Analyses of the ratings - Spotting the issues

WoodsierCorn696 · June 18, 2020, 3:31pm

https://aoe2.net/#profile-76561198118286822

I found out a new great profile. You win 4 out of 10 1v1s. Your position on the ladder? #24 of the world.

I would challenge someone like Viper and Daut start an adventure with a second account. Play only teamgaemes and win almost everything until they became 2.8k team rating. Then both start playing 1v1 for the first time on that account. I wont be surprised if one of those accounts will hit 3k for 1v1.

If some current team game player can you the ladder at #24 without even playing the real top names, then someone like the viper can you the ladder with a secondary account why higher.

This shows how broken the current ladder is.

anon642a · June 21, 2020, 6:16am

What harm does this cause?

WoodsierCorn696 · June 21, 2020, 11:52am

The source is unknown and dev still dont react. I hope they will find the issue.

anon642a · June 21, 2020, 5:03pm

I’ve heard that you are gaining or losing points based only on the highest player on the other team, and it seems to make sense given the results I’ve seen (beating 2x2.4k+2x1.2k as a 1.7k player in a 4v4 gains you a ton of points). Thus you’re gaining the max and lose the minimum. Would be nice if they made it based on the average, but yeah, I don’t see any real harm this is causing.

RoninSan7691 · June 21, 2020, 9:05pm

I don’t know the exact numbers. But when I win team games I get like plus 15-20 points and when I lose is minus 10-15… So obviously the ELO will get inflated even with a 45-50% winrate. Something is wrong!

Obuk8723 · June 24, 2020, 5:02pm

hey, thats a great thread i am seeing just now.
I made a similiar thread, but I know what the issue is.
check herE: TG-Matchmaking - we get unbalanced game bc elo calculation is wrong

its basically a terrible calculation that values the indivduals elo over the average elo when rewarding points. the good news is, that its an EASY FIX if someone reads this.

my assumption of the formula is sth like this:

the current formula is sth like:
+/- 0 for win/loss of balanced games, +/- 3 for deviation of AVERAGE elo, +/-20 for deviation of RELATIVE elo

a better formula would be sth like:
+/- 10 for win/loss of balanced games, +/- 10 for deviation of AVERAGE elo, +/- 5 for RELATIVE elo

WoodsierCorn696 · July 1, 2020, 9:38am

I mind have an idea why ratings are inflated.

At Voobly (and probably HD too) it seems like a team game is considers as 1v1 to find the total amount of Elo win/lost by the teams. That number seems to be distributed around all players.

At DE everyone seems to be compared to the average elo of the enemy. This will result in larger deltas for all players in the first place.

Lets look at some examples:

Example 1

Voobly
4v4 - all 1600. Winning team gets +16 elo (distributed around all players), so they all get +4 elo.

DE
4v4 - all 1600. Every player is 1600 and this will be compared to 1600. The result is that everyone on the winning side will gets + 16 elo and everyone at the loosing side gets -16 elo.

There is a problem with imbalanced teams (not everyone having the same Elo).

Example 2

Voobly
4v4 - one team is all 1600. Other team is 3x 1500, 1x 1900. Winning team still gets 16 elo to distribute. If that is team 2, then the 3 players with elo 1500 get a little bit more than 4 Elo (probably 5), will the higher rated player will just get 1 point. In total still the 16 points. In the other team everyone lost 4 points. Also a total of 16 points. So everything stays in balance.

DE
Seems teams as at Voobly:
If 4x1600 is still the loosing team, they all lost 16 elo. 4x16=64 elo in total.
At the winning times things are different:
1500 rated players will all get + 20 elo.
1900 rated player will only get + 5.
In total this will be + 65 elo.

The difference in this example is just 1 elo, but it is already an inflation of 0,25 per game for each player. Thinks can be even more extreme.

Make it 3x1400 + 1x2200 and they will even get +72 elo, while the loosing team only lost 64 elo.

In the end we even hadnt looked at teams of non equal strength, which can even make this kind of thing even worse. We had for some time the issue of unbalanced teams (fixed issue some months ago). Those unbalanced teams made things even worse.

So the issue is that total elo win isnt equal to total elo lost. This makes rating inflated and needs to be fixed.

Another point

In the first games an the ladder you will get more points. There is some kind of multiplyer. If the average rating is above the starting rating of a new player, they will on average move up to the average rating. This means also that they will gain more points, then they will loose. This number will be multiplyed be a factor. This will also lead to and inflation of the ladder.

I have to say: I do like this feature. It makes sure starting players will get there true rating more quickly. But this only works if the ladder is not inflating or deflating. Since the ladder is inflating, it works like a multiplyer on the inflation as well.

Conclusion

We kind of found the two main reasons for the inflated ladder:

winning elo isnt equal to losing elo.
New players works as multiplyer on the inflation if there is inflation at the ladder.

Fixes

Why dont we just use the calculation at Voobly and HD as base? It will fix issue #1. That will be the main reason why there is inflation.
You can still add some multiplyer for new players so they will move up or down more quickly. That is not really an issue when the inflation of #1 is fixed.

If they change the way how the calculations work, then i think it is a good idea to recalculate the rating of every players again. I know there are APIs where you can get the full history of a player. That makes me conclude the devs will have such list too. They just need to re evalute the elo of all players from all games.

WoodsierCorn696 · July 5, 2020, 8:43pm

This is posted at the upcoming AoE update. To makes thinks clear: This about AoE, not AoE II.
I am not really into AoE, but it seems like AoE had also some issues. The devs even decided to reset the ladder.

I hope they will fix the AoE II Team game ladder also at some moment. I dont know if the devs will also reset the current rankings on AoE II after they fix everything.

I also found the following picture about the team game ratings.

WoodsierCorn696 · July 10, 2020, 9:32am

2 months age i started this threads. Until today still no response from the devs.

WoodsierCorn696 · July 19, 2020, 10:28am

RM 1v1: 50% percentile: 1020 (almost 1000, almost no inflation)

RM Teamgames: 50% percentile: 1350. The average rating is already inflated by 35%.
Note that: At May 10, when i started this thread, it was just 1275. So the ratings are inflated by 6% in just over 2 months! Given the current inflation we will hit 1600 as average team games rating! This is really bad design of the team game rating and something needs to be done to make the team game ratings reiable again!

I see they added death match support.
50% percentile for 1v1 DM is about 1050. This is also probably due to not many players playing death maps. Seems like not that much inflation to the 1v1 ladder, which is the same for 1v1 RM. So also no issues with the 1v1 DM ladder.

With DM Team games we see the same as RM Team games. The ladder is really inflated. The 50% percentile is already about 1425. This means an inflation of 42,5% in yust 9 months since release. This is almost the same result als the team game rating for RM.

This shows us that there is an issue with BOTH team game ladders. They are hugely inflated and the devs needs to fix the underlying issues so ranked team games (RM and DM) will get a better experience.

WoodsierCorn696 · July 21, 2020, 9:17pm

I dont know why, because i didn’t read anyrhing about a change of elo, but it seems like that devs messes up 1v1 elo. I heard stories where winning player looses elo in some cases and loosing player gets elo.

If the devs have a look at that i hope they also look at this issue.

hydrogen6099 · July 22, 2020, 12:06pm

The ELO system was designed for 1vs1 games and does not work for team games. It works well for the players on the extreme i.e. Viper will always have a high team ELO and some pleb will always have a very low ELO.

The system is designed to take into account the difference in ELO points but in a team game if you do that you are doomed. How does affect to the winning percentage the existence of a much rated player ? The ELO variance inside the team ? Top rated ? Lowest rated ? Number of players with higher rating in one team ? Premade teams vs non premade ? yada yada yada etc. etc. the list could go on and on.

On top of that then there is the point you are complaining about here which is how you distribute the points after a win/lose game.

WoodsierCorn696 · July 25, 2020, 9:06am

I know that elo is designed for 1v1 and needs to be adopted to team games. Currently there are multiple alternatives that are suitable for team games. I hope that the devs had a look at this when they decides on this elo calculation. So i hope those things will never be an issue. Still you have the difference between premades and solos. Given the same skill level, the premade team will mostly win by superior communication. In the end i think team elo is less reiable then 1v1 elo, but still good enough to use, if there are no issues in the calculation, like i already point out.

I still hope from a respons of the devs on this subject. Still waiting from May 10 till now…

WoodsierCorn696 · July 31, 2020, 5:58pm

The median of the TG RM rating is already inflated by an addition 10 points since 12 days ago. For TG DM it looks like even an addition 15 points. I really hope the devs will correct this issues pretty soon.

WoodsierCorn696 · August 7, 2020, 7:46am

I would love to see a response from the devs. This issue together with the lag/desyncs is killing TGs.

breeminator · August 7, 2020, 2:26pm

If nothing else, they should at least account for the difference in median rating when carrying ratings across between 1v1 and TG. I saw an account with nearly 2k 1v1 ELO where the person had only played 2 1v1 games with a 50% win rate. They had 2.2k TG rating, so I assume what happened is they played TG first, and it carried an approx 2k rating across for their first 1v1 game.

phoenix1089 · August 7, 2020, 3:38pm

Why does it matter so much to you? This whole thread is basically a monologue from you.

The inflation affects everyone. The outliers are minimal. So what if there’s one guy with a very high 1v1 elo from a couple games? It’s not affecting 99% of the other players?

So what if TG elo is inflated? I would happily bet all those top elos are arranged teams so they can happily face each other while the other 90% of the community still faces moderately fair matches?

The single elo for 2v2,3v3,4v4,random teams and arranged teams all being bundled into a singular number is blatantly skew as it is. It doesn’t matter about inflation. Its a completely skew number anyway.

Do you lose elo if you drop from a match but your team still wins afterwards?

WoodsierCorn696 · August 8, 2020, 11:50am

I’ll will try to answer your questions, with posts of other people, to show it is not only me. There are many threads about this issue. It is just that i use this thread frequently. Have a look at all the quotes below of other users. Please let me know if you still think it is only me after reading all the comments. I think all these quotes will give you the answers to your questions. There is also a recent thread at aoezone at this moment about this issue. I post the url below all quotes.

Rating calculations on aoe2 (start rating)

The simplest thing for it to do, if it wants to make use of the existing information on a player’s level, is to start them at the same percentile. E.g. if they are 35% of the way down the TG leaderboard, then they play their first 1v1, start them at the ELO that is 35% of the way down the 1v1 leaderboard.

You’d still have the problem that TG ratings can be distorted if people often play with teammates who are a significantly different standard, but it would at least account for the large overall differences between the ratings scales.

Starting everyone at 1000 is an option, but it means that people lose ELO when they play against a better player who is working their way up. Transferring rating across is an attempt to solve this problem.

UprightAtol98:

Teamgame Elo has been inflated for a long time but I feel like it just keeps getting worse.

At some point I was TG Elo 1750 and Rank 5000.
A few month later I was also TG Elo 1750 but Rank 10.000.
That means, either the number of people playing ranked doubled (which is not the case) or: the Teamgame Elo of everyone is increasing!

That does not make any sense or at least that is not how an Elo system should be working.

I even saw a lot of people who had only a 1v1 elo of about 1000 but a TG Elo of 2000.

Right now it seems your Teamgame Elo depends more on the amount of games you play rather then your winrate…

Because the TG Elo is broken Teamgame Tournaments use the 1v1 Elo of participants to rate their skill group…

Will it ever be fixed?
Will the Devs at least sometime mention what is wrong with it?

TG Rating vs 1v1 Rating (Matchmaking)

Hi Devs,

I am a 1000 ELO player on 1v1 (17,000 ranking in the world) and when I play team games I play with my friend who is much better, 3000 ranking in the world (around 1400 ELO).

Our team game ranking is around 10,000 in the world which works mostly as he usually carries our team, even though our team game rankings and elo ratings are very similar. Because we are playing with people around the 10,000th best in the world he is better and I am worse so it balances.

But when I play on my own, my team and I usually get crushed, because DE thinks I am much better than I am because I normally play with my friend who is much better than me and so I have this “inflated” rating. Likewise, when he plays on his own he normally wins, and so it is not so fun. What’s worse is when I play with other friends who are very new to the game and don’t play so often and are much worse than me DE assumes I am very good because of my “inflated” 10,000 in the world ranking, and so we always get stomped on.

Also Team Game rankings seem very unreliable, my TG ELO is 800 higher than my 1v1, but sometimes I see people who are much better than me at 1v1 with TG ELOs the same as their 1v1 ELO.

So, I think it would make sense to factor in the 1v1 rating as well as the TG rating, because the 1v1 rating is so much more accurate to how good a player is. I hear chat on streams saying that they’re 1300 1v1 and 2200 TG and stopped because they play against people like Slam and it’s not balanced even though according to the TG ratings it is balanced.

Taking into account 1v1 as well as the team game rating would undoubtedly help, and maybe my friends who are much worse will come back to play with me. And maybe I would occasionally play team games on my own because I won’t always get stomped. For players who don’t have a 1v1 rating, then I understand just only using their TG rating in the matchmaking equation makes sense.

tl;dr - DE thinks I’m a top player because 80-90% of the time i play together with my high rated friend, so whenever I play on my own, or even with my newbie friends we always get stomped and so have stopped playing. TG rating is very inaccurate.

TG-Matchmaking - we get unbalanced game bc elo calculation is wrong

Ya, but MM is supposed to be randomly creating these games, so highly unlikely that he will play 20 games in a row that are above his ELO. Random games means that he will sometimes be over and sometimes under the average ELO of the game.

Even if he gets 20 games in a row with a higher ELO (again unlikely) then winning 10 and losing 10 means he deserves his ELO to increase a lot more than decrease. He didn’t break even at his level, he broke even at games higher than his ELO by 300 on average. This is not the problem with current ELO.

Many of us have noticed that we gain more points upon winning on average, than we lost upon losing on average. There lies the problem.

Another issue I hope devs would resolve when they (eventually, if ever) work on fixing ELO, is to make the starting/average ELO the same for 1v1 and TG. I can’t count how many noob games I got kicked out of, because the hosts get scared of my 1500 TG ELO, though it is the same as ~ 1000 ELO in 1v1. This is annoying.

TG-Matchmaking - we get unbalanced game bc elo calculation is wrong

I play mostly team games and I’ve noticed that when your team wins the higher ranked teammates gain fewer points than the lower ones, and that when your team loses the lower ranked players lose fewer points. The gain/loss also depends on the other team of course, but the important point is that the algorithm compares your individual scores with the other team’s scores.

Here is the problem: Suppose you have one strong player teaming with one weak player. The algorithm is probably going to give them opponents with an average skill level somewhere in the middle. If the matchmaking is roughly fair*, they should win roughly half the time. But the strong player is going to be losing a lot more points each time (s)he loses than she gains when the pair wins, and the weak player is going to gain a lot more points with each victory than (s)he loses when they lose.

If they play multiple games their elo is going to converge, and the stronger player will effectively be giving away elo to the weaker player. This will of course cause mismatches in skill level if they ever decide to play with other teammates or enter the queue alone.

*You don’t actually have to assume fair matchmaking for this argument to work. Just note that the expected elo change is p(win) * gain - p(loss) * loss. Playing together means you have the same p(win) and p(loss), but the higher skill player will have a smaller gain and a larger loss, so there is no way both of these expectations can be zero.

Team ELO

Team ELO is clearly not working and something needs to be done. What? I don’t know, but I know what it looks like: a more appropriate team ELO as a result of your contribution and performance in team games. Poor performance should result in a disproportionate penalty-- not to stick it to the guy who gets pwned every game but, when it happens and happens often, their ELO adjustment needs to find a more appropriate and competitive opponent for that person going forward.

Length of matchmaking is a total non-issue at the moment. If I were to guesstimate some numbers for my experience I’d say fully 85% of my team games require me to carry 1, 2, or 3 allies against premades composed of foes approximately at or beyond my level. I’d wait a half hour just to avoid beeing teamed with people with whom victory was never possible.

Three thousand elo players?

As anyone can see the team games ladder has reached 2.8k players sooner than i predicted, 3 months early, so maybe 3k elo players would arise within september or october. I read few guys said"there are not enough players for that"well the evidence suggest the opposite ■■■ there are 200 players above 2500 and 500 players above 2400 and they get higher everyday.

I think it has been enough of the high inaccurate elo team game players, it has been a stacking party, the top players are not even in the top, they are 200-300 points below, there are 1500 1x1 players in 2700, how is that accurate?

The ELO algorithm for team games should be different, players should not be winning 17 points for a 4x4, at best 8 points should be enough, giving half the points you are giving right now should stop the inflation or slow it down.

Stacking players to get higher ranks is a thing, not a conspiration against players with friends, i don’t know if it is possible but if the system detects the same players on a team or in the same party for 2 games in a row or more it should give them less points per winning match.

2600 has been the roof in aoe since its beginning, it was a number that reflexed skills and was a really hard to achieve only for top players, now it is just a clownish party, there can’t be a thousand points of difference btw a player on his 1x1 and Tg elo and right now that is real, just check the guys in top 50-200 90% of them they are 1k points above their 1x1 elo.

Please devs take some actions before reaching the 3k elo players.

Three thousand elo players?

Well, some time ago a guy was complaining in here that his elo was unbelieavebly high in TG - he was 2193, IIRC. Also, he was waaay below that in 1v1, not to mention he considered himself as an “above average” player at best. I won’t call names coz i don’t remember then lol. He mentioned “the game threw me vs the best player of the world” as he reffered to an 2.2k opponent.

As for the difference in 1v1 and tg elo “because he is best in tg”, i don’t believe in - in these proportions, at least. You can beat TheViper and MBL when with a friend, but loses to an moderate AI in 1v1? (poetic license for the metaphor pls).

I believe in op, TG elo inflantion is real. Some kind of pattern always existed, keeping numbers close - even Voobly and HD elo ratings, provided you played enough, weren’t different that much.

Also, i suspect that the disparity on the speed you achieve higher elo can affect matchmaking - for low elo startes, at least - but this is based only in my guts.

Three thousand elo players?

The problem has been dicussed before in other threads. Sum of points gained by the winning team is often larger than the sum of points lost by the losing team. Therefore new points are created out of thin air, which violates the rules of ELO-systems. The effect is that the average rises, so inflation of the worth of ELO points and the top will get higher and higher. This does not have to affect the MM-system negatively for active players, if it’s well designed. But if a player doesn’t play for a long time and comes back, then his ELO will be too low for his skill level due to the inflation.

I already got suspicious of this fault early on: on my first 1v1 game I started with 1000 ELO, but on my first team game I started with an ELO of 1318. So likely the average at that time was 1318 and the system gives a new player in the ladder the current average ELO, which is a correct mechanism, except that the average ELO should stay on 1000.

FinalBucket3743 · August 8, 2020, 3:17pm

I saw: 3k elo is reached by modri

UprightAtol98 · August 7, 2020, 8:44am

Teamgame Elo has been inflated for a long time but I feel like it just keeps getting worse.

At some point I was TG Elo 1750 and Rank 5000.
A few month later I was also TG Elo 1750 but Rank 10.000.
That means, either the number of people playing ranked doubled (which is not the case) or: the Teamgame Elo of everyone is increasing!

That does not make any sense or at least that is not how an Elo system should be working.

I even saw a lot of people who had only a 1v1 elo of about 1000 but a TG Elo of 2000.

Right now it seems your Teamgame Elo depends more on the amount of games you play rather then your winrate…

Because the TG Elo is broken Teamgame Tournaments use the 1v1 Elo of participants to rate their skill group…

Will it ever be fixed?
Will the Devs at least sometime mention what is wrong with it?