Matching Rank to Ability

[duckfandad]

I play 6 Nimmt! alot, and am wondering about the manner in which rank evolution is calculated.

It appears that evolving up is very slow and evolving down is unforgivably fast, and that there is little or no change for those who place third place out of five.

If the calculation considered the possibility that a rank of 100 may simply indicate an expert on their way up to 400, then a rank of 300 ought not be docked as much for a loss because they actually had a low percentage chance of winning in the first place.

I would recommend a calculation possibly corrected thus:

Opponent rank effect on rank
< 100 25%
< 200 50%
< 250 75%
All Else Same as current

Then those at the bottom ranks would improve dramatically in wins against high ranking players. Those at the top could play low ranking experts without fear of dramatically losing their rank. And corrections to relative ranks would be made as experts played each other. This would make game playing more enjoyable for all.

Thank-you for your consideration.

[FSKFSK]

>Those at the top could play low ranking experts without fear of dramatically losing their rank.

This is unfortunately a flaw in the ELO system. ELO assigns a 90% win probability when a player is ranked 400 points above their opponent. There are many games on BGA where you will never get a win rate of more than 65%-75% against an opponent with basic competence. It is almost impossible to get and maintain a high elo on games like that.

Instead of using the fixed ELO distribution, there should be a different distribution for each game. Games with a significant luck component should use a different distribution.

[duckfandad]

I do not believe that the fault resides with the ELO system itself:

https://en.wikipedia.org/wiki/Elo_rating_system

But in how it is naturally applied in games at BGA. ELO was originally created for Chess tournaments. The very nature of chess being one on one meant that your ELO rank was only modified in relation to the skill of a single competitor. Apart from having a real bad day, in which case you would likely bow out of the game, your new rank would be an accurate indicator of your skill in relation to your opponent.

Also, apart from unknown prodigys, the chess tournaments in which you enter will likely have persons with skills and ELO rank similar to your own. BGAs strength as a platform is that persons of varying skill levels are all welcome at every table. One just needs to adjust the ELO system to allow everyone to have their ranking climb at a reasonable pace to a level which is correct for her/him.

The best adjustment is to assume that three-fourths of the persons below 100 are actually experts on their way to 400. Then, even a person ranked at 150 ought only be docked 25% of the norm for losing to a relative prodigy at ranking 75. OTOH, the person at ranking 75 should be increased by 100% of the norm for besting the 150. In this way, the lower ranking players can rise based on the accurate skill ranking of their opponents while the higher ranking players are relatively unharmed by the incorrect skill ranking of their opponent.

[Silene]

The good thing is: unknown prodigy already get adjusted to proper elo faster by getting a high k-factor in the beginning.

So what's missing right now is: a strong player who plays against a new player (someone who has high K-factor) should in that game get a low K-factor instead of the normal one. Which means that in case of a win, he gets only 2 elo points instead of 4 (which is barely a difference).
But in case of a loss (which might easily happen when the other player is new but has big skill from playing outside BGA or just because they're very talented), they'd only lose 10 points instead of 20.

It also makes it harder to "feed" on rookies or friends.

I think this asymetrical, opponnent-dependent K-factor adaption would be a great improvement of the system. It also lifts motivation to allow apprentices to your tables because the result couldn't be that painful.

[Meeplelowda]

So what's missing right now is: a strong player who plays against a new player (someone who has high K-factor) should in that game get a low K-factor instead of the normal one. Which means that in case of a win, he gets only 2 elo points instead of 4 (which is barely a difference).
But in case of a loss (which might easily happen when the other player is new but has big skill from playing outside BGA or just because they're very talented), they'd only lose 10 points instead of 20.

Another feature of this is that it takes advantage of a quantity that's already being calculated anyway, so hopefully wouldn't require as much of an overhaul to implement.

[duckfandad]

I think this asymmetrical, opponent-dependent K-factor adaption would be a great improvement of the system. It also lifts motivation to allow apprentices to your tables because the result couldn't be that painful.

While I was suggesting a quick and dirty adjustment to allow skilled persons with a lower score to rise to their correct rank without unfairly dragging equally skilled persons to drop to an incorrect rank, I can see the value of a more extended adaptation.

Each player could be assigned a personal K-factor which remains with the player instead of being estimated for each game:

- A player just starting out would receive a K-Factor of 40, allowing them to rise rather quickly.

- As the player rises above 100 their personal K-Factor would drop to 30

- above 200....drop to 20

- above 300....drop to 10

- above 350....drop to 5

As the player is able to maintain a high ELO, the K-Factor can gradually rise (or be manually raised?).

Both the ELO score and personal K-Factor could be advertised with each player. A player skilled enough to maintain a high ELO ranking along with a higher K-Factor would be recognized as a superior player.