Is Ranked MMR Broken?

Recently, I have been playing Solo Queued Ranked Matchmaking, and haven't had much luck getting a decent MMR yet. I just finished another Solo Que Match, linked below, and have confirmed a suspicion I've had about how MMR works. Below are some of the Rankings I've had recently (don't laugh please, I have friends who already do that).

These 4 values are the only values I've had for my MMR Rating in my last 10 or so games. I find it rather weird that in spite of having roughly 20 Matches after the 10 "Calibration" Matches, my MMR seems to never leave these 4 values. If I win, it goes 1 value to the right, if I lose 1 value to the left. I remember it jumping by about 100 after my 1st non-Calibrated Match, but from there, it's just the same values over and over.

Now, I considered the possibility that this was due to Win/Loss of a Match being the primary consideration in determining a change in my Ranking, and that I needed to exceed a set value for my performance to affect the changes, e.g. if I did stunningly, my Rating would increase at a higher rate than normal if I won, and perhaps even increase a little if I lost. This brings me to my latest match, linked below.

http://dotabuff.com/matches/478724457

In my Team, we had an Earth Spirit who seemed to have no idea about basics like "moving back when you're outnumbered", an Alchemist who admitted it was his 1st time playing the Hero and a rather...something... Pudge and Morphling.

Realising my Team had little chance to win, I figured Windranger would be the best Hero to Support with, since I could Escape from most Ganks and build a Force Staff to help my "Positioningly Challenged" Team members. I started off by buying the Animal Courier and Observer Wards as any good Support will do, skipping Sentry Wards this game, since I was expecting they wouldn't be worth the cost on my tight budget.

In the end, I had a high amount of Assists, some successful early Kills, and only 2 Deaths, the 2nd being a Juggernaut with Aghanim's Scepter using Omnislash on me and Alchemist while we were in the Fountain, at Full HP, so you can only really count the 1st Death against me. Because of my lack of Deaths, I ended up with the highest Xpm on my Team and 2nd highest Gpm, not much, but surprisingly good given the circumstances as well as the fact I was the Hard Support.

After performing at what I can't see as anything other than better than the rest of the other 9 Players, I ended up going back from 2205 to 2180 again. Please feel free to watch the replay and tell me I'm wrong.

Is it my imagination, or is the Rating System based SOLELY on Winning or Losing? Was I playing badly enough to deserve a flat decrease in Ranking? Or is MMR completely broken? Is anyone else's MMR drifting between static values in the same way as me?

Afraid I don't have an awful lot to add as I find the ranked system pretty confusing too. Most of my calibrating games were a wash (1 leaver after 5 mins, 2 were nobody even managed to buy a courier, terrible compositions) but somehow I still ended up with a 4k rating. I'm pretty convinced it takes at least something from your pub MMR.

As for moving up and down, mine has a similar pattern, up max +29, min -27 and somewhere mid twenties inbetween. It seems to have some relation to K/D/A and farm/xp, but it's only a few points here and there. I don't know if it really understands support play, or more likely works on the idea that good supports will win more games.

Unfortunately it tends to promote saying "MID!" so that some numpty on your team doesn't screw it up, and dominating the game with a snowballer if possible. Still you can end up on a team with 4 carries or games where nobody wards, midlaners who don't know how to deny, players who dont know how to get rid of tower agro...and this is supposed to be in the top few % of players!

Hell I played one earlier where I had to buy an Animal Courier as mid Tinker, and bought wards all frigging game. Luckily my opponent was pretty clueless and I got the farm I needed anyway. Our safe lane Pudge and Mirana were obviously above such things, while our other heroes were Lifestealer and Riki...somehow, we won, but it's still shocking...some days I really can't believe I'm not in low priority or something!

Dotabuff

I really think that doto starting mmr is heavily based on number of wins or games, cause I have ~800 wins and 4100 mmr, when my friend (I play with him like 80-90% of the time) have ~350 wins and 2500 mmr. If all players would start with same mmr, we should both have 4100 or 2500.

Same with team mmr. Teams that started with 5 players, where each have like 1000 wins, can lose 3 games and still end up with 5k mmr.
I started my team when we all had like 200-300 wins, we have 80% winrate and 2300 rating. http://dotabuff.com/teams/654419
So volvo fix ur **** please?

I started with 4700 MMR when I should've started with 5500~ just because I lost some of the first games.

The system doesn't work that well really.

I have nearly 700 wins, but most of those will be in a party, since I almost never played Solo once I found friends to play with, until Ranked Matchmaking came out. Can't help but think this is how MMR is calculated.

I read somewhere that the way its supposed to work is that that you will lose or gain points directly based on the win. However, as I understand it, the amount it adds or subtracts should be based on the skill of both your team and the enemy team. If you are high skilled, and your four allies are low skilled, and your enemy is medium skilled, you should lose a medium amount of points if you lose. However, if your entire team is high skilled, and your opponents low skilled and you win, you gain low amount of points, or lose high amount if the opponent win.

This is a similar system used in Starcraft 2, and I would've hoped that it used some form of that, as it helps alleviate problems when matched with lower-ranked players.

But I'm starting to feel like this is not the case at all...

It just doesn't make sense to me. Won 7 out of 10 in the calibration matches, and got just over 3k for my rating. Bit annoyed as I've played over 1000 games, but thought it can't be that hard to rise in the ranking system.

Recently played a Bounty Hunter match, where I went solo offlane, and trashed the enemy early-mid game. But as I didn't go the battlefury build I had to rely on my carry to do something in teamfights. All my Luna carry did was get initiated on and then panic bkb and ult just before she died... So I try and transition into the carry for the team going butterfly, and almost do it, but there's way too much damage for me to stand up to. I lost the match but don't see how I played badly and should be rated down for the game.

http://dotabuff.com/matches/470326695

Can anyone see what I did wrong in the game?

It doesn't mean much to me, I gain or lose the same amount of MMR whether the win/loss is legitimate or if someone AFK abandons/disconnects and the team gets stomped from there, which happens far too often in ranked games.
I suspect it is broken, it really all seems to hinge on whether you win or lose and that's about it.

So i asked a couple of my friends to check out my games and stuff. They all said i should have like 3600-4000 MMR but i have 2800. And because of that i stomp EVERY ranked game and have won 6 of the last 6 ranked games played. Volvo plz

Really can't see anything you did wrong that game Timmay, other than being on a team with players who failed. Unfortunately that's the crux of MMR, the numbers aren't always accurate, the way people play particular heroes doesn't always work, team composition, balance, counter-picks, strategy, etc.

It's impossible for any system to get all these things right, DOTA must be a very complex game to try to analyse statistically, as there are a lot of things that don't make sense. Players NOT getting farm/xp, getting killed, can actually be good play and beneficial for their team. (e.g. Vengeful Spirit swaps out stunned carry late game and saves them, gets killed instead)

I think what they've done is paint some very broad strokes and gone with "if they win, they're helping their team", which obviously leads to allsorts of frustrating situations and frankly cr*p players getting reasonably good ratings (the number I've seen can't just be a statistical blip). However, it's probably also the best way of getting people to be relatively unselfish.

There are several flips sides to this though:

- If they told people exactly how the system works, people would find ways of gaming it and try to artifically boost their rating.

- Ratings will be wrong sometimes, but they're more likely to be right. Giving people the option to recalibrate would kind of be fairer, but again is open to abuse.

- If they told you the relative ratings of your team during the picking stage, would it alter your picks? Would it lead to lower ranked players being shoved into supporting roles? Would you let 1 go mid? Or should people stick with their strongest area?

- Trying to put a single number on somebody's skill level is inherently inaccurate, especially in a team game, even more so when people play in different "positions" with different abilities.

For example, in football/soccer, if Wayne Rooney is an 8 playing at striker, how does he compare with Joe Hart (a 9) playing in goal? (numbers made up for the sake of argument). Their impact on the game might be completely different, who's more important, or better? It's impossible to say, as it totally depends on the situation.

What if they switched roles - how should they now be rated outside their usual position? How about if you had two 10* strikers but no goal keeper, would that be better or worse? Same in DOTA if you have 4 carries and no support? Or a great support player who's a **** ganker going mid and a carry playing support who doesn't know how to ward properly or gets bored and rushes a mek instead?

The best MMR is ever going to be is a general guide to somebody's level, yes it's nice to have a higher number next to your name, but personally I'm much more interested in getting good games...

Would still like to see some kind of mode where you can queue by role :)