Some weaknesses of the current FFA rating system

In 2p chess, the rating tends to predict the final outcome quite accurately. If a 2100 plays against a 1800, it's highly unlikely he would lose. Whereas in 4p chess it's MUCH more complex, as there are plenty of human interactions and numerous possibilities.

Typically, you may be 2800 here and still VERY EASILY lose against 3 1800 rated players and finish 4th as they would mostly be afraid and would attack 3 vs. 1. (And even in anonymous games you may easily be in this case as the guy in front understands nothing about the 4p chess basics (cf. https://www.chess.com/clubs/forum/view/basic-ffa-aspects), would in many cases attack you or weaken you (finishing himself 3rd, cf. https://www.chess.com/clubs/forum/view/to-betray-too-early-means-3rd-place-in-95-of-cases), and you are 4th like that).

As direct consequence, most 2400-3000 players would never join a 2000 or 2100 queue as they know that if they win, they would get maybe 5 pts whereas if they lose, it can be -50, and the mathematical expectancy is just very very bad .

In itself it's not a problem, but because of that a) weaker players don't progress as fast as they should, b) stronger players quite often have to spend hours waiting for a game of their level. Even when you're only 2400 this problem exists already. So sometimes you're fed up of waiting and click just on any FFA. And of course it's a stupid idea and you get punished.

I joined a 1800+ game today; the ratings involved were 1918, 1933, 1993 (and I was 2376, yes, I dropped from 2450-2500 recently). 2-3 years ago 1900 was still a decent rating but now it's mainly bull..it. Unfortunately. And in addition I was green

So the sides cooperated almost correctly (with several opportunities missed and some mistakes, but still, they cooperated), whereas my opp played very passively, missed 1001 opportunities I created, failed to eliminate the queen of a side when I created this possibility, etc. And of course I finished 4th. One just CAN'T win a FFA with 2 cooperating sides and a passive opp.

I'm punished with a -31 pts, it's totally my fault, shouldn't have joined this low level game. But the problem is (cf. above) that the current FFA rating system is not good for the community as it implies that we have LESS available games and LESS opportunities to learn and to progress for weaker players.

How to solve this issue?

That CAN be done. ONCE AGAIN (it's more than 1 year that I've been proposing this), let's create a rule limiting the loss/increase in points for games where the delta between the strongest and the weakest players is beyond 200. The result of such games (as much less relevant) should be discriminated, typically by dividing the points variations by 10 for instance. 

Like that, stronger players would much easier join lower levels' games as they won't risk more than 2-4 pts of their rating. There would be more games and the overall activity would become more fluid. We won't be anymore in a situation where it's impossible to find a game for 2 hours even if one wants to play just for fun (of course there are casual games, but in fact there are very few).

Cf. also https://www.chess.com/clubs/forum/view/another-proposal

Basically, introducing that would let players totally free: if they join a large spread game (like 1900-2400), they know that maybe the result of this game would be discounted (if the levels are too different). And if they join a game like 2400-2600, they know that it would be a fully serious game.

And also the payers who LAUNCH a game would be free to decide whether they want to accept players with very different ratings (in which case maybe the result won't be taken into account in full) or they want it to be pro.

In other words: this proposal would create a new class of games between standard rated and casual. Some games would still modify the ratings, but not a lot.

Not really going to comment on the post. However, you say you were green as if that's bad?

Green is the best colour.

1) You get to see your opp's opening move. If it's bad, you can abort.

2) Green has the highest percentage of wins. I think it's around 30%, but you'd have to check with Luke.

I agree with 1°, Tom.
I'm very surprised as for 2°. I remember about some stats GustavKlimt posted 2 years ago. Statistically green was definitely worse than red, and slightly worse than B and Y. Am I wrong?

(But personally IDC about being G, I think my results are totally the same for all 4 colours as I don't play classically)

I disagree with encouraging the practice of farming by lessening the punishment for playing in low rated queues.  If one wants to farm, they need to accept low rewards and high punishments.  I farm all the time, but I shouldn't be rewarded for playing with players far below my rating.  And let's not play games here; Red is of course the best color.

Grathienben, I think on the contrary that THE CURRENT system encourages some farming. I mean, no one would continue to farm if TEN TIMES MORE games are needed to reach any given level. Whereas now, if/when someone is good enough, the risks are not THAT high in reasonably high level games. I mean for instance someone with 2800 playing 2400 queues, not 1800 queues for sure

Thinking about it in Teams, RY is about +4 in the current setup.

In Omatamix, it’s only like +0.4.

Why am I saying this? For absolutely no reason! There’s no Teams in lower rated games! Seeing a damn 2100 doing opposite cooperation or even mimicking it is like seeing a four leaf clover or smth.

Rating changes are just broken, sure you shouldn’t get rating points for beating lower rated players, but the rating change really needs a floor (maximum rating change), probably calculated based on rating & Glicko RD…

But then again, if we’re looking at high level 2800+ games, at least new setup will be more tolerable for BG…

this is not a weakness of FFA, it is a weakness in glicko system. but it is difficult to call this a weaknesse, this demonstrates the most relevant player rating (not only the level of play, but also the player form). there is an alternative used for console games, in which you quickly get your rank, but then very slowly move to another (you will not lose much at one time, even with very weak opponents). for an experienced player with a more or less stable rank, this system was very good, as it smooths out the moments when player is tilted, but for newbies who progress quickly, it would be a quiet horror
what you propose will generate rating inflation, so it is not suitable here. need to look for or come up with new systems for multiplayer games that would fit well for FFA
can automatically reduce glicko of players who cross a certain rating boundary, as is done in classical chess. e.g. after 2500+, reduce the player's glicko to 20-40 from the usual 60-80, and then it will work similarly to the upper alternative. but this is a controversial issue

YES ABSOLUTELY.

One may lose or win a game for plenty of reasons without his level being really different from what it used to be just before (disconnections, total idiots in front, emotional people playing to destroy just you, new profiles of experienced players with 1500 but in fact 2400 FFA, etc.) and to have a variation of more than 20 pts just DOESN'T MAKE SENSE. +40? -35? Your force just can't vary that much from 10 minutes beforehand, that means that either your previous rating or your new one are not close to truth.

Certainly not worse than NOW!

And anyway it's feasible to address this problem by adjusting the system as follows: 

We create an ALPHA coefficient, which is the delta between the strongest and the weakest players of the board, divided per the average of the board.

For instance, in the game mentioned above, it would be (2376-1918)/[(1918+1933+1993+2376)/4]=458/2055=0,2229.

And we discount the changes by (1 - 4 ALPHA) for instance, here it would be: 0,1084. I.e. instead of -31,2, it would've been -3,3821 pts for me.

ALPHA would just show the coherence of the ratings involved. Typically imagine a game between you, Riba, Hest and Rojitto, with your current ratings. ALPHA would be: 0,030138, as your FFA Rapid ratings are quite close. So a game between you would mean a lot, pretty relevant. And the results wouldn't be discounted or rather, very few (just multiplied by 0,8787, i.e. almost taken as is).

You come from the premise that it is good for 4pc for 1800s to play with 2200s...I'm not sure that is the case. I would think that being demolished (as often happens look at crimsonlee) would hurt, not help 4pc and player retention.

(with this addition, of course games where ALPHA is more than 0.25 would become automatically casual , and I find it very good: such games are simply not relevant as for rating purpose...

Example: 1700, 1750, 1950 and 2800 play together. ALPHA = 0.5366. And it's logical: it would be a fully b...it game, with no relevance. Hence casual).

Simpler is better. I think in that case, 2800s should not be allowed to play 1700+ games altogether.

I also proposed that, some months ago (=games with more than 200 of delta between the strongest and the weakest => casual automatically). I was told it was a silly idea as above 2700 games are extremely scarce, etc.

I suggest that it gives the higher rated players a challenge

Nope, you can't give to 1st what 4th lost. Their levels vs. the average aren't the same, etc.

lol I was 2302 hyper bullet and played a bunch of 1/4 ffa standard games and (playing against players rated between 1400 - 2100) it took me about 20 games of winning and the sometimes getting second with @rk_killermac (he was playing a lot of this mode with me at that time) as first. I then played three games with a lot of lag and just down right bad playing and went from 2350 to 2183. And NO I am NOT asking for a rating refund (it sorta bugs me when people do that) I'm just trying to give an example for this particular aspect of your post.

LONG STORY SHORT: I agree

lol