Which matchmaking 'issue' is the most important to you? (See post for more details!)

1) AFKs in Champion Select Lobby 4,839 36.32%
2) Duo-Queue Elo Disparities in Ranked 998 7.49%
3) Skilled Ranked Players in Normal Modes 657 4.93%
4) Premade Matching 663 4.98%
5) Transitioning from Normal to Ranked Mode 1,337 10.03%
6) Free to Play Champions in Ranked Mode 799 6.00%
7) Random Champions in Ranked Mode 643 4.83%
8) Provisional Matches in Ranked 712 5.34%
9) Duo Queue Prevalence in Ranked 414 3.11%
10) Level Disparities 635 4.77%
11) Team Margin of Victory 1,628 12.22%
Voters: 13325. You may not vote on this poll

After Hours with Matchmaking and Lyte

First Riot Post
Comment below rating threshold, click here to show it.

Spiritside

Senior Member

04-10-2012

Now that I think about it, duo queueing shouldn't be allowed at all, but if it has to be, it should be for severely restricted Elo ranges, like +/- 100 Elo (or whatever's the regular range that the matchmaker allows).

I had to lane with a 1400 support player the other day (who obviously duo'd with one of our other teammates) when I was around 1650 Elo; it was miserable. I didn't even think the skill gap could possibly be so wide, considering only 250 Elo, but it was bad. The guy literally had no idea how to play against 1600s players, so we just gradually lost lane until the slow losses compounded into a complete loss.

There are bad 1600s players who probably could have played better than that guy. When I'm 1650 Elo, I don't ever want to be forced to play with someone who's 250 Elo below me against two people who are closer to 1700.

Duo queues have a lot of potential to make the entire rest of the team have a lower quality game at their own expense, if the two players are wildly different in skill level.

EDIT: The current system is like allowing professional sports players to play on the same team as amateurs. That situation can't ever be good. The "amateurs" will perform poorly almost regardless of the situation and the "professionals" are limited in potential by their weakest links. Sometimes the pros will be matched up against the pros in these lane matches, but other times you get pros matched up against amateurs and that's when really terrible, toxic play/raging occurs.

I think the main reason that duo queues can create a lot of noise is their randomized placement on a team. You could get the situation I talked about earlier, 1650 + 1400 Elo vs. 1700 + 1700 in bot lane, which shouldn't even be a possibility. Despite what a lot of people think, there is quite a noticeable difference in skill between every hundred or couple hundred Elo.


Comment below rating threshold, click here to show it.

isobold

This user has referred a friend to League of Legends, click for more information

Senior Member

04-10-2012

Quote:
Originally Posted by Alyssa Riel View Post
Do you think you could post those please?
1200+Elo: 245,681
1300+Elo: 141,118
1400+Elo: 71,228
1500+Elo: 31,099
1600+Elo: 15,549
1700+Elo: 7,763
1800+Elo: 3,920
1900+Elo: 2,066
2000+Elo: 940
2100+Elo: 397
2200+Elo: 178
2300+Elo: 65
2400+Elo: 23
2500+Elo: 3

This is for NA, today at 10pm GMT+1, solo-Q. Numbers for 1300 Elo are included in numbers for 1200 Elo, and so on ...
I would expect the total of current ranked players to be around 550k ... maybe Lyte will be kind enough to add in the real total ...


Quote:
Originally Posted by Alyssa Riel View Post
Personally i think shifting where new players start towards where the new player average elo actually is will help resolve the "elo" hell that gets players frustrated. I think that it is an easy to implement solution, and that is why i advocate it so often.
Except it's not a solution. The average of Elo is where ever you make people start. If you make it start at 500 Elo, roughly 50% of the players will be above 500 Elo and roughly 50% of the players will be below 500 Elo. Nothing changes. Those numbers have no meaning at all. And then, once you made the system start at 500 Elo, the average will start to decay do to Elo getting lost on every Q-dodge and every Elo-decay, while only little Elo is generated during "loss forgiven times".

"To any problem there is always an easy solution, and this solution is always wrong."
Shifting start Elo is such an "easy solution" which in itself can't work, as it has no effect. If someone suggests it, he is basically telling everyone who actually understands Elo, that he didn't understand what he is talking about. Now look how often it is suggested and what this tells us about the average people complaining about Elo ...


Quote:
Originally Posted by Alyssa Riel View Post
I've also suggested bracketed ranked matches based on rigid brackets, and braketed ranked matches based on your current elo.
Brackets wouldn't change anything, except how it "feels". Except if by brackets you mean: once you climbed a bracket, you can't ever drop again. In this case you would outright break Elo. People have to be able to drop in order to make any ranking work.

This kind of "no-drop" bracketing is the exact reason why SC2 ladders aren't working. Their values have no meaning at all and that's why they have to reset them every 3-6 month. While SC2 rating might "feel" good, it's simply not working. And AFAIK no one uses them to do actual rankings of players ...


Quote:
Originally Posted by Alyssa Riel View Post
Finally, I've even suggested setting a lower "floor" elo that players have to build from.
That's the worst idea. A lower "floor" elo would basically lead to the following. Bad players drop to the floor. From there they can't drop any further, but they still make their team lose. So on each of their loses, Elo is generated from nowhere, making the average Elo and the median shift up. This will make a trickle-up-effect of Elo do transfer Elo from the very bottom of the ladder to the very top, introducing LOTS of noise, till every single player is on top of the floor. Then the floor is basically gone.

If you make new player start at this floor, this trickle-up-effect will start at the floor, increasing the time it takes for players to reach their elo by 5-10 times on average compared to now. Only the worst players will have their true Elo from start. But since Elo will always build towards a normal distribution, hardly any player will ever be placed appropriately by starting on the floor.

To add insult to injury Elo has to be constantly generated in order to build towards the normal distribution. This Elo-generation happens at the expense of bad players. The worse a player is, the more he has to lose to generate enough Elo for the top-players. This system would be worse for most players in LoL, except the top 25%. Instead of MM guaranteeing a 50/50 W/L to any player who has reached his True-Elo, adding a floor would guarantee bad players to have up to 0/100 W/L, while really good players would have even better W/L-ratios than they have now. This would only stop, once the floor has become insignificant.


Quote:
Originally Posted by Alyssa Riel View Post
Obviously I can't see into the future, but we, the player community who care, have to at least try and help. The only way we can really do that is to keep tossing ideas out there. Someone is going to come up with something good eventually.
Some of us have to keep playing the bad guys as well, think a bit more about the math and tell you when some idea isn't going to work without us having to try a system, that can't even work in theory.

@Lyte:
On the European forums I'm maintaining a german Elo-Guide/FAQ. I'm collecting ideas to improve Elo and MM since approximately a year. Since I've heard only 3 good ideas:
1. Make Noob-Island last longer, have Elo move less while in it and don't display Elo to players while in it. This would help with the players perception of their own skill.
2. A really cool paper about how to match players that match in playstyle: http://www.thefengs.com/wuchang/work...ayerrating.pdf (this has been tested in WoW and it worked pretty well there)
3. Make Elo-loss/gain decrease with the amount of time played. A 20-minute-stomp should be worth more Elo than a 50-minute close match. In a 50 minute close match you already got a good match as reward and obviously none of the teams dominated the other. Thus you shouldn't be winning as much Elo as in a 20 minute match, where the winning team clearly dominated.

If any of you reads german, my Guide on Elo is here: http://euw.leagueoflegends.com/board...d.php?t=380759

I would be especially interested in hearing what Lyte has to say about the second idea. I guess he read the paper already in one of the earlier threads, where we linked it. So he might have some insights on this ...


Quote:
Originally Posted by Alyssa Riel View Post
Elo is much more suited to 1v1 play so it is only natural that this problem would exist in lol. I suspect that some modified elo system could be created that accurately represents player skill though.
This is stated on a daily basis, but I don't buy it. While I agree that being dependent on a team is prone to add noise, the actual reality proves the claim wrong. We have seen many Elo-Hell-experiments where players dropped on purpose by queue-dodging. In every single case those players were able to regain their Elo easily. I've collected some dozens of those experiments, since the start of season 1. In one case even a 1250 Elo player dropped to 0 Elo and had no issues to regain 12xx Elo again with a constant 75% Win-Rate. Excelling players even achieve to get 80-90% Winratios when dropping way below their Elo. You wouldn't expect a 100% Win-Rate of the better player in 1vs1-scenarios either. So yes: team dependency adds noise. But this noise is far less important than people make it out to be.

Generally this issue is overstated by people who are searching for an excuse for their own failures. Instead of searching where to improve themselves, they prefer to blame the team for them losing more often than they would like to. That's why we here this fake-argument that often. And sadly even players who are obviously not searching for such an excuse seem to buy it sometimes, because it looks logical at a first glance ...

Also I've yet to see a modified Elo system, that more "accurately represents player skill" than the current system does. I've called for ideas on hundreds of elo-hell-threads. I haven't heard a single good idea, except some detail improvements which I mentioned above. I would happily jump on any good proposal. But there hasn't been even one yet.


Comment below rating threshold, click here to show it.

Goodguy Hopper

This user has referred a friend to League of Legends, click for more information

Senior Member

04-10-2012

Quote:
Originally Posted by isobold View Post
Except it's not a solution. The average of Elo is where ever you make people start. If you make it start at 500 Elo, roughly 50% of the players will be above 500 Elo and roughly 50% of the players will be below 500 Elo. Nothing changes. Those numbers have no meaning at all. And then, once you made the system start at 500 Elo, the average will start to decay do to Elo getting lost on every Q-dodge and every Elo-decay, while only little Elo is generated during "loss forgiven times".
My previous post based on the numbers they gave us at the end of season one is that the average elo is not where players are starting. 1080 is over a full hundred points below the starting point, and that is a problem. i know it doesn't feel like a 100 point elo change is that huge, but jsut playing through it a couple times you can really tell the difference.

If i am mistaken and the average is not 1080 then i'd be willing to concede the point, but by my understanding that is the case.

Also sorry i asked you to get that data i could have gotten that alone, i had held onto the hope that you had more accurate numbers below the 1200's.

Lyte do you think we could get some player percentages and averages for the current elo system so we can make some more informed decisions?


Comment below rating threshold, click here to show it.

Lyte

This user has referred a friend to League of Legends, click for more information

Lead Social Systems Designer

Follow RiotLyte on Twitter

04-10-2012
76 of 362 Riot Posts

Quote:
Originally Posted by Alyssa Riel View Post
My previous post based on the numbers they gave us at the end of season one is that the average elo is not where players are starting. 1080 is over a full hundred points below the starting point, and that is a problem. i know it doesn't feel like a 100 point elo change is that huge, but jsut playing through it a couple times you can really tell the difference.

If i am mistaken and the average is not 1080 then i'd be willing to concede the point, but by my understanding that is the case.

Also sorry i asked you to get that data i could have gotten that alone, i had held onto the hope that you had more accurate numbers below the 1200's.

Lyte do you think we could get some player percentages and averages for the current elo system so we can make some more informed decisions?
Unfortunately, we cannot release exact numbers for many of the things you guys are discussing. It's not because we want to hide data from our fans, it's a business decision we have to respect. Once in awhile, like at the end of a season, some limited stats will be released.


Comment below rating threshold, click here to show it.

Doflamingo51

This user has referred a friend to League of Legends, click for more information

Member

04-10-2012

Still no fix on Normal draft prioritizing pre-mades


Comment below rating threshold, click here to show it.

isobold

This user has referred a friend to League of Legends, click for more information

Senior Member

04-10-2012

Quote:
Originally Posted by Alyssa Riel View Post
My previous post based on the numbers they gave us at the end of season one is that the average elo is not where players are starting. 1080 is over a full hundred points below the starting point, and that is a problem. i know it doesn't feel like a 100 point elo change is that huge, but jsut playing through it a couple times you can really tell the difference.
But you need to understand where this difference is coming from. If you make a fresh start, the average is 1200 Elo. Or what-ever you choose as starting Elo. It is only over the course of this season that this average is shifting down slowly. This would happen no matter which starting point you choose and isn't related at all to 1200 Elo. It would happen if you choose 500 Elo as starting point as well.
The average shifts down due to A) Q-dodge-penalties and and B) Elo-decay. And I'm repeating this for the 3rd time now. This should be common knowledge in this thread, if you want an informed discussion on the matter.
So changing the starting point can't ever fix the problem, as it's not the source of it or even related to it.


Comment below rating threshold, click here to show it.

calicfer

Senior Member

04-10-2012

Quote:
Originally Posted by Lyte View Post
I like the idea of a "ready up" system, but we would have to figure out ways to avoid it taking 20 minutes from hitting "Play" to actually getting into a game.
A ready up room that gives 1 minute before bans would be awesome.
Cool features would include static information like 5 favorite champions, ranked stats, favorite role, 3 most desired bans.
Other cool features would include selecting your champion select order. Buying champions for the sole purpose of trading is kinda bogus. (I admit it can add to the fun of draft pick.) Being able to select order would reduce the necessity of buying extra champions.

Mostly it would enable a team to force the other team to ban one of the "op champs" by preparing to first pick it. (there isn't enough time to communicate that before the first ban.)

I purposely ban at the last second to give my team time to discuss. That 1 minute of pre discussion would speed up bans by at least 30 seconds. It would speed up bans more with the static information.


Comment below rating threshold, click here to show it.

calicfer

Senior Member

04-10-2012

Quote:
Originally Posted by isobold View Post
But you need to understand where this difference is coming from. If you make a fresh start, the average is 1200 Elo. Or what-ever you choose as starting Elo. It is only over the course of this season that this average is shifting down slowly. This would happen no matter which starting point you choose and isn't related at all to 1200 Elo. It would happen if you choose 500 Elo as starting point as well.
The average shifts down due to A) Q-dodge-penalties and and B) Elo-decay. And I'm repeating this for the 3rd time now. This should be common knowledge in this thread, if you want an informed discussion on the matter.
So changing the starting point can't ever fix the problem, as it's not the source of it or even related to it.
que dodges should award elo to the opposite team. Especially in situations where the team was able to put together a better team composition, ban someones favorite champion, or correctly counter pick.


Comment below rating threshold, click here to show it.

Soup is lame

This user has referred a friend to League of Legends, click for more information

Senior Member

04-10-2012

I just remembered something that REALLY bothers me that many people complain about that SHOULD be on this list:

If I report someone in a match, and add them to my ignore list... WHY ARE THEY ON MY TEAM OR IN MY NEXT GAME WHEN I REQUEUE?!

Being trolled by someone for 30 mins is rough as it is, but then to shrug it off, requeue, and then see in chat "ahahahaha prepare to suffer again noob" but not be able to dodge because of ELO loss is quite possibly the worst catch 20 situation I have ever experienced in a video game.

I'm guaranteed to lose ELO if I dodge, or try to carry regardless (which if you've ever had an intentional troll, is impossible to do) on the sliver of hope that I might not lose ELO.

There has to be SOME way to avoid being matched with these cancerous-to-the-community players in consecutive matches.


Comment below rating threshold, click here to show it.

calicfer

Senior Member

04-10-2012

Quote:
Originally Posted by HayApples View Post
I think part of the draft/normal dichotomy discussed in point #5 is that both modes have critical weaknesses that need to be shored up.

Normal mode suffers from a lack of counterplay and a number of overtly luck-based factors. Team A gets lucky and gets favorable matchups versus Team B. Certain champs need certain summoners to counter them effectively, and so on. It also casts too large an umbrella with inclusiveness, leading to "worst case" scenarios where players are included who are hopelessly outmatched.

Draft mode suffers from problems as well. I have documented in a previous post that it takes stupidly too long to set up. I put a stopwatch and counted 8 minutes between "PLAY" and actual action/combat in an average game. If the game surrenders at 20, you spent literally 1/3 of the game time just setting the dumb game up. Secondly, if you regularly play in an arranged 3 or 4 group, ranked does not accommodate you, period.


Addressing any of these ancillary problems will improve the experience. I would start with blind pick as being the "default" game mode. The number of bad scenarios that blind pick enables is just staggering. You can make all the ranking adjustments you want, if the underlying structure flawed, it is all window dressing. Mirror matchups suck. Losing as rock because your opponent picked paper sucks. Having a 1900 ELO rated Malzahar sucks if he's playing first time Tryndamere.

To summarize, new and different modes are needed to better structure the players. It is as integral to the matchmaking process as the actual ranking system. If you go back to your roots, DOTA All-Pick has a completely open/freeform champ picking system and it produces better results than the supposedly evolved blind pick used here.
Being able to select "Try hard mode", "regular mode" and "learning mode" in normals with different hidden elo's for each would be awesome.

Also (in blind pick) revealing the other players champions before the loading screen and letting people select summoners runes and masteries accordingly would be awesome. (although it could lead to que dodging). (maybe flag them as in game before revealing the opponents champions??)

Could there be a matchmaking system with random champion select? Cause doing that on customs leads to horrible matches usually... maybe matchmaking wouldn't do much better though.