Dote Night: How Does Dota 2 Matchmaking Work?

Part of a miscellany of serious thoughts, animal gifs, and anecdotage from the realm of MOBAs/hero brawlers/lane-pushers/ARTS/tactical wizard-em-ups. One day Pip might even tell you the story of how she bumped into Na’Vi’s Dendi at a dessert buffet cart.

“We do not examine individual win / loss streaks or try to end them,” say Valve staff in a blog about matchmaking and matchmaking ranking (MMR). I believe them. Right now, I am particularly inclined to believe them because I’m thirteen games into a losing streak and Dota 2‘s matchmaking system has not intervened to make my digital wizard experience pleasant or easy even once.

The point of matchmaking as Valve describes it is to set up games of Dota such that each team has as close to a 50% chance of winning as possible, that a five stack of players (likely on external voice comms) isn’t matched against five individual players speaking different languages, and that the spread of skill is similar on each side. Essentially matches are supposed to be balanced and fun.

One of the main factors it takes into account for all this is the MMR. There are five different MMRs at the moment and they depend on the type of game you’re queuing for as well as who you’re with. You’ll have a number assigned for solo unranked games, party unranked games, solo ranked, party ranked and team games. Of these, solo ranked has been the cause of a lot of my recent misery.

Your solo ranked MMR shows up on your profile once you’ve played enough games and is visible to friends as well as viewers if you’re streaming said games. “It’s only a number” was my thinking approximately this time a couple of months back. “You’ll never catch me getting all stressed about THAT.”

Not long after this I was idling by my PC and contemplating some solo queue. “I know – I’ll see if I can bring my MMR up a bit,” I thought. I can’t remember the exact sequence of wins and losses which peppered the intervening period but this process of self-improvement has, if I remember correctly, now wiped roughly 200-300 points off my MMR. I’m worried that at some point I’ll queue for a game and then it’ll just drop me into Animal Crossing so I can wander about and gank the more non-threatening fossils for 45 minutes.

There’s a lot which has been written about the meaning (or lack of meaning) when it comes to MMR. There are guides on what types of heroes to play to give you the best chance of scoring a victory. They tend to involve assuming your teammates are a liability and choosing to play as someone who can either guide the match or who doesn’t rely heavily on others to excel. Essentially, everyone else is the problem. A fair few of the matches I’ve played or friends have told me about in solo ranked offer up a weird version of Dota – Destiny’s Child Dota where all the heroes who independent throw their hands/claws/paws up at me (oh look – there are ten of you – who knew?) But rather than retread those topics, I’m interested in why the number now matters in a way it didn’t before.

It’s not the lack of improvement that bothers me because by other measures I’m sure I’m getting better at Dota. I know I have a bigger hero pool, a better understanding of abilities, statistics and items, I can predict other people’s behaviour better and co-ordinate with the people I play with on a regular basis. The problem with the solo MMR number going down is the sense that I’ve spoilt something or made it worse. That number was fine until I started to tinker with it but now there’s a gnawing frustration – a sense that I need to repair the damage I’ve done. It’s a sensation almost identical to one I associate with trips to the seaside back home. I grew up on the coast and so weekends often involved trips to one of those tiny seaside arcades which had clusters of 2p machines.

If you don’t know what a 2p machine is, they’re maybe the same size as a fruit machine and make the same cheery musical racket but instead of glowing buttons there are shelves – some moving – loaded up with 2p pieces threatening to topple off the edge. All you have to do is post more 2ps through a slot at the top. By timing it right the one you add to the pile will dislodge the rest and cause an avalanche of coins to fall into the dispenser at the bottom.

What I would do is save up all the 2p pieces I got given as change or found in the street and then go to the arcade, experimenting with timings and angles to see if I could best the system and playing until I had no more left. Rarely I would stick a quid in the nearby coin dispenser and chuck more 2ps into the coin pusher, chasing the thrill of a win. It was always this point where it became a negative experience. Losing pennies was fine when I was playing with “extra” money I’d picked up as change or found on the pavement – I’d already mentally budgeted for that or written it off as the price of fun – but when the cost moved beyond that I felt a rush of guilt, of panic.

The instinct was to pump more money in with the goal of breaking even. Deciding to quit once there’s a cost attached is hard. In economics and business the phenomenon is known as the sunk cost fallacy. The sunk cost is the resources you’ve already invested into a project – usually time or money or both. The fallacy, as Freakonomics’ Stephen Dubner puts it, “is when you tell yourself that you can’t quit because of all that time or money you spent. We shouldn’t fall for this fallacy, but we do it all the time.” You might also know the same concept as throwing good money after bad.

As a kid I was pretty good at knowing when to quit even though it went against my instincts. I knew when I was better off spending the money next door on an ice cream or saving my 2ps for the local carnival. I’m not sure if you guys had those either but we would throw coppers, ostensibly “for charity” but really just to show off our aim, at the passing floats. I think it’s because of this that the local beauty pageant winners ended up rolling past wearing pretty dresses and sitting inside protective cages made of mesh and netting.

As an adult I feel like I’m at a similar point but with time and emotional energy rather than 2ps. I’m trying to decide whether to leave my solo MMR the hell alone instead of trying to game the system. The time I’ve already invested is lost but rather than continuing my trips to the land of Destiny’s Child Dota I could be spending that time far more enjoyably. Perhaps learning to skillshot Miss Whitstable and her attendant princesses – or at least their netting – as they sail past on a flower-strewn trailer.

34 Comments

  1. alw says:

    Penny fall machines, those were the best thing ever. The trick is to see how the pennies are stacked around the edges. If they’re edge to edge, they can push other coins off, but if they’re overlapped, there’s little to nochance of that.

    (This tip brought to you by a misspent childhood in various coastal arcades)

  2. BananaMan3000 says:

    Practice new heroes, try new strats, etc in unranked games so it doesn’t cost you MMR if you screw up – when you play solo ranked stick to heroes you are experienced with and that don’t need a sympathetic team to work well.

    I think you’re right about solo ranked having a bias (at lower levels anyway, that most of us player at) – if you feel your skill is above your current MMR then the best way to improve it is to try and take hold of the game with a powerful carry/snowball hero. Playing a great support game means you can still lose, playing a great carry – the chances are much smaller, they’re called carries for a reason :)

    I’m not sure Dota 2’s matchmaking really works any better than randomly scrambling the teams however, considering the majority of games turn out to be one-sided (this could be due to snowball encouraging mechanics though) – I’d be interested if anyone with a background in statistics could say if there was a way to tell the difference between a system that is able to purposefully constructs teams that each have a win chance of 50% versus just scrambling them every time.

    Personally I’ve been at the same MMR for the last 100 or so games (win one, lose one, win one, etc) and feel it probably represents my overall skill level (across lots of games) pretty well. Most people forget the games when they play horribly and remember that one time they went 20-0 with a carry so this obviously means they’re awesome (forgetting the other team’s players were appalling that game). Dunning Kruger gets the better of the vast majority of people frustrated with their MMR.

    The blog post below has a really astute breakdown of how MMR/Dota 2 games work statistically and what to do to improve. It’s very clever in pointing out that the way you increase your MMR is by winning games that you statistically weren’t supposed to. It’s geared towards the 4-5k MMR range but makes a lot of sense.

    Check it out and good luck!

    link to 26andstupid.com

    • Synesthesia says:

      This is an underrated post. I never play MMR, because, as you say, people tend to play heroes they usually play, and that usually means not thinking enough into team composition and counterpicks. Playing in pubs has been extremely fulfilling so far, with not much to lose, everyone plays their best, including the picks, without overly thinking about the outcome.

    • sebmojo says:

      you write excellent, funny and thoughtful posts, pwarr. I’m thinking of transitioning my regular gaming group into DOTA2: is starting out against bots the best idea? when’s a good time to transition to actual people?

      • Banyan says:

        Yes. I would recommend everyone start with reading Purge’s Welcome to Dota, You Suck guide (link to purgegamers.com) and play a few games against bots. Even if you’re familiar with other MOBAs, Dota is very… fiddly… and you will probably want to know, for example, how to summon, upgrade and call the courier or the locations of the side shops and secret shops before you worry about playing actual humans.

      • BananaMan3000 says:

        Play a few games against bots yep – until you’ve figure out how to last hit (really important) buy items and use the courier, what some of the items do, how towers work etc

        Don’t play against them forever though, the bots play differently to humans, they’re great at last hitting / laning, judging when to go on you and try a gank, but terrible at most other things and tend to group up as 5 as soon as the mid game arrives. Remember when you start out the people you’ll be playing against will be new players too. Try to find out which heroes are simple to play (use google to search for good starting heroes) – some are massively more complicated to play than others.

        Good luck and have fun, once you’re over the hill and starting to understand the game a bit rather than be bamboozled is will become really rewarding.

    • Furious Adam says:

      In my opinion the MMR needs a lot of polishing. My friend started to play the game a few months ago and I’ve been stalking a few of his games as a spectator. He’s currently level 13, he’s a genuine beginner and not a smurf, yet he’s getting enemies that have over 3k games already. And the 50% / 50% chance of win/lose on both sides is a fairy tale. The closest thing is probably that it ‘tries’ to bring both sides to the 50%, but always fails. One in a 1000 games is perhaps a statistically valid 50% / 50% . Let us assume it works all the time, even then it’s still just a number that only suggests a performance and outcome but not necessarily manifests in reality, meaning that you SHOULD have 50% to win and lose, yet the players will distort that number significantly during gameplay. There at this point the only thing we can say for certain about MMR is, that it tries to make it a fair game but fails in reality, because it’s impossible to determine the outcome, only the chances, which people will change in the game itself. For instance, there is a statistically better team vs a worse one. The worse one becomes victor at the end. This was unexpected. Why did it happen? Many many factors play here. The better team is only better, because they were lucky enough to get good teammates in more games than when they get bad ones. Doesn’t make them better players. Only their stats are better. I’ve also had games where chances were pointing at our victory and in fact we were controlling the game. BUT we lost in the end, because people in my team altered mentality and grew overconfident and started feeding 1 by 1. Regardless how hard I screamed at them to stop solo play and regroup and finish the game, they didn’t. So much for MMR.

      PS.: I stopped playing solo ranked a long time ago, as it makes no sense to me. Too much luck factor for me to get a normal team. I want to win with skill not luck, as I want to lose against higher skilled people and not because my team is subpar.

    • Everblue says:

      At my MMR level (2k and rising) I would disagree absolutely that playing a carry or snowball hero is a good idea.

      I would suggest playing a lane support hero – since the other bastards on your team are all going to pick carries anyway, playing lane support means that at least one carry on the map will get proper farm (the one you are laning with), and you hope that carry will win the game for you.

      I suggest Abaddon as a good hero for this, but Witch Doctor is also good for example as is Jakiro.

      If you pick a carry then you may be without any supports at all…

      • Banyan says:

        I assume that I’m in a low MMR and people who insist on playing carries don’t seem to understand that two supports with disables can end the game before the hard carries start becoming effective. For example, Witch Doctor and Vengeful Spirit rotations to gank the lanes transitioning into tower pushes can put you so far ahead that you’re raxing by 30 mins, esp if the other team is all carries.

    • jonjonjon says:

      dota is so unbalanced at the moment. its stomp or be stomped. doesn’t matter if its pro games, the first page of live games or the trench most games are just boring stomps. i’m sure some of this is the result of buffing every hero every patch. you can’t buff 80% of the heroes every patch and think things will be fine. there comes a time where the damage creeps too high and gets out of hand. i think dota needs some drastic changes maybe time for someone other then icefrog.

  3. Lord Custard Smingleigh says:

    This is why I obsess over my characters’ stats, but never, ever look at my own stats. I can feel a sense of accomplishment over my character leveling up, but evidence that I, personally, have dropped a level? That can sod right off.

    • Banyan says:

      This is why I’ve never played ranked. I know I’d feel obligated to play Witch Doctor (my best hero) until I got burnt out on him. Unranked single draft, least played, or random draft is the only way I can play this game without losing my mind.

      • proppaganda says:

        Yup, random/single draft is the only way i can play now. I dont even bother playing ranked as I have a career and will never become a pro dota player (as I am 24, too close to the retirement age :P) But i understand why people are into it. I love competition, but I don’t love getting stressed about it.

  4. Gnoupi says:

    Having recently switched to DOTA 2 from LoL, I notice an annoying issue:
    I’m constantly matched up with groups of friends. And those groups are usually composed of the following :
    – One veteran who played a thousand games
    – One or two of his friends who just started and have no clue what is going on.

    And same on the both sides. With me and maybe another player of my level (12-13) to fill the gaps.

    It’s very frustrating, because while I understand it tries its best to “average” the level of the group, in practice it means you will lane up with someone who doesn’t really know what to do (not that I’m an expert, but I have more or less an idea of what is going on), which is average when you play support like me. But the worst is that no matter what you do, DOTA 2 is a game where one person can carry the entire game on his own, if he gets fed. Cue the veteran player(s) accumulating gold, and stomping all over you, no matter how hard you try.

    So that’s where I am so far. I hope it changes with the increase in level, because it’s very frustrating to be thrown in the middle like that on every game.

    • sophof says:

      This sadly appears to be on purpose. If I read between the lines a little in Valve’s post regarding matchmaking, they purposefully fill in the game with single people who are the average MMR of a team. So whenever you play with a group of people, you’ll always be in the middle.

      I always play with friends and this truly seems to be the case. We rarely get a complete newbie or some sort of super-dota player.

      • Gnoupi says:

        Indeed, I’ve just read their post, and indeed, it makes sense from the “matching parties with other people” point of view. From this, it seems that as my own level and skill increases (eventually…), it will smooth out by itself.

        I wish though they had left the “solo only” option in matchmaking, for now.

  5. jama says:

    I noticed that by often switching between heroes I did more damage to my MMR than good. By recently focusing on just one single hero (in my case Ember Spirit) I managed to snatch a lot more wins than usually, subjectively speaking. Maybe it’s because Ember contributes a lot to team fights but at the same time has a nice escape mechanism in case you need to get the hell out of the fight (because, after all, you cannot trust those pubs. Their anger blinds them.) Feeling that you have an impact in team fights—even if you end up losing—helps.

    It also helps that Ember is still really fun to play even after 20 (more or less) consecutive games.

    Now, what hero to try next.

    • shagohad says:

      I would suggest trying Kunkka,

      once you get used to landing boats and torrents (this is without x mark) you can win teamfights, initiate, and run all over in the midgame. Get a fast battlefury pretty easily then flash farm and carry the game. Also doesn’t really need to be played mid so if some guy goes invoker last pick you can just adjust your build a tiny bit.

      Also the hero is hella fun when you get the hang of him :)

      edit: you can also play him support if your team goes 5 cores but this is a bit harder to be successful with

      • jama says:

        You just sold me that hero

      • Fiyenyaa says:

        Also laning against Kunkka who is investing in his tidebringer skill early is the worst experience in the world. So he has that going for him.

      • Everblue says:

        Silencer is also pretty great – you can start out as a support, teamfight like a monster midgame, and even carry a bit late on if you don’t die.

  6. wengart says:

    I use solo matchmaking to test out heroes I specifically want to play or solve my itch to play a certain hero. So if I feel like playing Broodmother I am going to pick Brood regardless of team comp.

    I play much more seriously in Random Draft which is what I consider the main competitive mode. My MMR has managed to stay at 2,700 or so though even with my stubborn picking.

  7. Rizlar says:

    So true. I do not play DOTA or any DOTA-like (wizard bashment rave?), but have experienced exactly the same thing with both 2p gambling machines in amusement arcades and competitive multiplayer games, vis a vis Plantside 2 and it’s utterly meaningless yet brutally visible kill/death ratio. :'(

    Down with this sort of thing.

  8. coeco says:

    This entire article = link to steamcommunity.com

  9. coppernaut says:

    I feel you. I’ve been fighting my solo MMR lately. I was so proud when I made the leap from 3200 to 3500, but now I’m back to 3400 and yeah, it frustrates me a little. It shouldn’t. Solo MMR is a really bad way to show skill level since you’re reliant on 4 teammate strangers and 5 enemy strangers to turn the odds into your favor somehow. My friend ground out almost 1k MMR over the course of months with Ursa to achieve 4K solo MMR. The amount of time and frustration he put into it makes me cringe, and I always think of that when I’m deciding to play solo ranked or not. I have way more fun in party ranked because at least then there’s some “real” teamwork going on, whether it be 2, 3, or 5 players partied together. I always have a sense of nervousness when I que solo ranked, but with players I can trust it’s far more relaxing.

  10. Billzor says:

    Thanks for this Pwarr.

    I have mixed emotions for MMR myself. Purge, of purgegamers and the TI4 Newbie Channel fame, taught me how to play Dota and I still look to the guy for tips and direction, but in some of his recent youtube casts he’s referred to the quality of the match being a direct result of playing with low mmr people. A friend of mine who helped get me into playing dota referred to solo ranked matches as “more of the same bullshit as unranked matches, just with a number,” meaning ranked matches still had spotty decision-making by your teammates and toxic assaults in chat for not doing what someone else thought you should do.

    I think I have a low solo mmr (1999 or something) and if that number is meaningful then part me does not want to be one of those people that makes someone as good as purge feel like it’s a crumby match, but on the other hand the idea of increasing a number through a challenge is appealing. On the other hand I feel appalled that this number creates a social division among players and being referred to as “3k trash” just makes me want to throw up.

  11. Valkez says:

    Personally I feel like their true mistake is the amount of points rewarded/removed at the end of the MMR games. The system doesnt take into consideration any individual values, and thus at the end the 5 who lost lose 25 MMR points, while the 5 who won receive 25 MMR points.
    I’ve played games where I owned for the entire early/mid game only to fall behind in the late game and lose due to my teammates doing close to nothing except dying. In the end I lost the same amount of points as them. The other way around also happened, as I was playing support and was mostly only dying just to be carried and ultimately win, to receive the same amount of MMR as my other teammates.
    On top of this, it also makes the statement “The teams are balanced. (Each team has a 50% chance to win.)” sound like a joke, because if that is true and correct, how’s one supposed to rise his MMR when every game adds or removes the same fixed amount of points (25)?
    In short, one is either (un)lucky or is stuck to whatever MMR value he ended up after the first 10 games.

    • Kitsunin says:

      Actually it makes sense. The idea is that you have a 50% chance of winning based on your MMR. In other words, if your MMR is lower than your actual skill level, you have a higher than 50% chance, if it’s higher, you will be more likely to lose, but if everyone in a given game is around their sweet spot, then yes, it should be 50/50. If you are at your sweet spot of 50/50 games, then you should stay right around that area because you should be gaining MMR just as often as you lose it.

      Unfortunately it can take a long time to reach that sweet spot, because you are only making a one tenth difference in any given game. Therefore a couple of other players being above or below their MMR will cause you to lose or gain without regard to your own skill…particularly likely at low MMR ratings. Still, it does mean that slowly, everyone is gravitating towards the correct MMR.

      I’m not arguing it shouldn’t take into account personal performance, I do agree that a better system will get people into that “sweet spot” more quickly. I’m just saying that the system does work…eventually…

      • Valkez says:

        That would work in a solo game, but in Dota the team as a whole counts a lot more than the single player on his own, plus the game is quite unforgiving, thus a bad player is most likely to bring down his whole team than a good player to win the game. I doubt anyone can say to be on his “sweet spot”, because our solo MMR value is determined from the people we are matched with rather than our skill. You are as likely to get a lucky streak of good english-speaking teammates as you are of being matched with a 4m premade of russians who completely ignore you. I believe they should start taking into account individual statistics such as gpm for carries, number of last hits / denies, how much a support character contributed to the team, etc. and modify the MMR points granted/lost according to such statistics.
        Perhaps such a system would still have its flaws, but right now I feel like we’re given a rank based on how “lucky” we are with our matchings. Hell, they gave us a crappy tool to measure our best hero’s performance with the TI4 compendium, why cant they measure our impact on a solo-queued ranked game?

        • Kitsunin says:

          Actually, even if you only have a 10% impact on a game’s outcome, and even if you are matched with completely random people (and not based on their MMR, like you are), the law of large numbers shows that if you play enough, you will eventually reach the statistical average of wins based on your skill compared to that of others. This means that, even if matchmaking doesn’t even use your MMR, you will still always eventually reach the correct MMR for your skill level.

          Again, I’m not even arguing that the system is good, I’m simply saying that if you play enough to reach the appropriate MMR, it is true that you will have about a 50% chance of winning most games.

        • BananaMan3000 says:

          You’re making classic mistakes in misunderstanding MMR.

          You have to view things from a statistical point of view over a number of games, which is the only way the system can measure your skill, because you are only 1 player out of 10, but the final result is shared by everyone. The “bad” player is just as likely (in fact more likely if you discount the possibility of yourself being the bad player, since there’s 5 “other” players on the enemy team and 4 on your own) to be on the opposing side as they are on your team. This actually is crucial to understanding how MMR gain works. If you play well you’re still only 1/10th of the players in the game playing well – the chance of this 10th being the bit that shifts the result from a loss to a win is small – this however is where you gain true MMR, the rest is just noise caused by the randomness of your team/opponents. It’s why if your MMR is 3400 but you think you’re 3500, it’s going to take a lot of games for your slight skill gap to overcome the randomness of the other factors.

          Measuring other stats like gpm isn’t a good way of measuring actual skill. If the other team are terrible and let you freefarm the entire game, this would give you a high gpm but it may be nothing to do with your own skill. Also you may have a very high gpm but make terrible decisions in teamfights (or not turn up at the right time) and cost your team the game. Everyone has played on a team with a carry with one-more-item-itis that will do nothing but farm even in the late game and won’t turn up to fights while the other team is gging you. The needs of the game and the “skills” needed are different in every game and fluctuate throughout, so measuring a small number of specific values isn’t particularly useful.

          Different heroes excel in different ways, some of them played well will end the game much earlier (meaning people have less kills at the end of the game, lower final levels etc). Some carries win games not by getting kills individually but pushing incredibly hard. Measuring stats would also lead to people gaming the system (imagine supports spamming useless wards constantly, doing nothing but stacking camps all game, etc) to increase their MMR, doing things which in the varying context of priorities in each game may not be useful. The only thing worth measuring is the final result, win or loss. Over enough games, the huge variables of your opponents and teammates are noise, and your actual skill is (slowly) revealed.

    • jonjonjon says:

      this is a terrible idea. there is no way to gauge individual performance. let me guess you’re a carry picker and you think all that matters is your K/D/A. what does it matter if you go 25/0/5 if you lose? people will discover what ever you’re using in your mmr algorithm and and they will exploit it to gain mmr. people will be more worried about farming/exploiting mmr then winning the game.

      how are you going to rate the dazzle support who didnt get any kills barely any assists but saved the carry all game allowing them to get kills? you could have an initiator who keeps going in and setting up easy kills for the team but dies and has a negative K/D but played much better than the carry who benefited and mopped up easy kills. what about a support who kill steals and steals the carries farm? a zeus who just spams his ult to do hero damage all game. there are games where the winning team all has negative K/D’s because the other team turtled in their base all game. having a positive K/D and a big networth because you afk farmed all game while the other team took your towers and rax doesn’t mean you had a good game. there is no way to judge individual performance in an algorithm and in the end all that matters is if you win or lose.

  12. vivlo says:

    i didn’t play dota in a long time (for reasons independant of my will), but from what i hear the MMR system is actually well balanced ; as it’s aimed at determining your global level and if you try to cheat with it i.e. by boosting it with a streak of games with your main hero, you will be smashed down the ground (once you have the bad idea to pick any other one).

  13. The Sombrero Kid says:

    It’s important to view your mmr as currently inaccurate and you must play a game to move it towards it’s correct position if you are to play ranked matches healthily imo.