually make it easier to beat a good player in a 4 player game than in a 1v1.

I understand the ELO is a respected and long used system... in chess... which is a two player game. But somehow in the transition to games with 3 or more, y'all messed something up. But, whenever anyone points the issues with how it is implemented in games with 3 or more, you guys trot out the "it's what they use for chess", like that's some magical thing. But it's not, because chess is always 1v1.

[Romain672]

But, according to the link I was shown, the K factor eventually ends up at 20, not 10 as you used. Not only that, but the vast majority of players are above 100, so for at least your first games (and likely beyond that), you'll be winning games against people with a higher ELO than you (given you start at 0). So, in all honesty, you'll be at 200 relatively quickly even without exaggerating things through the K factor.

I just verified: https://boardgamearena.com/table?table=344424762
And so I confirm what i said k-factor is 3 times lower after lots of games.

Besides, earlier in this discussion, when I was pointing out the folly of getting less ELO for winning a 4 player game than you do a 1v1 despite the fact that, all things being equal, it's statistically harder to win a 4 player game than it is a 1v1. Thus, one could expect to climb the ranks more quickly playing 1v1.

That was dismissed with "eventually players will normalize at their appropriate ELO regardless of whether they get there playing 4 or 2 player games."

So, which is it? Are you willing to have certain players show an inaccurate ELO for a period of time or not?

I'm the one which talk about that point later on, so yeah I just agree with you there.

But you just want to limit that inaccurate ELO.


But it's really hard to find a good system. What would you propose?

1)
Use the alternative ranking of Complexity/Strategy/Luck/Interaction
About my previous post again let's compare 6 nimmt! and Secret moon both at 8 players:
- 6 nimmt! has way more luck at 8 players, but you still got a good control of what you can do and limit your bad games
- Secret moon is more of a team game, and so if you get in the wrong team, or are unlucky, you could just lose

Or at least it's how I feel on those.
6 nimmt! has 3luck, 4interaction. Secret Moon has has 3luck, 2interaction.
So those ratings with my feelings on those game don't work well together (higher interaction = more focus = give more elo to higher elo player, but so 6nimmt! should be considered to give less elo to higher elo player (compared to secret moon)).

2)
Remove the reduction of k-factor in high number of players games.
That will lead to what I said previously:
High player games would get a big swing of elo, and elo will said nothing. When you gain +120 elo per first place, and max elo for 99% of players is 300, your elo will mainly said how well you did in your last 3 games.
I really like the elo atm for high count games.

3)
Remove the reduction of k-factor for 3 & 4 players, and make it low for 5 players:
This would increase the swing of elo for 3/4 players games.
Nobody answered my last question though, and since I'm not fully sure of that part, that's annoying:

To answer that: in non-interactive games, like Yahtzee, what happen for players who plays exclusively high count? Are they competitive?
If yes, I should be right.
If no, I should be wrong.
(but don't check the top players of Yathzee to answer that, if 99% of people plays 2p, you will get top players who plays 2p ofc)

But you would still got the problem detlefchef11 talk from the start: high (or medium) interactive games would still be best at 2 players because of the focus of others players (I remember when I was playing Legendary Inventors at 3+ players, when others people were faced with two similar choices between helping me (the highest elo player) vs helping the other players most of the time they choosed the other one, and I would do that too).

4)
Separate the elo.
It's something talked regularly. But while it separate the community, Separating the lowest count of players and others could help too.
But that just reduce and move the problem.
Instead of playing 2p people would just play 3p for other ranking.
And in games like "6nimmt! logical" that wouldn't solve the problem (since one option change radically the game).

5)
Ignore the problem.
That's an option Which have the advantage of the simplicity. We can understand the system at this point.
It could be not the case if you complicated it.

[Romain672]

This is the original game in question https://boardgamearena.com/table?table=451030890
As you can see, I won a 4 player game that included a player ranked well ahead of me and yet got less points than I did for winning this 1v1 game against someone that was ahead of me by less ELO than the high ranking player in the 1st game https://boardgamearena.com/table?table=448669185

Then, of course, came the strawman arguments using extreme examples where the inclusion of extra players would actually make it easier to beat a good player in a 4 player game than in a 1v1.

Thing is, the system assume there was more randomness on this 4 player game since others players could impact the winner of the game.
As a result, the system make you win and lose less elo.
From that logic, everything work.
(ie my previous post who ask for solutions)

I understand the ELO is a respected and long used system... in chess... which is a two player game. But somehow in the transition to games with 3 or more, y'all messed something up. But, whenever anyone points the issues with how it is implemented in games with 3 or more, you guys trot out the "it's what they use for chess", like that's some magical thing. But it's not, because chess is always 1v1.

Nobody messed up anything. The chess system is done and work really well for 2 players.
You just cannot translate it for 3 players.
Especially when you want the system to be: simple, work for any number of players, work for team game, work for game with low luck, high, low interaction, high interaction, differents settings. And want single elo ranking for all that.


And again, I agree with you outside of mostly the maths parts.

[detlefchef11]

I recognize that some games it doesn't really matter if you get 2nd or 3rd or 7th because you didn't win. Other games it matters more. So, as much as I lean towards winner take all (I think it's the American in me , because we hate ties as well. not saying that as a point of pride, btw), I can see the argument for awarding ELO points for 2nd place in 4 or more player games, 3rd place in 6 or more player games, or 4th place in 8 player games. I can't, however, see why 3rd place in a 4 player game should not lose ELO (even if only a little) unless they're below 100 and y'all want to stick with getting people up to and over 100.

But, mostly, what needs to be recognized is that, in the vast majority of scenarios, it is harder to beat 3 other people or 4 other people or 7 other people than it is one. That's that many more people who can get lucky or be better than you. So, at absolute least, you shouldn't get LESS ELO for winning a multi opponent game.

That is really my main point. That, when all this was being worked out, I'm amazed that someone didn't say, "wait a minute, you actually get more ELO points for winning games that are statistically easier to win. We need to fix that."

[detlefchef11]

This is the original game in question https://boardgamearena.com/table?table=451030890
As you can see, I won a 4 player game that included a player ranked well ahead of me and yet got less points than I did for winning this 1v1 game against someone that was ahead of me by less ELO than the high ranking player in the 1st game https://boardgamearena.com/table?table=448669185

Then, of course, came the strawman arguments using extreme examples where the inclusion of extra players would actually make it easier to beat a good player in a 4 player game than in a 1v1.

Thing is, the system assume there was more randomness on this 4 player game since others players could impact the winner of the game.
As a result, the system make you win and lose less elo.
From that logic, everything work.
(ie my previous post who ask for solutions)

I understand the ELO is a respected and long used system... in chess... which is a two player game. But somehow in the transition to games with 3 or more, y'all messed something up. But, whenever anyone points the issues with how it is implemented in games with 3 or more, you guys trot out the "it's what they use for chess", like that's some magical thing. But it's not, because chess is always 1v1.

Nobody messed up anything. The chess system is done and work really well for 2 players.
You just cannot translate it for 3 players.
Especially when you want the system to be: simple, work for any number of players, work for team game, work for game with low luck, high, low interaction, high interaction, differents settings. And want single elo ranking for all that.


And again, I agree with you outside of mostly the maths parts.

So, you actually think it was easier for me to beat the expert at the table in that 4 player game than it would have had I played them 1v1? Because that's what it sounds like and what others have exactly said. While I recognize there are things that poor players can do that muck things up and, I suppose, could disadvantage one player over another. But there's no saying which player is going to get screwed. If a Raven pops up in the tray and the newbie that goes before me doesn't realize it's gold and passes on it, then I'm the one getting the advantage. Of course, it could just as easily happen for the other good player and I'd be looking at a sure loss.

Isn't it better to build a system that works in games where everyone is at least competent? Where there may be different skill levels, but at least everyone gets the gist of how to play?

So, say you're a 350 ELO. You've got two games. One is a 4 player game against a 450, 325, and 275 and the other is 1v1 against the 450. Which game do you have a better chance of winning before the game starts. You have no idea of turn order, cards, or whatever elements of luck are in the game.

[silentProtest]

So, you actually think it was easier for me to beat the expert at the table in that 4 player game than it would have had I played them 1v1? Because that's what it sounds like and what others have exactly said. While I recognize there are things that poor players can do that muck things up and, I suppose, could disadvantage one player over another. But there's no saying which player is going to get screwed. If a Raven pops up in the tray and the newbie that goes before me doesn't realize it's gold and passes on it, then I'm the one getting the advantage. Of course, it could just as easily happen for the other good player and I'd be looking at a sure loss.

Isn't it better to build a system that works in games where everyone is at least competent? Where there may be different skill levels, but at least everyone gets the gist of how to play?

So, say you're a 350 ELO. You've got two games. One is a 4 player game against a 450, 325, and 275 and the other is 1v1 against the 450. Which game do you have a better chance of winning before the game starts. You have no idea of turn order, cards, or whatever elements of luck are in the game.

... so you know, elo isnt perfect, even more so in luck involved games or games with high tie tendency. it is just the best we've got. also do look up how it works, and the idea to use that math for it.

and what I understood just from skimming - you compare 1v1v1v1 and 1v1 with a high elo player as opponent. those things, generraly speaking, just dont compare. it usually is a different balance. not sure how much counter play there is in wingspan though, if you dont play vs ai

Edit: who says you had a chance vs that high rank in a 1v1 ?? xD

[detlefchef11]

So, you actually think it was easier for me to beat the expert at the table in that 4 player game than it would have had I played them 1v1? Because that's what it sounds like and what others have exactly said. While I recognize there are things that poor players can do that muck things up and, I suppose, could disadvantage one player over another. But there's no saying which player is going to get screwed. If a Raven pops up in the tray and the newbie that goes before me doesn't realize it's gold and passes on it, then I'm the one getting the advantage. Of course, it could just as easily happen for the other good player and I'd be looking at a sure loss.

Isn't it better to build a system that works in games where everyone is at least competent? Where there may be different skill levels, but at least everyone gets the gist of how to play?

So, say you're a 350 ELO. You've got two games. One is a 4 player game against a 450, 325, and 275 and the other is 1v1 against the 450. Which game do you have a better chance of winning before the game starts. You have no idea of turn order, cards, or whatever elements of luck are in the game.

... so you know, elo isnt perfect, even more so in luck involved games or games with high tie tendency. it is just the best we've got. also do look up how it works, and the idea to use that math for it.

and what I understood just from skimming - you compare 1v1v1v1 and 1v1 with a high elo player as opponent. those things, generraly speaking, just dont compare. it usually is a different balance. not sure how much counter play there is in wingspan though, if you dont play vs ai

Edit: who says you had a chance vs that high rank in a 1v1 ?? xD

LOL, I've got my giant killer badge or whatever they call it, so yeah, I've got a chance.

That's the thing about Wingspan, there's more counter play than many think, but it's not ALL about that.

But, I used to play competitive variations on Risk at another site. There is obviously plenty of counter play in that game because you're literally attacking someone nearly every turn. There's also plenty of luck, both in the initial drop and the dice. And multi-opponent games play vastly different than one opponent games as there is diplomacy and trying to lay possum while everyone else beats each other up. While 1v1 (or 2v2, etc) there's none of that for obvious reasons.

None the less, I found their system simple and effective.

First off, it was winner take all (despite the fact that one could argue, in many games (not all, but many) some players were eliminated early and then there was basically a second game with 2 or 3 players, then 2).

You simply divided each loser's rating into the winners and multiplied it by 20. So, a player with a 3000 rating who beat someone with a 1500 rating would get 10 pts but if it was the other way around, the player with 1500 would get 40. Of course, the other player would lose that many points, so it was always a net zero sum. And you do that for every player, so if you won a game with a lot of players, you would win a ton.

They also started you off a 1000 rating points and didn't protect you from losing them. Presumably, you would quit before you got to zero and, given the above ratio, you'd have to lose A LOT and to bad players to ever get to zero.

They had ranks associated with different levels of rating points, so it was fun to go for. That's not unlike here where, if you're at 295, you really want to get over 300 and if you're at 495, you really want to get over 500.

But I liked that system.