One Weird Trick to Raise MMR That Raijin Thunderkeg Doesn’t Want You to Know About

In recent news, we have this tale from the PlayDota forums of one man’s harrowing tale of taking a 2900 rated Dota2 account to 5400 in the course of 144 games with a 122-22 (.847) record.  Some might say that this disproves the existence of ‘ELO[sic] Hell.’  But of course for some people it didn’t, and today I want to focus on one of the objections.

Juice, the player behind the experiment, prefers to play mid and carry and did so throughout the experiment.  A common reaction then has been that all this experiment proves is that if you want to raise your MMR, you need to pick ‘high impact’ heroes, preferably mids.  Juice seemed to prefer Ember Spirit, Storm Spirit, and Templar Assassin, though his picks overall were actually quite varied.  In any case, I want to examine the idea that maining these snowballing mid heroes are some universal secret to raising your MMR.

1. Let’s start by looking at the overall win rates for some mid heroes:


Outside of our three exceptions, things don’t look so good for traditional mid heroes, and when people recommend picking a high impact mid, Zeus and Necro aren’t typically the heroes that get brought up.

Of course it’s true that some of these heroes do perform better in higher rated games.  Ember, Storm, Tinker, Invoker, and Templar Assassin are all among the heroes with the greatest win rate improvement in Very High games (roughly >4000 MMR).  But at the same time, none of these shifts are strong enough to get any of these heroes to +50% in that category.  Furthermore, these heroes are only nearing that range of success in the hands of people with the mechanics to make it to >4000 play.  If you can’t make it to 4k using the heroes you’re comfortable with, it’s extremely unlikely that you have the mechanics necessary to make any of these heroes your ticket out of the so-called trench.  At least right now.  If you aspire to learn everything there is to know about one of these heroes, then good, go for it.  Just don’t pretend it’s going to be the fast fix that will carry you to the MMR you really deserve.

But I feel that this simple win rate examination fails to capture the entirety of what’s going on here, so…

2. Let’s step back and ask ourselves what exactly it is that hero win rates measure in the first place.

If you’d like to know, I can tell you that in your universe you move freely in three dimensions that you call space. You move in a straight line in a fourth, which you call time, and stay rooted to one place in a fifth, which is the first fundamental of probability. After that it gets a bit complicated, and there’s all sort of stuff going on in dimensions thirteen to twenty-two that you really wouldn’t want to know about. All you really need to know for the moment is that the universe is a lot more complicated than you might think, even if you start from a position of thinking it’s pretty damn complicated in the first place. I can easily not say words like “damn” if it offends you.

-Mostly Harmless

Think of a hero win rate as a giant array of outcomes condensed into a single number.  Two key dimensions within this array represent the probability that the hero will influence their team into a positive gold/XP state and the probability that a hero can take a positive gold/XP state and convert it into an even more dominant position.  The heroes that we think of as supports tend to produce good outcomes of the first type while the heroes that we think of as carries tend to produce good outcomes of the second type.

Of course it’s more complicated than that.  Lich likely has one of the highest public win rates in a large part because you can stick him into any dumb pub composition and he’ll accomplish something in the laning phase.  Meanwhile, Shadow Demon has one of the worst public win rates because his early impact is extremely dependent on having someone take advantage of Disruption initiation.  On the other end of the spectrum, Anti-Mage has possibly the strongest capability to convert GPM into total dominance, but his mere existence on a team makes it less likely to get to those stats.  Viper has a much weaker gold conversion but has a dramatically higher win rate because he lanes well in general.  Spectre provides a somewhat more puzzling counter-example, as she, like Anti-Mage, is a weak laner, and yet she still has had a consistent >52% win rate for several patch periods.

Anyway, We’ve got this idea of snowball or momentum heroes, so where do they fit in?  We could say that they’re just a lot like carries since carries snowball too, but that feels off.  Items do matter for heroes like Templar Assassin, Storm Spirit, and Outworld Devourer, but not really in a way that’s comparable to Anti-Mage or Spectre.  Anti-Mage and Spectre aren’t dependent on momentum; they’re more like time bombs, with their rate of farm determining how much time you have to disarm them.

What I suggest is that the success of these snowball semi-carries isn’t about reaching some kind of item-based end state like it is with a traditional carry.  Instead, the goal of a snowball semi-carry is to create a gap between their damage progression and the HP levels of the opposing team.  The most straightforward example of this is Outworld Devourer’s ultimate.  Sanity’s Eclipse does damage based on the difference between OD’s intelligence and the target’s.  If OD gets an item and level lead, Eclipse will do huge amounts of damage in a large area of effect.  Conversely, if OD falls far enough behind it won’t do any damage at all.

The central unifying feature to this class of heroes appears to be a specialization in burst damage that tends to be at least as level dependent as it is item dependent.

  • Templar Assassin can use the free damage and armor shred from Refraction and Meld to melt low HP heroes.  Damage items like Desolator and Daedalus can be useful but aren’t strictly necessary and often skipped/delayed in favor of pickups like a Blink Dagger.
  • Storm Spirit thrives on hitting that point where any lone traveler is dead within the duration of Electric Vortex and Orchid.
  • Ember Spirit gets right click damage scaling off of a 5-second AoE nuke with a very long range, but struggles to keep up with the sustained damage output of a true late game carry.

If any of these heroes get ahead, they can consistently blow up out of position targets and put their opponents on the defensive with a 5v4.  If they fall behind, or even just fail to stay ahead, they can find themselves put into more prolonged fights that their kits are not as well suited for.

And that’s fine.  Part of being a hero that depends on momentum is the downside of potentially becoming dead weight if you lose that momentum.  The first choice you make in any Dota game is trying to pick a hero with a strong array of outcomes, but those arrays aren’t fixed.  Wisp’s array in pub play is pretty terrible, but his array in competitive play has been stellar.  Just like Shadow Demon, if you draft a positive environment for the hero, they can drastically outperform their performance in essentially random pub environments.  Snowball mids are just another class of hero where a good environment is essential to success.  Subsequently, their overall pub win rates kinda suck because pub teams in general are tremendously more self-centered than even the most marginal of competitive teams.

 3. So if we accept that environment influences a heroes success, why can’t MMR (and matchmaking in general) itself play a huge role in shaping the expected environment?

Imagine that we have Juice, a ~5500 rated player, queuing into his first match on a 2900 account and playing Templar Assassin.  What do we know about the match conditions?

GPMYou probably don’t recognize this graph from over a year ago, but it shows the distribution of game ending GPM for players in Dota’s three skill brackets back in 6.74.  I’m going to run with the assumption that the shape of these distributions hasn’t changed dramatically since then, partially because it’s likely true and partially because I don’t want to go through the trouble of making an updated version right now.

The blue distribution is Normal games, which roughly correlates to < 3000 MMR.  The red distribution is Very High games, roughly > 4000 MMR.  Green is everything in between.

Thanks to the Dota2 blog we also have an idea of the shape of the overall distribution of MMRs.

5% 1100

10% 1500

25% 2000

50% 2250

75% 2731

90% 3200

95% 3900

99% 4100

With a ~5500 rating, the games Juice plays on his main account would be deep in the tail end of the red distribution.  Based on the chart, we’d expect his games to fall around the 400 GPM per player mark per average.  And what do we find when we take the average GPM of all the players in his last 10 games on his main account starting with 646057416?  An average GPM of 406.41.  We don’t have a fair measure the average GPM of the games at the start of this experiment because Juice’s presence in those games inevitably warps the results, but if we trust that my 6.74 data holds, it’s likely somewhere between 300 and 325.  XPM likely has a similar, if somewhat less pronounced gap.

For a hero based on momentum this is a dream environment.  If your potential as a hero is based on the difference between your stat progression and theirs, then with 100 GPM worth of a handicap that difference will practically never go away.  On top of this, you get a guaranteed 1v1 lane against a player you almost certainly massively outskill, and there’s virtually no risk of ganks since you’re in a bracket where supports, if they even exist, largely have no idea what they are doing and believe that only mids gank.  You give a top .1% player all those advantages and it’s not surprising that the immediate result is this:


So yeah, if you’re a player playing 2500 MMR below your true MMR, snowball mids might indeed be the fastest, most reliable way to maximize your winrate and MMR growth, but for the 99% of players playing at a relatively accurate MMR, there’s absolutely no reason to believe that these heroes are a fast fix to a positive win rate.  Sure, with enough practice a hero like Storm Spirit can get you up there, but the operative word here is ‘practice.’  You’ll have to actually put in the effort to improve as a player, and you’ll be better for it than a player just trying to Flavor-of-the-Month it up the ladder.

4. Epilogue: The Origin of an Idea

Suppose that instead of MMR, matchmaking just threw 10 people randomly together into teams to create games for you.  What would these games look like?  Well to start we need to add ourselves, so let’s assume that we’re a vaguely defined ‘above average’ player with an MMR somewhere between 3500-4500.  The other 9 slots would tend to be distributed pretty similarly to the distribution earlier from the official Dota 2 blog.  This means that on average, 8 of the other 9 people are going to be below 3000 rating, and 5 of 9 are going to be below 2250.  So if you’re a 4000 player in one of these games, most of your games are going to going to be you as a ~4000 rated player, and a bunch of other people 3000 or below.  In other words, the games from your perspective would look an awful lot like the game’s Juice went through on his experiment.

I’m not going to claim to be an expert on the social structure of original Dota, but it’s likely that basic custom games were pretty similar to this as the normal distribution has this pesky tendency to show up just about everywhere.  This means that if you were a decent player in the Dota days, then just about every game you played was similar to the ones in Juice’s experiment.  Is it any wonder that ‘play a snowball mid’ is common knowledge?  It would have really been true in almost every game.

For better or worse (hint: better), matchmaking has killed this, but Dota players in their opinions are often the slaves of some defunct patch period.  You can see the same effect when it comes to win rates.  In a completely random system, a 60-80% win rate would be evidence of the equivalent of a 3000-5000 rated player.  With Dota2’s matchmaking, win rates are no longer a direct predictor of player skill.  But you still have this crowd longing for the regular assurance that they’re nominally above average that you would get in a completely random.   And so you get complaints about ‘socialist matchmaking’ forcing 50% win rates.


42 Responses to One Weird Trick to Raise MMR That Raijin Thunderkeg Doesn’t Want You to Know About

  1. Jimmydorry says:

    You said a lot of things, but either the links were not clear and you did not make a strong enough opinion, or I am too far in the “trench” to comprehend them.

    You said it yourself that the mid position gives the best chance for sucess, due to their typical snowball nature. You also said that support skill and team coordination is typically low in pubs until you reach those higher brackets (i.e out of trench).

    It would stand to reason then that when you don’t go mid, you are essentially flipping a coin to determine how much of a disadvantage you are starting with.

    The crux of the argument people have is that as a support, you have a lot less control over the game than you do as a carry or mid. By not addressing this, i don’t see what point you are trying to make in this article.

    You can influence the game as any hero, but the real drive is gonna be snowballing character and then the timebomb, most of the time. The laning stage is short, and it is very easy to lose all yours gains in a single good rotation from mid, putting you on the backfoot fir the rest of the game.

    You discussed a hypothetical scenario of random MM (which I should point out works in larger team games like WoT). It unfortunately does not have much bearing on actual dota games, but I guess could be used as a weak support for the psychology driving player’s understandings.

    Preemptively, i would like to state that this should not be dragged to the highest tier of play… We were talking about raising MMR “out of the trench”, weren’t we?

    I propose the following thesis:
    If you play a mid hero at or above your skill level, you are likely to improve your MMR or remain stayionary. If you play a support at or above your skill level, you are essentially flipping coins and skill won’t even factor in until you are playing significantly above your level.

    • phantasmal says:

      It’s plausible that risky, snowball mids give you the best chance of success IF you massively outskill the average MMR of the game. For the vast majority of people this will not be the case.

      It’s really likely that when they do go mid, these players will have more success with heroes like Zeus and Death Prophet (in 6.81 at least), as they appear to be more resilient in the face of unfavorable environments and unreliable levels of personal skill.

      The flip side of this is that ‘maining’ a hero like Storm Spirit might, in the long run, push you to improve your game more than a hero like Zeus does. This should be considered a more valuable benefit than finding a short term fix to temporarily game 300 points of MMR, but instant gratification.

      Slark comes up often as another example to raise your MMR, but he’s a far more plausible general case than a snowball mid. He has early timings, can get solo kills, and farms above average and gets value out of gold at an above average rate. But more importantly, he’s really resilient. He’s passable as a safe lane farmer, good in many kill lanes, passable as a mid, and even relatively viable as a suicide solo at most pub levels of play. If you’re solo queuing this kind of lane resilience is far more valuable than it is in a professional captain’s mode match where teams actually plan and coordinate their picks. And conversely, most snowball mids struggle when they don’t go mid. If you slam down your usual Storm Spirit pick and someone immediately randoms Queen of Pain, someone’s likely going to have to deal with a suboptimal lane on a hero that doesn’t perform well in suboptimal conditions.

      And finally, support. A support is just a hero that doesn’t need a farm lane to have an impact on the game. That doesn’t mean that you automatically have an impact on the game by virtue of playing a support.

      If you’re playing a support in < 3000 play, your role isn't to vaguely stand behind the safe lane carry and complain internally every time he misses a last hit. Your role also isn't to patiently ward and deward, because no one in these games does anything interesting in the first 10 minutes on a regular basis for wards to catch, and lol at the idea of de-warding the wards that the other team will almost never have (But still use wards when you can, just don't think of them as -Essential- to the same degree that they are in games where people are actually proactive. More importantly, don't think that being ward bitch alone entitles you to win more than 50% of your games, because it doesnt).

      No, if you're playing support in the trench, your role is to create kills. If you cannot do this in your current lane, then you should be looking to other lanes. If you regularly create kills, you will move up as a support.

      But a lot of people who want to play support actually want a role where they can just follow someone else's initiative, support it, and feel like they're contributing. They might even be capable of maintaining a 50% win rate at a surprisingly broad range of MMRs because they do a passable job at finding a competent player and doing things that help them.

      But that isn't good support play. Support is the most proactive role in the game. You scout out and create kill opportunities. You find and create farm opportunities that aren't just lanes handed to you. If you do these things, you can absolutely carry your way out of the trench. If you play a passive, follow the leader style, then yeah, you're going to go nowhere. But that doesn't mean that support is a no-impact role; it means that when you pretend to play support you're a no-impact player.

      • Jimmydorry says:

        While you raise some good points, I feel they weren’t specifically aimed at my reply, ignored my thesis, and were specifically aimed at a player far outskilling his opponents or generally already playing at a high level.

        If you play support at your MMR, you are flipping coins. The majority of games finish in the mid game, which gave the biggest emphasis on carries and mid player, so two random variables. If you take either of those other roles, just by being consistent, you have already increased your chances of winning.

        By playing these roles, you have the bigger impact come crunch time. I didn’t mean to say supports have no impact, they just have far less, especially at lower MMRs.

        I am willing to bet that repetitions of this experiment, but with the person maining a single role each time, would go a long way to demonstrating this.

      • phantasmal says:

        In 6.80 Normal bracket games, 5 of the top 10 win rates were supports (Lich, Abaddon, Crystal Maiden, Vengeful Spirit, Warlock) while only one was a carry (Skeleton King). Someone, somewhere will complain about me labeling Warlock as a support but that’s explicitly how he fits into low level play. He’s bulky with a good, long autoattack and Shadow Word works on both ends of attrition laning, which is a pretty dominant strat at bad levels of play. Two of the other top 10 (Necro and Centaur Warrunner) are two of the very few heroes that consistently outperform in lower level games because they both can occasionally win (bad) lanes simply by existing. And as for Skeleton King, people are now using him as a support so does that make it a straight 6 of 10 victory for supports over their CS consuming brethren?

        You’re right that the majority of low level games should essentially be finished by mid game (and when they aren’t it is essentially a 10-man symphony of mutual failure), and support heroes specialize in doing this. Take Lich and just Sacrifice -> Frost Blast nonstop and chances are you’ll start winning lanes regularly against people that don’t bring regen to lane, and wouldn’t know how to courier one out if they even have a courier which is basically a coin flip. Hell, if you just Frost Blast a single creep twice a minute that alone effectively equals the average CS per player in many of these games. Put that into the legendary sub-20 minute Mek in a Normal game and Chain Frost and you’ll be teamfight MVP in no time.

        But you are there to make the other team miserable, you’re there to do it as soon as possible, and you’re not necessarily going to get anyone’s help in pulling it off. Going Leshrac in a low MMR game is largely an exercise in frustration because no one is going to set up your split earth and no one is going to help you have the creep waves to push towers with edict. Shadow Shaman would be a more preferred choice for the role because he’s much more self reliant.

        If you’re good at farming, then yeah, (self-sufficient) farm converting heroes will get you out of Normal more quickly. But most people who are actually good at farming don’t stay in Normal very long (or even get to spend time there at all if accelerated placement has its say). But support does just fine if you accept that you’re there to create kills. Also that you should absolutely be looking for farm every chance you get because Normal players leave so much on the plate that it’s not even funny.

    • box says:

      All you have to do is look at the pro stats. Ember has a win rate of 42% in 66 games in 6.81. Psychologically people have huge confirmation bias. If you want something to be true you will only notice data that proves it. You won’t remember the games ember lost and you will keep thinking about the ones he won.

      • 6.81 is still new, so I’m not convinced that Ember is underpowered. Even if he isn’t top-tier any more, he could still be situationally good. In that case, if pro players adapt and only pick him in games that favor his play-style, he could eventually have a large winrate (say 65%) despite the huge 6.81 nerf.

  2. DivineOne says:

    It’s a good analysis, but I think you’re missing a fundamental concept that I think is key to effectively outperforming people you’re matchmade with.

    The notion is the degree of interaction or coupling between your play and your teammates/opponents. Consider a highly farm dependent carry like spectre, and a ganking semi-carry like QOP. Even if winrate averages to ~50% over all games for both heroes, spectre depends much more highly on the perfomance of both the rest of his team (to let him farm) and his enemies (to fail in ganking him) than QOP. We would expect highly skilled players to consistently win more on QOP than spectre due to the lack of these player interactions.

    Mid naturally cuts down on player interactions due to the 1v1 nature, and the gold and xp bonus. But, heroes like 6.74 lycan and 6.80 terrorblade are additional examples of non-mid heroes that could consistently do well regardless of the rest of their team.

  3. TwelveEleven says:

    Interesting read, swiftending/juice is now actually rated above 6k on his main account, so he’s in the ~top80 players of EU. I think his experiment is nice and it shows that ELO hell doesn’t really exist, but his experiment went way faster than it would for anyone who’s rated above average.

    He also played hero’s HE is good with, if your best role is support or off lane. You should be playing that role to get higher.

  4. Excellent little article. here. :)

  5. dejecaal says:

    While I’m not sure what the point of the entire post as a whole is, I think the last couple of paragraphs raise a somewhat interesting issue.

    In the past, before matchmaking systems, “getting good” at competitive video games like this periodically rewarded you by reminding that you actually are pretty alright at the game, when you ended up in a game against people worse than you.

    Assuming Dota2 had a perfect matchmaking system (it really doesn’t), games like that shouldn’t ever happen. The only indication that you’re any good at the game would be an awfully impersonal and abstract number, which is a lot more unsatisfying than the very concrete indication of being decent that is playing against worse players.

    To make matters worse, imperfect matchmaking, such as the one Dota2 has tend to make everything even more unsatisfying. This is something that could have (and has had, numerous) full blog post dedicated to it, but in short, matchmaking tends to result in generally more unsatisfying games than the random pubgames used to. Because of the nature of the game, all it takes is for one matchup on the map to be unfair, and the game will significantly suck for the rest of the map.

    For an example of what I mean, let’s imagine one the games Juice played while he was ranked incorrectly by the matchmaking. Every other lane had roughly equally skilled 2900MMR players playing against each other, except in midlane you had a 2900MMR player getting his shit packed in by a 5300MMR player. This one single “error” in the matchmaking with a single player essentially being matchmade into the wrong game ruined the match for everyone by making it horribly unfair. You may argue that situations like this are rare, but in a huge matchmaking pool like this that constantly gets new accounts games like this are bound to happen very often.

    Something exactly like this was much rarer in the random pubgames on, thanks to them being random. Matchups in those games were rarely fair, but the difference is that instead of one or two players in the game being too good (or too bad) for it, everyone was just completely randomly skilled and there would be no baseline to compare them to. Thus in my experience they would often turn into games where each side had a skilled player or two doing really well trying to out-carry the opposing team’s skilled players. And frankly, I think that was a lot fairer than the modern matchmaking systems we have nowadays.

    • phantasmal says:

      The unavoidable question is “fairer for whom?” A certain percentage of the (relatively) skilled players might be happy with the arrangement because the disparity works in their favor more often than not. They might hate matchmaking by comparison because it sometimes creates games where they feel helpless in comparison and resent the reminder that they’re incapable of controlling their own destiny 100% of the time. Both of these factors might be points of frustration, but I reject the idea that they are “unfair.”

      More importantly, developing as a player involves playing people stronger than you and figuring out how to beat them. did not provide an effective environment for this, and it created and reinforced bad habits and incorrect attitudes towards the game, such as massively overvaluing the importance of mid and underestimating the high risk/high reward nature of certain heroes. For the competitive evolution of Dota, matchmaking as the default way to play is indisputably superior than random lobbies, and at least for me, that’s more important than any consideration of fairness.

    • Pitlord or riot says:

      I think your judgement is blinded by nostalgia. Bnet pubs were a complete mess. The games we get in Dota2 are way more balanced.

  6. Gerbil says:

    Cool post! One beef I have is that you’ve equated the distribution of skill of all players in Dota with the distribution of skill in the average Dota game. Certainly it must be true that better players play more than their lower skilled counterparts, so I’d imagine that a random matchmaking system would only have something like 1-3 players in the sub-2250 bracket as opposed to 5.

    • phantasmal says:

      It’s plausible that such an effect exists, but we have no idea what the magnitude is, and it’s probably not quite as large as you think it is. In fact, one of the criticisms of my own distribution tests is that I was finding the distribution of games and not players. Still, Normal (< 3000) were repeatedly around 80% of the games played. On the other end of things, Valve says only 1% of players are above 4100 (which should be around the region where VH games begin) and yet I would find that roughly 2-3% of games where VH. So yeah, it's likely that better players play more often, but the effect of this might be completely negligible compared to the sheer population size of < 3000 players.

      And then on top of that, higher skill players might have played more often, but then might have opted to play more often in in-house leagues instead of random games.

  7. Crow says:

    Here’s the problem that I see with this analysis (and to be fair it was somewhat mentioned in the article). I’m not a high skilled player, but I believe that I should generally be gaining MMR because I tend to play better than 90% of people in my bracket. Generally there is only one other player as good as I am in the game. Sometimes he’s on my team and sometimes not.

    What I’ve experienced is that I’m not so good as to completely obliterate my lane, but I do win it handily. I come out with more CS, GPM, and a better item progression than who I’m laning against. Usually I’m able to get 1-2 kills on them too.

    The problem I face in about 50-60% of my games is that the people in the other lanes do so poorly that I am left without a team to back me up. On top of that, they give the enemy team enough of a gold and XP advantage to negate me winning my lane. With only a modest/zero advantage, I’m left trying to corral a herd of cats into doing something productive with their heroes. When they inevitably fail to understand the basics of how to win a game of Dota, I’m left to essentially try and 1 v 5 the enemy team, which you can imagine is very difficult without a huge lead as you described in your article.

    So even though I could complete evenly with someone 1000-2000 MMR higher than me, I will probably never reach that point in my life because I simply don’t have competent team often enough to gain MMR at a reasonable rate.

    It’s a simple numbers game. Due to my job and internet constraints I can only play about 10 games of Dota a week. Lets just make it easy and say this averages out to 500 games per year. If I have a 51% winrate and all of my games were even (+/- 25 MMR points per game) then I would gain, on average, 125 MMR per YEAR.

    This is why people say they are stuck in the trench. Without a significant win streak in excess of the normal 50% +/- 2% winrates, you simple cannot advance to a higher MMR at a reasonable rate without playing dozens of games per week (i.e. as a full time job).

    • Jenesis33 says:

      I understand where you are coming from, however what you are saying is assuming your team dont have bad players. They always have good players and beat your other teammate which are shit. This is not possible, bec MM make you play in a similar skilll bracket on both team. So yes in some game you will loss other lanes (due to bad teammate) but in some games, your teammate might beat the other people on lane. If you can’t have enough influence to win those games which is on the lossing path, then you don’t derseve to move up.

      Just look at this from a stat point of view. Both teams have 5 players.. you are the only constant factor. YOu are good.. (assuming this and You are better than anyone else you are playing with).. This means the worst possible case for your team is 4 bad players with You. However the worst case for enemy team is 5 bad players. So from a probility POV, you will always win more than you loss, provided you are constant a good player.

      Of course, playing more games will help to even out the data.

      Just because you are 90% time better, does not mean you should win 90% of teh game. But if you are indeed better than rest of everyone on the field 90% of the time, your win rate should be definitely much higher than 51%.

      I just hate how everyone complain about shitty teammate, but never rearlize enemy team can have shitty people too. Its the same coin just two side..

      Just remmeber, you are teh only consatnt factors in this, and if you are indeed much better, then you should win more..

      • JoaoMelao says:

        Maybe so, but things are much more polarized then that.

        Imagine for example that 50% of the games your team feed, so hard that your advantage can’t come back from that, and the other 50% of the games the other team feeds making it super easy for you to win. You are still a constant, but if games are that polarized maybe being just a little above average never leads you to getting your mmr up.

  8. estate_kebab says:

    I notice you blink to one of my threads on the forced 50% and its implications. Now in your last paragraph you bring up a randomized matchmaking, and I know you aren’t saying thats what I recommend, and you’re ambiguous on whether you’d support such a thing, but one thing to make clear is that one of the recommendations that grew out of that thread was a randomized matchmaking- but one very different from what you are talking about.

    Right now it takes 10 players of as equal skill as possible, then arranges them into a game to get the MMR sum on both sides as equal as possible, which ‘forces’ the 50% winrate, and does so by handicapping good players. The better you are at the game, the worse your allies will be compared to opponents. Thats just a mathematical fact of how matchmaking works, because that bell distribution means it can’t keep giving you 9 other equally skilled players.

    The proposed change would be to do that same first step- find 10 as closely matched players as possible- then put them into random team assignments in the game. Not stacking the teams to force a 50% winrate. Thats substantially different, because you’re talking about players getting 80% winrates because they are 5500 and the other 9 players are 2250 on average, because they were chosen at random out of all the players. That wouldn’t happen in the aforementioned change. You might have a 5500 player with 9 other players who are 4500 on average, but thats only because the top 9 other players available at the time had 4500 average.

    Now, would such hypothetical changes to matchmaking change your central point about snowballing mid heroes being the most high impact role for a good player when playing with people less skilled than him? I didn’t really see anything affirming or denying that in your article, and I don’t think it would necessarily change anything. What it WOULD do, for sure, is make playing other roles than solo mid less hostile to good players by comparison. I have MMR somewhere around the same as Juice. Say I go into the same games as him, and play a #5 role hard support and take no last hits, low exp, buy wards and courier. Then I am leaving the game largely in the hands of my allies, and I know due to how matchmaking works that my allies are worse than my opponents on average. I lose most games. And my experience with statistics reaffirms that, I have 40% winrate on most support heroes and 60-80% on high impact roles.

    The last thing I need to bring up is something you seem to ignore with your statistics, and poisons your analysis. You talk about heroes like ember and phoenix and invoker and so on having an “MMR shift” that is favorable to high skill, yet their winrates are clearly below 50% even in ‘very high’. Yet anyone who has played high level pubs will instantly tell you that heroes like ember run wild with extremely high winrates in the top of the top games, and in 6.79 earth spirit was absurdly overpowered and considered instant win and games were just a race to pick him first. Yet he didn’t break 50% in ‘very high’ statistics. The reason why is clear: Your “very high” statistics have too low of a skill floor and the majority of games even in ‘very high’ are still between relatively unskilled players, and winrates at low skill don’t reflect the actual strength of a hero, but the skill needed to play it- heroes like earth or ember with lots of buttons and nonobvious triggered effects and combos you need to know will be awful in the hands of bad players, overpowered in the hands of good ones, and your “very high” is mostly bad players. Unless you have a way to get an even higher skill bracket cutoff, your statistics are poisoned by this. Heroes that are cancering up the top level pubs shouldn’t have sub-50% winrates, or we can’t trust your statistics. Lies and damned lies.

    • phantasmal says:

      Well that kinda worked. Eh well, close enough.

      I included the link because of the absurdity of invoking ‘socialism.’ That’s like accusing the NBA of socialism because they won’t allow the Miami Heat to play against random JV squads, preventing them from realizing their -True MMR-. Those bad players you talk about? There’s more of them, and collectively, they buy more hats than you. Welcome to capitalism. Please enjoy your stay.

      As for MMR shift, 6.79 Earth Spirit showed an almost unprecedented amount of improvement between Normal and Very High. Sure the y-intercept was low, but that degree of a shift is a big sign that something was up with the hero. Ember, Phoenix, and Invoker demonstrate much smaller shifts (though Ember and Phoenix’s could have grown towards the end of 6.80, I don’t have those stats), so I was less concerned that we were seeing a repeat of Earth Spirit. Point is kinda moot though since Ember got hit pretty hard in 6.81.

      Really though, if you’re around Juice’s MMR, the issues you’re facing aren’t because of the mechanics of matchmaking; it’s because there are so few players at that MMR that matchmaking has to choose between less-than-tight MMR distributions or > 15 minute queue times. There’s an inevitable trade off there, because even if you would gladly accept a 30 minute queue for a tighter MMR distribution, other players would be pissed off by them. Allowing you to opt out of short queues in favor of tighter games will lengthen other players’ queue times even if they didn’t opt out. Your suggestion doesn’t address any of this. It just says you feel inconvenienced by the current compromise, and other people should have to put up with permanently worse games to make up for it.

      • estate_kebab says:

        There is an inevitable tradeoff between match ‘fairness’ and ‘balance’, yes. And right now the game opts entirely for the latter at complete expense of the former. The more trouble the game has finding 9 other equally skilled players to you, the more good players are handicapped by having games stacked against them. Its not exactly a revolutionary and heretical viewpoint to say that maybe, just maybe, matchmaking should take into account the fairness of games being handicapped or not in addition to having them be equal chance to win on both sides. And maybe it could strike some kind of compromise solution between them. And no, its not only apparent for any 5500+ MMR master race- logically, as per the bell curve distribution, its proportional (not necessarily linearly) to your deviation from the median. Its a factor for 4000 MMR players, less so than 5500 MMR players, but still a factor

        But socialism? Its taking away winrate from the haves and redistributing to the have-nots. Its punishing people for doing well and rewarding them for being incompetent. And while you might agree with this on a philosophical level for human welfare, I think its highly misplaced in a competitive game. A game that you’re supposed to play to *win*.

        I am being inconvenienced to fill in for other people’s inadequacies. It is not selfish of me to ask to be entitled to just to just the fair fruit of my own labours and sweat off my brow, not more than I’ve earned. The fact I’m open to compromise is nothing short of altruistic. If valve’s matchmaking has limited factors that reduce match quality like player availability and MMR spread, they should be borne equally across all skill levels, not be purely on the backs of us skilled meta forging bourgeois

        oh and
        “Build creators”
        “No amnesty for LPQers”
        “Put up a fence between USA and russian servers”

    • pzkw says:

      Your entire premise is flawed. Let me link some MSPaint work just dripping with finesse to demonstrate why your ideas are total fallacies:

      Even if your outlandish concept of a deliberate conspiracy by Valve to suppress your genius is correct, it’s correct for everyone. That means that in your system, the other team is in exactly the same position and your statements remain totally meaningless. That doesn’t mean it’s a rigged game, it means it’s a different type of game where two giants are playing against minnows. Perhaps it’s high time that instead of bitching you learned to topple the other giant, if you truly believe your shit to be true.

      I genuinely don’t believe that you believe a word of the shit you drivel though. I find it implausible unless you have a level of cognitive dissonance that borders on insanity. There are so many things you have to completely ignore in order to reach your conclusions about conspiracies to suppress your MMR and force a 50% winrate on everyone that I don’t believe you could seriously believe a word you type.

      The far, far more plausible explanation for all of the things you notice is that statistics other than those of the ELO system itself that only measures weighted wins measure irrelevant information (I’m directly poking a stick at something as ridiculous as DSR right now, but it holds true for other systems like TSR and TrueSkill) since literally only a single thing matters in the entire game – which ancient dies first – the path taken there doesn’t matter, but some are better worn. In this context, the reason people tend towards a 50% winrate is because their MMR has been correctly measured by the system and they’re consistently paired with similarly skilled players – you wouldn’t expect anyone’s MMR to significantly deviate unless their skill significantly deviates – and it does.

      To give a really, really simple example, I recently got a flu and found life/game hard. I was playing noticeably worse – missing last hits I wouldn’t normally miss, missing disjoints I wouldn’t normally miss, not able to intuitively track and recall cooldowns in and out of teamfights and so on. The result? My MMR dropped by about 500 over the course of the month I was sick. It’s implausible to suggest that Valve forces a 50% winrate on people when winrates deviate with skill changes, as in this example, and as in the article we’re discussing – old mate didn’t stack and yet somehow ended up with 85% winrate before he was consistently paired with players of his own skill – how socialist is that!

      Your anecdotes about different roles giving you different winrates are similarly irrelevant and much more simply explained by you being better in some roles than others. In this case, you’re a better core player than a support player – your analysis into why you think you lost those games as support (relative lack of ability to influence the game) reveals precisely some of the biases that probably restrict your ability to perform in that role.

      In any case, like most of the stuff I’ve ever read that you’ve written, it smells strongly of confirmation bias and a high degree of intellectual dishonesty. You bat aside and ignore any of the information that doesn’t suit you and lurch through convoluted explanations that suit the conclusion you’ve already come to rather than seeking alternative and simple solutions that require far fewer totally unsupported assumptions.

      • pzkw says:

        The post got rid of my link because it was wrapped in HTML:

      • estate_kebab says:

        I’m glad to hear you got the flu.

        I didn’t read half your post, and I don’t have any grand plans to scroll back up. But I thought I’d let you know that your entire premise for your post, that my premise is flawed, is flawed. Funny how that works. What I am discussing is not a grand conspiracy, but simple mathematical fact that could be no other way given how the matchmaking works. And the simplest way for you to learn what you’re blathering on about without me wasting *too* much of my rather valuable time on an ignoramus is for a thought experiment. Try your hand at something I would present to a 10 year old child and see if you can draw the conclusions.

        Take the following 10 players:
        “Quit”: 6379 MMR
        “Hustle”: 5067 MMR
        “Taco”: 4607 MMR
        “Tom”: 4384 MMR
        “Dis”: 3997 MMR
        “Killer”: 3943 MMR
        “Bing”: 3908 MMR
        “Luo”: 3905 MMR
        “Tips”: 3893 MMR
        “Aha”: 3892 MMR

        Your job is to sort these 10 players into two teams so the total MMR on both teams is as equal as possible. Can you complete that task? Take your time. Sort them into teams and come up with a balanced game.

        Now tell me, what is the average MMR of the 4 allies on “Quit”‘s team? If you were “Quit”, and were in this game, what would the average MMR of the allies it gave you be? Just the allies, not yourself. Now what is the average MMR of his opponents?

        Your answer should come out to be 3900 Allies and 4400 Opponents.

        Was that a grand conspiracy? Did the game hand him worse allies than opponents because he slept with Gaben’s wife? Will you pretend it was just random bad luck?

        Its mathematical fact, the better you are, the worse your allies will be compared to your opponents. The more skill variance in a game, the more you are handicapped by giving you mouthbreathing allies and competent opponents. Of course the inverse is true for bad players, they get helped by this system. Players like you, at the expense of players like me.

      • Col. Asdasd says:

        IcyCalm? Is that you?

      • Jimmydorry says:

        Well played kebab, not give any validity to you argument, my off the back of my hand team turned out to be true and screwed quit.

  9. abs says:

    This article doesn’t take the ‘Peruvian Factor’ into account. Nice try.

  10. pzkw says:

    A lot of these replies really miss the point of the article, which is that no position or role will really improve your MMR. The only thing that improves your MMR is winning more games over a long period of time – ie getting better.

    Part of that is learning how to pick and play flexible heroes that perform well in lots of lineups and conditions, part of that is making better decisions and executing better at all phases of the game and part of that is knowing how and when to take specialised picks to exploit or create opportunities.

    Arguments about hero specialisation and team dependence miss the point entirely. They are just another decision with their own set of risks, potential mitigations and rewards that as a player you assess on their merits and make a call on. Like all decisions, whether it’s a good or bad call is situational, and those decisions are just another skill.

    Arbitrarily “maining” a hero or role for hundreds of games worked in the case study because the player in question was really just adjusting the MMR of a new account to his real MMR. It offers no insight into MM except to demonstrate that better players get higher MMR. If it were really so straightforward to get 5.5k MMR, surely everyone at 5.5k MMR would be trying to pick a semicarry mid every game, which isn’t the case. So all we really get out of this is what we started with – if you want a better MMR, improve as a player and win more – there is no concentration of any role at high MMRs, if anything there’s far more mechanically strong players who are well rounded and flexible.

    • phantasmal says:

      Progressive taxation is a desirable, and as we’re likely to find out in the coming decades, necessary component of a market economy. This isn’t out of some ethical commitment to fairness, but simply because it calibrates the game so that each iteration outputs socially desirable outcomes. Similarly, the purpose of matchmaking is to produce interesting games. Any appeal to the perception of fairness is so secondary that it is effectively irrelevant.

      Your focus on win rate is like going to the gym and complaining that someone is replacing your 15 lb dumbbells with something marginally less pathetic because it makes it harder for you to do 1,000 reps per day. Really more to be pitied than engaged with.

  11. fragk says:

    This is all very interesting…

    But unfortuntely, it is all for naught.

    Statistics and theories are fine on paper. But they are not very practical, because they completely ignore one simple variable: luck.

    And yes, Luck can be the deciding factor in any environment. You can have a 50% chance to draw tails on coin toss, but you may lose 7 times out of 7. Add to that the chaos of 9 different variables (the other players) with all their own Luck variable and you’re in for a hot mess.

    Simply showing the ‘experiment’ with Juice proves how random things can be and how your team has a huge effect on win/loss.
    Maybe over 100 years the statistics win but unfortunately most gamers (and games) don’t live that long.

    There is only one true way to win games consistently: Adapt.
    You Adapt to the enemy team and your own. Planning head and adapting to the environment is what makes any living bieng survive.
    It makes you a far more reliable and versatile player in the end. Not just a “SS mid or feed” kind of player.

    • “Simply showing the ‘experiment’ with Juice proves how random things can be and how your team has a huge effect on win/loss.”

      Exactly the opposite. Juice went 24 wins, 0 losses when going from 3k to 3.5k MMR. No matter how bad his team was, he could always make up for any deficit by making his enemies feed harder than his allies were because of the huge difference in his skill. If someone isn’t winning 80% of their games at 3k MMR, they aren’t 5.5k material.

  12. Chris says:

    Phantasmal thanks once again for this fine article, and your continued work on this blog. I love it!

    So really your point is this – High impact heros are only high impact if you have a significant, or better massive, skill advantage. At the same skill levels, whatever flavour of the month resilient heros like necro, DP and zeus (or indeed a support) would actually server you better. That is as per the win-rate statistics.

    There is an argument to say playing those high impact heros may help improve you as a player even if you lose more games. Or that playing flavour of the month heros may inadvertently reinforce bad habits by being generally more resilient to poor in-game decisions.

    • phantasmal says:

      To be more precise, these high impact heroes are high impact in exchange for being risky, and we have a tendency to selectively remember their spectacular games while assuming that the risks don’t apply to us. And the risks actually don’t apply to people in games where they have a massive skill advantage, but matchmaking makes those games much more rare than they were in the past.

      From there we can go a number of directions, including the ones you bring up. The big ones though arethat that these snowball heroes will not be immediately effective (win rate positive over a large selection of games, leading to a big MMR jump) for most people, and that playing mid is also not necessary to move up in MMR.

  13. pzkw says:

    Cherry picked and ridiculous example caused by having players who couldn’t MM into their own tier because of lack of players.

    Keep trying.

  14. JohnTsunami says:

    Can we have a graph like this for 6.81?

  15. gator says:

    tldr; Mid Or Feed!

  16. robert says:

    Well i don’t know, i created a smurf account, cause my main account was only 4k or 4k and a little, and in the smurf i played good and won the first 10 games obv snowballing like crazy, tbh the 3rd game i played in that acc was very high skill already, anyway, once i got into very high skill i kinda started to lose, i won some i lose some, but then games were even more complicated, since my team was not competent enough or they just wouldn’t care since it was unranked, thing is i started to lose every game, like a real losing streak at would only stop for a win or maybe two, and then back again with 6 loses in a row. I finally got to lvl 13 and played ranked hoping my losing streaks would end, indeed the first matches i won, but then i lost 6 ranked in a row ( anyway i knew i was in a good tier cause i could tell by the players i had when i was TBA, and i even asked them what mmr they were) anyway after i lost those 6 games i could see my mmr, gratefully for me it was 4920 mmr, with an account with 26 wins and freaking 52 loses. I didn’t believe that was posible, i was wondering what would have been my mmr if i acutally won more than i lost. Sad thing is, i was excited about almost having 5k that i said okay lets win 3 games and im there, sadly i lost 4 and dropped to 4819. Now i’m scared of playing, since i lose almost every game i play with a bit higher of 33% winrate i’m not comfortable of playing any more ranked games in that account. Idk what to do honestly i was so close but yet so far. and i feel bad about losing so much, but i can’t help it, even when i play properly, game just turns up to be awful, or maybe not even that, we start winning, with a huge advantaje and then they comeback or we throw and lose, please somebody help me im going crazy here

    • phantasmal says:

      When you create a new account, the matchmaking system tries to estimate your account based on your playstyle. Because of the nature of the game, MMR takes a long time to move, so a solid estimate can cut weeks off of how long it would take to discover your MMR.

      But an estimate is an estimate, and in your case it has almost certainly overestimated your MMR. The result is it thinks you’re 4800, but you can’t cut it there, so your MMR is going to slowly decay until you hit a range where you hover around 50%.

      I know this sucks to hear because it feels like you were close to a milestone that’s slipping away from your grasp, but that rating is just an illusion from a calibration attempt that missed the mark.

      • robert says:

        that would be accurate if losing the games was big part my fault, which is still not the case, so idk, feels like there are scumbags even in the high tier bracket. I’ve won 2 games now, im close to 4900 again. Let’s see what happens next, hope i get a winning streak

  17. bob says:

    None of these so-called guides take stupid teammates into account. They’ll mention them, but your idea of rising in mmr is some utopian shit. The only way to get out of mmr hell is never to go there in the first place, so until you’re in the very high skill bracket and have a 60% win rate, don’t do your placement matches. I can’t support feeders who won’t listen and I can’t outcarry their feeding, it’s impossible. If you don’t think 4 people feeding every 2 minutes or so is impossible to beat, then you must be cheating because no one can outfarm the fact that the enemy team has as many player kills as they have last hits.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: