ELO SYSTEM

[Sabian80]

I like the separate elo based on the number of players....As far as doodads...

Why not remove elo gain when elo gap is too big ?

I noticed they only play ONE GAME on here! Congrats on being the #1 Caylus player..that is an achievement all in itself. But when You are ranked higher, it is safe to say that you didn't gain all your elo points from playing top ranked players...given your stats you most likely have played the game before you visited BGA.
That said, I answer your question with another question, if your an expert at Caylus, which clearly you are, then why only play 2 player games? If I understand correctly you can play upto 5. So while you are trying to maintain your #1 ranking by player less skilled players, sometimes you will lose, and that is the risk you run playing two player games with less skilled. If you are so compelled to play a person with such a gap, then you can manually TURN OFF ELO.

The problem is the win will not count and the game will not count for your record. SO you can turn off the elo when the gap is too big, and then you will truly be playing for fun....

[Doodads]

Hi Sabian80.

I'm not really Caylus number 1 player and the problem is still the stupid elo system.
Ickoicko is real number one player by far but is not ranked anymore because no one wants to play him for free loss.

I didn't really played the game before BGA, just few games with friend of mine. Just look at my first games it's terrible lol.

I prefer 2 players games because no luck involved.
When playing more than 2 players game in caylus (which is a no luck factor game) then if some player makes some mistake it can hurt you and give advantage to other player.
Less skill is involved.
I don't like this, sounds like playing chess with 3 players to me.

I don't like playing with players who have very weaker level than me.
There is no fun for me in free wins beating begginers or fake experts (cause of stupid elo system which tempt people to play mass secure games instead of learning how to play).
Because lot of players in fake top 10 caylus players avoid to play with me i m obliged to play games against weaker players.

With my idea there can t be top ranked players if they re not skilled and everyone would play with people with same level.
It seems pathetic to me to encourage people to rape weaker players instead of trying to improve themselves.

The elo system doesn't work at all at the moment.
I've noticed a problem with my idea for the first player. If he has 200 elo more than the second top ranked then he can't win elo anymore lol.
But exept that my idea seems really better than current elo system.

[Doodads]

Actually i have 2376 elo at Caylus.

If i win some expert (1800+ elo) i win only 2 elo points.
If i play someone with 1300 elo and win i still win 1 elo (which is ridiculous imho, it should be 0).

So, if my goal is to be top ranked what is the point taking risks playing good players when i can play super bad players and win almost same elo ?
The current elo system is weak and encourage people to play weaker players and run away when someone stronger wants to play with you.

My new suggestion is to turn on elo system only when you play someone who has rank 20 bigger or lower than you.
This way Rank 1 player would be obliged to play only top 20 players to win more elo.
And Rank 20 player could play Top 40 players to win or lose elo.

This seems fair to me.

[groupscheme]

My new suggestion is to turn on elo system only when you play someone who has rank 20 bigger or lower than you.
This way Rank 1 player would be obliged to play only top 20 players to win more elo.
And Rank 20 player could play Top 40 players to win or lose elo.

This seems fair to me.

Sorry, but that idea would fail miserably. It might work well for the top 40 or so, but what about the middle of the pack, especially in very popular games (i.e. 100 experts and several hundred good players)? A difference of 20 in rank would amount to a difference of few ELO points (or maybe even one!), while many players don't know their own exact rating. The result is pretty much every game not being rated, and so it would be impossible to judge a player's real ability (someone may be high Good or even Expert but appear Average because they couldn't improve their rating).

Since there's only an issue when the ELO difference (not rank difference) is high, why not make changes in those situations? Perhaps ELO swings after games should be rounded down, not up, to the nearest integer. BGA always rounds up to +/- 1 at least, but if you "deserve", say, +/- 0.1 based on the difference in ratings, there's potential for the stronger player to abuse the system unless it's rounded down to 0, even if the abuse is just a bunch of tedious grinding.

(In games with some luck, however, this wouldn't be necessary; there's a nontrivial chance of losing which would deter such matchups in the first place.)

[glitter and doom]

Actually i have 2376 elo at Caylus.

If i win some expert (1800+ elo) i win only 2 elo points.
If i play someone with 1300 elo and win i still win 1 elo (which is ridiculous imho, it should be 0).

So, if my goal is to be top ranked what is the point taking risks playing good players when i can play super bad players and win almost same elo ?

I don't think your calculation is correct.
You would win 1.401 elo points against 1800 player, and 0.082 elo points against 1300 player.
It is hardly almost the same elo.

[Jest Phulin]

If i win some expert (1800+ elo) i win only 2 elo points.
If i play someone with 1300 elo and win i still win 1 elo (which is ridiculous imho, it should be 0).

I don't think your calculation is correct.
You would win 1.401 elo points against 1800 player, and 0.082 elo points against 1300 player.
It is hardly almost the same elo.

The calculation is correct when one takes into account the system rounds up.
ceiling(1.401) = 2. ceiling(0.082) = 1. Note, Doodads did say that with the 1300 win gaining an ELO point is ridiculous.

[glitter and doom]

The calculation is correct when one takes into account the system rounds up.

The elo points are not being rounded up.
They are rounded only for display purposes and they are rounded to the nearest whole number, not rounded up.
The system is using decimal numbers for calculation purposes.

Example:
http://imgur.com/Tr9BI0F
http://en.boardgamearena.com/#!table?table=8427507

1831 - 1822 = 9

[Jest Phulin]

Then I am in error, and apologize.

[cctsai]

The calculation is correct when one takes into account the system rounds up.

The elo points are not being rounded up.
They are rounded only for display purposes and they are rounded to the nearest whole number, not rounded up.
The system is using decimal numbers for calculation purposes.

Example:
http://imgur.com/Tr9BI0F
http://en.boardgamearena.com/#!table?table=8427507

1831 - 1822 = 9

Dear glitter and doom,

May I ask for the source of the formula? In http://fr.boardgamearena.com/#!table?table=8856900, ickoicko(2457) play with Dominik123(1636), and we all know who wins. Nevertheless, the display result was +1/-1, while the effect on the display number is
ickoicko(2457->2458), Dominik123(1636->1634), that is, the game seems to create an ELO change of more than 1, even though the ELO difference among the players is larger than 800. This seems weird, isn't it?

By the way, can I make a second suggestion on the 60-day rule: how about a player will stay in the ranking only if the ELO change ever happened in the recent 60 days is larger than, for example, 10 or 15? (I am thinking this in a way of 2-player game, must need more subtle change in general) By this I mean if a player lose a game and lose 20 ELO points, then she/he is "refreshed." But if one player win and gain 2 ELO point, then he has to do this 5 times to prove his participation.

This hardly affects most players, however this manages to stop some players to play once every 2 months with a player with ELO rate like 1300, just to preserve ranking, which he/she might gain one or two years ago when people were not as skilled as now. I think this new setting can encourage some "serious participation" for players high on the ranking list.

[infamouswhiteknight]

I think when talking about the rating system it's very important to decide upfront if the discussion is about one particular game, or all the BGA games. The/an "ideal rating system" for a game like Caylus (a no-luck game when played with 2 players) is probably very different than the/an "ideal rating system" for a game like Race for the Galaxy (RFTG) (a game with plenty of luck involved).

The fact that BGA has a single rating system that works "fairly well" across so many games is actually quite a feat! If we want to have a discussion about improving the rating system (let's assume it's a purely academic discussion), we should make sure we know whether we're talking about improving the general rating system for all games, or if we're talking about how we could tweak the rating system for specific games. I think most of the discussion in this thread has been about tweaking the rating system for specific games, not about changes to the general system, or at least should have been. Issues like the one Doodads brought up about Caylus are Caylus specific: other games (even other pure-skill games, like Quoridor) don't have those same issues. Similarly, sub0ptimal mentioned an issue where some games have ratings that can vary widely in just 4-6 games; I'm not sure which games have ratings that vary so quickly, but certainly this sort of thing can happen in 10-20 games of RFTG easily, however this issue doesn't occur in games like Quoridor and Caylus.

All that said, I mostly want to talk about some of the issues brought up for Caylus:

It's probably true that there are some people who are quite a bit over-rated at Caylus, and have achieved this by "gaming" the rating system. However, I pretty much agree with one of the first points by Jest Phulin: no matter what system you come up with, people who want to "cheat" the system will find a way to do so. Either way, I'm not sure the situation is quite as bad as Doodads suggests anyway. While some of the Caylus players might appear to be very weak compared to their rank to Doodads, he is a very strong player, and may not realize that the "bad" moves they are making against him can be quite effective against weaker players. I've played a lot of games with Doodads trying to learn from him, and while it's true that I've gotten better at playing him, the strategies I am learning/using in order to compete against his style of play don't necessarily help me compete against players who make fairly different (maybe weaker) moves.

Also, I don't think there is much, or any difference between having a "cutoff" of 1 elo point gained between players of vastly different ranks, or 0. The relevant digits of the ranks we're considering are the hundreds. The time it would take to get a relevant number of points by playing so far below your rank that you only get 1 or 2 points per win would be massive! (Caylus games are often 45m-1h long, and maybe even longer if you're playing beginners.) If people are gaming the system, they are more likely doing it more efficiently, e.g. by finding players who they think are over-rated and playing them to win more like 10-ish points per win. There might be ways of solving this problem (like to have a tighter bell curve in the ELO calculations), but I don't think changing 1 and 2 point wins/losses to 0 point wins/losses is the answer.