Oh, just saw this. I got this formula by grabbing stats from the Siege Engineers ratings chart: https://ratings.aoe2.se/
The website is updated every day with new leaderboard data. I went to the project’s GitHub repo for the prior commits and grabbed the leaderboard data from the commits immediately prior to and immediately after the Elo change. Then I just paired up the old and new Elos for each player and did a simple regression. If you plot the points they’re in basically a straight line, with a correlation coefficient of over 0.99.
(Note there are some players who played games between the two leaderboard updates, so some of that data is affected by these games in addition to the overall Elo readjustments. But overall the approximate formula works.)
Inflation will be an issue again very soon, we are getting and losing more per win or loss, meaning that stackers will skyrocket back, how come that a match vs guys 30-50 points lower can take 22 points, the rating exchange doesn’t correspond with the players elo.
The easiest fix is to simply copy the old voobly or even the zone system by reducing the points exchange for 4x4, 3x3 and 2x2, the base shouldn’t be 16 but 8 for 4x4, 10 for 3x3 and 12 for 2x2.
I am losing much more games since the new ELO system. 
Idk, I think this will depend on the matchmaking how the teams are matched.
Depending on the matchmaking there will be inflation or deflation. It’s only dependent on how the teams are matched not WHO WINS. And yes there are matchmakings that will always LOSE elo.
This is expected for all players who have lower than average elo cause they need less wins to keep up their elo. High ranked players will have very high winrates (and therefore also kinda boring matches most of the time).
The biggest issue I see is that with this low ranked players will be turned off multiplayer and then mid elo players and so on. It’s even likely that at some point low elos will be alienated for having bad winrates.
This isnt true in practise. For most people there win rate is about 50%. Only the real bottom and real top will really have different winrates. Pretty much almost everyone who isnt in the top 5% or bottom 5% will have a winrate between 45-55% (if played enough games). You can also see this at the 1v1 ranking which is much more stable. Just look at the top 250. Only the top 50 really have winrates above 60%. Almost everyone else will have a winrate between 50-60%. Only exceptions are the few players with the least games in the top 250.
The current problem at the TG ladder is mainly that the TG ladder is a complete mess. The calculations were bugged from the start. The devs already change the calculations multiple times. But the ladder just need time to settle again based on the new calculations. People who were overrated in the old calculation will probably drop elo in the new calculation, so they have to loose elo and thus lose much more. This just needs time to settle and will only really settle if the new calculation is fine. I am pretty sure it will really take a long time before it settles. It needs to unwind all mess of the previes calculation. And this new calculation is terrible again. So i dont think it will become good with the current calculations.
It is true. I’ve calced it. If you have lower elo you will most likely be paired more often with players of higher elo and fight against opposing teams with higher average elo than you have. This means you need a lower winrate to maintain your elo and therefore you will end up with a winrate below 50 %.
And vice versa ofc: People with high elo will need high winrates to maintain it.
Ofc you can’t see it now cause the patch isn’t on for long, but it will happen.
This is correct. In the new system, no longer does the winrate of most players converge to 50%. Which means it is a stupid system.
Apologies I’m not saying you are wrong but would you be able to show evidence of this or at least the workings for theoretically it would be the case. Apologies it’s just I’m not sure I understand why this would be true for people who aren’t at the extreme ends of the Elo distribution
I can give you a way how you can calc it with the distribution you have:
A) pick a player with a certain elo (like 800)
B) advise 3 players of +/- 200 elo to his team (randomly chosen of the players in this range)
C) let this team play against teams with +/- 100 average elo
D) see how his elo has changed
You will see if you picked a player with less than the mean elo he will always gain elo. The reason for that is that the current elo system overestimates him. To “hold” this elo he will need to lose more than to win.
The reasons for that behaviour are also a bit more complicated than it is immediately visible, but I will try it:
The Elo distribution is kinda close to a standard distribution. That means if you go off the center you will always have more players in your proximity that are closer to the mean than on the other side. So with about 800 elo you will probably have like 3-4 times more players in the 800-1000 elo range than in the 600-800 elo range. Which means that a 800 elo player will preferably paired with players of higher elo. So he will in many cases in the matchmaking have a lower elo than the average of his team. This team is then paired with other teams of approximately the same average elo which means this player has in most cases also lower elo than the opposing team. This means he is expected to lose most of his games by the current team elo system. IF he would win more he would gain elo, but then his initial elo would have been wrong in the first place.
Was this understandable?
Edit: Maybe the effect will only be clearly visiable in a few weeks cause it takes time until the players will get a good representative elo for the new system. I expect that many players with lower than 1k elo will gain some elo and players above 1k will lose some before the discrepancy in winrates will be broadly appearant. Maybe it’s a good idea to safe some pictures of the distributions and compare them with the distributions in a few weeks. I expect the distribution to be narrower then. (Has someone saved the initial new team elo distribution?)
I think I get what you say. However, I have a question regarding this part:
If this player has most cases a lower elo than the opposing team, then he has most of the cases a teammate that has a higher elo than the opposing team (assuming approximately the same averaged team elos).
Reusing your words:
This means most of the time he has a teammate that is expected to win the game by the current team elo system.
Thus, combining your and my example, the elo system expects that the original player loses and his strongest teammate wins? Or does it matter that the teammates change from game to game and I’m not allowed to make these conclusions?
Can’t your reasoning also be made for 1vs1?
A) pick a player with a certain elo (like 800)
B) find an enemy with a elo close to 800 (randomly chosen). Because of the distribution it is more likely that he gets mateched with a player that is closer to the mean, i.e. a elo larger than 800. Thus, the original 800 player is expected to lose more often?
But are you not assuming here that they are always apart of the wider deviation team? Surely they are equally likely to be on the ±100 team to balence it out ?
I think this is an assumption isn’t it? As in they could have implemented a system that still provides an even spread of matches despite a skewed Elo?
In most cases yes. That’s what causes the issue. The system doesn’t regards the team as a whole anymore, but expects different winrates from different players in the team.
This is important. If the Teammatees don’t change every single team member elo will slowly converge to the mean team elo (and ofc all will have the same winrate). But if the team members change, players with lower elos will generally lose more games (cause they are expected to) and players with higher elos will generally win more.
That is true, but there isn’t the in-between step of team building. In Team games his team elo will be pushed upwards because of his teammates and therefore he will be matched with even higher players than in 1v1s.
But yes, if you look up winrates in 1v1s you will see that players below 1k elo will in generally have slightly less than 50 % winrate and vice versa. But it will only have a big effect on the edges of the distribution.
Yeah ofc, it’s an asumption. I just wanted to explain it understandably why this occurs.
Unlikely. I even don’t know how you want to implement a System like this… I don’t think it is even possible cause how do you want to provide this on the edges of the ranking? There are no players of lower or higher elos left to pair with…
It’s not the point. The point is that a player with lower than average elo will in most games be in a team with higher mean elo than himself and then ofc also fight against a team with higher elo than himself.
If the matchmaking is like this, the system expects him to lose more games than he wins. And ofc if the system expects him to do so he will to hold his elo. It’s actually that simple.
Maybe it’s easier to explain it for top players. If the best players in the world queue in team games they will be paired with lower elo players (as there are no better ones). This pulls down the mean team elo. Then their team will be matched with an accordingly high mean elo. Just eg if a 2k5 player is paired with three 2k1 players. The mean elo is 2k2. Then they are matched with a team consisting of 4 2k2 players. In this scenario the 2k5 player is expected to win like 85 % of the games (sorry i don’t have the calc right here currently). So the system already expects him to win these amount of games and if he doesn’t he would lose elo. And this in a Matchup the system thinks is “Balanced”.
The result of this calculation is that high level players will be vastly underrated cause they need a higher winrate to maintain their high elos. And they gain that higher winrate by beig vastly underrated.
And ofc it’s then vice versa for low elo players. They will be overrated cause the system expects them to lose more and they will lose more cause they are overrated.
Everyone nowadays does fast castle into castle drops and spamming of Unique Units, despite the feudal agression. I just lost a match in African Clearing despite killing about 10 villagers in feudal age, because the opponent went fast castle, made a 2 gold unit composition and kept castle dropping me. Something similar is happening in Nomad as well. Only Arabia seems to be dectnt nowadays. This game is somehow ruined.
don’t look too much into tg elo, it punishes you for playing right now… just look at tatoh losing 500 tg elo post rescaling while being maybe the best tg player active on the ladder. The more you play the more inaccurate it will be.