Another complaint about ELO

[silentProtest]

This is the original game in question https://boardgamearena.com/table?table=451030890

Let me calc for you:
original elo approx.:
351
118
159
466

K-factor (if 20 assumed base, which its prob not for all): 20 * 4/(2*(4-1)) = 20 * 2/3


gain for 1st:
from 2nd: 3,115
from 3rd: 3,668
from 4th: 9,147
total 15,93

gain/loss 2nd player:
from 1st: -3,115
from 3rd: 7,568
from 4th: 11,866
total 16,319

well the K-factors seems to not be 13,33333 for 4 players -.- [extra: prob there is a recommended player No. of 6 ...]
but you get my point I guess
... why it feels unfair to you stems from the harsh difference, that wins/loses against players with more elo (linear) resulting in way more/less elo gain/loss (non linear), which is just a demonstration of that, in this game.


PS: is there a way to quickly look up the end calculation window, instead of letting go replay to the end ?? [Edit: "Recording game results + computing statistics in progress..." *moderated to "freaking"?* awesome stuff!!!]


the rest is a general discussion about approxiability of skill level in games, and whether elo, and the way it is used here, is adequat. But that should be a discussion between mathematicians and data scientists, IF you want actual answers, instead of empty talk

Edit: fun fact, since the RecommendedPlayer# not null, he would have gotten even less elo in a 1v1

*Moderator edit: please refrain from swearing on BGA, this is a family-friendly platform.

[silentProtest]

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.

It's a *moderated to "very bad"* system compared to elo xD - no offense

(about winner takes all, its debetable - makes sense for some games more than others. Id argue for that in catan for instance, because it also changes the meta, and how top player play in specific situations...)

the charm in elo is that you can very easily asses your winning chances, just by the point difference.

70 elo difference, youre x1,5 more likely to win than your opponent (1,5/(1,5+1)=1,5/2,5=0,6=60% vs 40%)
120 -> x2 more likely
190 [120+70] -> x3
240 [120+120] -> x4
etc....

*Moderator edit: please refrain from swearing on BGA, this is a family-friendly platform.

[Silene]

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."

In both examples [(a) winning vs. 3 others (b) new player losing many points for becoming last] you completely forget about looking at the case that DIDN'T happen. However the result that DID happen could also turn out differently. You need to have a system that is fair for the expected outcome.

In the newly discussed case (b) the player had their elo-loss tripled because they had played less then 10 games so far (and you will see that if it wasn't tripled, the overall result of the game would've been net-zero. But if the player would've won this game, their gain would've also been tripled - catapulting them higher rapidly - and that's how it is fair for them. Don't forget that there were 2 (or 3 in this case) possible outcomes! The system makes new players - whose true skill isn't known yet - reach a good estimation about their skill much faster. Without the tripling of gains/losses for that player, there would be no way, that they'd be already on over 200 ELO after just 5 games (which they were!). So yes, in this game they lost a lot. But overall, they're approaching a good estimation much quicker.

In case (a) you say it's not fair someone winning gains fewer points by winning vs. 2/3 others than had they only 1 opponnent. Again - don't forget that the system also protects them from losing many points in the case (that did not happen) where they had lost.

It's like when you can choose to play a betting game.
Game A: 50% chance you win 100$ vs. 50% chance you lose 100$
Game B: 20% chance you win 80$ vs. 80% chance you lose 20$.

If you play Game B and win, you only gain 80 - less then if you win in Game A. Although in Game A it was much more likely to win. Does it mean it is better for you to choose Game A? No - it's exactly the same expected value (0). You can't judge only by one of the possible outcomes!

[Romain672]

Thinking again to the last game shared, the way to resolve your problem would be to create differents k-factor.
A: 700 elo
B: 700 elo
C: 100 elo
D: 100 elo
By default your k-factor against B, C, and D should be 13.3 (20* the formula of ufm page 6).
If you only fight B, your k-factor should be 20.

So if you beat BCD, you would get a total of 40 k-factor. But since AvsC & AvsD are negligeable, you will only gain let's said 14.
If you beat B, you would get a total of 20 k-factor.

The way to resolve it would be to give an higher k-factor on the match-up AvsB. So if you keep a k-factor of 20 for that match up:
If you beat BCD, you get a total of 46.7 k-factor. But since AvsC & AvsD are negligeable, you will only gain 20.7
If you beat B, you still get that 20.

But now let's assume that C did some dumbs moves, and as a result B won instead of A.
Should AvsB still got a k-factor of 20?
I'm good with both answers.

But that increase the complexity. And to the question why doing that? I don't think the complexity is worth for what it does which is controversial I believe.
Else you would need to find another method of calculation.

[ChiefPointThief]

So I mentioned trio earlier in the thread. In 3p winner takes all and there is no 2nd place. But what I didn't realize until today is that because the two losers tie the one with the most elo is losing points to the other player. Elo is typically low in this game and this explains why there aren't many players above 300 elo.

https://boardgamearena.com/table?table=463509936

If you look at this game sabbie loses 9pts for tying with chikky even though they both come in 2nd place.

https://boardgamearena.com/1/trio?table=463503913

Even in this game with an 100 elo player she loses 6.3 for tying with them. This needs to be fixed along with several teams games that still didn't get the team elo upgrade.

[Romain672]

Yeah I reported that problem in mid 2020: https://boardgamearena.com/bug?id=11297

Not Alone is worst.
It's a 1vAll.
So if you play as All, you could have some games where after you win, you are supposed to lose elo since you tie with all your team.

On that case, bga makes the elo loss up to 0. But that's not a great solution.

[silentProtest]

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?

just wondering, you got detailed responses... how come you stopped responding or defending your heartfelt point ?? O.o

[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?

just wondering, you got detailed responses... how come you stopped responding or defending your heartfelt point ?? O.o

To be honest, I gave up. Nobody has convinced me that the system is good, but I obviously haven't convinced anyone of the contrary. So I'm just going to drop it and move on. It is, after all, not that important in the big picture.

[silentProtest]

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?

just wondering, you got detailed responses... how come you stopped responding or defending your heartfelt point ?? O.o

To be honest, I gave up. Nobody has convinced me that the system is good, but I obviously haven't convinced anyone of the contrary. So I'm just going to drop it and move on. It is, after all, not that important in the big picture.

Maybe I missed it somehow, but you have not posted clear points you need fixed with elo; Besides not liking, very generally, how it handles 3+ player games here , and because you liked the system more with "big number goes bigger"...
Neither posted solutions to that things (except bigger number thing), or other described problems with elo, and the implementation here.

[detlefchef11]

just wondering, you got detailed responses... how come you stopped responding or defending your heartfelt point ?? O.o

To be honest, I gave up. Nobody has convinced me that the system is good, but I obviously haven't convinced anyone of the contrary. So I'm just going to drop it and move on. It is, after all, not that important in the big picture.

Maybe I missed it somehow, but you have not posted clear points you need fixed with elo; Besides not liking, very generally, how it handles 3+ player games here , and because you liked the system more with "big number goes bigger"...
Neither posted solutions to that things (except bigger number thing), or other described problems with elo, and the implementation here.

This is the last thing I'm going to post here because I'm honestly done with it. I do think BGA tried to take a system of ranking people designed to work in a h2h game and added a bunch of complicated things that made something that doesn't do a very good job in multi-opponent games. Of course, the complicated nature of it has a built-in defense of "you just don't understand it".

This is not my site and it's not my job to come up with a better system to rank players. Especially if there are so many disciples of the current system that I'd have to fight that battle. And, like I said before, I don't have to understand your system to point out that it's flawed. That's the part you're missing. If you taste some food in a restaurant that's so nasty, it makes you gag, you don't need to know how to make the dish yourself to point out that the food sucks.

At any rate, I've just decided to not care about it (as some here suggested I do) and play the games I enjoy here.

That is all.