Dear Lyte,

The theme of this post is "Can we not punish the whole team for the failure of one team member?" Or put another way, "Can we make it so my ELO is adjusted based on my individual performance rather than whether my team wins or loses?"

I have let the ideas people have tossed back and forth "simmer" for a few weeks as I've contemplated them and started doing ranked solo queue again. Here's some updated experience and thoughts:

My last three ranked games have been losses. Here's what happened in each game:
Game 1: I was Lee Sin top, won my lane quite handily and ended the game 8/3/3 with a loss. The loss was mainly due to enemy AD carry being fed and synergizing well with enemy Mundo.
Game 2: I was Alistar, which was not an ideal support for me; I prefer to play sona or soraka as support but our team comp ended up being super squishy so I felt forced to play Alistar. Our top lane was Akali and she picked first and was easily counterpicked (even though I asked to pick top since I was last and could counterpick). She ended up something like 1/5 and our jungler, jax, ended up 0/6 or something. I sat there as a support helplessly watching the game fall apart. I ended up 1/3/1.
Game 3: I was Kog'maw and I picked second. Somehow, after I picked kog'maw, the third pick on our team took urgot so we had dual AD carries bot lane with no support. Our urgot turned out to be horribly bad which made me think he is probably freefalling down the ELO ladder at this time. I finished this game 2/1/3. IMO, I was still a positive contributor to the team despite a disastrous loss.

My ELO has been in the 900's most of this time. I was up to like 975 ELO or something but now I am back down to 941. I go through stages where I randomly get decent teammates and randomly get people like the Urgot in game 3.

My record is now 62-68.

Here is my frustration:
I have a sense that my ELO is "too low" mainly because I find most of the games I play to be on the easier side (for example during laning phase). My ELO was at one time 1353 and those games did not feel overwhelmingly hard to me. So in other words, I feel like I can handle 1200-1300 ELO games just fine. However, when another lane is feeding it partially (or fully) negates my performance and makes me feel like my efforts are in vain and I am wasting my time.

What would be really cool.... is to not make the entire team suffer for the failure of one team member.
In other words, in game 1 where I went 8/3/3, it would make a whole lot of sense for my ELO to go _up_ even though the team lost. I'm not saying it needs to go up a lot.. I am just saying... judge my ELO on my individual performance, not whether my team wins or loses.

The benefits:
- people would feel more empowered and would feel like practicing would actually pay off
- people would be more willing to play support roles and not feel like they have to carry or their team loses
- if a lane is feeding, people wouldn't care as much "Ok we're going to lose but I'm doing well so my ELO still goes up". less rage.

rivanor,

I have thought a lot about how individual metrics can affect Elo but there are too many abusable cases. If you check back in this thread, we have a lot of discussion about why individual metrics are not a good way to adjust Elos.

For example, in you final bullet point you claim that if a lane is feeding, players would not care as much because they know they can still win their lane and gain Elo. What's the point of ever winning a game then? If a player simply ensures 100% of the time he wins his lane and does not care at all about his teammates and never helps out his other lanes, he theoretically will be the #1 League of Legends player in the game. Is that an OK system? I don't believe so.

I had a post about an elo formula a long time ago now(I want to say it was late Season 1), that seemed to get some support.

It basically had a sliding scale of factors where you always gained or lost Elo depending on the outcome of the game,but several factors lead it to make some people lose a little more, or less, or gain a little less than others.

You still play as you normally would(in theory of course) but depending on the variables you could lose and be less annoyed that you lost the same amount as someone who disconnected, or said in all "Come to dragon and get free kills" because you wouldn't. Same for winning. That person who disconnects still gets the elo gain if they don't get picked up by the system and put in for a leave.

I think you make it sound so black and white Lyte.

You could have a 50% team ELO and 50% personal ELO. So if you lose, but play perfectly, you don't go down ELO (-6/+6), but comparatively if you win but play horribly you don't gain any (+6/-6). This mean the average amount you climb or drop would be less (as you're unlikely to play perfectly or completely uselessly every game) so it'll be (+/-6)+(+/-6) every game.
The maximum potential gain or loss is the same (so you could adjust the total if you wanted the average to be similar).

It wouldn't just have to be about laning. You obviously already collect a bunch of interesting metrics for your Business Intelligence and post game stats, if you collected a few more you could probably extrapolate out how well someone played. Off the top of my head some interesting markers would be:

-Getting First blood
-K+A/D ratio (combining kills and assists then makes it fair for supports)
-CS
-Placing X wards in key map areas (designed so you can't spam 5 wards in -1 useful place)
-Last hitting baron/dragon (for smite steals)
-Number of blue or red buff/sgame (for good counter jungling)
-Highest damage done to champions (with tweaking around aoe)
-Some sort of damage taken from champions/deaths metric for displaying your survival rate in team fights
-Double/Triple/Quadra/Penta Kills
-Killing champions higher level than you
-First to hit level 18 in game
-Enemy jungle CS (once again for counter jungling)
-Turret hits taken/deaths for lower levels for pro tower dives.

It's hard to tell without know what you track, but you might already have a lot of this!

Apologies, the intent was not to make it black and white. We've just discussed individual metrics numerous times in the other threads and the black and white example was the easiest to point out to the new visitors to the thread.

A lot of the metrics you point out are team-oriented metrics and are things we are doing analysis on to determine if they can be usable for adjusting Elo. For example, there might be ways we can incorporate number of towers destroyed as a team to determine if your entire team deserves less or more Elo.

However, we have done some research on the individual metrics you listed and very few of them are great predictors of success or skill in League. Most of the individual metrics tend to promote gameplay that is not productive to winning the game. For example, getting to Level 18 first in the game does not necessarily predict a top level player--it does encourage some players to sit in their lane and attempt to farm to Level 18 as fast as possible.

Then you would still promote self-fish play instead of team-play, which is a no-go in a team-game. Lyte already said it in a nutshell. You don't want abusable cases in your ranking-system. ANY personal score will always be abusable, so personal score can't be part of the measurement. And they aren't helping anyways. If you are a positive contribution to the game, you will win more than you lose and hence rise in Elo.

IF you would ever find a measurement for personal score that can't be cheated, then it would still result in all players being in the same Elo-range they are in now, just maybe a bit faster. So there is little benefit to personal scores (just rating speed). Since their advantage is THAT small, you can only adopt them, if they are 100% bullet proof. The slightest disadvantage compared to Elo disqualifies them. That's why you have to take a look at the black&white-view of the matter.

Btw. Dominion proves perfectly that personal score is a bad idea. If I play to become MVP, my team loses >70% of all games, yet I'll be MVP in >75% of all those games.
Edit: I learned that the hard way btw.

It's not implemented on another server. This is a bug that we fixed in the latest patch actually.

I don't particularly agree. The goal behind my post above was that each game you play you have to prove you are excelling in each aspect of the game. Taking your example, if you stay top all game farming till 18 (first), then you may get +1 ELO, however you'll get negative personal ELO for not warding the map or contributing to your damage/deaths ratio in team fights or tower/dragon kills. I'm not saying, match one of these markers, gain 6 ELO, I'm saying do ALL of them and you get +6, you have to play _perfectly_ to get the +6.

In addition, your personal ELO score could only cancel out a loss, not take you higher, and if you get carried in a game but spend the game trying to achieve particular +ELO goals, you will not be contributing to your team, so will get the +6 for a win, but less than the +6 for playing a perfect game (ie you can't last hit dragon, kill tower, team fight if you're farming top, so you won't get +6).
The metrics would have to be finely tuned to include personal performance, which includes contribution to the team. Once again in your dominion example, you would get +6 for a perfect game and -6 for a loss, so you don't climb. If you get a win and a +3 for good game you get a respectable amount. If you win, but you don't do well, get carried and don't meet the personal ELO goals you would get -4 and +6, so only go up +2.

The reason I think this is because I don't really believe that if you contribute well to every game you will rise in ELO, as I think others failing to contribute is more of a factor in taking you down than that of your contribution taking you up, at least at the low ELO I play. So many people claim they don't feel they are in the correct ELO. Not all of them will be right, however having a personal contribution and a team contribution would put more responsibility on the player to play well, rewarding them, or not punishing them, for playing well, as well as the team result, which I feel would help actually place the player in their ELO.

Perhaps this should only apply to solo/duo queue.

If you match ALL of these markers, you will get exactly Elo. Since Elo already takes all these markers into account. And it does so perfectly, since Victory can't be cheated.


Not getting dragged lower is the very same thing from a mathematical view point than getting higher, if in fact you deserved to get lowered.


Actually Elo should ONLY measure your contribution to the team, which Elo already does. The more you try to make a personal score to be an actual team-contribution score, and the more you add in factors that promote "good" play, the more your proposal will be similar to the current system.

The issue isn't how to tweak it. The issue is your approach. You can't rate individual performance to deduce team-contribution. Sometimes it's better to go all-in on a bad plan than to splitt-fight over a good plan. You can't have the cake and eat it. Either you rate individual performance: then it can and will be cheated. Or you go with what's already there to rate the performance objectively by only taking into account what can't be cheated: victory.

If you are losing each and everygame with random teammates of your Elo, your personal score doesn't matter. You obviously aren't a good teamplayer, even if you add team-play-measurements and those display a good score.

You are trying to fix a problem that's not there. Elo isn't about a single game. The deviation of a single game isn't important to your total. So no fixing to it is needed. You are trying to introduce weaknesses to this system, without adding a strength that would be worth it.


You are missing the point. In Dominion we have the MVP system. The MVP system could show my personal score to be great, all the while my Elo drops further and further. What's the point of having a personal score then, when it's missleading?


But that's not about opinion. You are just plain wrong for mathematical reasons: http://lol.noamik.de/No-Elo-Hell.png
Elo IS working, it just needs time (50-100 games) to rate you appropriately.

If you are contributing well to every game, you WILL rise in Elo. Has been proven countless times. When I get dropped to 400 Elo, I win 80%+ of all the games all by myself. To me players there are worse than bots. Bots at least don't fail their combos. Bots at least don't go in 1vs3 for no reason. Bots don't waste gold on wards without ever taking a single look on the minimap. I could link you dozens of Elo-Hell experiments that display the same results. Lyte even introduced a fix now which will prevent Elo loss on queue dodge, since there are to many players who for fun dodge down to Low-Elo and go rampage among players there, tanking their Elo by dodging every second queue.


None of them is, if they played enough games to be correctly rated. In the german forums we just had a thread where low-elos in Elo-Hell gathered and played games against random gold players. Draft-mode, without voice communication. So far 8:0 for the Gold Elos ...

And to be clear: none of these matches was even remotely close. And none of the low-elo players was able to shine, even if only compared to his team-mates. On average team gold went 30k+ in Gold, double the cs on every position and double to triple the gold.