ELO SYSTEM

[Doodads]

Hi.

I don't know if something similar has already been posted here but i have a suggestion.

I see lot of players who are top ranked in some games but are not skilled at all.
They just avoid to play real good players and only play lot of games against lot of bad players and win like 2 elo each time.

Why not remove elo gain when elo gap is too big ?
It's simple and elo rank would reflect real players levels.

[Jest Phulin]

With any set of rules, there are those that will try to find a way around them. Changing the rules will just result in them finding new ways around them.

There are those that want to be top ranked in the ELO system. There are two main ways to achieve this: Be an excellent player, or "cheat." And by "Cheat," I mean find a way around the system as intended. You seem to think that being able to gain a few ELO from playing players ranked much lower is a cheat.

A few points:

First, the ELO rating system on BGA is intended to allow players of similar skill level to more easily find others of a similar skill level to play against, as this tends to be a more enjoyable experience. This is not hampered by people artificially inflating their ELO by only playing against poor players. They will not join tables with truly skilled players. Truly skilled players will still be able to find truly skilled players as opponents, even though there are a few non-skilled players with the same rank. These are the only people that will actually join a table.

Second, what is "when elo gap is too big"? In your example, the ELO gain is already only 2. At what point should it be that "an ELO gain of less than X is actually a gain of zero"? In order for non-skilled players to gain a lot of ELO against other non-skilled players, they have to play a LOT of games. Additionally, they have to have either some luck or some skill, as the ELO loss from losing a game against a lower-ranked player is quite large when compared to the gain from playing the same opponent. Lowering the ELO gain when the gap is too big is already built into the system.

Third, there are always ways around the system. One could easily set up several accounts, and manipulate standings that way. For example, accounts FakeAcct01 through FakeAcct10 are set up, and each account plays every other account, and the games are played such that lowered number accounts always win. Then, the "real" account comes in and plays FakeAcct01. They then look like they are a better player than 10 others. This will take quite a while to accomplish, however, and is truly only useful to those people that think that winning the game of life is having a large score on BGA, rather than enjoying the experience of playing.

As a side note, any system that forces players to play against their skill level is a "bad idea" (tm). I have seen many skilled players will take time to explain a game to a newb when asked politely. Preventing these games will result in new players not being able to see skillful moves.

[Doodads]

Thanks for your (long :p) answer Jest Phulin.

With any set of rules, there are those that will try to find a way around them. Changing the rules will just result in them finding new ways around them.

I don't see any way to cheat with what i suggested.

There are those that want to be top ranked in the ELO system. There are two main ways to achieve this: Be an excellent player, or "cheat." And by "Cheat," I mean find a way around the system as intended. You seem to think that being able to gain a few ELO from playing players ranked much lower is a cheat.

Yes i think that being able to gain few ELO from playing players ranked much lower is cheat at some point.

First, the ELO rating system on BGA is intended to allow players of similar skill level to more easily find others of a similar skill level to play against, as this tends to be a more enjoyable experience. This is not hampered by people artificially inflating their ELO by only playing against poor players. They will not join tables with truly skilled players. Truly skilled players will still be able to find truly skilled players as opponents, even though there are a few non-skilled players with the same rank. These are the only people that will actually join a table.

With my suggestion elo would reflect real level so this point would be solved.

Second, what is "when elo gap is too big"? In your example, the ELO gain is already only 2. At what point should it be that "an ELO gain of less than X is actually a gain of zero"? In order for non-skilled players to gain a lot of ELO against other non-skilled players, they have to play a LOT of games. Additionally, they have to have either some luck or some skill, as the ELO loss from losing a game against a lower-ranked player is quite large when compared to the gain from playing the same opponent. Lowering the ELO gain when the gap is too big is already built into the system.

I have no real idea but 200 elo gap seems good to me.
Someone with a real 1600 elo has no risk to play some real 1300 elo player.
He will always win.
More, sometimes some player can have 2000 elo level but just started to play. With my suggestion he would have to play players with same elo as him.
So no risk for a 1900 elo to lose to a new account.

Third, there are always ways around the system. One could easily set up several accounts, and manipulate standings that way. For example, accounts FakeAcct01 through FakeAcct10 are set up, and each account plays every other account, and the games are played such that lowered number accounts always win. Then, the "real" account comes in and plays FakeAcct01. They then look like they are a better player than 10 others. This will take quite a while to accomplish, however, and is truly only useful to those people that think that winning the game of life is having a large score on BGA, rather than enjoying the experience of playing.

About the ones who set up several accounts, it already exists.
Player like deszcz did that with maybe like 10 accounts. With or without applying my suggestion it will still exists, so it s not relevant point.

As a side note, any system that forces players to play against their skill level is a "bad idea" (tm). I have seen many skilled players will take time to explain a game to a newb when asked politely. Preventing these games will result in new players not being able to see skillful moves.

With my suggestion about elo system, it lead people to learn how to play instead of trying to play only weaker opponents.
This point seems important to me. It sounds more interesting to make people play better and better and try to learn more and more than taunt them to steal little elo cheating playing mass poor quality games.

I did myself explained caylus to begginers, and i won no elo points.
So people will continue to "teach" even if elo system is changed.

Also real begginers can't really understand skillfull move.
A good evolving curve is to play players from your level and you just learn correctly little by little.
You can also watch top ranked players games as example, especially if those players are really top ranked players

[Jest Phulin]

Strange. We seem to agree on a point-per-point basis, but we are reaching different conclusions. Fair enough. We have different opinions.

[sub0ptimal]

My issue with the ELO system, as implemented here, is the wide variance in ratings.

In one game I've gone from over 1700 to the 1400s in 3 days. I went from being in the top 10 to being below average. No worries, I'm not here for ELO points, but it's clear that I didn't become a novice in 3 days.

If you look at the Trueskill system that Microsoft created (http://research.microsoft.com/en-us/projects/trueskill), there is a confidence factor that relates to both how many games the player has played and how recent those games were. So someone who plays a lot of games should theoretically not go up and down very much unless they are improving their game (or failing to improve vs. the rest of the players).

Even in games with over 100 plays, the rating almost always seems to be just the results of the most recent 3-8 games played. You can easily shoot up 200 points by winning 3 or 4 games in a row. The end result is that the rating just shows how well the player played his last few games.

[cctsai]

Hello,

I'd love to share a link
http://forum.boardgamearena.com/viewtop ... f=9&t=3579
especially admistrator sourisdudesert's post, explaining some ideas and history.

So the problem raised by SonGoku there was that sometimes people keep evolving their strategy in a game, so that a 2000 ELO a year ago might just be a equivalent to a 1800 ELO right now. If the player who had 2000 ELO happens to not have played for a long time, a natural strategy for him or her to keep the ranking will be just to play with less-skilled players once every 60 days, which is something we might not want to encourage. This has some similarity with the problem Doodad pointed out, which can be really annoying in my opinion in games with little luck factor.

So I'd like to sell my idea of inactive bonus The idea is to add another "inactive penalty," which applies if an expert player does not play with any players with nearby or higher ELO points for quite a time (maybe something like not playing with more than 2 players with at most 200 ELO point less than him/her for 30 days, and becomes larger and larger over time). This penalty does not apply to ELO point - the ELO system stay the same. This "inactive penalty" may be shown separately in some way, and most importantly the ranking will be based on "ELO point - inactive penalty."

In this way, the ELO system remains stable, but players having good ELO points will be encourage/forced to play with their equivalents at least occasionally if they want to keep their rank.

[Sabian80]

Ok just to chip in a bit....I play quite frequently and have seen my ELO fluctuate quite frequently from good to average to expert than really it stays around good. The only way my ELO REALLY drops is when I finish behind a player of a less than average (lets face it 1200 ELO with 2 wins is not average), then I lose nearly a hundred points.

After watching this I soon find that there is little to no advantage to playing with anyone with a lower ELO/ type of player i.e if I am an expert I do not want to play 4 player Puerto Rico with 1 expert, 1 good and a really low average player. If the average player finishes ahead of any of the higher rated players they lose quite a lot of ELO; which is suggested given that "the rating system" projects them as a better player. Thus, the expert should finish ahead of the average. So, others that have figured this out tend not to play people that much lower than them in anything other than two-player. At least with two-player they have more control over the game.

For example: "firstgrader", and "Kadwa" are two of the top three Puerto Rico players on this site. If you look through the former's game history, you will see that firstgrader never plays a game with more than two players! In fact firstgrader posted a response to a question asked by matthallway just a week ago when Matt asked why the handle "firstgrader" only plays two player games, his response:

"firstgrader-It has do with the ELO system. If you play 3 or 4 player, and one of the players is bad, it can make the game more random, and the top player could come in 3rd or 4th and lose 100 ELO points in one game."
7 days ago

That is the top rated Puerto Rico player's own words! Furthermore, I have not seen any games of Puerto Rico between two of the top players ever! I see some players play 3-player Puerto RIco with all experts or good players, but never have I seen any of the top-five in the same game....

Granted it is difficult to win as much as they do even in two player, but once you learn the game, anyone can start over and only play two-player games for rating purposes only. The result has players like Doctor Ders, who changed to ressentiment, looked like a beginner, but had to tell people he was an expert to get into games. When he did he beat them and then they lost a lot of ELO.

In summary: the ELO system does have many flaws, however, the players that only play for elo, as one handle says, make it less fun to play. People will always be able to either play once every 60 days. Or only play players of lesser skill in two-player games to manipulate or maintain their ELO.

Suggestion: Multi-player games and games in real-time should have a heavier ELO impact than two player games, regardless of gap. Having a set point system for a 2 player victory, 3-player victory and so on, in addition to the current system would allow players that play with more BGA members at a time, a chance at rankings and prestigious recognition. Also, I disagree that players should not have to play people of their own rank. I do think that is a good idea, but not only their own rank! The idea Jest, that there is no risk of a 1900 losing to a beginner is a ridiculous in a sense, because being ranked higher implies of having a greater chance of victory. If you do not want to take the risk, simply do not play versus a beginner, that was the lesson I had to learn the hard way.

[LudoH]

I quite agree with Sabiano80: the ELO system clearly takes the assumption that there is no luck involved. I think, this is related to the fact that the ELO system is the criteria for evaluating good players. The consequence of this is that this makes people encline to play against players of their own rank if luck is involved. As a consequence:

- in race for the galaxy, where there is quite some luck, people play with players of their rank and restrict tables this way.

- apparently the same happens with Perto rico. There is not much luck in PR. However, if you are 3 players: you are expert, there is a beginner and a third player, you can always have the bad "luck" that the 3rd player favors the beginner when taking actions. Even if the beginner is less good, if this happens a lot, you end up playing against 2 people and the beginner might end up in front of you ...

I have nothing against this in principle, and I do not want to change the ELO system, but I think it takes too much importance in BGA.

- first I think that the problem is that what is called "good player" is a player with ELO>1600. Thus if you want to stay good at a game with some randomness you'd better play with good players. I would suggest to remove this classification, or to adopt one that favors playing with different people: the ELO rating is the same, but to be a good player you should have ELO rating>1600 AND have more than 40% hit (numbers to be adjusted of course)
We could still have "pure ELO selection" with a criteria that says ELO>1600 for ELO addicts

- second I think that there should be a ranking that favors frequent and enjoyable players, so that people that are just less good but enjoy he games and with which you can have a nice game are not considered as "bad players". I personally tend more to impose a certain thumb-up and no time-out level rather than a high ELO rank in my tables

I know that this does not solve the original problem of the thread; which is more valid for games with no luck (and thus with 2 players), but for this, the idea mentioned above looks quite good to me: you may have to "confirm" your level regularly by playing at least every N days with somebody of your level.


Long post, plenty of ideas, what do you think?

[matthallway]

how about maybe seperate ELO per number of players? Not sure how hard that would be to implement but I believe they do it at that way on one of the Catan sites I play.. (ie. each player would have a 2 player rating, 3 player rating, etc. which would show based on the number of players the game is set for?)

[RippedFreak]

The flaw isn't in the ELO system itself but the ranking design that goes along with it. Being ranked "Good", "Expert" or otherwise should not hinge on the exacting ELO numbers of 1600+, 1800+, etc. Once a player reaches "Good"(etc) status they should retain that status for an extended period of time/number of games. This is especially an obvious flaw in cases such as have already been stated where luck is a large factor. Because of very bad cards, dice, etc you can lose 20 games in a row, just as easily as you can win 20 games in a row with the opposite luck.

When a good player with many games under their belt loses a slew of games It's not as if they suddenly because a worse player, it's just that the luck factor likely didn't go their way. In the current system now if you drop from a 1900 ELO to a 1400 ELO you go from "Expert" to "Average" or even below that. The way it should be is that once you attain status of whatever, you should keep that status with only minor consideration from ELO. Use ELO to determine that status initially but only gear it on a amount/time factor of someone dropping below the current system for a specific amount of games or over a specific time frame, say 30 or 40 games or 2 weeks. Whatever number really doesn't matter, it's just the fact that you won't be stuck playing beginners and very poor players because you lost a dozen games to horrid luck.

Such a change could also be used to eliminate such status from those who achieved "Good", "Expert" or otherwise but never play games with ELO anymore from being ranked. It serves no purpose other than to stroke one's ego to keep status and is really pointless.