The Science Behind your Skill Calibration.

MMR CALIBRATION

"i won like all 10 of my games but i got given a low MMR why is that?"

"Im way better then my MMR why is it, that i cant change it?"

"i played all my games and i won heaps and im good but i got given a **** MMR why is this?"

Now im sure all of you have heard these statements before, yes?

hell even ive been guilty of saying them. BUT i didnt just sit there and whine, no i went out and researched and read and did a whole lot of asking to figure out.....

WHAT THE HELL IS MY MMR? AND HOW IS IT CALIBRATED?

---

MATCH MAKING RATING

"What is Match making rating?"

MMR is the logarithmic system used by valve to derive who you should be matched with when queuing.

FALSE FACT1# calibration games give you your MMR.

WRONG, when you first start DOTA2 you are given a BASIC "HIDDEN" MMR of 1000. (source*Valve)(Source*SubredditDotA2)

this "HIDDEN" MMR stays with you from your first game.


FALSE FACT2# Everything in game affects your MMR.

WRONG, the only thing that affects your MMR is WINS or LOSSES. The amount it changes is dependent upon TWO things.

1). YOUR MMR
2). YOUR OPPONENTS MMR

When matched against players who are of lower MMR than you, IF you win you shall receive an increase in your "HIDDEN" MMR by an amount based upon the difference between the two HIDDEN MMR's. usually since you are Higher then your opponents the amount is quite small. on average it will be around 1 point per 50 mean average difference. IF the MEAN AVERAGE DIFFERENCE is below a certain threshold usually 500 MMRMAD the increase is a static 25 points. (Source*Valve subreddit) (Source*DotA2 Dev page)

if you are matched against players with a significantly higher Mean MMR, the amount your receive will be larger.

i.e. : a 2.5k MMR Mean team defeating a 4.5k MMR mean team: players will receive an increase of 40 points on average.

---

CALIBRATION GAMES

CALIBRATION GAMES DO NOT SET YOUR MMR

Claibration games are used by the Logorithmic system NOT to determine your MMR but to ASSESS the VALIDITY of your HIDDEN MMR.

Your Calibration games are INDEPENDENT of your HIDDEN MMR.

they are used as a ANCHOR point for the Algorithm to differentiate between your HIDDEN MMR and your MMR given by the calibration games.

EXAMPLE 1#

you have a HIDDEN MMR of 1.5k

your calibration games give you an MMR of 3.3k

your FINAL MMR will be sitting between 2.1k and 2.5k as per the MEAN MMR AVERAGE.

this is why you may have won all your calibration games but still have been given a horrible MMR.

THE REASON?

this account is most likely your only one yes?
this account is the one you started on yes?
Your were pretty **** at the start hey?
you lost a lot of your original games hey?

NO **** WE ALL DID!!!!!

the reason why is because your hidden MMR would have taken a beating from you learning to play.

so even though you have improved so ******** freaking much it still takes a lot to get that HIDDEN MMR up there and unless you have a +60% win rate its not going to increase at a huge rate.

---

HOW TO GET A PROPER CALIBRATION

1). Create another account.
2). play games and because you should be more skilled then most of the people. (barring smurfs) you will win a lot of games.
3). on average you should be sitting at a +60% win rate.
4). your HIDDEN MMR wont take such a beating.
5). Do your Calibration matches.
6). receive your proper MMR.

---

Any way guys i hope i have helped and maybe a lot of the posts will stop about MMR.

if the devs want to sticky this maybe ? i dont know but i think it might help people who feel annoyed at their MMR.

---

SOURCES

http://dev.dota2.com/
- MMR
- Matchmaking
- Calibration
- MMR Calibration
http://www.reddit.com/r/DotA2/
- MMR
- Matchmaking
- MY MMR SUCKS
- HOW DOES MMR WORK?
https://developer.valvesoftware.com/w/index
- DotA2

Thank you very much for the info, Smuggels.

Just a side question; I have been trying to increase my solo-ranked MMR recently but my average is a 50% win/ loss ratio. Whenever I win a game, I always lose the next game, then win, then lose, etc...

I mentioned this to a friend and he said that if you win, you will be teamed up with lower skilled players in the next game and if you lose, you will be teamed with higher skilled players and that Valve/ DotA does this deliberately in order to keep things at that 50% mark.

In your research, have you come across anything like this?

It just seems uncanny that I very rarely win or lose two games in a row but consistently win one then lose one. I also notice that the quality of my teammates can vary greatly from game to game, between my team having no idea about how to play to people who are clearly higher skilled than me and the opposing team, resulting in an easy win.

Please let me know what you think, when possible.

i know exactly where you are coming from but

in this instance and in a lot of others...

"Correlation does not imply causation"

yes it may seem like it is intentional but it in fact is anything but.

First off :"YOU WILL ALWAYS BE MATCHED DEPENDING ON THE MEAN MMR"

this will be if solo ranked the MEAN MMR of 5 players is 2.5k then will be matched with another MEAN of 2.5kmmr give or take 50-100 MMR as you increase in MMR you will naturally increase who you are matched with BUT it goes off MEAN MMR not INDIVISUAL MMR.

the reason this happens is part of a much larger chain of events.

for example.

Player A has never played dota2 before but when he started he had friends that did thus he played with them and they basically stomped his early games enough that he could stave of the inevitable hidden MMR beating. he then does solo ranked and gets a reasonable score due to a slightly higher hidden MMR which is not directly indicative of his actual skill. now that he is doing solo ranked he is consistently losing.

Player B has played many many many games and in fact has good skills yet because he trudged through pubs beating his hidden MMR to hell he receives an MMR that is not indicative of his actual skill. when playing solo ranked he is winnning games when placed with team mates who have similar experience but losing when Player A shows up.

these two players are most likely the types of people you will meet in solo ranked.

BECAUSE of the HUGE amount of players the MEAN amount of Player A and Player B will be around the same amount so you will get matches with skilled players who are rising and then matches with players who are falling.

this has the side effect of creating a VARIABLE BUBBLE of skill with teams either having a majority of player A or a majority of Player B

you have to EXCEED that skill bubbles threshold to actually increase repetitively which is where PLAYER C arrives.

EXAMPLE 2:

Player C has a horrible MMR of say 1kmmr but because his ACTUAL skill is quite HIGH he/she can basically carry a team to a WIN. because of his relative HIGH skill he transitions skill bubbles at a consistent rate of 4 wins to one loss. As he ascends from the 1kmmr to the 2kmmr his ascension rate reduces to a 3-1 rate. He can still carry the team but it requires more skill to do so. now he has transitioned into the 3k mmr his ascension rate slows to a 2-1 ratio. soon he will arrive at a VARIABLE BUBBLE OF SKILL where it is harder for a SINGLE PERSON to CARRY a team as the comparable skill of Player A and player B are too close. thus he will now sit at a 1-1 W/L ratio.

as you can see it is not an actual effect of the SYSTEM it is a SIDE EFFECT of the PLAYERS.

I don't know how much time you spent searching all this information but thanks for doing it and posting this here. I'm sure it will help more people to understand these things. Also as a side note +rep given!

tbh mmr isn't a numerical rating of your skill level until you get to about 4.5K or above, before then its basically what level too which you can carry your team/have a very positive impact on the game.

For example if we take a professional player who calibrates at 1K, he can quickly and easily make it up to about 4K without even losing a game because that's the point in which he/she can completely carry a game up to.

When thought about that way the ranked match making system is fair enough.

I'm certainly not saying I like the system, its absolute trash but with people constantly complaining about not being able to raise their mmr due to **** teams that's ********, if your losing at a skill level which you think is far below yours then lets face it, your not as good as you think.

But thanks for doing the research for us smuggels, its some nice information to have on hand, I might go make a new account and calibrate it, should be a learning experience :)

Good post Smuggels, I think you explain a lot of the important points - especially about the "forced win rate of 50%" - it's simply not true, the idea is every GAME should have a 50/50 chance for either team.

Think there's 3 things that majorly effect people's MMR thinking:

1) People judging their own skill only by their best performances, not in combination with their worst.

2) People don't realise what an impact relative skill has on the game - I've occasionally tutored people well below my MMR and easily won lanes 1v2/1v3 that I wouldn't have a prayer in against equally skilled opponents. If you should be ranked well above your MMR, you should be regularly dishing out a beating in lane.

3) Fleet's point about "positive game impact" is a good one. I've got to 4k+ MMR playing probably 70-80% of my games as support, often hard and/or solo support. I haven't often been mid, I haven't often been carry - but I've helped the team win. Filling the roles that need doing, doing them well, and communicating effectively is more important than having 5 players who want to "carry" their team.

MMR is only an overall measure of your win rate - there are many, many skills that contribute towards that. Some players are only good with specific heroes, specific roles. Some players have good technical skills (last hitting, ability usage, actives). Some players have fast reflexes. Some players make a lot of good decisions, have good positioning and can adapt their item builds to suit the situation. Some players have a positive communication impact. Some players are good at co-ordinating with their team mates even when they don't know them. Some players can make good strategic calls and get their team to go with them. Some players have a good understanding of what to pick to counter the opposition and/or synergise with their allies. The list goes on...

So this is not 100% CONFIRMED ...BUT

MMR is Region based.

so for each Region the MEAN MMR is completely different.

for example.

US the MEAN MMR is around 2.8k.
Where in SEA the MEAN MMR is hovering around 2.6k.

so if a SEA player who is 2.6k would play in US there is a loading applied of 200 MMR.

i only recently learned this through the Dev.page and its bloody huge i think.

depending on the MEAN MMR of the region each player would TECHNICALLY have a VARIABLE MMR.




i bet mine in russia is over 9000.....

exactly.

the algorithm uses the cumulative MMR of 5 people to make a team. it doesn't take into account whether they are the PLAYER A type or the PLAYER B type or PLAYER C type . it only goes off the MMR and it tries its very best to get a MEAN MMR as close to each other as possible.

thus you can have a team of 4 great players and one horrible or the opposite.

but at its core you should have a relatively common skill level. (with exceptions)

actual Player skill should not be read as MMR it is SUBJECTIVE.

Wow didnt expect logarithmic system to be something in English.

for the most art very good points. a few things:

Your smurf winrate likely won't be much different from your main. I'm about 100 games into a smurf, but it only took 10 before I was playing similar level games to my main. Since then I have a 54% WR versus 51% on main. Winrate also means nothing for mmr, it only signifies the overall direction you're moving.

Region thing sounds interesting, not what I expected but it really shouldn't affect anything, I would expect that USE/USW as well as EUE/EUW/RUS share MMR since they share a common leaderboard.