To the MMR team:

Ed2Cute·Oct 31, 2014·1 votes·1,222 views

Why is it a good idea to determine someone's personal performance based only on whether the team they were on has won or lost? It's a little like trying to determine "Z" from the formulas "B+C=A, D-H=A, FxL=A, R/E=A" where "Z" is your skill, "A" is your MMR, and all other letters are blue and purples teams. They're all variables and have varying outcomes and you're trying to determine a constant without actually factoring in the constant.

In 5v5's, 3v3's, or even high Diamond to Challenger tiers, the MMR system is fine. There're far less variables and to be that high up you actually do have to be close in skill level. You can more-or-less decide how good someone is based on the team performance (mostly in 5's or 3's). But you can't really do that when there are thousands of people in the same division and you almost never see the same person in a game twice. Every game is a gamble and the win or loss can possibly have nothing to do with your actual performance.

30 Comments

Matthias9119Oct 31, 20145 votes

It's true that in any individual match, a single player's ability to determine the outcome is limited. Over many many many many many matches with the only constant being that single player, though, the other factors even out.

In some sense, if "MMR" in solo queue is supposed to represent "how good are you at helping a random team win?", it's optimal to base it on how often teams that you're on win.

Then there's a more practical concern, which is that if you start rewarding players for individual stats you will inevitably distort how people play the game. Most of the straightforward statistical measures that are usually suggested (KDA, CS/minute, gold earned/minute, kill participation, objectives taken, etc.) either punish certain roles, overreward certain tactics, discourage risky plays, or pigeonhole players into "meta" roles. Or all of the above.

Ed2CuteOct 31, 20141 votes

It's true that in any individual match, a single player's ability to determine the outcome is limited. Over many many many many many matches with the only constant being that single player, though, the other factors even out.

They don't even out. They can't. Hypothetically, if you didn't know any of these beverages I'm about to name, that would be like you being able to pinpoint the Kool-Aide flavor added to Dr. Pepper, Coca-Cola, Pepsi, Mountain Dew, Sprite, and Sierra Mist. Again, it's hypothetical if you've never heard of them and didn't know their chemical make-up at all. Everything is different between them. You don't know which taste is the Kool-Aide and which is the soda. And over a million cans, you're going to find differences and similarities in all of them, but you'll never be able to 100% pinpoint the Kool-Aide.

In some sense, if "MMR" in solo queue is supposed to represent "how good are you at helping a random team win?", it's optimal to base it on how often teams that you're on win.

Not necessarily since there are undeniably those players will disregard any advice you give them. I've come into conflict with MANY players who are mad just because I'm trying to call shots and they think shot calling is for LCS players only and it doesn't work in solo-queue. I mean, there's a reason the LCS does that. It's a good strat to have the team on the same page and the best way to do that is to flip to one person's page.

Then there's a more practical concern, which is that if you start rewarding players for individual stats you will inevitably distort how people play the game. Most of the straightforward statistical measures that are usually suggested (KDA, CS/minute, gold earned/minute, kill participation, objectives taken, etc.) either punish certain roles, overreward certain tactics, discourage risky plays, or pigeonhole players into "meta" roles. Or all of the above.

That issue already exists. That's why champions have a 35% pick rate. In over 1/3 of all games a single champion is present because their gameplay is so rewarding. Or that's why, in solo-queue, you will almost never see a Trist solo-top, Sona+Ahri mid, Teemo jungle, and Malphite solo bot vs "The Meta." And pigeon-holing is already a thing. You know, when someone is last pick by tries to say, "I'm a bad jungler" and lo' and behold every other role is taken so he HAS to jungle?

Matthias9119Oct 31, 20143 votes

They don't even out. They can't.

They do if you assume:

A player's "skill" has some impact on the outcome of each match, and that having more "skill" gives a bigger positive impact (or smaller negative impact if you want to model below-average skill as being a negative factor).
The other people filling in on both teams are selected randomly and uniformly -- that is, any given other player being considered to put in the match has an equal chance to fill in any of the 9 remaining "slots" in the match. (This makes the effect of those players look like a normal distribution.)
Any other random factors that might influence the outcome of the match (like pick order, blue vs. purple side, a player getting randomly disconnected, crippling server lag, etc.) occur randomly and are not stacked against anyone in particular.

Then a particular player's chance of winning a particular match would probably be proportional in some way to:

(player skill) + 4 x (teammate skill) - 5 x (opponent skill) + (random factors)

You might argue about things like whether teammate and opponent skill should be considered additive, etc. but I hope you'd agree it would look SOMETHING like that. In general, having better teammates makes you more likely to win, and having better opponents makes you more likely to lose. Random factors are random.

Given the assumptions I outlined, the average of (teammate skill), (opponent skill), and (random factors) will each converge on some fixed value as the number of games played tends towards infinity. You're just as likely to have good teammates as bad ones, and just as likely to have good opponents as bad ones. Therefore the only difference in how various players are expected to perform in the long run is their own skill. Because of that, it justifies deriving "skill" by looking at winrate over many games.

Not necessarily since there are undeniably those players will disregard any advice you give them.

True, you can be dragged down by bad players or AFKs, just as you can be carried by someone much better than you or win due to an AFK on the other team. You still have some influence on the outcome of the match by virtue of your own play and your ability to attempt to influence your teammates. Over many matches any positive or negative influence caused by the other players must even out, since the same set of players is appearing as both your teammates and opponents.

That issue already exists. That's why champions have a 35% pick rate. In over 1/3 of all games a single champion is present because their gameplay is so rewarding. Or that's why, in solo-queue, you will almost never see a Trist solo-top, Sona+Ahri mid, Teemo jungle, and Malphite solo bot vs "The Meta." And pigeon-holing is already a thing. You know, when someone is last pick by tries to say, "I'm a bad jungler" and lo' and behold every other role is taken so he HAS to jungle?

You're talking about a totally different issue, which is whether there are a limited number of strategies that maximize your team's chance of winning the match. This is irrelevant to whether matchmaking/MMR is "fair", because the same set of strategies are available to all solo queue players.

If you rate people on individual performance, the winning strategies are the ones that maximize your personal score, not necessarily the ones that maximize your team's chance of winning the match. The more heavily you weight individual performance versus the team win/loss, the stronger the potential incentive to do things that improve "your" score but damage the team's chances of winning. This creates perverse incentives that can cause bad situations in-game. For instance, if you lose less LP/MMR in a loss by having a better KDA and you believe the game is likely to be lost no matter what, you are incentivized to let the enemy team win rather than making risky plays to attempt a comeback. Especially if you can also lose less LP/MMR by doing things like getting CS or taking an undefended tower in top lane while they 5-man push your base.

Ed2CuteOct 31, 20142 votes

They do if you assume:

A player's "skill" has some impact on the outcome of each match, and that having more "skill" gives a bigger positive impact (or smaller negative impact if you want to model below-average skill as being a negative factor).
The other people filling in on both teams are selected randomly and uniformly -- that is, any given other player being considered to put in the match has an equal chance to fill in any of the 9 remaining "slots" in the match. (This makes the effect of those players look like a normal distribution.)
Any other random factors that might influence the outcome of the match (like pick order, blue vs. purple side, a player getting randomly disconnected, crippling server lag, etc.) occur randomly and are not stacked against anyone in particular.
Yes, it does. And, generally speaking, you should have an impact. However, we all know a wonderful term coined by gamers called "Getting carried." Which refers to one person making all the effort to win or someone who was a total detriment to the team but they still managed to come out on top.
Given that every game is different people that could be in different moods on their smurfs, the variables stack up too much to produce a constant.
They're not "stacked" in a particular favor, but they do effect you regardless. Just because they could happen to anyone doesn't mean that it happens to everyone an equal amount of times to provide a level playing field.

(player skill) + 4 x (teammate skill) - 5 x (opponent skill) + (random factors)

It's more like MMR=((p+4t)-(5o+r))/g. p=player, t=team, o=opponents, r=variables, all divided by g=games played. Well, that's what you're saying. And the formula would be extended MASSIVELY because you have to assign values and add them up to determine the single value of r. But that's not how Riot works. Riot does MMR=W/L and assigns you a bracket.
I don't think it will work like that because that's a flat number. It's per game until you do g, but in my points above, that wouldn't work either. If anything, determining a player's skill via current MMR system is like when a parabola on a graph gets infinitely closer to the edge but never actually reaches it (I forget the real term). Only Riot's parabola has a much more slight curve and is much farther away from the edge.

Given the assumptions I outlined, the average of (teammate skill)

You can't average teammate skill because the players are always random and depending on whether you win or lose, you gain a new set of eligible teammates for better or for worse.

Over many matches any positive or negative influence caused by the other players must even out, since the same set of players is appearing as both your teammates and opponents.

No, it doesn't have to even out. There's no calculation in Riot's MMR that says, "Well he had and AFK on his team, so we need to make the next game have an AFK on the enemy's team." It's like trying to achieve a true 100% average after you've already failed one test. It's a blemish that will always be there.

If you rate people on individual performance, the winning strategies are the ones that maximize your personal score, not necessarily the ones that maximize your team's chance of winning the match.

If it's based solely on whether you win or lose, that promotes the winning strategy. If you do not choose to conform, you're putting your team at a disadvantage. If it's based on individual performance, it encourages you to play better and smarter. If you're team hasn't already lost the game without losing the nexus, then you're in good shape. But if your team as a whole goes 15/37 where 10 of those kills are yours in top and you only died 3 times, then that game is lost. There's no shame in admitting defeat. Trying to justify mmr only being win or loss by saying risky plays are a good thing isn't... valid in my opinion. I mean, passive play is already so rewarding. There's no downside to just saying, "Okay, I'll sit under tower and farm forever."

Matthias9119Oct 31, 20141 votes

I'm going to try to paraphrase the arguments you're making -- please tell me if I'm misrepresenting what you meant.

The random factors might not even out for a particular player.

This is true, but it becomes extremely unlikely that an individual player will have a significant net amount of "luck" as they play more and more games. Over 25 or 50 games, sure, you could get screwed by disconnects on your team or uncooperative teammates or whatever. Over 100? 200? 500? It's going to even out for almost everyone. Bad luck in individual games gets balanced out by good luck in others. What matters is that on average your chances of being helped or hindered by your teammates/opponents/random factors evens out.

You don't like adjusting the system's estimate of "skill" solely based on wins and losses.

If your probability of winning an individual match (let's call this W(x) ) is proportional to (<your skill> + <random factors>), then your expected win rate over N games [g_1, g_2, ..., g_n] should be:

( W(g_1) + W(g_2) + ... + W(g_n) ) / N

If you substitute in the terms, you get:

( (<your skill> + <random factors_1>) + (<your skill> + <random factors_2>) + ... + (<your skill> + <random factors_n>) ) / N

Or:

( (N * <your skill>) + (<the sum of the N random factors>) ) / N

And the key thing here is that as N gets big, the sum of the N random factors becomes very close to zero in expectation. This leaves your skill as the only significant factor in your winrate. This is true whether the random factors are due to the influence of teammates and opponents, r something intrinsic to the game -- as long as it is expected to even out over a large number of matches.

You feel that an individually strong performance means that a loss is not that player's fault. (And also, presumably, that a weak performance means a win was not due to that player.)

Conceptually, this can be true. For instance, if a player deliberately sabotages his team by AFKing or intentionally feeding, you can argue that any positive result there should not count for that player (Riot actually does this for AFKs; if the team wins the 4v5 it's still a "loss" for the AFK player) and a negative result should have less or no impact on that player's teammates.

Perhaps you could apply that in a less severe way in less extreme cases, but it is very difficult to directly tie individual stats to the team's performance in a robust way. Just having a lot of kills or a lofty KDA doesn't mean you were killing the right players at the right time, or that you actually used that advantage to improve your team's chances of winning.

You argued that a player with good stats didn't contribute as much to the loss, but to me that's backwards. If a player is consistently losing despite having "good stats", that means that the stats aren't actually what matters.

"I always win lane and then my team throws, why are my teams so bad!" Is a common complaint. But "winning lane" is not some magical guarantee of winning the game. What's actually happening is that these players are not having enough mid-to-late game influence to actually close out the match. Their early-game dominance might lead to pretty stats, but that doesn't mean anything if it doesn't actually win the game.

Then that leads into the last argument...

You don't think that individual scoring would influence gameplay, or perhaps that gameplay is already being influenced in some way by Riot.

Riot effectively "scores" you right now based on whether your team was successful in destroying the enemy nexus. They do not give you any kind of bonus or penalty based on any decisions you made in-game. You are rated solely on how effective your team's strategy and execution is.

While the rules of the game itself may penalize certain decisions, that's just... how games work? Complaining that League "punishes" you for not having a dedicated jungler is like complaining that chess "punishes" you for carelessly losing your queen.

Ed2CuteNov 1, 20142 votes

What matters is that on average your chances of being helped or hindered by your teammates/opponents/random factors evens out.

There's no law or MMR prompt that dictates evening the odds. Therefore, you can't say that they just even out. Flipping a coin a million times does not mean you should get heads 50% of the time. It's a 50% chance every flip and the results can tip either way, however drastically.

( (N * <your skill>) + (<the sum of the N random factors>) ) / N

Using this, we're supposed to solve for, let's say "W," the probably of winning. Ideally, W should equal 50%. So N=# of games, Y=your skill, R=sum of the random factors. (NY+R)/N=.5 If we give skill a rating out of 10, we can maybe say a Gold 5 player is at like a 6. So new formula is (N6+R)/N=.5 We can also give a number of games. So maybe 50 games. (300+R)/50=.5 So we multiply both sides by 50 and get 300+R=25 So the negative factors (skill, dc's, afks, etc) have been given a value of -275. So let's see what happens over more games. N=200. So (1200+R)/200=.5 1200+R=100 R= -1100 The problem with this formula is that WHAT is R? We have a value for it, but R is the sum of something. And it must always be negative. So afks, dcs, everything is given more weight and pulls you down farther than any positive. And that's to balance out someone achieving too much too fast. Fucking mathematicians man. Lol.

becomes very close to zero in expectation.

Becoming close to zero isn't zero. And that marginal difference enough to drop you bit-by-bit. And since it's a team game, the MMR formula is basically calculating at which division you are good enough to single-handily defeat every opponent on the enemy team (and, to a degree, which champion is over powered as fuck).

Their early-game dominance might lead to pretty stats, but that doesn't mean anything if it doesn't actually win the game.

No, you have to look at the team's stats. A win or loss doesn't determine overall performance. It could have been a lucky play at Baron that won it or a simple mistake. But, again, if the Riven is like 10/3/7 and every other lane is like 2/6/0, the Riven tried to contribute, but a single person can win the entire game. One person has a limited influence on the game. It can be a big influence like forcing their jungler to camp you so your jungler can help lanes, but if your team gives the enemy a triple kill when they gank, you can't force them to play better. And you shouldn't be punished because your team underperformed and couldn't use or influence effectively. And it's a rather large punishments because it's always much easier to dwindle a percentage than it is to increase it.

While the rules of the game itself may penalize certain decisions, that's just... how games work? Complaining that League "punishes" you for not having a dedicated jungler is like complaining that chess "punishes" you for carelessly losing your queen.

I'm not complaining that. I'm saying that it wouldn't change based on personal or W/L stat. Either way you want to use your "Queen" effectively. Either way, if you play a careless strategy, you're going to lose and be punished for it. The only difference is that you can't be as severely punished if your top laner TPs into 5 of them when your jungler and ADC are already dead for god knows why and thus you lose the game. It saves you from having to deal with other people's mistakes. If you want to compare it to anything, it's like going into a team battle on Halo with someone who continually kills his teammates. It purposefully lowers your teams score and either takes them longer to win or can easily turn the tides and make you lose horribly. That's not fun and it's not your fault. What are you gonna do, kill him back and lower your score twice as fast? The way Halo deals with it is that it will kick that player, then the team is still stuck with fewer people on the team. And that the team that had someone get booted SHOULDN'T lose that exp/LP/whatever Halo has. It's simply not fair that they're punished for one ass-hat.

Matthias9119Nov 1, 20141 votes

There's no law or MMR prompt that dictates evening the odds. Therefore, you can't say that they just even out. Flipping a coin a million times does not mean you should get heads 50% of the time. It's a 50% chance every flip and the results can tip either way, however drastically.

I'm going to guess you haven't taken much statistics. If you have a fair coin and flip it a thousand times (not a million), you only have a 5% chance of having more than 525 or fewer than 475 heads. In a million trials, you have a ~98% chance of being between 499000 and 501000.

Could there be one incredibly unlucky guy who plays 1000 matches of league and gets an AFK on his team 300 times and not once on the enemy team? Sure. But that is going to mess up any rating system you use, unless you very heavily weight individual performance over team outcomes.

( (N * <your skill>) + (<the sum of the N random factors>) ) / N

Using this, we're supposed to solve for, let's say "W," the probably of winning. Ideally, W should equal 50%.

No, that analysis is not correct. Maybe I was unclear.

The sort of formula I'm outlining there would apply if you were taking ALL players and just mashing them into games together. Looking at what actually happens with Elo-like systems over time is more complicated, because they try to match up players who are close in (estimated) skill.

If most players' estimated skill is reasonably accurate, this means that the higher your estimated skill, the stronger your teammates and opponents are. If your actual skill is much higher than the estimate, you'll tend to dominate games and win a lot more than you lose, raising your estimated level of skill. If your skill is lower than the estimate, you'll tend to be dominated in games and lose more than you win, lowering your estimated level of skill.

Let's say we rate all players from 1-10 in skill. If you're really a 1 but you're rated a 5, your win probability would be proportional to:

(1 + 4 x 5 - 5 x 5 + (random factors)) = -4 + (random factors)

(your 4 teammates are all around estimated skill "5", as are your five opponents.)

If you're actually a 9 but you're rated at 5, your winrate should be proportional to:

(9 + 4 x 5 - 5 x 5 + (random factors)) = +4 + (random factors)

In reality this is probably not a linear relationship, but the general idea is that your winrate will be a lot higher if you're underrated than if you're overrated. If your estimated skill rating (ie, MMR) is where it "should" be, you'll win about 50% of the time.

What that means is that if you're winning <50% of your matches over many games, your current MMR is (probably) too high. And if you're winning >50% of your matches over many games, your current MMR is (probably) too low.

becomes very close to zero in expectation.

Becoming close to zero isn't zero. And that marginal difference enough to drop you bit-by-bit. And since it's a team game, the MMR formula is basically calculating at which division you are good enough to single-handily defeat every opponent on the enemy team (and, to a degree, which champion is over powered as fuck).

Being close to zero "in expectation" means that as you play more and more games the random factors tend to even out.

Also, I assumed that "R" could be positive or negative, not just negative. You can win by random factors like getting carried or having a DC on the other team, or getting the "OP" champ on your team. Those things happen about as often as you lose to due to those same factors.

I'm going to write another post on the rest of this, because of post character limits. :-)

Ed2CuteNov 2, 20141 votes

you only have a 5% chance of having more than 525 or fewer than 475 heads. In a million trials, you have a ~98% chance of being between 499000 and 501000.

Statistics do not equal probability. And the probably of getting 500 heads in a 1,000 coin flip session is like a quarter of a percent (I looked that one up). And the higher your trials, the more likely you are to get 1,000,000 heads in a row. That's the Law of Large Numbers. The way you look at it would be like watching a game of craps and noticing that the 4 on the dice came up 10 times while the next closest was the 2 that came up 6 times, therefore 4 has the highest likely-hood of come up. Which isn't true. Each side has a 1 in 6 chance of coming up.

(1 + 4 x 5 - 5 x 5 + (random factors)) = -4 + (random factors)

Ok so I'm guessing the base model for your formula are like (Player)+4(Team)-5(Enemy)+(Random)= -E+R? I'm not really sure what the -4 is supposed to be. But you can't have a negative probability. And your equation doesn't actually factor in any random factors. And simply because you're a 1 doesn't mean you will lose the game. And back to my earlier point, there's no formula, law, program, etc in place that will give you a win for any circumstance out of your control that lost you the game. And your assumption of "evening out" means that someone else's game gets fucked up too, and it's a HUGE circle of people getting endlessly fucked up by random factors.

Those things happen about as often as you lose to due to those same factors.

No, they are as likely to happen on your team as often as their team. But that doesn't mean they do happen like that. And the moment one of them happens, it instantly brings your MMR lower. That means you'll get worse people on your team (and the enemy), however we know they're not all equally skilled. So it's enough to put you in the decline.

I'm going to write another post on the rest of this, because of post character limits. :-)

I hate the limites. ._.

Weighting individual performance very heavily is, to me, a very bad idea for the reasons I outlined before. It is very likely to punish and reward specific playstyles, and sometimes encourages players to sabotage their team's chances of winning in order to preserve or pad their own stats. You're also extremely likely to not rate creative or unusual playstyles properly, even if they are actually effective.

But you think it's totally fine to rate and individual's performance based solely on the end result? It's not fair. Solo-Queue in a team game isn't balanced. A player will never be where they should be. Like I said, 5's and 3's are totally fine. However, a solo-queue player is punished or rewarded unfairly and there is absolutely nothing in life or games that governs good luck and bad luck to be equal.

In a team game your ability to coordinate with your team and lead them to victory is part of the "skill" involved, IMO. There is no "you", you are part of a team and you win or lose as a team.

You're only a team if you work together. If they won't listen to you or cooperate, then you're not playing a team game.

But so does everyone else, and short-term luck (good or bad!) does not have a significant impact on where you'll end up being rated in the long run.

Yes, so does everyone else. That's why the most common memes associated with League are along the lines of, "Solo-Queue is cancer." And it does have a huge impact because it's harder to rise than it is to fall. You can win 20 games in a row and get 28LP per win, and the moment you lose one you lose like 15LP and your next win gets you like 18LP. Instantly trying to hinder your rise as much as possible without coming on too strong.

I would be okay with Riot reducing all the LP and MMR gains/losses in matches that had an AFK player or an intentional feeder.

That's what I was suggesting. Or introduce retroactive LP control based on reports.

Matthias9119Nov 2, 20141 votes

Statistics do not equal probability.

Statistics is the study of probability.

And the probably of getting 500 heads in a 1,000 coin flip session is like a quarter of a percent (I looked that one up).

That might be the probability of getting exactly 500 heads. What you care more about here is the probability of getting X or more (or X or less) heads out of 1000, which is a lot smaller as you get away from the expected number of successes. (see, e.g.: http://stattrek.com/online-calculator/binomial.aspx)

And the higher your trials, the more likely you are to get 1,000,000 heads in a row. That's the Law of Large Numbers.

The Law of Large Numbers states that "the average of the results obtained from a large number of trials should be close to the expected value, and will tend to become closer as more trials are performed." (http://en.wikipedia.org/wiki/Law_of_large_numbers)

In layman's terms, it's a lot more likely to flip a coin 10 times and get 10 heads than it is to flip a coin 1000 times and get 1000 heads. And it's even less likely to get a million heads in a row. The more samples you have, the less likely it is that the sum of all the samples diverges greatly from the most likely outcome (ie, the expected value).

In other words, good luck and bad luck even out over time.

Ok so I'm guessing the base model for your formula are like (Player)+4(Team)-5(Enemy)+(Random)= -E+R? I'm not really sure what the -4 is supposed to be. But you can't have a negative probability

That function would give you something like what Elo does -- a relative rating between the teams, taking any random factors influencing that match into account. If you take the random factors out this is one way of extending Elo to teams.

To transform those "ratings" to an actual win probability you'd have to plug them into something like a logistic function: http://en.wikipedia.org/wiki/Logistic_function (see also uses of logistic functions in Elo ratings at: http://en.wikipedia.org/wiki/Elo_rating_system#Theory).

No, they are as likely to happen on your team as often as their team. But that doesn't mean they do happen like that.

If they are as likely to happen to either team then the law of large numbers says they will even out over time.

But you think it's totally fine to rate and individual's performance based solely on the end result? It's not fair.

The objective is to win the match, not to rack up pretty stats for yourself. How good you are at racking up stats is not relevant; what is relevant is how good you are at winning matches!

Any system that takes individual stats into account is making some sort of value judgement about how you win the match. This is, IMO, an extremely bad idea because it will encourage players in solo queue to play for stats rather than always trying to maximize their team's chances to win.

Solo-Queue in a team game isn't balanced. A player will never be where they should be.

You may bounce around due to random luck, but you'll eventually converge on your correct rating.

...there is absolutely nothing in life or games that governs good luck and bad luck to be equal.

The law of large numbers says that your good and bad luck will even out in the long run.

You're only a team if you work together. If they won't listen to you or cooperate, then you're not playing a team game.

No, you're playing a team game with an uncooperate or unskilled team. If you don't like random matchmaking, form a premade team. They're doing the best ratings they can given random matchmaking.

...it's harder to rise than it is to fall. You can win 20 games in a row and get 28LP per win, and the moment you lose one you lose like 15LP and your next win gets you like 18LP. Instantly trying to hinder your rise as much as possible without coming on too strong.

That's not how it works. LP is a whole different beast, because it's basically a pretty set of ranks put on top of MMR (which is hidden). Your LP gains/losses depend heavily on your current MMR.

Ed2CuteNov 2, 20141 votes

Statistics is the study of probability.

No. Statistics is the study of the passed. Statistics concludes that heads is more likely to come up if the trial results produce 6 heads out of 10 flips. Probability states that, given a fair coin, heads has a 50% chance of coming up each time and so does tails, regardless of the outcome.

That might be the probability of getting exactly 500 heads. What you care more about here is the probability of getting X or more (or X or less) heads out of 1000, which is a lot smaller as you get away from the expected number of successes.

If you broaden your search for an expected result, you get a higher likelihood of success. And the site you linked only shows that you have the same likelihood for getting 500 or more heads as you do getting 500 or more tails. Because every flip is 50% and you are not even close to guaranteed a certain result either way.

The Law of Large Numbers states that "the average of the results obtained from a large number of trials should be close to the expected value, and will tend to become closer as more trials are performed." (http://en.wikipedia.org/wiki/Law_of_large_numbers ) In layman's terms, it's a lot more likely to flip a coin 10 times and get 10 heads than it is to flip a coin 1000 times and get 1000 heads. And it's even less likely to get a million heads in a row. The more samples you have, the less likely it is that the sum of all the samples diverges greatly from the most likely outcome (ie, the expected value).

Yes
No. I mean, yes you're more likely to get 10 heads in a row than you are 1,000. But your analogy is irrelevant. A fair coin has a 50% of landing on heads or tails, therefore over N trials eventually we will reach the 50% mark on results as the average. However, that's an infinite case. We do not have infinite games to play during a season in League. And it's much more likely that we will not have a 50% after just 100 games.

http://en.wikipedia.org/wiki/Elo_rating_system#Theory

"Performance isn't measured absolutely; it is inferred from wins, losses, and draws against other players." And as we all know, you can infer incorrectly. Riot does this by ignoring information.

If they are as likely to happen to either team then the law of large numbers says they will even out over time.

Sure, but we don't have enough time. If your N is 500 games, there is no part of "Law of Large Numbers" that says if you won 280 games, you will have to have lost 220 games. That's the Gambler's Fallacy. Law of Large Numbers is infinite and only when it reaches the expected value is your study complete. You can't give it a value and expect the gods of probability to even it out for you.

The objective is to win the match, not to rack up pretty stats for yourself. How good you are at racking up stats is not relevant; what is relevant is how good you are at winning matches!

How good you are at racking up stats is the only relevance to how good of a player you are. In a game that is 100% team oriented, the win or loss will ONLY factor the team's overall skill COMPARED to that specific enemy. It is not a solo-effort and therefore you should not be treated as if it's all your fault. Calling someone in Bronze terrible isn't justified just because they're in Bronze.

The law of large numbers says that your good and bad luck will even out in the long run.

In the indeterminably long run. If we squish one season down to a day and the average time of games is about 40 minutes queue and champ select take about 10 minutes, you have 50 minute segments to fill the day. So you can play ~29 games a gay. Law of Large Numbers doesn't dictate that you will have lost 14, won 14, and there was a draw. That's a Gambler's Fallacy.

Pt.1, I have to post another. I'll do it in reply to this one.

Ed2CuteNov 2, 20141 votes

No, you're playing a team game with an uncooperate or unskilled team. If you don't like random matchmaking, form a premade team. They're doing the best ratings they can given random matchmaking.

I said 3's and 5's was fine. And they are doing decent. It's not hard to see what bugs or issues may have effected the game and it is DEFINITELY easier to do retroactive LP gain/loss via reports. But they are labeling it all wrong. It's not a skill rating. Ignoring every factor that determines one person's skill is not a valid way to determine one person's skill. IE: trying talk about how sweet kool-aide is judging by how red it is. Two irrelevant factors and you're trying to draw a conclusion from it.

That's not how it works. LP is a whole different beast, because it's basically a pretty set of ranks put on top of MMR (which is hidden). Your LP gains/losses depend heavily on your current MMR.

It is how it works. LP gains and losses can very drastically, but it's always harder to increase your LP gain. And LP gain/loss is dependent on current MMR, so it's harder to gain MMR than it is to lose it. In the same way the average of (100+100+100+99+100)/5 =/= 100. At a certain point, you have to ignore the passed. Now, it's true that if you suddenly start winning 85% of your games, you can increase your MMR and LP gain, but it's much harder to do that than to lose 45% of your games and have your MMR and LP gain diminish.

Matthias9119Nov 3, 20141 votes

It is how it works. LP gains and losses can very drastically, but it's always harder to increase your LP gain.

If you stay in the same division but you improve your MMR, your LP gain goes up. Not complicated.

And LP gain/loss is dependent on current MMR, so it's harder to gain MMR than it is to lose it.

LP and "rank" are derived from MMR -- whatever your current LP/rank is has no impact on MMR gains or losses.

In the same way the average of (100+100+100+99+100)/5 =/= 100. At a certain point, you have to ignore the passed. [sic]

This doesn't even make any sense.

Now, it's true that if you suddenly start winning 85% of your games, you can increase your MMR and LP gain, but it's much harder to do that than to lose 45% of your games and have your MMR and LP gain diminish.

Generally, if you win >50% of your matches your MMR goes up, and if you win <50% of your matches your MMR goes down. There are some really weird extreme cases where this might not happen, but you'd have to win only close matches and lose ones where you were heavily favored. (This is unlikely to happen in Ranked queues because most players are more consistent and serious there, and it won't usually give you heavily lopsided matches to begin with.)

If you're losing most of your matches that's evidence that you are rated too highly, and conversely if you are winning most of your matches you are probably rated too low. The more you are misrated the more lopsided it will be one way or the other -- if you suddenly gave an average player a Challenger-tier MMR they'd probably have a 10% winrate until they dropped down quite a ways. And the bigger the sample size the more convincing the evidence, which is why MMR adjusts fairly slowly once you complete your provisional matches.

Ed2CuteNov 3, 20141 votes

If you stay in the same division but you improve your MMR, your LP gain goes up. Not complicated.

What. The. Fuck. NO SHIT! That's we're saying the same thing. That is a fact. What you're not recognizing is that it is easier to drop than it is to improve. One loss weighs more than a win. THAT is not complicated.

LP and "rank" are derived from MMR -- whatever your current LP/rank is has no impact on MMR gains or losses.

Are you troll right now? They are directly correlated. You win more, you gain MMR, you gain LP. They all feed into each other so they are ALL impacting each other.

"In the same way the average of (100+100+100+99+100)/5 =/= 100. At a certain point, you have to ignore the passed. [sic]"

This doesn't even make any sense.

It's basic math. The average of four 100's and one 99 does not equal 100. It is literally an undeniable truth.

If you're losing most of your matches that's evidence that you are rated too highly, and conversely if you are winning most of your matches you are probably rated too low. The more you are misrated the more lopsided it will be one way or the other -- if you suddenly gave an average player a Challenger-tier MMR they'd probably have a 10% winrate until they dropped down quite a ways. And the bigger the sample size the more convincing the evidence, which is why MMR adjusts fairly slowly once you complete your provisional matches.

Nonononono. You can have a winstreak and a lose streak in the same MMR. It's evident in Challenger/Master ranks. At the tip top they can have 15 wins in a row and the next day lose 10 in a row. It happens all the time. I see it personally with people like KEITHMCBRIEF. Just because you're on a loss streak doesn't mean your MMR is too high and vice versa isn't true.

Matthias9119Nov 3, 20141 votes

Listen... this has been an interesting conversation, but if you're going to start putting words in my mouth and then arguing against basic statistics I'm not sure it is going anywhere at this point.

No. Statistics is the study of the passed. [sic]

When you take an academic class on these topics the course is usually called "statistics". Anyway, irrelevant.

Statistics concludes that heads is more likely to come up if the trial results produce 6 heads out of 10 flips.

No. That's inference. Of course, what Riot is trying to do is to infer the actual skill level of all the players in the Ranked ladder as accurately as possible based on how they perform.

If you broaden your search for an expected result, you get a higher likelihood of success. And the site you linked only shows that you have the same likelihood for getting 500 or more heads as you do getting 500 or more tails. Because every flip is 50% and you are not even close to guaranteed a certain result either way.

...I mean, yes you're more likely to get 10 heads in a row than you are 1,000. But your analogy is irrelevant. A fair coin has a 50% of landing on heads or tails, therefore over N trials eventually we will reach the 50% mark on results as the average. However, that's an infinite case. We do not have infinite games to play during a season in League. And it's much more likely that we will not have a 50% after just 100 games.

I really don't think you understand the law of large numbers.

Yes, in any finite numbers of fair-coin-flip trials you are unlikely to get exactly 50% heads. But as the number of trials increases it is less and less likely that the number of heads diverges significantly from the expected number. In the long run, short-term luck is irrelevant.

"Performance isn't measured absolutely; it is inferred from wins, losses, and draws against other players." And as we all know, you can infer incorrectly. Riot does this by ignoring information.

Whether a team wins or loses depends solely how the players on each team in the match perform. (Barring things like catastrophic server failure, etc.) No information is lost by only considering wins and losses, because they are derived solely from the players' skill in the long run.

What you might be able to argue is that if you did have extremely robust statistical measures that had a causal relationship with winning, you could converge on correct estimates of players' skills more rapidly than by just considering wins and losses. However, this appears to be very hard to do, and doing this would impact how people play the game (which is, IMO, extremely not good).

Sure, but we don't have enough time.

The more games you play, the less good or bad luck in any individual game matters. Is it possible you could play hundreds of games and have a huge amount of net negative luck? Yes. But it's very, very unlikely, and the more games you play the less likely it is to be a significant factor overall. And bad luck could cause you to have really bad personal stats game after game after game, too, or to lose even with a premade team! (Let's say you keep having horrible ping spikes and DCing.)

How good you are at racking up stats is the only relevance to how good of a player you are.

STRONGLY disagree. "Stats" are not a good measure of "ability to win games" unless it's something that is absolutely necessary to win and has a provably strong causal relationship with winning. Most stats in League are far too abstract to use like this, because there are so many different strategies possible and it's so easy to pad stats in wins or preserve them in losses.

If you come up with some clever strategy that results in you personally having "bad" stats but still winning lots of matches, you shoudn't be punished for it. As an example, proxy Singed play can heavily influence matches but usually results in the proxying player having poor KDA numbers even when successful. Arguably being a 0-CS support does the same thing to a lesser degree.

OnegarionNov 3, 20141 votes

I've got to jump in with this great discussion since he doesn't seem to understand what is going on. I'm going to use Matthias9119 last post and Moonbat's post 2 ago with quotes about this. I'm also going to try and make it a little less mathematical overall.

Statistics concludes that heads is more likely to come up if the trial results produce 6 heads out of 10 flips.

First league isn't set up in groups of 10 matches causing you to rise or fall. League is based on an idea that we will have an infinite number of matches. Just because you may play 1 match a day or something doesn't take away from the law of large numbers. This involves frequency which is a measure of how often something happens. SO when you include frequency in your view of this you can then measure how quick the law of large numbers starts to really appear. If I play 20 games a week and Matthias9119 plays 10 matches a week its going to take me half the time for the appearance of this. I'm going to see my MMR get to a fixed point faster.

When we look at frequency in terms of League we can also assume that with increased frequency we also have an increase of player skill by a small percentage through repeating the same processes. So this as well puts in a positive addition to your equation that you both have discussed.

No. I mean, yes you're more likely to get 10 heads in a row than you are 1,000. But your analogy is irrelevant. A fair coin has a 50% of landing on heads or tails, therefore over N trials eventually we will reach the 50% mark on results as the average. However, that's an infinite case. We do not have infinite games to play during a season in League. And it's much more likely that we will not have a 50% after just 100 games.

You kind of have proven against yourself here. With each game you play you bounce around to eventually get to a middle ground. MOBA's at their heart are meant to get every player to 50% WR. SO this is our expected goal. We expect that when a player has reached their actually MMR that their WR will stay at 50% over the course of many games. This is why we can't look at 50% in the short term. Atm I personally have 506W/503L in ranked giving me a 50.1% WR. Presumably we can assume that atm this is where my MMR should be, but I won 2 out of my last 3 ranked games so this must be false......Wrong. You can't dismiss the large number of games for a current trend. You have to always look at your total games and not just your last 10 ranked games in your history. You will never get an accurate result from it because it has way to many false information and incorrect routes. So your MMR is closer to "right" after playing 20 games instead of 10. This is why people can gain/lose a lot of LP and subsequently MMR because the system has a wide range to look at. It has to play more games to know where to put you.

Say you seed at 1500 MMR. Next game you play makes you win so you go up 100MMR (grossly inflated of actual results) and your next game is a loss causing you to lose 100 MMR. So the system not has a basis, albeit small, of where to put you. So it thinks you are now between 1500-1600 MMR. Your third game is a loss causing you to lose 90 MMR. Now the system corrects itself saying that maybe you are between 1401-1500. You win gaining 98 MMR putting the system now to 1499-1598. As you can see the system gets closer and closer to a number as you fluctuate between winning and losing. As you play more each win/loss has less overall impact on the systems range. Only a big increase of win/loss will cause the system to correct largely. If you have played 100 games and go on a 5 game win streak the game will increase a lot more than if you won 3 loss 1 and then won 2 more. With the above examples your MMR would be considered to be the average of the high and low range. Ofc as the ranges close in the MMR average gets to be much more stagnant. This is how you can stay at a similar MMR even though you may be on a winning or losing streak as well.

OnegarionNov 3, 20141 votes

How good you are at racking up stats is the only relevance to how good of a player you are.

This by far is one of the worst cases for league. You CAN NOT DO THIS. There is no viable way of calculating this. I have heard many different models out there with some being halfway decent. The main problem though with each system is setting a value to make each role viable. You can't say that a kill is worth X for an adc, but Y for a support. You can't say an assist is worth less than a kill because that would then inhibit supports from moving up. You can't use damage taken as a tank because you can't account for whether that damage was received right or not. I could tank tower shots walk away and do it again and farm stats or the same with some champs. Now i'm making a deal with the enemy to let me tank a lot of damage from then in turn I trade a couple kills to them to boost both of us up.

Ultimately this game isn't won based off killing the enemy a lot. I've seen games won with less than 10 kills in the entire game and I've seen some where there is close to 100 kills. Honestly I feel like the game with 10 kills shows a better skill overall than the game with 100 kills. It shows overall smarter play and that players where not randomly dying or causing huge bloodbaths. It shows a better understand of whats going on with the ability to stay alive as a whole and not feeding the enemy senselessly.

Then you have to consider the number of plays that are made that don't get recorded by stats. That player facetanking the tower to take it. The split push that let your team sneak a baron/dragon. The Leona that landed a perfect 5 man ult. Yea the leona gets credit for the assists, but the contribution to the team was much more than 5 simple assists. How about that initiator that goes in without the team and causing your team to lose? Well to stats its simply one death and doesn't mean a whole lot overall, but thats they one move that caused your team to lose even though you had more gold/kills/towers taken/etc. That player is actually benefited because they made the losing play, but only get hurt for having the 1 kill and maybe getting more points for damage taken.

There is absolutely no way that having the most stats makes you win. To include sports into this. I've watch both football and basketball games where 1 player had all the stats (QB throwing all the TD, Point guard with all the points) and their team lost. Yet no one likes to talk about the fact that the QB was against an ailing defense (random factors) and had their best player benched due to injury causing him to get a favorable matchup for their TE and letting him throw to him all day with no contest. Yet his team's defense was much worse than the enemies and they were outscored. Their entire season is based only on W/L and no one seems to complain thats its not right. It doesn't matter if half the team is good and the other half bad and those are organized teams. Just think if it was random every week where players would play from the basics of what team to what position under what coach during what time. This is the way league is. It gives us much more randoms, but if a player is a good WR he is always a good WR. Just because he changes teams doesn't make him a worse player. Maybe the QB can't throw as far or as accurate, but it doesn't change the WR talent. THe same is true for league. Just because you ADC/SUpport suck doesn't mean you a top laner are any worse. You get credit for being good by having your good score, but its ultimately not what matters. Its all about the W/L.

Ed2CuteNov 3, 20141 votes

If I play 20 games a week and Matthias9119 plays 10 matches a week its going to take me half the time for the appearance of this. I'm going to see my MMR get to a fixed point faster.

That's not technically true. Since the Law of Large Numbers doesn't have a rule about how many times something can happen until you get your desired result.

You kind of have proven against yourself here.

Not really. I'm talking about a working correction system. Which I don't believe Riot is using. So I believe the results are skewed by random factors either way.

This by far is one of the worst cases for league. You CAN NOT DO THIS. There is no viable way of calculating this...

Riot did this with the FantasyLeague, remember? They can and have assigned points based on performance. And it didn't seem to change how they play. They still played for the win because there's and end-game goal in mind. People who troll or trade stat boosts, report them for abusing the system. We can track this the same way team builder works with selecting roles.

To include sports into this.

No. Players are rated as a team. "The Cowboys are the best team" or "The Angles will win the series." And, like I said, this system works for 3's or 5's teams. However, nobody says "Angels lost, John Erikson must be a terrible pitcher." (Random name) And he is paid based on how well HE performs. Not the team. Hall of Fame is awarded based on what HE did. Not his team.

In League terms. Gold doesn't mean I am Gold skill. Gold means this is where my teams have carried me. You know my division, not my story.

But, I mean, I do have likea 51% winrate or something right now. I'm 235 to 230 ish.

OnegarionNov 3, 20142 votes

That's not technically true. Since the Law of Large Numbers doesn't have a rule about how many times something can happen until you get your desired result.

While Frequency is not relevant when explaining the Law of large numbers its by no means independent of it. If i play more games the law of large numbers without a doubt manifests itself much more than if I play half as many in the same amount of time. There is not set golden number of where it really shows, but without a doubt playing 1000 games yields a truer result than playing 100 which factors in frequency in a game like League where MMR gets reset and not allowed to scale naturally to infinity without being set back manually.

Not really. I'm talking about a working correction system. Which I don't believe Riot is using. So I believe the results are skewed by random factors either way.

These random factors are accounted for though. AFK? In the equation you discussed earlier thats accounted for and has time and time again been discussed since by the odds and law of large numbers your opponent has a much higher chance of having and AFK than you which ultimately favors your equation and would overall push your MMR up by default. If we wanted to put a number to it lets say there is a .01% chance of getting an afk. So you have a .04% chance and the enemy has a .05% chance. While not a big difference playing 100 games you would have 4 games with an AFK and 5 games with an afk on the other team increasing your score and ofc that difference increases the more games played. I'm really not sure of what random factors skew it in such a way that the model is deemed unusable or invalid.

Riot did this with the FantasyLeague, remember? They can and have assigned points based on performance. And it didn't seem to change how they play. They still played for the win because there's and end-game goal in mind. People who troll or trade stat boosts, report them for abusing the system. We can track this the same way team builder works with selecting roles.

yes they did do this with the fantasy league because it didn't have an effect on the outcome of the game. Getting more kills didn't make a team win more in their actual game. A throw would still cause them to lose. The goal of a game of League isn't to have the highest score, but to win so you chose a way that wins the game. The goal of Fantasy league is to have the highest score so you pick the way that gives you the most points. When going through league there were clear roles/positions that were treated favorably like there are in Fantasy Football that will yield the most points and I can tell you that mostly that was mid laners or individually skilled players that always did well. If the same system was placed on normal games where win/loss wasn't the goal, but to have the best score the way league is played would be completely different. The main difference is the end goal of both of these systems on how you W/L (saying for emphasis).

And he is paid based on how well HE performs. Not the team. Hall of Fame is awarded based on what HE did. Not his team.

Yes this player gets paid based on his performance, but the problem with worrying about pay is that we aren't pros. We are normal people and we aren't on a consistent team. While I started the sports comparison I now have to attack it as well. You can't compare them to normal league as we see them on TV because thats not an equal comparison. You could compare them to LCS play and make a fair comparison. As soloque players we don't have a consistent team to base our skills off of to know how good we are. We base our skill off of how many wins we get and how well we perform. We have constantly changing teams and outcomes based on that. To draw a true comparison you have to include the end of what I said about the WR (long quote won't post, but last paragraph) and the shuffling of the league.

Honestly Hall of Fame has no baring on this discussion. Hall of Fame does take 1 persons accomplishments, but thats after they reach the top. Not the climb.

Ed2CuteNov 3, 20141 votes

What is [sic]?

No. That's inference.

Statsitics: the practice or science of collecting and analyzing numerical data in large quantities, especially for the purpose of inferring proportions in a whole from those in a representative sample.

I really don't think you understand the law of large numbers.

What the fuck? Right below this you agree and still think I don't know? What I'm saying with the section this one is in response to is that we don't have enough games to make Law of Large Numbers apply. Yes, we can have a lot. However, it is entirely possible (and likely) that people never reach this expected point.

Whether a team wins or loses depends solely how the players on each team in the match perform.

It is dependent the performance of the team. So it is unfair to judge a single player's abilities based on what is a team performance. That's why you pros that look at ALLIANCE and are like, "The team is made up of very inconsistent players. Froggen is their only consistent player." Or some shit. They recognize there is a difference between what a team can do and what a player can do. They will rate Froggen more highly than the rest of the team members. And that's why people make the distinction that Froggen is good when they say ALLIANCE is bad.

The more games you play, the less good or bad luck in any individual game matters. Is it possible you could play hundreds of games and have a huge amount of net negative luck? Yes. But it's very, very unlikely, and the more games you play the less likely it is to be a significant factor overall.

Right, my argument is that we don't have enough games. AND there's a formula at work trying to apply the gambler's fallacy. If you do super well, it puts you against harder opponents. You go against harder opponents, it's more possible you lose some. And then it can fall downhill from their. The opposite can happen. But they don't scale. If you have Platinum MMR and you're Gold 4, you don't skip to Platinum because you beat a bunch of Platinums. You gain 28 LP and call it good. Then the next game can have better people or different circumstances (like just playing out of your mind) and you lose it and it drops farther than you went up because the gambler's fallacy is trying to compensate for you progressed to fast with no decline.

If you come up with some clever strategy that results in you personally having "bad" stats but still winning lots of matches, you shoudn't be punished for it. As an example, proxy Singed play can heavily influence matches but usually results in the proxying player having poor KDA numbers even when successful. Arguably being a 0-CS support does the same thing to a lesser degree.

But they get that CS tho. A LOT. And as a tank, they get rating based on damage and assists. There are redeeming factors in whatever strat you choose that works. At the same time, someone shouldn't be rewarded for going 0/15/0 and afk jungle farm but his team won because the other team had a duo that DC'd at the same time for longer.

Matthias9119Nov 3, 20141 votes

What I'm saying with the section this one is in response to is that we don't have enough games to make Law of Large Numbers apply.

This is completely your opinion. The more games you play, the more accurate your rating should be -- that's what the law of large numbers says. Whether it converges fast enough for your liking is another question.

It is dependent the performance of the team. So it is unfair to judge a single player's abilities based on what is a team performance.

It's different talking about pro teams where they play many matches with the same teams against the same opponents.

I stand by my statement that if you have "good stats" but cannot win at a >50% rate over many games in solo queue, the problem isn't with the ranking system. It's that the stats you're looking at do not have a strong causal relationship to winning.

AND there's a formula at work trying to apply the gambler's fallacy. If you do super well, it puts you against harder opponents.

If you're doing really well over time, the most likely explanation is that you are underrated and should be going against harder opponents.

If you have Platinum MMR and you're Gold 4, you don't skip to Platinum because you beat a bunch of Platinums.

Actually, that exact thing can happen. You would probably skip at least one division the next time you promote if you really had a Plat-level MMR in low Gold. You can even skip tier promotions (like jumping from G2->P5) if your MMR is crazy high because you keep losing promo series.

There are redeeming factors in whatever strat you choose that works.

It's not hard come up with strategies that are not as effective at winning games but are much more likely to give you good personal stats in comparison to your team. Such strategies should not be rewarded relative to ones that are better at actually winning games, and if you think they should then we will have to agree to disagree.

A secondary problem is that there is no guarantee that every effective strategy will appear to generate good stats for every player on the team. It's very difficult to take into account things like sacrificing yourself to buy time for your team's CDs to be up in a lategame teamfight. If it works you win the game -- but if you get penalized for having a worse KDA than your team, nobody will do things like that in solo queue.

At the same time, someone shouldn't be rewarded for going 0/15/0 and afk jungle farm but his team won because the other team had a duo that DC'd at the same time for longer.

This sort of thing isn't a problem in the long run. If the strategy a player uses is ineffective, it will not work over a large number of games. Short-term good and bad luck is not relevant with a large sample size.

What you're not recognizing is that it is easier to drop than it is to improve. One loss weighs more than a win. THAT is not complicated.

This is not true in general of Elo-like systems. If you win at even a slightly over 50% rate over time in League you will steadily gain MMR.

Are you troll right now? [LP and MMR] are directly correlated. You win more, you gain MMR, you gain LP. They all feed into each other so they are ALL impacting each other.

LP and rank are a pretty set of labels on top of the MMR system. Matchmaking is done entirely based on MMR. MMR gains and losses have nothing to do with your current "rank" or current amount of LP. They may be correlated, but the causation only goes one way.

It's basic math. The average of four 100's and one 99 does not equal 100. It is literally an undeniable truth.

Okay, yes, I understand averages. Maybe I should say that I have no idea how this relates to the point you are trying to make.

Just because you're on a loss streak doesn't mean your MMR is too high and vice versa isn't true.

Over 5 or 10 games? Probably not. Over 25 or 50? Maybe. Over 100 or more? Yeah, it almost always does.

Ed2CuteNov 4, 20141 votes

Whether it converges fast enough for your liking is another question.

It's not my liking. It's literally how it works. The 50% the current system brings you to is forced. You win a lot, it tries to make you lose some. If you lose a lot, it matches you against people whom you should win against. It's the Gambler's Fallacy being forced to work.

It's that the stats you're looking at do not have a strong causal relationship to winning.

BUT NOBODY'S JUDGING YOU BASED ON YOUR ABILITY TO WIN! THEY'RE JUDGING YOUR SKILL AS A PLAYER IN EVERY ASPECT! WHAT ARE YOU NOT GETTING ABOUT THE DIFFERENCE BETWEEN SOLO QUEUE AND PRE-MADES!? Goddamn are you retarded? Like seriously. I'm sorry if you are and we can end the debate now, but if you're not you're making a convincing argument that you are.

Actually, that exact thing can happen.

No. You get maybe one skipped rank. But if you're beating Plat 2 and you're Gold 4, technically you should skip to Plat 2... if rank means literally anything ever at all.

Such strategies should not be rewarded relative to ones that are better at actually winning games,

If they are trying to preserve stats, that must mean they did well for most of the game. If they're trying to preserve stats, that means they're at risk of losing them because their team performed poorly. If the team performed poorly, the game is lost anyway. Let them win and move on. Life isn't riding on this game and it's foolish to bank on the fact that the enemy will make a dumb move that will lose the game.

If you win at even a slightly over 50% rate over time in League you will steadily gain MMR.

Every win isn't worth 50 MMR. It's inconsistent. That inconsistency makes it hard to rise than it is to fall. You don't have to prove you're bad, you have to prove that you're good. That's how life works.

LP and rank are a pretty set of labels on top of the MMR system.

Rank and LP gain mean nothing. We should have a 1-N leaderboard system like CoD or something has that just uses MMR. It's would be a much more valid ranking system and much more honest. That's what you're saying.

Maybe I should say that I have no idea how this relates to the point you are trying to make.

Because we're talking about the average of all your games, numb-nuts.

Yeah, it almost always does.

Fucking site your sources or shut up. I explained how people like Dyrus, KEITHMCBRIEF, Blem, literally any other challenger at max MMR can have a win and loss streak in the same division and MMR. Where the fuck is your source?

Matthias9119Nov 5, 20141 votes

...and I think we're done here.

Matthias9119Nov 1, 20141 votes

Their early-game dominance might lead to pretty stats, but that doesn't mean anything if it doesn't actually win the game.

No, you have to look at the team's stats. A win or loss doesn't determine overall performance. It could have been a lucky play at Baron that won it or a simple mistake. But, again, if the Riven is like 10/3/7 and every other lane is like 2/6/0, the Riven tried to contribute...

You are assuming that having pretty KDA numbers, or "better" numbers than your teammates, means you deserve to win or be highly rated. This is not a convincing argument, and baking such assumptions into the rating system is a bad idea. The objective is to destroy the enemy's nexus, not to play TDM.

Could a player actually perform optimally but be held down by bad teammates or other random bad luck in any particular game? Sure. Just like they could actually perform badly but win by being carried by a strong team performance or due to something like a disconnect randomly crippling the enemy team. This is irrelevant in the long run because your good and bad luck tends to even out.

If that player performs well over time, they will win more than they lose if they are actually better at influencing the outcome of games than the players they are being matched up with currently. If they continually get a high KDA but still can't win >50% of their matches, this does not mean they are being held back by their teammates. It means that whatever they are doing in the matches is not effective at actually winning, no matter how pretty their stats look.

OR they have been extremely unlucky. But it is very unlikely to be extremely unlucky over a large number of games. Unless you want to weight individual performance much more heavily than actually winning/losing, being extremely unlucky will always hold you back in Ranked play.

Weighting individual performance very heavily is, to me, a very bad idea for the reasons I outlined before. It is very likely to punish and reward specific playstyles, and sometimes encourages players to sabotage their team's chances of winning in order to preserve or pad their own stats. You're also extremely likely to not rate creative or unusual playstyles properly, even if they are actually effective.

While the rules of the game itself may penalize certain decisions, that's just... how games work? Complaining that League "punishes" you for not having a dedicated jungler is like complaining that chess "punishes" you for carelessly losing your queen.

I'm not complaining that. I'm saying that it wouldn't change based on personal or W/L stat. Either way you want to use your "Queen" effectively. Either way, if you play a careless strategy, you're going to lose and be punished for it. The only difference is that you can't be as severely punished if your top laner TPs into 5 of them when your jungler and ADC are already dead for god knows why and thus you lose the game. It saves you from having to deal with other people's mistakes.

In a team game your ability to coordinate with your team and lead them to victory is part of the "skill" involved, IMO. There is no "you", you are part of a team and you win or lose as a team.

Yes, in solo queue you have to deal with the vagaries of random matchmaking. But so does everyone else, and short-term luck (good or bad!) does not have a significant impact on where you'll end up being rated in the long run.

I would be okay with Riot reducing all the LP and MMR gains/losses in matches that had an AFK player or an intentional feeder. However, to not break the ranking system you have to reduce both the gains and losses. This isn't too bad if there is something "unfair" from the very start of the match. But applying that mid-match due to a ragequit or someone on the losing team snapping and starting to intentionally feed feels unfair to the team that was actually winning hard -- then you're getting punished for someone on the other team behaving badly.

RaptameiNov 1, 20141 votes

No, randomness will NOT even out the effect of your team.

Say you play 9999 games, 500 of these had an afk on your team, 500 of these had an afk on the enemy team. You play another game and it has an afk on your team. Even though you had 5% enemy afks and 5.01% allied afks ("see, it's balanced!"), you're still one game in the hole.

The game does not compensate for a game thrown by a teammate by giving you a free win. It just moves you down by 20 LP and randomness resumes. You are less likely to get that 20 LP back than to not get that 20 LP back.

And that's why this game needs individual scoring.

And if it's not possible to do so because there are too many factors to consider, change the game to remove those factors. In a deathmatch game, KD is a 100% correct representation of your skill. And really, the classic moba format isn't THAT important if it prevents correct ranking.

FrikgeekNov 1, 20142 votes

Except that your idiotic theory assumes that both teams have an equal chance of having an afk/feeder/troll. You forgot that you're on your team too. If you don't afk or troll then the enemy team has 5 possible trolls while yours only has 4. Random factors don't work for or against you with enough repetition. If you're really better than others in your rating, you'll climb the ladder. This has been proven by countless streamers and smurfs.

the classic moba format isn't THAT important

You can't be serious. You can't just butcher a strategic team oriented game and boil it down to a simple deathmatch. Long term strategic choices are why people like mobas, it's not just about mechanics. If you want to play a competitive game with a high mechanical ceiling and only immediate decision making then play a 1v1 fighter.

Ed2CuteNov 2, 20141 votes

Streamers duo 95% of their smurf games until they're in high ELO. Two people who drastically out skill their competition will probably win more. Not to mention that League of Legends is immensely imbalanced and it's possible for two bruisers to 2v5 and win the game whether their team wants to or not.

FrikgeekNov 2, 20141 votes

Except that some of them don't duo queue and still stomp. Also, you really shouldn't be talking about balance if you're low elo, of course you're going to get wrecked by bruisers when you can't pull off proper kiting.

Ed2CuteNov 2, 20141 votes

They also lose a lot more frequently when they solo-queue. Yusui wouldn't have a 90% win ratio if he were just solo-queuing. This is a demonstrated fact. And pros do duo like 90% of smurf games until high ELO.

There are some you can't kite at different stages of the game. In a 1v1, bruisers pretty assuredly win against any other role. And assuming one's skill based on ELO isn't valid. That's kind of WHY this post is here. There are people who WANT to be Bronze because it's "ELO Heaven" for them. Or there are people who make the mistake of duo-ing with a friends who's really not that good, afk'd, whatever and hurt their ranking. There are so many circumstances that make the current ELO system invalid.

OnegarionNov 3, 20141 votes

So why don't you duo then? Would make sense that if they duo and do well that you could too. While they duo usually its with players who are lower than normal. I've seen 2 diamond smurfs duoing but thats the strong majority in my cases. The strongest cases are people duoing with a friend through it just like others. ITs not against the rules or showing anything bad when they duo with a friend. If its taht big a deal or how theymove up do it yourself or its not a valid argument.

Ed2CuteNov 3, 20141 votes

I try to. But I never boasted about being Challenger skill level. And I don't have any Diamond - Challenger friends with smurfs to duo with. My friends normally range around the Platinum levels and I duo with them when I can. But not everyone always wants to duo. And we're not 100% to win (though I win most of them), and my solo-queue is definitely lower than a 100% win chance because of my limited influence on the game.

Also, please don't use any recent games as a display of my skill. I've only been using QWER for like 3 days-ish now and I need to still get used to it. I play like 4 games a day or some shit. Or ARAMs which, let's be honest, diminish your skill level compared to a 5v5 environment. Before I was just using the buttons on the Razer Naga.