BGA has a serious bully problem

[MarkSteere]

Even if chat gpt is right, there's no point in using it in a conversation like this. I want to discuss with a person, with his own opinion, which he has built himself and fully understands.

You're discussing it with me, a person. I've used ChatGPT as a qualified resource on mundane topics like Elo to help settle a dispute. Admittedly, I don't have a complete understanding of Elo, but it's clear by now that few of the other participants in this discussion "fully understand" Elo. I only posted the brief summary of ChatGPT's long explanation, all of which seemed reasonable. My opinion is that ChatGPT has drawn the correct conclusion in this instance.

I'd rather hear something accurate and informative from ChatGPT than the grasping, flailing, floundering, self-built opinions of humans, most of whom, in this case, have no idea what they're talking about.

[tbhp]

It is a flaw. The elo gain/loss is determined by a probability distribution based on the difference in player's ratings. For luck-based games, they should use a different probability distribution, one with a lower cap.

Elo says that a 400 rating point difference means a 90% chance of winning, with no cap.

For a luck-based game, it should be a number smaller than 90% and there should be a cap.

You are putting the cart before the horse, it seems. Elo calculation is based on the games that have happened, so the probabilities it tells you are correct. If a game is impossible to win 90% of the time, then that will be reflected in the results and therefore the elo, and no one will have a very high elo in that game.

[FrankJones]

ChatGPT...

Elo does not perform as well for luck-based games because it was designed around a skill-dominant outcome model. When variance dominates, the Elo probability distribution systematically overstates how predictive rating differences really are.

The issue isn’t that Elo “can’t handle luck,” but that it assumes luck is small relative to skill.

Well, with all due respect to chatGPT, its answers are not always correct.

Well it is very accurate on this topic. Me & others have shown examples and given data on this topic on several threads here.

I could give examples as well. I believe it is more nuanced than the chatGPT answer. In high-luck games, the variance is higher, which means it will take a larger sample size of games for player Elos to reflect their skill, but I still believe it eventually will.

I know I'm a 300 Elo player at Open-face Chinese poker, but it took an agonizingly long time to get there because of the high variance from game to game. But if enough games are played, the variance is reduced. In the short-term, sure, Elos might fluctuate wildly.

In several other games, low luck, I attained 300 quickly, within 10-15 games.

Even in games with little or no luck, winning streaks and losing streaks can occur. I'm not convinced any of this indicates a problem with the Elo system, but I'm open-minded toward this discussion.


The reasoning seems circular to me:

"players with a 400 Elo gap cannot win 90% of the time, therefore the Elo system is flawed."
Well, if the system is flawed, then how do we know there was actually a 400 Elo gap? Maybe the 500 Elo player was artificially high, riding a winning streak, and the 100 Elo player is actually a 280 Skill level player who hasn't played enough games yet for the Elo to reflect that.

I've never lost to a player 400 Elo below me, and I don't think I've ever beaten one 400 Elo above me. If I did, it was because my true skill level was a couple hundred points higher and I hadn't played enough games yet.

How many "high luck" games are there that even have players with a 500 Elo?

[FrankJones]

Even if chat gpt is right, there's no point in using it in a conversation like this. I want to discuss with a person, with his own opinion, which he has built himself and fully understands.

You're discussing it with me, a person. I've used ChatGPT as a qualified resource on mundane topics like Elo to help settle a dispute. Admittedly, I don't have a complete understanding of Elo, but it's clear by now that few of the other participants in this discussion "fully understand" Elo. I only posted the brief summary of ChatGPT's long explanation, all of which seemed reasonable. My opinion is that ChatGPT has drawn the correct conclusion in this instance.

I'd rather hear something accurate and informative from ChatGPT than the grasping, flailing, floundering, self-built opinions of humans, most of whom, in this case, have no idea what they're talking about.

I haven't seen anything to indicate that anyone in this thread "has no idea what they're talking about".

If you understand Elo yourself well enough to assert that others in this thread have no clue, you should be able to explain why in your own words rather than copying and pasting from chatGPT.

For the record, I have seen chatGPT wrong on matters of strict fact.

[FSKFSK]

Suppose I have an elo 400 points higher than someone, and it is a luck-based game where I can't win more than 75% of the time against a competent opponent. In that case, if I care about my elo, I should be a jerk and refuse to play someone who is rated substantially lower than me.

A system that encourages bad behavior is broken.

In luck-based games, it is very hard to keep an elo higher than 300. The reason is a flaw in elo, and not that "nobody is genuinely rated that high".

What happens in luck-based games is that your elo will bounce around from 200 to 400, based on whether you are on a winning streak or losing streak. Did I really become 100 points of elo worse just because I went on a luck-based losing streak?

[Meeplelowda]

How many "high luck" games are there that even have players with a 500 Elo?

I'm going to go out on a limb and say zero. People conflate "high luck" with being non-deterministic, i.e., having random elements as part of the game mechanics. Games that are truly high luck, in the sense that they are really just an elaborate way of flipping a coin, don't have players in the 500s.

[FrankJones]

Suppose I have an elo 400 points higher than someone, and it is a luck-based game where I can't win more than 75% of the time against a competent opponent. In that case, if I care about my elo, I should be a jerk and refuse to play someone who is rated substantially lower than me.

A system that encourages bad behavior is broken.

In luck-based games, it is very hard to keep an elo higher than 300. The reason is a flaw in elo, and not that "nobody is genuinely rated that high".

What happens in luck-based games is that your elo will bounce around from 200 to 400, based on whether you are on a winning streak or losing streak. Did I really become 100 points of elo worse just because I went on a luck-based losing streak?

What game are you referring to where you cannot win consistently against an opponent with an Elo 400 points lower?

Also, using filters to avoid playing players 400 Elo lower than you is not "being a jerk". I use such filters every time I play. So do many people. That's why the filters exist.

[FrankJones]

How many "high luck" games are there that even have players with a 500 Elo?

I'm going to go out on a limb and say zero. People conflate "high luck" with being non-deterministic, i.e., having random elements as part of the game mechanics. Games that are truly high luck, in the sense that they are really just an elaborate way of flipping a coin, don't have players in the 500s.

That's what I think, but we have people in this thread insisting there are scenarios in which an actual 400 Elo difference exists but the player with the higher-by-400 Elo cannot win consistently.

[tbhp]

I could give examples as well. I believe it is more nuanced than the chatGPT answer. In high-luck games, the variance is higher, which means it will take a larger sample size of games for player Elos to reflect their skill, but I still believe it eventually will.

Exactly. Elo can't guess by itself the luck factor of a game, it can only go with the game results it's presented with. If a person happens to get lucky and win 90% of their games in a game where winrate is normally capped at 60%, then their elo will appear higher than their true skill level, and will display a higher probability of winning than what is possible. But that's not elo's fault, and it doesn't mean the formula is wrong for luck heavy games, it's just that it's based on things that have actually happened, it shouldn't be taken as a 100% reliable indicator of what will happen.

That's how rankings for any game or sport are supposed to work. A player who defies the odds will get points rewards that reflect the reality of how well they did, not what you think they were likely to do. Rankings are based on results, not probabilities. You can derive probabilities from results, but that comes with uncertainty, as is the case any time you input real world data into a mathematical formula.

[FrankJones]

What happens in luck-based games is that your elo will bounce around from 200 to 400, based on whether you are on a winning streak or losing streak. Did I really become 100 points of elo worse just because I went on a luck-based losing streak?

So what if a player's Elo ebbs and flows? A 500 Elo player might see the Elo fluctuate between 400 and 600 based on a winning streak or losing streak. Why is this an issue and a flaw? Within which arbitrary bounds are you saying a player's Elo should remain?

And, no, you did not become 100 points worse. If your skill level is 400 and you go on a losing streak, it means one of three thigs:

1) You weren't actually a 400 Elo player
2) You went on an unlucky streak, and/or
3) You had a stretch of games in which you didn't play your best, for whatever reason.

All of this is normal, and none of it reflects any inherent flaw in the system.