General DiscussionZenoth's guide to understanding MMR

Alright, enough with the big hoo-hah and all the ignorance about this number and how it works. Too little people understand what it really means (or attribute too much meaning to it).

DISCLAIMER: Numbers are pulled out of my ass for the purpose of comparison purposes within this article.

First up, let us imagine an ideal world, where there is a possible way of assigning an average rating to each aspect of a player's individual skill, independent of heroes picked.

It would look something like this:
Awareness: 4250
Laning Mechanics: 5120
Game Knowledge: Relentless6200
Execution: 4920
Micromanagement Skills: 3450

Now remember these are average values, which are variable game-to-game depending on your mood, how much your mother is nagging while you play, or whether or not you are a man-child like Mason and throw a hissy fit everytime people refuse to do things your way, effectively making your rating -500 in each field that game as you TP mid multiple times to feed the opponent.

Using the example above the player has an average rating of 4788. However, this rating still doesn't represent the actual skill rating he is playing at in any particular game. By picking heroes that capitalize on Laning Mechanics and don't require much Micromanagement Skills, he can play at a higher level than his individual skill rating implies.

Real World implications: A 5k player can have his rating comprising of many different aspects. He could be 3500 in laning but 6000 in awareness, and lose to an average 4k player purely laning-wise. As humans we are mostly capable of spotting others' mistakes in areas we are more competent in, so the person with better execution will wonder why his teammates are not using their skills in fights, while those same teammates may be wondering how he lost his lane so badly. At 3k ratings the average could come from a wild mix of values, like 6028 Game Knowledge but 2308 Laning Mechanics, or a flat 3024 in everything. This leads to everyone finding that their teammates are far inferior to them, since they are likely to compare themselves to others ONLY IN THE ASPECTS THAT THEY ARE BETTER IN, despite being of a roughly equal skill level on average.

Now, individual skill is still but one small aspect of one's rating. Moving on, we have hero picks.

Different heroes capitalize heavily on different aspects of individual skill. OD and TA, for example, are heroes that benefit greatly from laning mechanics, which may put a player so far ahead that his weaknesses in Execution and Awareness are not glaringly obvious. This will effectively increase the rating of his play in that particular game, possibly from 3850 -> 4120. Alternatively, heroes like Lifestealer who are almost always able to get away with a Rage-TP are beneficial to a player with low awareness but high execution. Naturally, familiarity with the hero will also affect the player's skill with it, as being unfamiliar with skills and combos will effectively lower reaction time and hence execution.

But the most important thing about hero picks are two things - synergistic effects and neutralizing effects. A player who is only capable of playing Axe at say, 4860 rating, but has a teammate who picks Dazzle and lanes with him, might effectively be raising his level of play to 5213 for that particular game. Alternatively, one picks Lanaya and usually plays it at a 2894 rating, but the enemy player picks a Viper, who plays it at a 2580 rating. However, thanks to the laning advantage Viper has over Lanaya, the Viper player is able to play at a 3120 level this game while the Lanaya player is lowered to 2680.

So at this point we can see that even if individual skill may differ, the hero picks are actually enough to make a significant enough difference to render individual skill differences obsolete. Furthermore, different heroes capitalize on different aspects of individual skill, which may contribute to a significant difference in game.

Now Individual Skill + Hero Picks is still not all there is. There is one more factor, which I'll call Try-Hardness. Here's a scale on 1-10 for reference:

-1 - Mason level. Being a manchild, throwing a hissy fit and losing the game on purpose because reasons.
1 - Last picking a fifth carry and calling mid, telling your OD to fuck off to the jungle
2 - Picking heroes you are bad with just because you feel like playing them, not caring about the team's picks.
3 - First picking heroes easily countered just because you feel confident playing these heroes.
4 - Picking something that fits into the lineup even if it isn't necessarily something you are very confident with.
5 - Playing the "best" hero of the patch every game even if you may not necessarily be very competent on him.
?? - Random
6 - Picking your most confident hero which fits into the current lineup.
7 - Exclusively picking your best heroes regardless of enemy lineup.
8 - Only playing your most confident heroes, which can fit into the current lineup.
9 - Exclusively picking heroes that you are very good with and hard to counter, into situations that usually guarantee you a win.
10 - EE-sama level. Pausing games to set up quick buys/disconnecting and spectating your own game to find where the enemies placed their wards
11 - Fucken account buyers/sandboxie users/language preference abusers

As you can see tryhard-ness is a factor which also includes hero picks to some extent. Someone who "tries hard" will necessarily be playing at a higher level than someone who doesn't, even IF their individual skills are at the same level.

In any game, whichever team is playing at a higher rating after all the factors have been accounted for will win. Therefore, rating is absolute in the sense that whichever team has a higher total sum is going to win the game. Remember though that performance varies and all these numbers are usually only averages.

Where does the current MMR system come in? MMR reflects an average of EVERYTHING mentioned above. It does not just reflect individual skill, it reflects something like

MMR = (Average Individual Skill)x(Average hero proficiency)x(Level of Try-Hardness)

So two players with identical individual skill may have vastly different ratings if one only plays a small pool of heroes he is good with while the other randoms every game. Likewise, two players with identical MMR may have vastly different levels of individual skill. And two players with identical level of average individual skill may see fit to criticize each other simply because they are skilled in different aspects of the game, and compare themselves to others only in those aspects. Furthermore, in the context of the same game, there are many possible factors affecting individual skill, which causes MMR to over/under represent a person's level of play.

In short, Individual Skill, the factor which most people assume to be directly represented by MMR, is but one of the factors in the overall rating. It is correlated, but it is far from absolute in terms of determining which of the two is a better player.

FAQ:
Q: How do I raise my rating then?
A: Get better at the game (increasing various aspects of Individual Skill). Practice more with certain heroes (increasing Average Hero Proficiency). Tryhard more (simply not raging in pubs will prevent your Individual Skill from effectively dropping during a game)

Q: But my teammates all suck?
A: On average they suck as much as you. People are placed in the game with you on basis of having a similar overall rating, and as I have mentioned there are many possible reasons which leads to a person having a certain rating.