EPTYou don't control some stuff, true. But neither does any of your teammates or opponents. Your degree of lack-of-control in the case is equal, meaning the rating is unbiased. If you don't understand this, you're either an idiot, a troll or a very lazy person refusing to read the wikies on simple things.
[Patch 1.0.0.135: Fiora] General Discussion - Page 227
| Forum Index > LoL General |
Same rules apply, per usual. Please use the appropriate threads (QQ, Brag, Champion, etc) whenever appropriate. Keep the resident Banling content. Thanks. Happy Gaming. | ||
|
BluzMan
Russian Federation4235 Posts
You don't control some stuff, true. But neither does any of your teammates or opponents. Your degree of lack-of-control in the case is equal, meaning the rating is unbiased. If you don't understand this, you're either an idiot, a troll or a very lazy person refusing to read the wikies on simple things. | ||
|
robertdinh
803 Posts
On March 19 2012 23:21 TheYango wrote: OK, so wait. You're saying that over the 2 possible teams that someone could get slotted into, random mishaps are NOT equally likely to happen to a player on blue team as one on purple team? Please tell me which team you think has a higher % chance of mishaps so I can queue dodge the next time I get on that color team. It isn't about probability, it is about reality, and the reality of it is some bad players may get lucky, and some good players may get unlucky, and the games they are in will have skewed results, and those results will ripple out and skew other results and so on and so forth. On March 19 2012 23:23 BluzMan wrote: A statistical system that has any hopes of being objective can only draw conclusions from individual player's winrate, regardless of whether he performs in teams or solo, only consistency in that matter is needed (which is true for LoL since for any given ELO score, you only play one game mode). It's math and it's been discussed over and over. Introducing any other variable into the picture (your kills, farm and whatever other shit) means introducing bias. You don't control some stuff, true. But neither does any of your teammates or opponents. Your degree of lack-of-control in the case is equal, meaning the rating is unbiased. If you don't understand this, you're either an idiot, a troll or a very lazy person refusing to read the wikies on simple things. If you don't understand that the degree of lack-of-control is not equal, then you are either an idiot, a troll or a very lazy person. No one is completely equally skilled, thus their degree of control will never be equal. Saintvicious may be equally rated to chauster at one point, do they have the same amount of control over the game? The system rates them near the same at that point, but do their roles matter? Does the current meta matter? Does the current balance matter? What about the people they get teamed with or against? Does that matter? If they play on opposite teams and one gets grouped with personalities or playstyles that mesh better with him than the other? Did they have an equal amount of control? If one happens to be grouped with someone that dislikes him and gets trolled, did they have the same amount of control? As I said before, you have not even begun to consider the impact that the variables have on the game. You are trying to claim accuracy in a system where you do not even weigh the variables, you just assume it will all work out based on probability, but probability is not accurate here because that "probability" does not consider variables that are constantly changing. Someone solo queueing at 1500 on a monday at 2am, may have much different variables and probabilities for similar variables than someone queueing at 2pm on a tuesday. | ||
|
Twinmold
Sweden238 Posts
On March 19 2012 23:23 Two_DoWn wrote: He has made it pretty damn clear that its always your team. This entire train of thought is pointless. Lets talk about flowers instead. I like flowers, provided they dont give me allergies. Lulu uses lots of flower particles, right? I sure hope so  | ||
|
ManyCookies
1164 Posts
On March 19 2012 23:23 BluzMan wrote:\ You don't control some stuff, true. But neither does any of your teammates or opponents. Your degree of lack-of-control in the case is equal, meaning the rating is unbiased. If you don't understand this, you're either an idiot, a troll or a very lazy person refusing to read the wikies on simple things. Ah, the classic "If you don't agree with me, you're an idiot". Never fails. In seriousness, the statistical variation is going to be about 0 in the long term. However, short term (~150 games) the variation will be a significant enough factor that a sizable percentage of players will be allocated significantly below their ELO (and a roughly equal percentage will be allocated significantly higher than their ELO). For example, let's assume that a 1400 average ELO level player has a 55% Win-Loss ratio at 1200 average ELO, and then plays 100 games. This implies that after 100 games, the player is most likely going to have a +10 win/loss ratio, which is approximately the 200 ELO difference between the two levels. Using Binomial Distribution, we can determine the probability of each win-loss configuration (i.e 55 wins 45 losses, 54 wins 46 losses, etc.). At 100 games, she has about a 54% chance of having 55 wins or more, accurately reflecting her skill level. But she also has an 18% chance of having 50 wins or less, trapping her at that ELO level (or potentially falling down). With thousands of players at that ELO, 18% is a significant amount of "trapped" players. I don't know what you or Riot would consider an acceptable confidence interval, but about 5% (~two SD) of those players will be "trapped" at 250 games. | ||
|
red_
United States8474 Posts
On March 19 2012 23:23 robertdinh wrote: No one is completely equally skilled, thus their degree of control will never be equal. So would you say we attach a number to this difference in degree of control over a person's ability to win any given game, and call it Elo? Sounds good, do it up. | ||
|
robertdinh
803 Posts
On March 19 2012 23:27 red_ wrote: So would you say we attach a number to this difference in degree of control over a person's ability to win any given game, and call it Elo? Sounds good, do it up. Oh except that's not what elo attempts to do, elo attempts to establish a rating at which you go 50% in your games, but doesn't actually consider how much each player you get pitted against or put on a team wish actually impacts the game. | ||
TheYango
United States47024 Posts
On March 19 2012 23:23 robertdinh wrote: It isn't about probability, it is about reality, and the reality of it is some bad players may get lucky, and some good players may get unlucky, and the games they are in will have skewed results, and those results will ripple out and skew other results and so on and so forth. No, because the system is self-correcting. If a player placed correctly at his Elo loses 5 games in a row due to DCs, he will be placed with players that are worse than him. Which means that of the games that are not auto-lost due to DCs/trolls/leavers, he will win HIGHER than 50% of the games because he is a better player than the other 9 players in the game. If leavers/trolls/DCs happen equally often to both teams, this means that he will shift back toward his correct Elo. Likewise a player that gets lucky will shift downward toward his correct Elo because he will win less than 50% of those games at the Elo that he lucked out on. This self-correcting tendency becomes stronger the bigger the disparity between current and true Elo is--exemplified most strongly when watching Saintvicious play at 0 Elo, where the self-correcting power of the skill disparity is so strong that it even overcomes leavers. The only way to maintain a misplaced Elo over a large sample size of games is to repeatedly and consistently get lucky or unlucky, to a statistically improbable degree. | ||
|
robertdinh
803 Posts
On March 19 2012 23:30 TheYango wrote: No, because the system is self-correcting. If a player placed correctly at his Elo loses 5 games in a row due to DCs, he will be placed with players that are worse than him. Which means that of the games that are not auto-lost due to DCs/trolls/leavers, he will win HIGHER than 50% of the games because he is a better player than the other 9 players in the game. If leavers/trolls/DCs happen equally often to both teams, this means that he will shift back toward his correct Elo. Likewise a player that gets lucky will shift downward toward his correct Elo because he will win less than 50% of those games at the Elo that he lucked out on. This self-correcting tendency becomes stronger the bigger the disparity between current and true Elo is--exemplified most strongly when watching Saintvicious play at 0 Elo, where the self-correcting power of the skill disparity is so strong that it even overcomes leavers. The only way to maintain a misplaced Elo over a large sample size of games is to repeatedly and consistently get lucky or unlucky, to a statistically improbable degree. Your self correcting system doesn't even consider a fraction of the variables that go into deciding the outcome of a game of LoL. You still seem to assume that leavers trolls and dcs happen equally to both teams, they don't. They are random, and you don't play with the same people under the same circumstances over and over. It could happen 90% of the time to one player, and 40% of the time to another player. But players involved in the random elements will move on and play other games, skewing those results as well, until it all ripples out. | ||
|
ManyCookies
1164 Posts
And yes, simplifying to a binomial win-rate is valid for this discussion; accounting for every variable is uneccesary for this discussion. | ||
|
Shanba
Scotland144 Posts
Incidentally weight of games isn't going to make you any more likely to reach your "true" elo - it's not one of those things where you take repeated measurements and thus get a more accurate estimate of the true underlying value. It's more like a random walk, where the probabilities change as you get further from the true value, making it more likely you win as you go lower and lose as you go higher. So 500 games after you first reach your true elo and 600 games after you first reach your true elo you're equally likely to have returned there. Also possible (even likely) that after many many games you'll have had streaks of bad luck and streaks of good luck such that your bottom and top elos are very likely to be some distance apart - and on a smaller scale, you'll probably bounce between just above and just below your true elo on a fairly regular basis. It's worth noting that yes, it is possible that if the randomness were large enough, that you wouldnt be able to determine your own contribution. Imagine a game where even the best player in the world could only raise his win rate to 51% and the rest was determined randomly. Obviously it would be nigh impossible to rank such a player. The proof that the elo system works, that the randomness is small enough such that the individual contribution of players can be reasonably measured, is that numerous top elo players have worked their way back up the ladder over and over - whether its those people who did the whole "queue dodge to 0 to show that elo hell doesnt exist" or just the pro players leveling smurfs (like double triple and pentalift all being in the top 10 ranked at one point.) Also it's fairly likely that at some point in time some of the millions of players of league of legends have been a long way out of their correct elo bracket for reasons outside of their control. But given how bad people are at analysing their own play, the myriad cognitive biases which cause people to, among other things, rationalise their defeats as not their fault, only remember the games they played well in and lost while forgetting the ones where they played badly and so on, it's far more likely that you just aren't as good as you think you are. At the end of the day it is INFINITELY more useful to concentrate on your own mistakes and your own gameplay than worrying about the contributions of your team - which are almost entirely out of your control. You can't make the matchmaking system give you better teammates and give your opponents worse ones, but you can make sure that you are playing to the best of your ability. Most of the people I play with who complain about elo hell in solo queue or whatever are just looking to shift the blame for their own shitty play. Don't be that guy. | ||
|
robertdinh
803 Posts
On March 19 2012 23:36 ManyCookies wrote: Well, he is right. Although there's still natural variance for teammates and such, the overall win rate will generally go up as ELO drops. The extent of this advantage is questionable. And yes, simplifying to a binomial win-rate is valid for this discussion; accounting for every variable is uneccesary for this discussion. I'm sorry but accounting for variables when talking about the accuracy of a system will always be necessary. On March 19 2012 23:36 Shanba wrote: Elo hell discussions are so silly. Sure there's this whole glut of randomness surrounding all the other peopl in the two teams, but at the end of the day there's only a single constant in all the games and that's you - and the randomness is equally likely to be against you as it is to be for you, so yah. Incidentally weight of games isn't going to make you any more likely to reach your "true" elo - it's not one of those things where you take repeated measurements and thus get a more accurate estimate of the true underlying value. It's more like a random walk, where the probabilities change as you get further from the true value, making it more likely you win as you go lower and lose as you go higher. So 500 games after you first reach your true elo and 600 games after you first reach your true elo you're equally likely to have returned there. Also possible (even likely) that after many many games you'll have had streaks of bad luck and streaks of good luck such that your bottom and top elos are very likely to be some distance apart - and on a smaller scale, you'll probably bounce between just above and just below your true elo on a fairly regular basis. It's worth noting that yes, it is possible that if the randomness were large enough, that you wouldnt be able to determine your own contribution. Imagine a game where even the best player in the world could only raise his win rate to 51% and the rest was determined randomly. Obviously it would be nigh impossible to rank such a player. The proof that the elo system works, that the randomness is small enough such that the individual contribution of players can be reasonably measured, is that numerous top elo players have worked their way back up the ladder over and over - whether its those people who did the whole "queue dodge to 0 to show that elo hell doesnt exist" or just the pro players leveling smurfs (like double triple and pentalift all being in the top 10 ranked at one point.) Also it's fairly likely that at some point in time some of the millions of players of league of legends have been a long way out of their correct elo bracket for reasons outside of their control. But given how bad people are at analysing their own play, the myriad cognitive biases which cause people to, among other things, rationalise their defeats as not their fault, only remember the games they played well in and lost while forgetting the ones where they played badly and so on, it's far more likely that you just aren't as good as you think you are. At the end of the day it is INFINITELY more useful to concentrate on your own mistakes and your own gameplay than worrying about the contributions of your team - which are almost entirely out of your control. You can't make the matchmaking system give you better teammates and give your opponents worse ones, but you can make sure that you are playing to the best of your ability. Most of the people I play with who complain about elo hell in solo queue or whatever are just looking to shift the blame for their own shitty play. Don't be that guy. Your proof regarding top players does not consider the variable that they have reputations and clout. They have more control even in champ select than an anonymous player. | ||
|
Two_DoWn
United States13684 Posts
| ||
|
red_
United States8474 Posts
Edit: T_D, I like the cut of your gib. You've got moxy kid. | ||
|
mr_tolkien
France8631 Posts
On March 19 2012 23:16 red_ wrote: robertdinh you seem to have missed the opportunity to introduce me to these 'really great players' stuck at 1400. I was aware that they are obviously not famous because of Elo Hell, and was giving you the chance to shine light upon them, get them known, so that they can no longer live their plight. So, if you would please, I am awaiting the information of these players. <- still stuck on that one personnaly. Waiting for the names so I can become a hipster fan ! EDIT : On another note, I love magniolias. They're magnificient. | ||
|
nojitosunrise
United States6188 Posts
On March 19 2012 23:37 robertdinh wrote: Your proof regarding top players does not consider the variable that they have reputations and clout. They have more control even in champ select than an anonymous player. You do realize many 2k+ elo players have unknown smurfs right? The top players will eventually reach the top of the ladder....regardless of the time needed to get there. Elo hell can't exist if that happens. | ||
|
billy5000
United States865 Posts
what do you guys think of the lifesteal quints? I figure with those, 3% from mastery, and 6% from 2 dblades, you pretty much have a wriggles without the constant wards, armor, and a proc that's usually useless in lane (unless maybe you're trying to push your lane hard as fiora). I'm thinking about buying them with wriggles nerf coming up :\ | ||
|
robertdinh
803 Posts
On March 19 2012 23:40 mr_tolkien wrote: <- still stuck on that one personnaly. Waiting for the names so I can become a hipster fan ! Do you think I would expose great players to scrutiny just to satisfy the curiosity of players who don't consider all of the variables and probably use solo queue elo as a measuring tool and consider it more accurate than it actually is? On March 19 2012 23:40 nojitosunrise wrote: You do realize many 2k+ elo players have unknown smurfs right? The top players will eventually reach the top of the ladder....regardless of the time needed to get there. Elo hell can't exist if that happens. Really? Because it seems like every time they "smurf" they actually tell people in game who they are and to let them have whatever champion they want, or word gets out because they are actually streaming it and not smurfing at all. Show me a 2k elo player that played a smurf account in complete anonymity and got back to the same elo in nearly the same amount of games played that it took his main account. | ||
|
ManyCookies
1164 Posts
On March 19 2012 23:37 robertdinh wrote: I'm sorry but accounting for variables when talking about the accuracy of a system will always be necessary. Let me rephrase that. For the purposes of this discussion, we can assume stuff like AFKs, ragers, leavers, feeders etc. are effectively random and can be represented as a binomial win-loss rate. The same patterns/variance will still emerge using a binomial system and drastically simplifies the problem. Err, if you're polling if people like Chocolate Ice Cream, you don't need to know what flavor the people who don't like Chocolate Ice Cream like. The information is there, but you just need to know if they like Chocolate Ice Cream or not. FUCK YEAH, SIMPLE COMPARISON THAT ADEQUATELY COVERS THE POINT. | ||
|
Two_DoWn
United States13684 Posts
| ||
|
ManyCookies
1164 Posts
| ||
| ||
