The goal of this thread is to clearly explain what MMR means. I am hoping to appeal to newer players that are not familiar with the system, as well as provide analytical detail on not-so-obvious parts of MMR. Feel free to correct me if I misspeak, I am not a statistician nor trained in even developing systems like this. The thread will explain:
-
What MMR tells you about your skill.
-
Why MMR is practically impossible to be cross-compared server to server. Adding in the example of why Excalibur lost all his MMR.
-
And finally, what you can do to raise your mmr. This question will never go away. New players, or new reddit users, will constantly ask this. There’s no reason in becoming annoyed with these new users for asking a question YOU asked yourself at one time. Feel free to simply not read this part of the post if it is old to you.
Dota 2’s MMR system is an Elo rating system
Here is the system I am describing, if you want more in-depth detail. The Elo system is flawed for team based games, but it is one of the best methods created for determining a players’ individual skill in a relative population. The system is simple - If you win, you go up MMR, and if you lose, you go down MMR. Based on the “balance” of the game, will determine whether you lose more or less Elo points. Specifically, if you are heavily favored to win, you will win much less points than normal, and vice versa.
Okay, now that the simple explanation is out of the way, what does it mean? What does your MMR rating actually mean? Well, for one, it certainly doesn’t mean that your MMR is a display of your absolute skill. As an example of this, it would be dumb for one to say “I’m 5500 party MMR. Therefore, I am 5500 solo MMR.” The solo MMR is relative to solo queue, and specifically the skill of an average player in solo queue. This is why it is true that just because you are good at solo queue, does not mean you will be remotely good in professional Dota. Last time I saw Dendi, he was floating around 5.9-6k. That would be near the end of the leaderboard ladder on EU, so that must certainly mean that Dendi is inferior to all other players above him, right? Of course not, that’d be a ridiculous statement, as his performance in professional matches would almost surely outclass 99% of the players on the leaderboard. This reinforces the fact that your MMR is relative, and specific, to the mode you are playing it in. Does this mean Dendi’s solo queue skill is below all of those players? Yes. But lucky for Dendi, solo queue skill isn’t what wins you boat loads of money and fame ;).
Leading me to my next concept, the last argument transitions into the fact that MMR cannot be accurately cross-compared to different servers
Yes, I am from USE. No, I am not white-knighting USE MMR ranges, I am simply stating how MMR works and thus it will seem like I am defending my home server. I am not. It could be the reverse, and I would still be writing this. Things that affect MMR distribution are: Relative skill levels of players, volume, and scoring system. Let’s take a look at volume. Why does the volume of players matter? It matters because a higher volume is how you reach lower distributed MMR ranged games, assuming that there is a bell-curve spread of skill. The line of best fit for the scatterplot of skills will get closer and closer to a true bell-curve. A player on EU, at 5.8k, could very well get in a game where everyone is 5.5k+. Does this mean their game is “more fair” based on the system’s calculations? No. Does this mean that they have a higher chance of winning? Maybe, maybe not, depends on the individual. We see a whole lot of 5.8k players-6k players on EU leaderboards, maybe this is because it is more natural to play as a coordinated team. There are more players, so naturally there will be more players at the same MMR. The assumption that a lower MMR distribution yields better teamwork and coordination might be wrong, but I doubt it. Working together with this same MMR players may yield a faster increase in solo queue skill as compared to the NA counterpart which is learning how to stomp a game with a bunch of 4ks on your team. In other words, the EU server playing conditions may favor a more wide variety of players to being considered “good”.
Going further into detail, this is how NA games go at tip top MMR: You are 6k MMR
You are put on a team with 4.5ks, and you are against a bunch of 4.8-5ks. Does this seem more appealing to you to play in, than a game where everyone is the same MMR, or at least much closer MMR? If you are a player that plays Storm, and heroes that snowball, then yes it probably is more appealing to play versus a bunch of scrubs. They can win by themselves. Look at the NA leaderboards - Dragonfist is a TA player. If I were to try to guess, I’d say Dragonfist would go down on EU servers, but not because he’s bad. But because he plays like an NA leaderboard player. We’ve seen EE go up when migrating to EU servers, and that’s testament to EE’s more team-oriented style, in my opinion. NA leaderboards promote a more one-dimensional type of play - Stomp your opponents or lose the game. A direct quote from a 6500 on NA: “If I don’t smash my lane, like literally get a kill by 5 minutes, we lose. The clock starts ticking against me past the 5 minute mark, and if I don’t do shit almost immediately, I can’t salvage the other 2 lanes that will lose 70% of the time. To be honest, it was easier to climb 6-6.5k than 5.5-6k, because my opponents get considerably worse.”
So what does this say about EU vs NA?
It means nothing in terms of skill, only that the playstyles are completely different, and thus the MMRs cannot be “translated” nor compared. A 5800 on EU could be 5k on NA. Or he could be 6500 on NA. No one knows. To propose what some people were saying “Excalibur’s MMR dropped because people on USW are horrible” is the same as saying “Elo hell exists”. People were literally stating that Excalibur was losing mmr because his team was bad. This is wrong. The system still calculates the same fairness of a game as it does on EU, he was not being matched any different. What did change was the MMR distribution. Excalibur was not used to playing with people at 3.8-4.1k, and thus was playing incorrectly based on what was required to be a 6700 on an NA server. If he goes back to EU, he will almost surely go back up to where he was, and that is testament to the fact that EU players aren’t “better” than NA. Again, claiming that a player loses MMR in an NA server because NA players are bad, is literally saying Elo hell exists - Excalibur cannot reach 6700 MMR because of his teammates. Given time, it’d be perfectly expectable for Excalibur to adapt to the playstyle requirements, and climb again to 6700+ on the NA servers. That’s part of being a good player, adaptability.
Gaining MMR is as simple as improving your skill at an accelerated rate, AND playing a lot of games
You need to play a lot, for a big sample size of games. 20 games isn’t enough. 100 games isn’t enough. I’d say closer to 300-400 games of solo queue is enough to put you around where you belong. In order to gain MMR, you need to improve your skill faster than the average player of Dota 2. This means that you could remain the same MMR for 5 years but still be 10x better than you were when you started. You could last hit much better, know all the heroes and their abilities, and know how to use them, but not go up in MMR. This is probably the most frustrating thing in the world for some players, because they simply don’t understand. How is it possible that I don’t go up MMR, after I am clearly better than I was when I started? Bottom line, most everyone else is learning those things too. Let’s not consider the select group that decrease in skill, mostly because they don’t play, they aren’t very relevant to this argument because I’d estimate the sample size of those players to be extremely small and almost insignificant. The players that do play, consistently, and focus on improving will constantly improve. Each game will add a little tick on your absolute Skill-O-Meter. But on your relative Skill-O-Meter, you may go down or up! If your friend went up 2 ticks on the absolute skill meter, you would go down on the relative skill meter, since you only went up 1 on the absolute skill meter. Get it? You need to play a lot of games, amass games really, and focus on hyper-learning in the games you do play. Learning faster than the average player is how you beat the average player. The faster you learn, the faster you get better, and the further your MMR is away from the average player. It is not simply about just getting better.
The term “get good” used by people who like to ridicule people who ask how to raise their mmr, should be refined to “get good, faster”.
Miseryy
Post a Comment