Stuck In Elo Hell: Part One

Avatar
This should be a mandatory read, and not only for LoL players.
Avatar
Good league article. Now to make it happen...
Avatar
You might say common myth busted, but from my personal experience the frequency of satisfying games between when i was 600-900 Elo and now that I'm over 1000 has jumped dramatically. Still get trolls or raging players, but the frequency has gone from every other game to where I can actually play a whole night without having a problem (if i'm lucky heh but at least it's possible).Now some of my teammates/opponents would still have bad games like we all do, and yes still some early surrender spamming guys but they dont start out doing w/e they can to make their respective team lose. And that improves the enjoyment of playing regardless of elo outcome.
Avatar
I understand how you feel and there are two reasons for this.First, skill is relative. You may feel that games are more satisfying because you are playing better than you did before. When you were 600-900, you were playing at a 600-900 skill level. That means 1000 elo players seemed good and 500 elo players seemed terrible. Now that you are 1000, your range has been lifted. You will consider 1200 players to be good and 800 elo players to be terrible. You can say in hindsight that you could feel "the difference" in skill, trolls, feeders, etc. but in reality, that's just you getting better. A 1400 player might be someone you aspire to be. To me, he'll be nothing more than a feeder. Skill is relative.Second, I can tell you that what you believe is the exact opposite. You actually have more ragers, trolls, and feeders the higher you go. The closer you get to silver, gold, platinum, and diamond, the bigger assholes you will get. They will rage if you don't pick the right champions. They'll literally afk and make for 13 minute ranked games. The number of feeders and trolls will never disappear with elo and that was the point I was making. People believe if they get into higher elo, they'll have a better time. They won't. It doesn't go away.I was trying to save this for my next article but I thought I'd respond. Maybe I'll include it anyways
Avatar
Warning: effortpost!I'd like to add some insights from some experiments I did to understand a bit more about the fact and myth of Elo hell. My day job involves a lot of statistics and mathematical models, so for fun I ran some simulations using LoL's Elo framework. Everything in the OP's article is true, but it's important to add some depth to some of his statements (which have been stated by others before).The first caution is that the realities of the Elo system for those with high Elo (and very low Elo) is different than the majority of us, so their perceptions are naturally a little different. In every simulation I performed those with Elo's in the upper tail of the "skill distribution" rapidly rise very close to their true Elo and stay there with little variability (and the same is true for those that are extremely unskilled). Let's parse the following:"Based on Elo Hell assumptions, the enemy team will have 5 feeders, but your team will only have 4. This doesn't mean you'll win every game. It just means over a long period of time, you will eventually have a winning percentage."True. But what exactly is a "long period of time"? In my tests, if you're in the skill range encompassing 90% of players, it's a very, very long period of time. It's on the order of 1000s of games, far more than a player can play in a single season. It is very common to see simulated players with a true Elo of 900 spend hundreds of games at 1400 Elo, and vice versa. Granted, given infinite time, your average Elo is your true Elo, but in the relatively small slice of time that a season contains, you can't know if you're on an overrated or underrated streak. In statistical terms, the variance of your true Elo from your actual Elo is incredibly large, and this trend diminishes the more extreme your true Elo is. If you are a perfectly average player, the variance is so high that I'd argue your Elo is essentially meaningless. One has to rise to levels of about 1600-1800 Elo before there is any reasonable certainty that you're "above average" and that's about as specific as you can state it."If you have 200 ranked games and you're stuck in 1100 elo, maybe it's time to take a deep breath and look at yourself in the mirror."And this is the sort of statement misconceptions about Elo lead to. This statement needs an additional phrase: "look at yourself in the mirror and realize you're not 2200 Elo". A person stuck at 1100 Elo for 200 games could easily have a true Elo of 1600 and never see it. All we really know is that he or she is not a top player, nor one of the worst players.In my opinion, if Riot continues to use the Elo system as it's currently implemented and you're not a >1800 Elo player, just have fun and pay no attention to it as it's effectively meaningless. You're not at a professional level of skill, and that's about all you can say. The variance just dwarfs any incremental improvements in skill until you start to reach those upper echelons.What can Riot due to improve the system? In my opinion, the problem stems from the matchmaking system. In an attempt to create two teams with roughly even Elo comprised of roughly evenly skilled players it grinds the separation of differently skilled players to a halt. A roughly analogous situation occurs quite often In the machine learning realm, where the number of rounds of simulation is so high that decent estimates of the true parameters of a model cannot be done in a reasonable timeframe. One way to address it is to allow extreme departures from current estimates to be put to the test. This could be achieved by simply relaxing the matchmaking algorithm to encompass a larger range of Elo around your current estimated Elo (say +/-90% confidence). The noticeable difference is that the Elo bump or loss for a single game can be much larger from time to time. On average it will still be +/-12 or so, but from time to time you might play in a game where +/-50 Elo is at stake. In my simulations, this type of strategy drastically decreases the variance between true and estimated Elo (one spends more time very close to their true Elo), and you arrive at this state much more quickly.So does Elo hell exist? Sort of. The Elo system is a zero-sum game. For every person in "Elo hell", there's one playing way above their true skill level. If you play enough games you will spend equal time in both of these positions.All the other advice in the article is excellent, because the most important thing is to have fun, improve, and be a positive member of the LoL community. I only differ in the notion that most of us can use Elo, as it's currently implemented, as any sort of accurate measure of our skill level or progress.Good luck Summoners!P.S. I didn't want to get into the effect of the 1200 Elo "starting rank" that first-time Ranked players begin at. Maybe I'll post about this in more depth another time. Briefly, my simulations actually show pretty clearly why Riot chooses this particular number. However, there is a detrimental effect (especially to those who have a true Elo of 1200) as it inflates the variance I talk about above even further. Unfortunately, there is no easy solution.
Avatar
well as the article is all about the reader changing the power is yours!
Avatar
I have no frigging idea why this automated matchmaking system was implemented in the first place. When I was playing DotA 1, Gunbound, and Gunz, the matchmaking was lobby based. I.e. more or less random. And now these bastards from Valve and Riot think they can do better. Now that I moved on to Dota 2 every single game is a fucking stomp either by my team or the enemy team.
Avatar
There lies the problem :D
Avatar
Effort post, but completely devoid of any supporting evidence and flies in the face of statistical reality. What "model" did you use? What elo system did you use (since Riot's is not disclosed)?Riot state that their elo system is 98% accurate after ~200 games -- what evidence other than "my simulations" can you offer? You basically /can't/ have the information you're claiming you have, as you neither have Riot's raw data nor an accurate model to work from.I'm calling a massive pile of BS on this until you show something for it. All it is at the moment, is a very wordy "elo doesn't matter".n.b. >1800 elo is essentially top 1% -- if you had any in-depth knowledge of statistics you'd know how silly it is to claim you can only accurately measure the outliers.Terrible post.
Avatar
I think the greater point, is that the concept of "The only reason I am not 2k elo is other people" is demonstrably untrue. I can name you a few players at 2.5k elo who have been banned for terrible behaviour; it's simply less common, not absent.
Avatar
The equations are available for anyone to read, which is what makes your post remarkably stupid; one can simply read Wikipedia to realise you're full of it (I'm guessing you're suffering a bout of Dunning-Kruger, and can't see from the evidence presented that you're wrong).Specifically, the Chess Elo system uses a K-value which is varied based on matches played and current Elo, but is otherwise static. League uses a fluctuating k-value weighted by predicted outcome which acts as a modifier in a **manner not documented by Riot anywhere** (as indeed both secret and patented) so you cannot have a "model" which "proves" your theory. If you want to prove the Riot model to be inaccurate, doing so would require experimental evidence, not theoretical as you have nothing to test. This is math 101, "math guy."It also uses a different distribution model based on team games, not the 1v1 Elo system deployed and documented elsewhere.I'll also point out, again, that the Elo ranking system is more accurate when NOT looking at outliers. This is also well documented. Christ man, didn't you even Google "elo ranking system" before writing nearly 1000 words of gibberish?If we assume that the LoL system weights players equally (not known) then a player with an elo of around 400 higher than the team average will have a ~75% winrate, meaning he will climb ~7 elo per game - meaning he will half the difference between him and his ~true elo~ in around ~30 games, and will curve off to close 75% of the gap in ~50 games.Where then, are you imagining standard deviations 4x higher than one actually sees?I could go on and :effort: further but no matter what I say, you'll simply say "I know I am right" (like you just did above) and you will never, ever be able to show a single thing for it because it's nothing more than "elo doesn't matter because it's not accurate" -- something that is demonstrably true in every *real* model of elo you could care to run. This isn't about "believing Riot" it's about believing in the application of a well known and understood equation.What do they teach a "math guy" these days, if not to show their working?
Avatar
Thanks for your feedback. I understand some of your assumptions and concerns such as "how long is a long period of time?" For the sake of the article, I didn't want to give a specific number since elo itself hosts a wide range. The article was meant for the reader to reflect upon oneself.Now as to your other points "A person stuck at 1100 Elo for 200 games could easily have a true Elo of 1600 and never see it", that is exactly the point of the article. I'm not sure if you understood what I was trying to say because I wasn't just babbling for paragraphs on end. If you're capable of 1600, but you're 1100 after 300 games, then you need to able to recognize this by looking at yourself. You have to tell yourself, "I'm not 1600 yet, maybe I'm doing something wrong." You have to be able to point out your mistakes instead of blaming matchmaking for your 1100 elo.Which brings me to my final point: Blaming the matchmaking system isn't going to make you better. You are ultimately blaming the elo system and the matchmaking system Riot has implemented. It doesn't matter what you think about their system and it doesn't matter how much you complain about it. You can't change that. The one thing you can change is yourself.No matter what your elo is, you should stop pointing fingers at other obstacles like bad teammates, Riot games, the matchmaking system, etc. If you want to get better or rise in elo, you need to absorb the blame and look at yourself. Thank you for the quality post, but it makes me wonder if you are starting out with the right attitude
Avatar
I'm happy to have a civil discussion with you, I'm not sure how the hostility is helpful.In my model, I used a static K-value of 25 which is consistent with the Elo change you see in ranked games (while this is documented, it's easy to determine empirically). I included the additional buff to K for the first 10 games played to account for the "placement" phase. I did not include any additional "fluctuations" that you speak of, but I've never seen any evidence in ranked games that this actually occurs other than during the placement phase. If you could be more specific.The player generator creates a population of players with a true Elo based on a Gaussian distribution, where the mean Elo is 1105 with a standard deviation of 260. This creates true Elo values that are consistent with the percentile information that has been released (e.g. 1500 Elo is the top 5%). I concede the true Elo distribution is probably not Gaussian, but my parameter values are very close, and changes to the parameters or the use of other distributions don't affect the conclusions.After 200 games played, the simulation shows that the 95% confidence interval is about +/- 350 for those who have a true Elo of 1200. This is not at all consistent with your quoted value of 98% accurate (which is not meaningful since you didn't include a confidence interval - 98% accurate within what level of error?). In addition, it's unclear to me how Riot would actually be able to determine this, since the "true Elo" of a real player is actually never known. Simulated players are the only way to make a measurement. If they've released results of their own simulations, please point me to them, as I would love to see them.In any event, my basic assertion still stands. If a player with a true Elo of 1200 will have an under- or over-estimated Elo of below 850 or above 1550 10% of the time, is it really worth putting a lot of stock into?I'm interested in hearing specifics on how Riot's "proprietary implementation" of the Elo system can overcome this inherent variance. Even a theoretical example of how this could be achieved would interest me. I certainly can't think of any. Your use of the term outlier in this context is also a bit strange, and I don't understand exactly what you're arguing. I merely stated that those at the extreme ends of Elo population distribution (i.e. the "pros") suffer less variance in their estimated Elo. An outlier usually describes an extreme measurement, not known values. There are indeed "outlier" measurements (I'm looking at a simulated 1338 Elo player rated 2009 Elo at this moment), but I never made any mention of this type of extreme in my original text or what I've written here.
Avatar
We totally agree on how a player should improve and behave. We also agree that Elo hell is not the result of feeders and trolls. I'm just making the argument that using Elo as a measure of your progress isn't a very good idea. Yes, it's an objective measure; but the variance is so high for the vast majority of players that it's hard to separate bad luck and poor skill on an individual level. So I suppose we differ on what Elo hell is the result of. While you imply it's from an actual lack of good play and improving will get you out, I'm arguing it's an inherent quality of the Elo rating system itself and improved play may or may not get you out.An interesting aside is the presence of "Elo heaven", which no one seems to ever complain about :P. Anyway, I think the LoL community seems to have this obsession with Elo that really isn't warranted except for the highest ranked players where the rating system works fairly well. Just play your best!
Avatar
~500 more words where you have still failed to provide:- A method- Resultsand have stuck to repeating your conclusions.I'll instead nitpick."In my model, I used a static K-value of 25"This is incorrect." I included the additional buff to K for the first 10 games played to account for the "placement" phase"This is incorrect." I did not include any additional "fluctuations" that you speak of, but I've never seen any evidence in ranked games that this actually occurs"K-value is non-static and determined per team, per game, based on Riot's model of your percentage chance to win. So, incorrect."The player generator creates a population of players with a true Elo based on a Gaussian distribution"This is widely considered a bad idea when ranking players."After 200 games played, the simulation shows that the 95% confidence interval is about +/- 350 for those who have a true Elo of 1200"This is inconsistent with every known model of elo (much, much too high)."This is not at all consistent with your quoted value of 98% accurate (which is not meaningful since you didn't include a confidence interval - 98% accurate within what level of error?)"It's the confidence in the predicted result (that is to say, the binary of win-lose per game) the model deploys."In addition, it's unclear to me how Riot would actually be able to determine this, since the "true Elo" of a real player is actually never known"Sure it is, it's the elo a player achieves when he fails to statistically differ from where the model places him - it will always have a confidence, but it's double-digit."In any event, my basic assertion still stands. If a player with a true Elo of 1200 will have an under- or over-estimated Elo of below 850 or above 1550 10% of the time, is it really worth putting a lot of stock into?"Your assertion doesn't stand, as you have failed to show this - you're just restating your conclusions as though they are evidence.I'm stopping here, as it's plainly clear you have nothing to actually show, since even a screenshot of your "model" would shut me up -- yet you're unable to show your method or results.
Avatar
Well, you've pretty clearly been trolling, but just in case anyone reading hasn't figured it out yet, let's talk about this one last time.>- A methodWhat details do you need? If you wanted to reproduce my work I've provided all the values needed to do so. I mean, it's a simulation using Elo and I've provided the specific values to plug in. If that methodology is foreign to you, any standard text on machine learning will contain the necessary background."In my model, I used a static K-value of 25">This is incorrect.No, it's correct. Typical ranked match results in 12-13 Elo gain or loss and since matchmaking tries to create two teams with a 50:50% chance of winning we can deduce it's K=25. But here's a source just to put it to rest: http://leagueoflegends.wikia.c... ("It appears that League of Legends uses a similar system of changing K values: K appears to start around 100, eventually leveling out to about 25.")." I included the additional buff to K for the first 10 games played to account for the "placement" phase">This is incorrect.See above." I did not include any additional "fluctuations" that you speak of, but I've never seen any evidence in ranked games that this actually occurs">K-value is non-static and determined per team, per game, based on Riot's model of your percentage chance to win. So, incorrect.Percentage to win has nothing to do with K. Percentage to win arises from the player's Elo. K, the expected outcome, and the actual outcome of the match are used to determine how your Elo should change as a result. K is static (except for the placement phase), the expected outcome is not (although the matchmaking algorithm attempts to find a composition where the expectation is 50% win/loss)."The player generator creates a population of players with a true Elo based on a Gaussian distribution">This is widely considered a bad idea when ranking players.Source?"After 200 games played, the simulation shows that the 95% confidence interval is about +/- 350 for those who have a true Elo of 1200">This is inconsistent with every known model of elo (much, much too high).Source? I have a source for you. Current, ongoing discussion on Reign of Gaming about Elo (http://www.reignofgaming.net/b.... The author of that article says standard deviation is appx. 200. My simulation says s.d. is 238, and from an Elo cross section with a higher variance. So very consistent between the two of us."This is not at all consistent with your quoted value of 98% accurate (which is not meaningful since you didn't include a confidence interval - 98% accurate within what level of error?)">It's the confidence in the predicted result (that is to say, the binary of win-lose per game) the model deploys.Gibberish."In addition, it's unclear to me how Riot would actually be able to determine this, since the "true Elo" of a real player is actually never known">Sure it is, it's the elo a player achieves when he fails to statistically differ from where the model places him - it will always have a confidence, but it's double-digit.More gibberish. True Elo cannot be "known", only estimated with a certain degree of confidence. That's exactly what my simulation is doing; it's determining what that degree of confidence is. For a player with a true Elo of 1200, it's +/-350 (95% confidence)."In any event, my basic assertion still stands. If a player with a true Elo of 1200 will have an under- or over-estimated Elo of below 850 or above 1550 10% of the time, is it really worth putting a lot of stock into?">Your assertion doesn't stand, as you have failed to show this - you're just restating your conclusions as though they are evidence.Well, like any experiment, you have no choice but to assume the person isn't just making up results. I've stated the method, and the results I obtained. So feel free to reproduce it and see if you get the same result.>I'm stopping here, as it's plainly clear you have nothing to actually show, since even a screenshot of your "model" would shut me up -- yet you're unable to show your method or results.Well it's impossible to show a screenshot of the model. I don't think the word means what you think it means. I could show a screenshot of the _results_. It's a plot of estimated vs true Elo. I'm not sure how this would shut you up. It shows exactly what I've stated in a visual form. The s.d. at 1200 Elo is 238, and gets smaller at high and low Elo. Looks sort of like an elongated ellipse along the diagonal. What you should really be asking for is the several megabytes of raw output. Probably just easier to run a simulation yourself?Anyway, it's been fun talking to you. Trolls can be helpful. Your psuedo-intellectual blabbering led me to find that RoG article confirming my results (and my conclusions). GG, Summoner.
Avatar
See, this is where you unravel. I'm now quite sure you HAVE tried to do some math, or at least research, on this (however flawed it might be) but you're making such an effort to snort derisively, that any message is lost.See, it hardly matters whether the SD is 80 or 180 -- the law of large numbers here is going to show that the lack of precision in the rating is not going to cause any particular bias, and Riot's internals suggest that after a triple digit amount of games, the rating system is accurate unto itself.What you, in fact, seem to be suggesting is "Elo rankings, when used to evaluate team games, are meaningless as the variance prevents an Elo figure accurately reflecting individual skill" (a long worded, 'Elo doesn't matter').Now I am in fact, open to that idea in a theoretical sense, but the problem is, this is original research. Original research that you refuse to elaborate on beyond "I know what I know." This is poor play.When I say "show your model" what I am asking for is the method you used to perform this. You know, what I would need in order to do this myself. Or anyone else who is interested. This, you have thus far failed to provide.I'm assuming your method is:1 - Create players based on a normal distribution2 - Take 10 and sort them into two teams of 53 - Average the elo per team4 - Determine the winner5 - Apply the K value to shift player Elo's6 - Repeat at 2.This would be deeply flawed, and you haven't even given enough information to know what you may have done here, trivially:1 - how many? Did you create a population or sample? This is important.2 - how? Riot's matchmaking is not documented. The variance per team, and between team, is crucial3 - This isn't what Riot does, so how did you vary or control this?4 - How?5 - How did you determine the K value, per game? (Remembering Riot's isn't static) Presumably this feeds back into 3) - so how many games were played at which K-variance? Perhaps you determined a relationship between K, and the number of games needed to get an accurate elo?"Well, like any experiment, you have no choice but to assume the person isn't just making up results"No, since one can *repeat* the experiment if the method is known."I've stated the method"No, you haven't.At this point I will remind you of the central tenet of your "simulation"> In my tests, if you're in the skill range encompassing 90% of players, it's a very, very long >period of time. It's on the order of 1000s of games, far more than a player can play in a >single season. It is very common to see simulated players with a true Elo of 900 spend >hundreds of games at 1400 Elo, and vice versaThis is what you need to show, wherin you state that elo will never matter for 90% of players because it won't be representative of their play. Interestingly, this is also rebuked by the article you linked which "proved me wrong".I would be interested in looking at your method, perhaps then there's a discussion to be had.I pretty much expect more trolling and "I know because I know" and exactly zero information on what you supposedly did, and wild claims for why you can't simply copy and paste the functions that made up your model.
Avatar
"Well, come on, I'm not going to cut-and-paste 250 lines of C++ code on a comment section ffs. Perhaps themittani.com will let me present a complete article where I can show graphs, pretty pictures and equations. I will look into this. However, like I said, this is not a particularly complex model. It would probably qualify as a homework exercise in an upper-level undergraduate machine learning course. Any other little details in implementation aren't going to affect the outcome - just do it. If you get a smaller variance than I reported, by all means tell me what assumptions you made to achieve this. To be frank, your complaints about this just seem to imply you don't have any working knowledge in this area. If you did, you'd have opened up your text editor, coded this up in an hour, and have run it."Yeah, pretty much as I said it would go - you can't show what you know, and will forever create new and exciting reasons as to why you can't simply paste a full method, beyond "my model" "my simulation" and "in my tests". A highschooler might be impressed away by this, but I see it for what it is.You haven't adequately given a method, still. Simply saying "The simulation does this, honest!" and going on to state because it performs this unseen task, the unseen results are valid.Fantastic self actualisation!Now show something for yourself.I'll accept a pastebin link to whatever you feel you can do to show you've done anything other that restate the same "conclusions" based on "your model." No, I can't personally program in C++, but I have someone who can pick through it.This, of course, is ignoring the fact that you can simply write out the functions performed in a combination of English and maths .. but I'll give you the option to take the easy route so you can't just claim :effort: and not do it.Once more, I will again remind you that your original argument was that a player won't be able to reach his "true elo" in a normal amount of games based on the variance in the elo rankings - where you said "A person stuck at 1100 Elo for 200 games could easily have a true Elo of 1600 and never see it" and qualiified it in your 'model' which showed "[a] simulated 1338 Elo player rated 2009 Elo"This is where it is complete and utter nonsense, and lacks any evidence. You can graph standard deviation against means all you want, but until you show something like the above you're just a guy writing words to support words he has already written.Remember, I am only asking you to show how you know what you claim to know; this really shouldn't be hard for anyone who has graduated HS and the only logical reason this won't be possible is if it doesn't exist.So, again, expect a lot of reasons why you can't show anything, and chest-beating about how everyone else needs to go off and program their own undocumented simulations instead of asking you to copy and paste something.
Avatar
Addendum: I should just clarify about the matchmaking, as I could see how this could be misinterpreted. I generate a "queue" of sorts, by sampling from the simulated players, and playing matches until everyone has played at least the required number of games. The matches come from the queue by simply sorting by estimated Elo and making chunks of 10 players to form the match. This has the effect of generating matches where the players are as equal as possible. So I didn't want to imply I just randomly throw players of wildly unequal skill levels into a match together, only that I do not impose any implicit limit on this, as Riot appears to (e.g. by fluke, in this queue a chunk of 10 were 200 Elo apart so they don't get to play and have to keep waiting). In practice, this never really happens except for extreme Elos where the player population is smaller (which is probably why IRL these players typically have to wait so long). This is probably over-engineered. If I change the sampling size, or even use the entire set of simulated players at once, there is no effect on the variance. The specifics of how the "queue" is simulated doesn't appear to be a contributor to the variance in estimated Elo.
Avatar
Ya, "elo heaven" exists and can be explained by almost the same words as "elo hell" can (which implies the origin and the reasoning behind their existence is basically the same), yet all I hear about is "elo hell" and not even a single word about "elo heaven" before you mentioned it. So the LoL community is a glass half empty kind of bunch, I would presume.
Avatar
No idea what you're talking about, but the matchmaking system on ranked is not random because they match you with people around the same elo, or should I said division, as you. In normal games however, its pretty much very random who you're up against (it does calculate vaguely on how many Normal Wins you have), just like the WC3 DotA lobby. Again, not really sure what you're talking about at all, I just struggle to see why DotA is any different from LoL when it comes to unranked matchmaking.

Wutswrong is a 2100 Elo Season 2 player. You can follow him on Twitch.tv here.

If there was ever a deep, frustrating limbo of online gaming, it would be the "Elo Hell" of the League of Legends community.

Nearly every LoL player has heard about Elo Hell. The victims (such as yours truly) form a tight bond based on our experiences. "I always get someone who feeds! Goes 0/4 by 10 mins. WTF am I supposed to do?"  I've been there. "It's always a 4 v 5. Someone always has to fucking disconnect or ragequit." I know that feel bro. "God damn trolls in all my ranked games. Fuck you instalock Heimer."  Yeah, I've seen that too. "I called mid but this piece of shit just took mid and now he's trolling and feeding." All too common.

So: that's why I'm writing this article. I was a 1900 Elo player in Season 1 and a 2100 Elo player in Season 2. I know the secrets, and I want to help - but you'll have to bear with me and work hard to become a better player.

I'm starting with the man in the mirror

Pro Tip 1: If you continually blame others, you will not see your own errors and will not improve.

There are ten players: five on your team and five opposition. Now, let's think about this for a second. Every game, there will be 9 players you cannot control. No matter what they do (feed, carry, etc), their actions are completely beyond your grasp. However, there is one person, one player that you will have full control of every single game. That's right, it's you. You are the most consistent factor in the game.

Based on Elo Hell assumptions, the enemy team will have 5 feeders, but your team will only have 4. This doesn't mean you'll win every game. It just means over a long period of time, you will eventually have a winning percentage. If you have 200 ranked games and you're stuck in 1100 elo, maybe it's time to take a deep breath and look at yourself in the mirror.

Biggest factor in the game: Attitude

In my two years of League of Legends experience, I've concluded that your personal attitude is absolutely the most important aspect to becoming a better player. If you're reading this article right now and you're thinking, "This guy doesn't know what he's talking about. He's never been there. He doesn't know," then you are not starting off on the right foot. For the record, I was stuck around the 1200-1300 range for my first 200 ranked games in season 1. I often play on 700-1100 elo accounts when I duo queue with my buddies in that range. I know what it's like, and it's the attitude that needs to change. If you want to get out of elo hell, you must stop pointing fingers and take in all the blame. Yes, all of it.

Pro Tip 2: Whenever you die, assess why you died and what you could've done differently. Even if it wasn't your fault, ask yourself why you died.

Every death, you should be pointing that finger at yourself. Don't say "teammates baited me into that fight". No, YOU baited you into that fight. Don't say "WTF where was my team? I pinged like 10 times!!!"  That is your own failure of situational awareness. Do not ever, EVER blame anyone except for yourself in every death. I understand, sometimes you may feel like it's not your fault. Many times, your death will, honestly, not be your fault at all. But that isn't how you get better. Every death, every blame. Whether it's situational awareness, decision-making, mechanics, positioning, or communication, you must criticize yourself and yourself only.

Pro Tip 3: If you don't have anything constructive to say, don't say it at all

Are you the type of player that says, "Real jungler. GG" when your jungler fails a gank? Do you say, "This fucking bot lane is trash" when they go 0/5? In champ select, do you cry when someone picks an underplayed champion or leaves open an overpowered champion? You can completely change the dynamic of your games by altering your attitude on others. Believe it or not, telling your teammates how bad they are won't make them better. Instead of telling your teammates that they're trash, tell them what they need to do. Don't add a nasty remark at the end of that criticism either.

[]

WE. COULD. HAVE. CAUGHT. HIM. AND. KILLED. HIM. YOU. ARE. A. TANK.

If you want to rise in the ranks, you must have a positive attitude and a critical mindset during each ranked game. You have to realize that you are a big problem in every loss. This is obviously not an easy feat since raging comes naturally to gamers. If you can accept the fact that people get outplayed sometimes, it'll make your experience better.

Pro Tip 4: If you lose a ranked game, there is ALWAYS something you could've done differently to impact the game's outcome.

I know there are those of you who are thinking, "No, there was no way we could've won this game."  You'd be right. Some games you simply can't win. But instead of getting roflstomped into a 20 minute surrender, it would've been a back-and-forth 35 minute match. And perhaps instead of a 40 minute loss, you could've had a 40 minute win. Again, the only person you can fully control is you. Telling your teammates how much they suck won't make them better. There is always something YOU could've done differently.

Common Myth Busted

There are less feeders and trolls as you get higher in elo.

This is a common misconception among those who are stuck in elo hell. There will always be feeders and there will always be trolls. It doesn't go away no matter what elo you play at. I have dozens of these pictures ranging from 1200-2100, but I will make my point with just this one.
[]

Remember, the attitude you have for yourself and the attitude you present towards others will make you a better player. After reading this article, take the time to reflect upon yourself and implement these steps to change your outlook on the game.

[name_1]
Wutswrong is a retired semi-pro League of Legends player (NA) that boasts a 1900 elo in season 1 and 2100 elo in season 2. Follow him at http://www.twitch.tv/wutswrong