BGA has a serious bully problem

[Ceaseless]

It is possible to have a system that fairly rates people for luck-based games, but that system isn't elo.

And what is that system?

Someone asked "what does elo measure"? If someone has an elo of less than 100, they're new to playing that game on BGA. They may have already played in person and know the strategy well, or they might have only played a few times. If someone has an elo of exactly 100, that means they probably aren't very good at the game. I've seen people with an elo in the 200-300 range make boneheaded plays. The cutoff for "good player" varies by game.

The cutoff for "good player" varies by person. What some players consider good others consider basically beginner. Also, everyone is going to make boneheaded plays as they're playing games.

[FSKFSK]

I already gave the answer. You have to use a different probability distribution than the one elo uses. That's one method. Another method is to calculate a rating using all of a player's games during the rating period, doing a curve fit. That avoids the "you need a winning streak to get number one arena" problem.

[FrankJones]

It is possible to have a system that fairly rates people for luck-based games, but that system isn't elo.

Someone asked "what does elo measure"? If someone has an elo of less than 100, they're new to playing that game on BGA. They may have already played in person and know the strategy well, or they might have only played a few times. If someone has an elo of exactly 100, that means they probably aren't very good at the game. I've seen people with an elo in the 200-300 range make boneheaded plays. The cutoff for "good player" varies by game.

Is there a good alternative?

And, yes, players in the 200-300 range will make bad plays. That's why they aren't in the 500-700 range, right?

[FrankJones]

It is possible to have a system that fairly rates people for luck-based games, but that system isn't elo.

Someone asked "what does elo measure"? If someone has an elo of less than 100, they're new to playing that game on BGA. They may have already played in person and know the strategy well, or they might have only played a few times. If someone has an elo of exactly 100, that means they probably aren't very good at the game. I've seen people with an elo in the 200-300 range make boneheaded plays. The cutoff for "good player" varies by game.

The ranges for "average, good, strong, expert, master" are arbitrary and defined titles set by BGA. They probably figured it was better than nothing. But for any specific game, yes, players should use their own judgment. Obviously, a player rated 275 where the world #1 player us rated 300 is probably better than just "good", and a player rated 203 where the world #1 is rated 1100 is likely not "good."

[FrankJones]

I already gave the answer. You have to use a different probability distribution than the one elo uses. That's one method. Another method is to calculate a rating using all of a player's games during the rating period, doing a curve fit. That avoids the "you need a winning streak to get number one arena" problem.

ceaseless was probably asking for a specific example with a proper noun, such as "[specific name of a different model] would be better than Elo".

[FrankJones]

So, in this thread it has been stated (I'm paraphrasing from memory, not quoting verbatim):

1) Elo is better for zer0-luck games,
2) because it actually can be used to predict who "should" or "will" win.
3) Elo is bad for high-luck games because of the variance and volatility.


Sooooooooo ,

Suppose two players play:

Player A: 47 games, 47 wins, 486 Elo.
Player B: 34 games, 34 wins, 596 Elo.

What does Elo "predict"? Because I see no way anyone could conclude anything beyond "It's probably a 50/50 between two players who are likely both higher skilled than their current Elo indicates".

And yet one player stands to potentially lose 13 Elo or gain 7 Elo, while the other stands to potentially gain 13 Elo or lose 7 Elo. In a zero luck game.

In other words - Elo isn't perfect for zer0-luck games either. Because, a player's Elo in a zero luck game is not guaranteed to reflect the player's underlying ability.

I'm not sure what you're trying to show. You have to distinguish between ELO as a measure of past performance (always possible) and ELO as an approximation of a skill-meter. In the later case you have to wait until the value stabilizes (which is probably not the case for any of the players of your exemple) and you have to assume that measuring the skill with a single value is possible and that the formula you've used to calculate the ELO gives a good approximation of that skill-meter.
Even if the ELO-values in your exemple reflect perfectly the skill of both players that doesn't mean that B will necessarily win : The skill-meter only describes an average outcome -- if they play repeatedly against each other (without learning = changing their skill) B will win more often than A.

What I was trying to show is:

1) The flaws and limitations, if any, of Elo on some-luck games have been overstated, and
2) the exactness of Elo is no-luck games has been overstated.

[FSKFSK]

Massey's Method is something specifc.

https://datascience.oneoffcoder.com/massey-method.html

There are a couple of other curve-fit based methods on that site.

BGA does not currently use margin of victory, but you might get better ratings using margin of victory. You also can just use +1 for a win and -1 for a loss if you want to ignore margin of victory.

[Fuchur]

What I was trying to show is:

1) The flaws and limitations, if any, of Elo on some-luck games have been overstated, and
2) the exactness of Elo is no-luck games has been overstated.

Ad 1 : So you're saying that ELO can be used to guess the outcome of future games ? This is clearly wrong in general (it isn't always possible to measure the "skill" with a linear scale) and especially for highly luck-based games (If winning is 100 % luck-dependent, everybody has the same, arbitrary skill, say 0 or -1000 or + 1000 000, and information about past performance doesn't have any predictive value.

Ad 2 : There are some games where ELO gives pretty good predictions (but the only exemple i know of is chess).

[FrankJones]

What I was trying to show is:

1) The flaws and limitations, if any, of Elo on some-luck games have been overstated, and
2) the exactness of Elo is no-luck games has been overstated.

Ad 1 : So you're saying that ELO can be used to guess the outcome of future games ? This is clearly wrong in general (it isn't always possible to measure the "skill" with a linear scale) and especially for highly luck-based games (If winning is 100 % luck-dependent, everybody has the same, arbitrary skill, say 0 or -1000 or + 1000 000, and information about past performance doesn't have any predictive value.

Ad 2 : There are some games where ELO gives pretty good predictions (but the only exemple i know of is chess).

Where did I say Elo can be used to predict the outcome of games? It can give an approximate likelihood of the outcome, but it would be an approximation. Even in chess or other zero-luck games.

Elo is useful for providing a record of what has occurred, and for establishing a leaderboard, as commonly used on board game sites.

The actual outcome of any one game depends on a lot of factors.

[ChiefPointThief]

I have played a lot of spades here. The system isn't perfect, but I don't think it's all that bad. Doesn't it just take the average Elo for each player on the team, and use that when calculating the Elo change after a win or loss?

No. I just explained to you how team elo works. Luka Doncic avg the most points in the NBA. Yet his team (lakers) were just 13.5pt underdogs to the spurs (they actually lost by 28). What team elo does is say "oh Luka doncic is better than player b on the spurs so he is 90% vs player b, 80% vs player c etc." Thus Luka Doncic would lose rating as a player (elo) in a game that he wasn't really favored in because the other team is actually better overall. In addition, if Luka Doncic were able to win with an inferior team then his teammates would benefit most out of the victory (if it worked like elo). So once again there is separation between players in certain games but the broken rating system (elo) being used to rate these players strength are constantly converging them closer than they should be.

What exactly is it you wish for Elo (here on BGA) to accomplish?

I took the time to explain that to you in detail already.

May I ask, why do you play players with much lower Elo than you, thereby putting your Elo at risk in a high-variance game?

There are only 16 players at that game rated 400 or higher. With 2 million total games played. Only one player is higher than 472, and nobody is higher than 500.

You answered your own question already.

Ummm.

posted by you literally on this page:
"We both played at an elite level throughout the season but took drastic falls at some point playing game by game trying to get a few more points than the other. Having a system in which you can play at an elite level over hundreds and hundreds of games over months and to lose all of that within 20 minutes on a bad streak in a 4 luck rating game is insane & an improper rating system."

You proceeded to quote me not saying what you said I said I'm confused

???

You said, "and to lose all of that within 20 minutes[/b][/i] on a bad streak in a 4 luck rating game is insane & an improper rating system" and then when I quoted you saying that, you said I misquoted you.

Just because there are only 16 players at 400 and above does not mean you need to play against people rated below 200. Just set the filter to 300 and above.

Firstly, I mentioned that I peaked at 550 (which no one else has) once because you made a comment suggesting that my win % playing against good players and above doesn't mean I'm a good player. It seems rather silly to me to suggest that someone who has accomplished as much as I have isn't a good player. You then applied this number to a bunch of other things suggesting that I feel as though my true ranking is 550, what I feel others true ranking is, & that I went from 550 to 380 in 20 mins etc.