Board Game Arena

Hello everyone,
I think this is going to be a very common topic and you guys already had lots of discussions, but since I see no effective improvement there, I think we should start discussing it again – about the ELO and ranking in every games.
I almost play Tokaido all the time here, so I can’t make sure 100% that this problem is in any game else, but while checking top 20 best players, I recognize a lot of them are very “inactive” (don’t play in most of the time, only manage to play 1, 2 games per 60 days to keep trophy) – or still cheating, by many ways. Take examples here:
- Ze fox: he only plays 10 games from July 2015 to now (1.5 years) – 1 of them was unranked. And everytime he played, he gained back the trophy (top 5 and above), no matter the results was. Is it fair to others when they played a lot and sometimes get very high ELO but not enough to overtake his place? (FYI, half of top 20 players or more only play 2-player game, which is easier to gain high ELO comparing to one plays 3p or 4p game).
You can check his history game in Tokaido here: https://en.boardgamearena.com/#!gamesta ... me_id=1003
- Journeyman: he is on the top of Tokaido ranking from maybe 1.5 or 2 years, with most victories. But the problem is, he only plays unranked games!! Or to be exact: check his history game in Tokaido: https://en.boardgamearena.com/#!gamesta ... me_id=1003 . Here you can see, from February 25th to today, he played 170 games and 167 of them were unranked! Moreover 3 ranked games of his were with beginners! (#28919182, #29383654, #28668073). I played with him some games (but only in ELO off mode, he simply refuses to play ELO on with any good player), he is not a bad player, but how can he become the best player in Tokaido with such kind of playing?
Here is about Tokaido only, I think we may see this problem in some games else like Stone Age or Tzolk’in (any game that has variants from 2 to 4-6 players may happen). So… is it fair for others who plays the more-player variant and win a lot, but can drop down easily by losing 1 or 2 games? As you know, the ELO calculation for multiple players (like 3-6 players per game) is using by adding expected score of each player and combinating them to have a results. Therefore the chance of losing ELO for players playing in these games is much higher than those who only plays 2-p game; not counting the fact that 2p game will have luck factor reduced (most of the games).

I think we all see this problem, but how to solve it? I have some ideas about these things, hope to see your ideas as well.
1) Ranking is for players who play “frequently” enough, not only play some games in order to keep the trophy. We should restrict the rules by adding these conditions:
- Having played at least 15 games in each 60 days (or 5 games in 60 days with turn-based mode). I think 1 game in 4 day with realtime game and 1 game in 12 day with turn-based are enough; more games could lead to argument like the game will be very long,…
- 80% of these games are ranked – we just simply don’t want the trophy belong to anyone who is coward and refuse to play ELO on with any others. 20% left maybe unranked in order to study or exhibition, this number is (I think) good enough.
2) Best player should play with good enough players to prove their ability, not playing with beginners to earn small ELO but enough to keep trophy. So how should the “good enough players” be defined? Using the expected score when playing a 2 player game: 1/(1+10^(Ra-Rb)/400), a game which is still not totally imbalanced if the expect score of the higher elo player are not exceed 0.85 (or even 0.9), otherwise the game is one-side and we’ll all know who is the winner. By calculating, the expect score listed above is about 300-400 elo difference between players. Therefore, we can restrict the player’s selection by: Expert (500 elo above) must play with good and above, Master (700 elo above) must play with strong and above (all are 300-400 elo difference), or even some games which has players from 900 elo and on, we can set a new name for them (like Grandmaster in chess), and restrict to play with expert and above.
However, the biggest problem here is playing tournament, where everyone can join and a master still can play with a beginner and keep trophy this way. I’ll think about a solution later.
3) As some of players here know, BGA has changed the elo calculating way by changing the coefficient K from this January. This leads to a problem: some players who played before 2017 but stop playing since 2017 will have higher rank with one plays from 2017 or still playing from 2016 to 2017. You’ll understand this feeling when playing Tokaido 4-p game and now it’s difficult to reach expert or even master but very easy to fall down… So, recalculating ELO for these players who played before 2016 and try to use this trick to have a high place on the ranking might be an great idea; though it takes time and energy to do clearly.
These 3 ideas may not solve all, but of course in the near future we can think about more solutions to stop some “coward” players to have a high place on the ranking, and let some who deserves these places have them. I’m not a big fan of caring my ELO or ranking, but when we use it as a method of defining players, we should use it effectively and reduce the problems as good as we can.

Hope to see any improvement from BGA soon. Thanks for reading!
lachimi

I definitly agree with all of this (except the 3, there are some arguments against too, but this is not the most important subject).

It has been already discussed in the forum, we did some research and identify that for a large number of games this is true.

This is quite difficult to find a solution that satisfy everyone : those who wants to play few games, and those who want to play a lot, but we will make the system more favorable to those who play a lot. I can give you a date for this but definitly this year.

sourisdudesert wrote:I definitly agree with all of this (except the 3, there are some arguments against too, but this is not the most important subject).

It has been already discussed in the forum, we did some research and identify that for a large number of games this is true.

This is quite difficult to find a solution that satisfy everyone : those who wants to play few games, and those who want to play a lot, but we will make the system more favorable to those who play a lot. I can give you a date for this but definitly this year.

I seriously think that the 3rd argument is a big deal. The changing value of coefficient K may not a big problem if we count discrete games but when we sum up and (even take the average of results), the changing of ELO is very big. I should have taken an example of number of experts in Tokaido (the game I played) but I didn't take any pic to be an argument so... very hard. But I remember that, before 2017, there were about 65-70 experts and above everyday in Tokaido, but now, there are only about 40-50 experts and above. Given that some of them are retired, but the fact here is, with this ELO calculation we're applying, it's much more difficult to reach a new level (strong to expert or good to strong...), but easier to lose a better rank. It leads to the fact that some players who only plays game in 2016 will have higher elo when being converted to the new system - comparing to who plays in 2017 only or keep playing from 2016 to 2017.
(I'll calculate my elo in caylus with new system - I quit this game very long time ago and though have no intention of playing back, but it's interesting to see the difference of elo there).

Just saying my opinion here to prove that changing ELO system will lead to many pros and cons; but I understand that it's not easy at all to calculating all the old games by this new system. However, I'd be pleased if we can do it.

Anyway, thanks so much for listening these ideas and I hope that date of changing thing will come as soon as possible!

lachimi

Does the change of K value means there are less ELO gain/loss in each game, so that if players play more games, they can eventually reach the elo level of high ranked players?

Jessicannot Sp eak wrote:Does the change of K value means there are less ELO gain/loss in each game, so that if players play more games, they can eventually reach the elo level of high ranked players?

Given that the probability of winning a game is still the same; let's convert it into elo through the change of coefficient K. K=40 before 2017, suppose that your ratio is 6/1 (6 win/1 loss), each win gains 20 elo and each loss lost 20 elo (equivalent to the difference of elo between 2 players is 0), while using K=20, each win gains 10 elo and each loss also lost 10 elo.
So after 7 games, in the old system, you'll gain 100 elo, and only 50 elo in the new system.

You're right, it seems that if players play more, they can reach the elo level of higher one. But the problem is the probability of winning a game is unstable; and even in many player game, the chance of having a bump is much smaller. Therefore, it's hard for players now to reach the elo of higher players.

Overall, you can see an example here - my elo in Tokaido: https://en.boardgamearena.com/#!gamesta ... finished=1
Here I spent 12 games to be from 492 elo to 707 elo (215 increased) - 18 average. What happened if I have this winning streak today? I think I'll have only like nearly 600. It seems no way to reach master

Thanks
I guess the problem is especially serious in some highly luck-based games (eg: senet) because players used to gain high elo by playing with high elo players, and a few winning matches allow them to have high elo so very likely high ranked players wont play elo anymore
as the winning rate cannot be so high in these games even by playing with complete beginners, it is difficult to gain high elo again because it is nearly impossible for players to win in a row and loss their elo from time to time. as they gain less elo in each game, they may have difficulties in achieving high..
and about your (3), there is 200 elo bonus to old players, making players who play their first game after the new elo system a bit discouraging to achieve high..

OH, but why when you are playing with the same 3 players, you win 14 elo when you are 1st while you lose 27 elo when you are 4th?

Jessicannot Sp eak wrote:Thanks
I guess the problem is especially serious in some highly luck-based games (eg: senet) because players used to gain high elo by playing with high elo players, and a few winning matches allow them to have high elo so very likely high ranked players wont play elo anymore
as the winning rate cannot be so high in these games even by playing with complete beginners, it is difficult to gain high elo again because it is nearly impossible for players to win in a row and loss their elo from time to time. as they gain less elo in each game, they may have difficulties in achieving high..
and about your (3), there is 200 elo bonus to old players, making players who play their first game after the new elo system a bit discouraging to achieve high..

It depends on the game you play. If that game is highly luck-based, playing some with top players (even with 3-4 players) and wins all can help you gain a lot and become the best player (this normally happens in Saboteur, Secret Moon, Love Letter,...) then quit that game is at least enough to keep trophy in 60 days; however, it's difficult to earn it back. Luck is a double-edged sword. If that game is mainly tactics-based (and 2-player game moreover), you can simply play enough games with beginners or in the level that you think you can beat them easily to climb slowly to the top;...

Yes, 200 elo bonus to old players are their advantage. One of the biggest problem here. So I think my 3rd argument is reasonable

Jessicannot Sp eak wrote:OH, but why when you are playing with the same 3 players, you win 14 elo when you are 1st while you lose 27 elo when you are 4th?

I forgot to explain, the actual score will be calculated by taking the real score subtract the expected scores. In a 4p game, the elo will be calculated by supposing you have 3 2p game with each other; so the correct formula is: K*(N/(2*(N-1))*(The actual score - 3 expected scores); with N here = 4 (a 4 player game). Therefore the elo while playing 4 player game is not reversed, leads to the difference of finishing 1st and 4th when you play with exact 3 other players.

Here is an example of calculating elo, you can check the formula there to understand : https://docs.google.com/spreadsheets/d/ ... sp=sharing

About the formula (N/2*(N-1)), there was already an article about this: https://en.boardgamearena.com/#!forum/v ... =3&t=10959.

I hope this explanation is easy enough for you to understand.

lachimi

so when the N is greater, there is less elo gain/loss after the game?
And in the post, some players suggest about suggested number of players. as tokaido have 4 as suggested number of players, so when 5players play this game the N is still 4, right?
About the example, it looks great but I still dont understand how do you calculate 'the expected score of certain rank with others'. do you mind further explain ? thx