For the 1000th time, we really need an ELO filter in Arena

[dschingis27]

It's not like only the last 20 games count. I was a bit exaggerating here. But particularly in Arena, with K=40, they make up the major influcence on your score. For Elo it's more like the last 50 games are relevant, but still the last 20 games have much higher influence on Elo than the games before.

[Jellby]

K=1 or K=40, as long as it's constant, doesn't change much, it's just a scale factor for all score changes (ignoring the fact that you can't go below 100). What matters is that the K changes between your first and later games.

[dschingis27]

K=1 or K=40, as long as it's constant, doesn't change much,

It does. I will try again to explain below.

it's just a scale factor for all score changes

Yes, exactly. That is exactly my point. The scale factor shouldn't be the same for all games.

Of course it makes a difference if you first lose 10 games and then win 10 games or vice versa. The recent games count more for your current Elo/EAS.

You can also think about if the K factor were 1000. In such case, you would always have low Elo if you lost your last game. The game before almost doesn't matter, not even speaking of the games before. In contrast, if K=1, the Elo score changes so slowly that a meaningful Elo is only reached over a very long period of time, really many games would matter in how your current Elo looks like.

Why are there different K factors assigned for the first 20 games you play when you learn a new game? Exactly so that the speed of Elo changes is adjusted. The K factor adjusts the Elo change per game. But per game means very different things for 7Wonders and Through the Ages.

I want to stress here: The K factor does not just define how strong Elo jumps up and down, it is indeed a marker for how quick older game results get forgotton.

[ElThoesen]

I do not like facing 100 ELO opponents. It isn't unreasonable to be able to remove this. There's very little upside for a win and tons of downside for a loss. All risk and little reward compared to similarly ELO'd opponents.

[RicardoRix]

It sounds like you're advocating for the K-factor to be reduced even further when players have played lots of games...

Is the lower limit 20?
Does anyone know when the K factors change?

If you raised a suggestion for this it would get some votes.

Maybe people wouldn't even mind when playing low rated opponents.

[cowboy_dan]

I do not like facing 100 ELO opponents. It isn't unreasonable to be able to remove this. There's very little upside for a win and tons of downside for a loss. All risk and little reward compared to similarly ELO'd opponents.

ELO won/lost is proportional your probability of winning. So, if you play ten games against opponents you should beat 90% of the time, you’ll, for example, lose 18 points once, and gain 2 points 9 times, and be right back where you were. If you’ve played a lot of games, your ELO should be pretty close to your “true level”, and so it shouldn’t change much.

[dschingis27]

It sounds like you're advocating for the K-factor to be reduced even further when players have played lots of games...

No. That wasn't my point. I want to have different K factors for different games. Low K factor for short and random games like Can't Stop, large K factor for Terra Mystica and colleagues. It's absolutely obnoxious that one game of Can't Stop counts as much for Elo as one game of Terra Mystica.

ELO won/lost is proportional your probability of winning. So, if you play ten games against opponents you should beat 90% of the time, you’ll, for example, lose 18 points once, and gain 2 points 9 times, and be right back where you were. If you’ve played a lot of games, your ELO should be pretty close to your “true level”, and so it shouldn’t change much.

The thing is, the order in which wins and defeats occur plays a role. If you played 2 games against opponents with identical Elo, you will have greater Elo if you first lose and then win instead first win and then lose. This effect grows larger if you think of streaks, 3 defeats and then 3 wins will give you greater Elo than 3 wins and then 3 defeats.

Furthermore, Elo and particularly EAS are not converging. Elo has a tendency to jump around your true value, but is not necessarily close to it. In games with a strong luck component like Can't Stop, your Elo is bound to bounce in a range of 150 or more Elo poins, for 7Wonders I'd say the range is roughly 100 Elo points. For EAS, you can double these ranges. It just naturally bounces up and down following your current level of luck. You never know how close your Elo value is to your "true value" or how much it is influenced by recent outcomes of chance. (If we want to mitigate this bouncing, we have to chose smaller K factors.)

[RicardoRix]

It sounds like you're advocating for the K-factor to be reduced even further when players have played lots of games...

No. That wasn't my point.

..... (If we want to mitigate this bouncing, we have to chose smaller K factors.)

[It's absolutely obnoxious that one game of Can't Stop counts as much for Elo as one game of Terra Mystica.

You can't compare ELO between different games. I'm not sure why you'd need to.
I think it's ok for Terra Mystica to have ELO 600 for good players, whereas Can't Stop will be 300. Some people use this as a gauge to measure the luck element of the game. Maybe you could even use this for your idea of different K factors for different games.
You want to make changes to improve the competitive side and accuracy yes, but it doesn't have to tally across games.
ELO is just a number indicator for comparison of players of the same game.

[dschingis27]

It sounds like you're advocating for the K-factor to be reduced even further when players have played lots of games...

No. That wasn't my point.

..... (If we want to mitigate this bouncing, we have to chose smaller K factors.)

[It's absolutely obnoxious that one game of Can't Stop counts as much for Elo as one game of Terra Mystica.

You can't compare ELO between different games. I'm not sure why you'd need to.
I think it's ok for Terra Mystica to have ELO 600 for good players, whereas Can't Stop will be 300. Some people use this as a gauge to measure the luck element of the game. Maybe you could even use this for your idea of different K factors for different games.
You want to make changes to improve the competitive side and accuracy yes, but it doesn't have to tally across games.
ELO is just a number indicator for comparison of players of the same game.

Maybe you can re-read my previous points of the current thread to understand my point.
I don't want to compare Elo between different games. I understand that ELO is just a number indicator for comparison of players of the same game, but I want to make this comparison more appropriate. I want to make Elo appropriate for the characteristics of each game. The main point is: You shouldn't have the same recency bias in Elo in each game. For Terra Mystica, it makes sense to measure your skill with the last 50 games you played and not so much the games before because you have quite a learning curve and there is not so much randomness effecting the game results, making 50 games a good sample size. For Backgammon, or Can't Stop, this is totally different.

The more luck plays a role in a game, the more matches have to be collected to have a proper skill determination. You can't measure your Backgammon or Can't Stop skill in just 100 games, not even approximately (and certainly not if you overweight the last 50, 20, 10, 2 games even more). Yet with current K factor, the games before your last 100 games will be almost irrelevant for your current Elo. That's a problem. When I filter for 'Good or Above' players, I want to get 'Good or Above' players and not the average players who only got lucky in recent games.

[dschingis27]

It's also stupid that Elo bounces in 30 minutes of Can't Stop as much as it bounces in 10 hours of Terra Mystica. My skill of Can't Stop (, Backgammon, ...) certainly doesn't change by 100 Elo points in 30 minutes.

The fact that one single game of Can't Stop or Backgammon changes your Elo by 20 points, or changes EAS by 40 points, it is just pure nonsense.