Why ELO doesn't suit to BGA

[fedetsrm]

Hi, i don't really like the forum part of GBA ( i prefer playing part ), but somethimes i take a shoot... last time i find this discussion and i have read a lot of interesting thinks. In fact i thing i'm involved in this argument, i'm one of that player that "defend" the ranking position. Till now i've more than twenty top 20 games, now i play this games only one or two at month and only with best players, is unfair? I don't know but when i play i put "elo on" so my level is always on the move. Other top players are fine with this system, wanna reach the top? just play and win, i did it in lots of games...

This is always better to defeat the first player, but in case - as second player - you feel this is impossible I see nothing's wrong to defend its second place, even in real life.
However, the new "ranking" system we have in mind only rewards the #1 player at a game. So it may solve this issue too.

If ranking system change in this way probably for me is impossible to reach the same results, get first position in more than 3-4 game is probably impossible. It takes a lot of time and fatigue for me and i will be very sad if in one days i'll lost everythings.
Last things: lately many players use the tournament to get easy points. Ex. If i organise a tournament of Senet ( i'm #1) i can gain 200-300 points, but there are some tournaments that allow you to earn 5000-6000 elo points ( like haggins) and only 5-6 players involved.... If you take out the awards for top 20 players only tournaments will decide the general ranking?

Thank you for your patience, I am not a good English writer

[Shaq Jenkins]

I think ELO works wonderfully for 2 player games. It's usually accurate and informative.

ELO for 3+ player games, especially ones based on luck and cooperation, is where I see lots of problems.

Because of ELO, people will stop playing a game because they want to protect their ranking, which is not good for the site in general, especially for rare games that people don't play very often. Instead of 60 days, I think rankings should only last for 30 days, and if you're in the Top 20, you can only keep your position if you play someone ranked similarly as you.

[latexrhombus]

I want to start by saying I do not see any reason to change the ELO system for 2 player games. The system works and is intuitive. With that in mind, I do not see the point in creating a leaderboard system. It does not add anything, but makes things more complicated and will have its own set of problems (some of which may be even worse than the problems we face now in 3+ player games in the current ELO system).

I do believe changes should be made to the ELO system for 3+ player games, however, and I believe the problems can be addressed by very simple changes.

For starters, the winner of the game deserves to gain large amount of points (no need to change this from the existing form). Winning a game with more players is more difficult, so it warrants more reward.

Players in 2nd through last place however should only lose points in accordance with ONE LOSS (even if you are 5th place) based on the ELO rank of the player who WON (1st) relative to theirs. This is because they have only lost one game. It is incorrect to assume a player placing 3rd played worse than a player placing 2nd, simply because if a player is playing to win they may take greater risks that actually hurt their finishes place. Players should still be encouraged to take these risks, however, because it is in the spirit of the game.

The final part of my suggested change is that players in 2nd through last gain points equal to the sum of all ELOs lower than them divided by their own initial ELO multiplied by the normal factor, but then DIVIDED by a predetermined but significant value such as 5 (does not have to be 5, it is an example). What this accomplishes is it still awards small point bonuses for better placements, but allows the majority of gains to fall with the victor encouraging "play to win" gameplay. As suggested above, I do not believe a lower placing non-winner should LOSE more points (they should all get 1 loss based on the winner's ELO), but this system will just let them REGAIN LESS. This prevents those situations where coming in last (especially when playing to win) will not result in the loss of MASSIVE ELO points (that is just not fair).

[gdBlaze]

Just trying to come up with some ideas for solving some of the problems we see with some games and elo.

Here's an idea that came to mind.

1. Make the levels based on percentage of players rather than an arbitrary ELO number. Perhaps the top 20 percent are Master, the next 20 percent are expert and the next 20 percent are good players. The remaining 40 percent are regular and beginners.

2. Make the Levels more 'sticky' using this general formula. Each month, the top 20 percent of a level move up to the next level and the bottom 20 percent move down. Obviously, Masters can't move up and regular players can't move down.

3. When a player plays against other players outside his level then there is a maximum ELO that he can win or lose. Something like plus or minus 50 percent for each level (so, if a master plays a good player and wins 15 elo, he only wins 3 elo (minus 50 percent for master vs expert (7 elo) and minus 50 percent for expert vs good (3 elo)). Since the levels are sticky, players at different levels aren't really competing against each other for position. This should encourage players at higher levels to play against players at lower levels.

Anyway, this would need a lot of fleshing out but would address a few of the items that have been mentioned as problems with the current system.

[sprockitz]

Lots of thoughts here, but I'll briefly add my 2 cents.

With 90+ games on this site and as one who plays a lot of them, I believe shortening the 60 day period to maintain an active ranking would be a bad move, and forcing them to play someone similarly ranked to keep it would be even worse...just not enough time for people to do it when they play lots of games.

Elo in multi-player. Each game is seen as a series of 1v1 matchups. At first I thought reducing the weight of matchups not involving the winner should be reduced. Whether reduced or eliminated, however, creates a major problem where a player who cannot win will now try and help the highest Elo player to win. They'd do this b/c losing to a high rated player means less rating points lost for the individual. So I think it is okay as is, no system is perfect.

I don't put much stock in rankings/ratings for luck based games as let's be honest, those are mostly achieved via table restriction if not more sinister methods Point is if luck is too high in a game it becomes very difficult to distinguish skill, would take way too many games, and a system that doesn't heavily weigh recent performance.

As for seasons being used to rank players instead of Elo, I'm not a huge fan of the current proposal.
First off, what is winning (if this is 1st place, then it makes players only want to play 2p games, if not then it encourages ganging up on the Master to make sure you get the big points).
Also, games not played as much would have issues trying to find unique opponents.
Finally, the exponential nature using power of 10's puts too much emphasis on opponent's Elo and makes playing the highest level possible so important.

In a fully turn-based world, I think a league system or multi-round tournament system would be more effective for ranking players. (think a multi-stage round robin tournament, if we did championships every 3 months you could have 3 sets of round robins with a month to finish each round, less popular games would likely only need 2 rounds). Again this works better for 2p, but has some possibility for more (though the 1 month time frame probably needs extended for games with more players).

[demiurgsage]

Elo in multi-player. Each game is seen as a series of 1v1 matchups. At first I thought reducing the weight of matchups not involving the winner should be reduced. Whether reduced or eliminated, however, creates a major problem where a player who cannot win will now try and help the highest Elo player to win. They'd do this b/c losing to a high rated player means less rating points lost for the individual.

I think this problem can be solved. Let the player who didn't win receive a loss against an average opponent (sum of losses against each opponent divided by number of opponents) not against the winner. And the winner gets a victory against each opponent (what is equal number of opponents victories against average opponent). After that a small bonus for 2nd (3rd) places can be added if needed.

[Liallan]

Instead of 60 days, I think rankings should only last for 30 days, and if you're in the Top 20, you can only keep your position if you play someone ranked similarly as you.

I'm sure people playing turn-based and taking over a month to finish a game will appreciate that. Not to mention what sprockitz said - that's difficult for anyone who plays a lot of games.

Players in 2nd through last place however should only lose points in accordance with ONE LOSS (even if you are 5th place) based on the ELO rank of the player who WON (1st) relative to theirs. This is because they have only lost one game. It is incorrect to assume a player placing 3rd played worse than a player placing 2nd, simply because if a player is playing to win they may take greater risks that actually hurt their finishes place. Players should still be encouraged to take these risks, however, because it is in the spirit of the game.

ARG!! I've played plenty of games where the people coming in 3rd or 4th are indeed worse players, and not necessarily someone who took a risk. Not to mention that if I can't win, I'm going to go for second - I'm going to try to do the best that I can, and you don't want me to get rewarded for that. If I don't think I can catch the winner, I'll start trying to hold others back. There are games where I got really good at coming in second but could never quite win, and through more and more play would finally be able to win. But apparently the ability to get close to it has no meaning to you, and I would never get anything out of doing the best I can, and be able to do better than 3rd or 4th or 5th. There are other games where I am quite terrible and come in 3rd or 4th a lot, and not because I was taking risks that didn't pay off, but because I'm terrible at the game. I shouldn't be awarded as much as someone who can do better than me. Not to mention that winning a game does not always involve risks, and all risks do not involve falling way back if they don't pay off.

Furthermore, the site should not be encouraging how people want to play games, but only that they don't cheat in some way.

[latexrhombus]

ARG!! I've played plenty of games where the people coming in 3rd or 4th are indeed worse players, and not necessarily someone who took a risk. Not to mention that if I can't win, I'm going to go for second - I'm going to try to do the best that I can, and you don't want me to get rewarded for that.

Firstly, you clearly did not understand (or try to understand) what I said as 2nd place would still gain back more points for each player placing below them (just not to the degree they are now). The point is to disincentivize playing for 2nd, not to demonize a 2nd place finish. Please do not be so alarmist.

Secondly, if you are ever so much worse than the person in first place that you have NO chance to win, you do not deserve comparable point gains to them as is sometimes the case in the existing system. This is obviously most extreme in 2-player games since you either GAIN points (for winning) or LOSE points (for coming in 2ND).

Let's take a game of Caylus for instance. If I was to play a worse player head-to-head, there is a very high probability they would lose, resulting in me gaining points and them losing points. If we were to play again by adding another 2 additional players (thus making it a 4-player game) and that player was significantly better than the other two and ended up finishing second, s/he would very likely GAIN points for essentially playing just as well as when he LOST points in the head-to-head matchup. As a result, there are many circumstances where the existing system actually favors playing for 2nd place finishes.

Caylus is a highly skilled game, so it may not be as frustrating, but in games like Love Letter (which are often cited), these high placement incentives can actually ruin the game for someone who is just trying to enjoy themselves.

[demiurgsage]

Top players say that if rating becomes more competitive they will not have enough time to compete in every game.
I think the solution in change BGA's leaderboard.
Now you score in the leaderboard for achievements in every game. It's not fair as most of the players don't have time to play every game. The solution is to score in the leaderboard for best 5 or 10 achivements in all games. But some games are more popular than others and to get an achivement in these games is more difficult. So achivements for the games should be normalized by multiplying score by the square root of number of active players in the game.

[latexrhombus]

I'm still not convince a leaderboard would add anything (except more complexity). Also, what types of achievements did you have in mind? If the reason for the leaderboard is time to compete in every game (which I'm not sure how allow that), achievements could not be things such as "Most Games Played" as these would clearly be time-dependent.

Let me know what you're thinking as I'd like to add to it. Just as it stands, I'm not sure I see the benefits.