Get rid of ELO. Get a TrueSkill system.

SexInTheCivic·Jul 29, 2015·5 votes·1,826 views

The system of ELO ranking is so broke, especially with 2/3 and 3/5 series matches. You can't control the team you get, and we all have seen games where one player single-handedly will carry his entire team to a win. Why do his teammates deserve as much LP as him? We all have games with trolls where when their pick rolls around and ADC is left open they choose to try and duo mid as if it was their right. If you're not able to carry one or two bad players, as much of a team game as this is, you shouldn't be slammed with a hard LP loss because you didn't want to surrender in a ranked game.

This takes me back to when I used to play Halo 3 all day long, 4 seasons straight. They had a system, which is patented by Microsoft called, the TrueSkill System. In a sense, every game you play, a program takes in how well you do, if you win or lose, the ranks of everyone on your team and compares them with the ranks of the other team. So if you have a team that's really good, let's say a team where everyone is in Gold, versus a team where everyone is in bronze, the bronze team off the bat will not lose as much LP as they would versus another bronze team, and if someone on that bronze team goes 15-0, they might actually gain LP, seeing as how they went against a much harder foe and managed to come out with a good score.

The only reason I can understand the logic behind the current ELO system is that old tiresome argument of, "it's a team game, and if you can't help your team win, you deserve to lose." If someone consistently is playing well, they deserve to rank up.

20 Comments

Sightless66Jul 29, 20153 votes

So if you have a team that's really good, let's say a team where everyone is in Gold, versus a team where everyone is in bronze, the bronze team off the bat will not lose as much LP as they would versus another bronze team

This is already true. You get more LP playing against higher skilled teams, and less playing against less skilled teams.

if someone on that bronze team goes 15-0, they might actually gain LP, seeing as how they went against a much harder foe and managed to come out with a good score.

Here's the problem with this. Going 15-0 is not the only way you can be playing well in this game. Enabling someone else to play this well. Say I play a tank jungler, and I camp midlane hard in order to get my midlaner fed, while giving up my own farm and levels and leaving myself open to counterjungling and invading by the enemy jungler to do so. My midlaner may get very fed and have a very good score. His good score is based on my actions, but it would be very hard to measure.

A good Leona may be one that initiates fights and gets a pick on the fed enemy carry, but dies in the process of killing him and doesn't get assists on the rest of the enemy team. Her score would be terrible, but her value to her team would be immense. She could easily be the best player on her team by a wide margin. How would a system reward her for these kinds of plays?

That's the problem with these systems. I've never seen a proposal that could accurately measure and reward actions that support the team's chance of winning. They always support one class or role over others. The fair thing about the MMR system is that it doesn't support any given class over other classes. You're rewarded based on whether or not you win and based on the average skill of your opponents. If you're good, you'll win more often and against more skilled opponents. If you aren't good, then you won't. That's fair.

Trueskill worked in Halo because virtually all performance there is individual and easily measured. My KDA is largely independent of my team's actions. I'm minimally reliant on my teammates to make my own performance better. It's very easy to measure the worth of a player in those games because the numerical stats can pretty accurately sum up the worth of a player. "Player X has the best KDA, so player X is the best". That statement usually works in Halo. It doesn't work well in League. You have so many different categories that you have to judge people on that the only way to make this kind of system is by making arbitrary judgments about which categories are more or less important.

If someone consistently is playing well, they deserve to rank up.

If someone is consistently playing well, then they will win more often than they lose. Any individual game may be won or lost due to teammates, but if you play enough, you will win more if you play better.

SexInTheCivicJul 29, 20151 votes

{quoted}

This is already true. You get more LP playing against higher skilled teams, and less playing against less skilled teams.

Here's the problem with this. Going 15-0 is not the only way you can be playing well in this game. Enabling someone else to play this well. Say I play a tank jungler, and I camp midlane hard in order to get my midlaner fed, while giving up my own farm and levels and leaving myself open to counterjungling and invading by the enemy jungler to do so. My midlaner may get very fed and have a very good score. His good score is based on my actions, but it would be very hard to measure.

A good Leona may be one that initiates fights and gets a pick on the fed enemy carry, but dies in the process of killing him and doesn't get assists on the rest of the enemy team. Her score would be terrible, but her value to her team would be immense. She could easily be the best player on her team by a wide margin. How would a system reward her for these kinds of plays?

That's the problem with these systems. I've never seen a proposal that could accurately measure and reward actions that support the team's chance of winning. They always support one class or role over others. The fair thing about the MMR system is that it doesn't support any given class over other classes. You're rewarded based on whether or not you win and based on the average skill of your opponents. If you're good, you'll win more often and against more skilled opponents. If you aren't good, then you won't. That's fair.

Trueskill worked in Halo because virtually all performance there is individual and easily measured. My KDA is largely independent of my team's actions. I'm minimally reliant on my teammates to make my own performance better. It's very easy to measure the worth of a player in those games because the numerical stats can pretty accurately sum up the worth of a player. "Player X has the best KDA, so player X is the best". That statement usually works in Halo. It doesn't work well in League. You have so many different categories that you have to judge people on that the only way to make this kind of system is by making arbitrary judgments about which categories are more or less important.

If someone is consistently playing well, then they will win more often than they lose. Any individual game may be won or lost due to teammates, but if you play enough, you will win more if you play better.

So it sounds like you're saying the assist system needs to be fixed. Why wouldn't she get the assist if she's helping out so much jumping on them? I'm saying, if your team beats a harder team, you deserve more LP regardless. You make the comparison to Halo being able to look at a KDA and say, oh he has the best, so he is the best, but you're claiming in League you look at the stats and say, well that guy might be 3-10, but I'm not entirely sure if he's better than that 10-3 guy. Your logic is flawed, but I see where you are coming from. If you are an asset to your team, your stats will show it. If you're a jung who camps mid, you are bound to get an assist or kill. If you camp mid to help your team win, and you win, you will be rewarded with LP regardless.

All I'm saying is I don't know why I should have to dodge, because the 5th person on my team thinks a duo mid is a good idea since they are a bad support. Something stupid like that. Why do I deserve to lose LP because I dodged, and on top of that, have to wait 10 minutes to play again. If I stay in the game and we lose, but I am able to hold my own, why should anyone on my team lose as much LP as the guy who trolled because he didn't want to play the last open spot that 4 other people asked him to play.

[{quoted}](name=Koala Frenzy,realm=NA,application-id=3ErqAdtq,discussion-id=B8KgahXF,comment-id=0000,timestamp=2015-07-29T16:03:06.096+0000)

Why should I focus on winning if all I have to do is get a postive KDA and then afk at the fountain?

That's the beauty of the report system, and the TrueSkill system. A system like that could easily monitor when you get your skills, like the game already does, and then if it sees you not getting cs, not getting kills/deaths/assists, and frankly, not even moving, your rank for that specific match goes down. Imagine a graph. You get a kill, you have a line going up, you kill 10 minions, your line goes up again, you get an assist, boom! Imagine what happens? Your line goes up. Then you die, now your line goes down. You go out, and die again, the line goes down. Now you're mad and rage quit, your line continually goes down, seeing as how you are not dealing damage to anything or moving. It's quite simple.

[{quoted}](name=404Kat Not Found,realm=NA,application-id=3ErqAdtq,discussion-id=B8KgahXF,comment-id=0001,timestamp=2015-07-29T16:09:02.703+0000)

You dont seem to consistently be doing good, as you have a 40% ranked win rate.

I don't really understand the relevance in how good/bad I am doing, but ok.

{quoted}

The problem is Solo Q, and you being punished for other teamates doing badly, while having no way to dodge matches without penalty, no recourse for AFK's and Feeders, and getting teamates based on a flawed ranking system as opposed to matching your skill level.

Team based rating systems are fine for actual teams, but with random teamates personal skill HAS to be a factor in ranking.

This is what I'm talking about. Solo q's, it can go either way. Why do you think they call bronze Elo hell? I have played with so many good people and when asked, "why the hell are you in bronze when you clearly run teams over by yourself?" I get the same response. "Elo hell gravity is to strong." I don't have a team of 4 other people consistently on, and who I trust enough, to play as a team with. AFK's can't be solved.

But saying the Elo system is not a problem is completely bogus. I could go into every game and act bad, and make other people lose because of it. A Trueskill system would work so effectively. LoL already records how much damage you do to buildings, cs, KDR, assists, and how much damage you do to other champs. If you're playing Sona where your damage might not be extremely high on champs because you're a support, or your CS might be weak because you don't farm, it's extremely possible to weight assists to fix that problem. Seeing as how it's such a team game, assists would be, and should be, extremely influential in how much LP you earn or lose. If you can go 15-0-30 in a game, that seems to me like you are a pretty good team player, and individual player... am I right? You had 15 kills and no deaths, extremely good game. On top of that you managed 30 assists. Also extremely beneficial to your team. Now if you go 15-0-30 and you have someone on your team who goes 1-17-10, and because of that you lose, do you really deserve to be hit with a 25 LP loss?

KazaashiJul 29, 20152 votes

{quoted}

This is what I'm talking about. Solo q's, it can go either way. Why do you think they call bronze Elo hell? I have played with so many good people and when asked, "why the hell are you in bronze when you clearly run teams over by yourself?" I get the same response. "Elo hell gravity is to strong." I don't have a team of 4 other people consistently on, and who I trust enough, to play as a team with. AFK's can't be solved.

People call it Elo Hell because they are stuck in a cycle of blaming thier teammates or Riot, instead of focusing on themselves and getting better at the game. Elo Hell only exists in your mind.

But saying the Elo system is not a problem is completely bogus. I could go into every game and act bad, and make other people lose because of it. A Trueskill system would work so effectively. LoL already records how much damage you do to buildings, cs, KDR, assists, and how much damage you do to other champs. If you're playing Sona where your damage might not be extremely high on champs because you're a support, or your CS might be weak because you don't farm, it's extremely possible to weight assists to fix that problem. Seeing as how it's such a team game, assists would be, and should be, extremely influential in how much LP you earn or lose. If you can go 15-0-30 in a game, that seems to me like you are a pretty good team player, and individual player... am I right? You had 15 kills and no deaths, extremely good game. On top of that you managed 30 assists. Also extremely beneficial to your team. Now if you go 15-0-30 and you have someone on your team who goes 1-17-10, and because of that you lose, do you really deserve to be hit with a 25 LP loss?

Yes, you didn't win the game; therefore, you obviously made some mistakes that caused you to lose (barring an AFK or DC). If you work on those mistakes, you will win more of those games in the future, and you will climb.

I'mma past what I said earlier:

Amount of CS doesn't matter, towers destroyed doesn't matter, and kill participation doesn't matter, as they are just a means to an end.

Sure, doing those things consistently and skillfully is important, but there is no set amount of any of those things that makes you win the game. Only blowing up the other teams Nexus will do that, and it is up to you and your team to figure out how to cobble those other factors together for a win. That is the skill that is tested in Ranked.

Sightless66Jul 29, 20152 votes

I'm saying, if your team beats a harder team, you deserve more LP regardless.

You already do. This is already in the game. Beat a team with a higher average MMR than yours, and you get more LP.

So it sounds like you're saying the assist system needs to be fixed. Why wouldn't she get the assist if she's helping out so much jumping on them?

She gets one assist, because she hit one person. That's all she should get, because that's the only champion that she affected. The problem is, that one assist is worth more than 5 assists from a Taric who just pops his W on the enemy team and does nothing, but there is no way to measure that contribution with any stats avalivable in the game. Tell me, how would you measure whether or not a Leona jumping on a fed target was a skillful play deserving of reward just with the avalivable stats?

So, yeah, the assist system would need to be fixed for your new system, but you can't fix it. You can't make a new system that works well enough to do the job adequately. That's the entire problem. Your proposal is predicated on the existence of a fair value-assessment system which just does not exist. You're going to have to make judgments that will benefit one champion over another. It's unavoidable.

But wait, it's not just assists. What about a split-pusher? In some games, the split-pusher puts a ton of pressure on, makes a bunch of space for their team, and single-handedly win the game. In other games, the split-pusher can be a dead-weight accomplishing nothing but allowing the enemies to steamroll over their team while they won't help. They can have exactly the same stats in both the good and the bad games, but the value of their choices changes based on the situations in the game. How would you measure that?

You can not just ignore this problem. It's the reason your system won't work. This has to be addressed and completely solved before any discussion of a trueskill-style system can possibly begin.

You make the comparison to Halo being able to look at a KDA and say, oh he has the best, so he is the best, but you're claiming in League you look at the stats and say, well that guy might be 3-10, but I'm not entirely sure if he's better than that 10-3 guy. Your logic is flawed, but I see where you are coming from. If you are an asset to your team, your stats will show it.

No, my logic isn't flawed, because the stats don't always show your contribution. The stats in this game show who's getting the kills or doing damage, but that only shows the value the source of damage. I want you to take a look at champion.gg sometime and look at the average stats for the jungler role. You'll find that tank junglers trend towards being lower than damaging junglers in most stats, even if their winrate is higher. You'd need a way to fairly compare those stats.

Stats don't show the full story of contribution. There are going to be games where characters with the lowest stats contribute the most. Certain roles are almost guaranteed better stats than other roles. That doesn't mean those players are more skilled. It just means that the numbers are inadequate.

I mean, just think about it for a second. If I'm playing a tank, my job can literally be "get between my carries and their damage and make sure I die before they hit my carries". Dying in a teamfight can be a sign that a tank protected their carry amazingly, but your system would punish them for that. Is that acceptable to you?

If you camp mid to help your team win, and you win, you will be rewarded with LP regardless.

Yeah, but the problem is that this guy would be more deserving of the bonus LP than the guy who just gets fed off of the backs of his allies, but the less skilled player would be getting rewarded. That's a fuckton of bias that you aren't going to be able to control for. There is no reason to introduce that problem. That's why the current system is the best. If you win, you get MRR. If you're better, you'll win more. There is no bias in that system.

All I'm saying is I don't know why I should have to dodge, because the 5th person on my team thinks a duo mid is a good idea since they are a bad support.

If this really is "all you're saying", there's a better solution. Implement pre-game reporting and allow punishment for these players. That solves the problem without introducing a ton of new ones. Quicker, simpler, more elegant. Better in every way.

The MMR system works fine. If you're good, you'll win more games than you lose and raise. If you're bad, you'll lose more and fall. There will be some games you're guaranteed to win and some you're guaranteed to lose. That's fine. You will rise if you're good. If you can't, then you just aren't good enough. All of those people you see in Bronze saying that they can't raise because of elo hell? They just aren't good enough. If they're really better than their rank would suggest, they'd be winning more games than they are losing. There really isn't anything more to it than that.

Final thing. Don't reply to every person on the thread in one comment. Reply to them directly, or they're unlikely to ever see it. Bad forum etiquette.

Koala FrenzyJul 29, 20151 votes

Why should I focus on winning if all I have to do is get a postive KDA and then afk at the fountain?

RISE Shlack123Jul 29, 20151 votes

If you are 4/0/5 but there are 40 kills in the matchup, your KDA shows a poor kill participation and thus you don't get any points. There are many factors they can use to determine your performance.

Koala FrenzyJul 29, 20151 votes

{quoted}

If you are 4/0/5 but there are 40 kills in the matchup, your KDA shows a poor kill participation and thus you don't get any points. There are many factors they can use to determine your performance.

Why do I need kill participation? The game could be a blood bath in top lane getting all those kills while bot lane is relatively peaceful. I should give up my lane and gravitate towards battles just to get LP?

RISE Shlack123Jul 29, 20152 votes

Because the TrueSkill system the OP is talking about takes all these things into account. Damage to champions, CS, tower damage, kill participation, healing, kills, deaths, assists - all taken into account.

If you aren't participating in the fights, the system could be set up to see this and try to account for it elsewhere. Are you getting a lot of cs? Are you getting towers or objectives while your team distracts? All of these factors would have separate weights and give you a skill score accordingly. So if you get a good KDA and then choose to sit in the fountain for 20 minutes, you won't have any of the other metrics the system looks for and thus you won't get a good skill score.

It's simple. You're just being thick.

Koala FrenzyJul 29, 20151 votes

{quoted}

Because the TrueSkill system the OP is talking about takes all these things into account. Damage to champions, CS, tower damage, kill participation, healing, kills, deaths, assists - all taken into account.

If you aren't participating in the fights, the system could be set up to see this and try to account for it elsewhere. Are you getting a lot of cs? Are you getting towers or objectives while your team distracts? All of these factors would have separate weights and give you a skill score accordingly. So if you get a good KDA and then choose to sit in the fountain for 20 minutes, you won't have any of the other metrics the system looks for and thus you won't get a good skill score.

It's simple. You're just being thick.

I'm not being thick, I'm finding obvious abuse points for a system that can never work in a game that has to account for strategic gameplay and support players. I've been in a game with an eve that never laned, jungled, or csed and just roamed the map ganking every lane over and over and we won off that. How does the system perceive that play in gold as different from typical bronze 5 poor CSing? How about the support janna whose only job is to ward and keep her ADC from dying? Does ward drop count count? What stops someone from going to the fountain with a sight stone and spamming them?

It is impossible to gauge individual play in a team game purely on statistics.

RISE Shlack123Aug 1, 20151 votes

Metrics are compared to league averages, other players, and length of the game. These are easily measured metrics, and many other games use similar programs, such as TrueSkill. Can it be abused? Of course it can. There are many aspects of this game that get abused - one example is smurfing.

Will there ever be a perfect way to measure a player's performance? No. But do you honestly think a simple win/loss is a good measurement of an individual's skill? I certainly don't.

Every single action that takes place in this game is measured on the backend. Hence, all of these numbers can be calculated and measured against the rest of the team, the enemy team, the league averages, and the length of the game to give a more accurate measurement of performance. That Janna's peels are measured by shielding done to allies, skillshots landed on enemies, and where wards are placed, and how often they are placed. That Evelynn's kill participation and damage to champions would be a heavy factor - if a jungler has 80+% kill participation, that will offset them having lower CS.

If one metric is much higher than another, it will outweigh it. High kill participation will offset low CS and still end up positively. It really begs the question as to whether or not you understand what metrics are.

KãtãrinãJul 29, 20151 votes

You dont seem to consistently be doing good, as you have a 40% ranked win rate.

Cenobite AzazielJul 29, 20151 votes

People fail to comprehend what is MMR and that LP is based on the teams MMR. the system that you want is the current system. ELO means nothing.

KazaashiJul 29, 20151 votes

Ranked isn't meant to determine your individual skill. It never has and never will be. This is a team game. Ranked is meant to test how well you can co-ordinate a team of 5 people to blow up the other teams Nexus before they blow up yours. That is literally all that matters in the game.

DrNovaJul 29, 20151 votes

The problem is Solo Q, and you being punished for other teamates doing badly, while having no way to dodge matches without penalty, no recourse for AFK's and Feeders, and getting teamates based on a flawed ranking system as opposed to matching your skill level.

Team based rating systems are fine for actual teams, but with random teamates personal skill HAS to be a factor in ranking.

KazaashiJul 29, 20151 votes

{quoted}

The problem is Solo Q, and you being punished for other teamates doing badly, while having no way to dodge matches without penalty, no recourse for AFK's and Feeders, and getting teamates based on a flawed ranking system as opposed to matching your skill level.

You only lose 3-5 LP and 10-15 mins of your time for dodging a Ranked match. Sure there is a bit of loss, but that is to prevent frequent dodging.
AFK's and Feeders are balanced out if you play enough games. Assuming that you are not one, there are only 4 slots for your team to have one and 5 slots for the enemy team to have one. That works out to a 45% chance for your team to have one and a 55% chance for the enemy team. The Law of Large Numbers states that as your sample size gets bigger, your expected results will fall closer and closer to the statistical predictions, so if you play enough games, the AFKs and Feeders will balance out, or possibly be in your favor.
The "Flawed" Ranking system is an opinion, not a fact to base an argument on. See below

Team based rating systems are fine for actual teams, but with random teamates personal skill HAS to be a factor in ranking.

It is testing your personal skill. Your skill in organizing a group of 5 people to blow up the other teams Nexus. That is the only skill that matters.

Amount of CS doesn't matter, towers destroyed doesn't matter, and kill participation doesn't matter, as they are just a means to an end. Sure, doing those things consistently and skillfully is important, but there is no set amount of any of those things that makes you win the game. Only blowing up the other teams Nexus will do that, and it is up to you and your team to figure out how to cobble those other factors together for a win. That is the skill that is tested.

AtheosisXJul 29, 20151 votes

I would assume that this would use an algorithm like the mastery score system to determine how well you did. If so, such a system would be as broken as the current system, possibly even moreso. The mastery scoring system is a joke.

KVbqbFsC8eAug 1, 20151 votes

Any system that uses solely wins/losses in a 5v5 game is flawed. According to Riots logic, JJ Watt is a silver player. Hah!

Koala FrenzyAug 1, 20151 votes

[{quoted}](name=The Big Piece,realm=NA,application-id=3ErqAdtq,discussion-id=B8KgahXF,comment-id=0006,timestamp=2015-08-01T08:52:04.208+0000)

Any system that uses solely wins/losses in a 5v5 game is flawed. According to Riots logic, JJ Watt is a silver player. Hah!

The NFL is like the LCS. Each game and player is individually analyzed. Watt had to climb through elo hell of highschool and college football to get there and also had the benefit of each game being analyzed by scouters. That can't exist in a video game accessible to everyone with an internet connection.

MathbalnaseAug 1, 20151 votes

Teemo prances across the map, spreading shrooms and triggering the dreaded Global Taunt. His enemies, infuriated by the antics of the Satan Squirrel, make it their life's mission to hunt him down and kill him. No matter the cost. Teemo dies. Repeatedly. He gets no kills in the process, and is soon worth so little gold to his murderers that they would be better off spending that time farming cs. Meanwhile, Teemo's allies are pushing top and soon take the inhibitor, the resulting pressure allowing them to push other lanes as well, and soon after, to win the game.

All of Teemo's game stats (K/D/A, c$, damage to enemy champions) are abysmal - he's barely distinguishable from an intentional feeder - but his contribution to his team allowed them to push with virtual impunity. By what metrics does a TrueSkill system fairly analyze Teemo's contribution and adjust his LP gains accordingly?