How Competitive Matchmaking and Rating Works (Season 19)

can you elaborate on this a little bit?

I saw once that you typed out MMR as +0.08 or -1.12

this is human readable. and there is a known formula that converts MMR into SR. So with that formula available, wouldn’t MMR be “human readable?”

It’s more like +0.0834281. Technically human readable, yes. But then, binary is also technically human readable. It’s not pleasant. SR is much nicer to look at.

1 Like

That’s cool you also placed that report from unnamedUser about the leaver SR mods. I never really note my gains/losses but I’ll try and note then more often going forward. I do know that there have been losses and gains of teens vs the usual ~24 but I don’t remember the exact team setup from those matches.

If it’s an undeclared change it’ll be great to be able to firmly state it even if blizzard won’t.

I cannot wrap my head around “Standard Deviation.” (with the 68/95/99 rule, why can’t one standard deviation contain 70 percent of the data? why must it contain 68% to be counted as one standard deviation? I’ve watched many of these videos and still don’t get it, oh well)

So I’ll just ask a simple question.

Do you have a guess as to where -1σ and +1σ is on the SR scale? My guess would be -1σ is around 1850sr and +1σ is around 2850sr. And 68% of the population resides in between 1850 and 2850.

Edit: I see that the 68/95/99 rule is heuristic. Is it safe to assume that this is the rule that is used in the MMR calculations?

  1. Take a Calculus class (up through integration)
  2. The formula for the normal distribution is given at Normal Distribution -- from Wolfram MathWorld, equation (1).
  3. Integrate x from -1 sigma to +1 sigma → You get ~68%. This is the area under the curve, from -1 sigma to +1 sigma.

I suppose we could define +/- 1 sigma to include 70% of the population, but no one does that, and it makes the math uglier (which is why no one does that).

The conversion is roughly 5000 SR per 6 sigma, with 0 sigma = 2350 SR. So,
+1 sigma = 2350 + 833 = 3183.
-1 sigma = 2350 - 833 = 1517.

2 Likes

The user in question is Xion. It’s in the reference (If you have a leaver in comp do you suffer the same decreaase in SR?). I’ve since confirmed it with my own games.

1 Like

You sound like a conspiracy theorist to a degree, but who knows. It’s the same company that made World of Warcraft a game of grinding as far as I know so them using the same player methodology of trying to keep you playing isn’t insanely far fetched.

I wonder why they decided to place new accounts in the middle. And more importantly if alternatives were seriously considered.

Or is it when you look at all the facts, new accounts should start in the middle with out a doubt?

2 Likes

This is a much trickier question than you give it credit for. First, realize that Blizzard doesn’t know if a new account is a smurf or a genuine newbie, so assuming that new players are bad is a bad assumption.

Second, the current system has a zero sum SR economy. As one person gains, another person loses. This means that wherever everyone gets started at, that will be the middle. They could call it silver, gold, or bronze, but the average player will always be close to the starting SR. (The reason that I say close and not exact is because of players leaving the game: this throws the average off a bit.)

However, it is possible to design a system with a different SR economy. Let’s say everyone started at 0 SR, but people could not fall below 0. This means, after a game at zero SR, some players gain, but no one loses. You’ll then get a distribution that looks more like a poisson than a gaussian. Hearthsone is kinda like that:

Most players cluster in the lower ranks (and feel bad about themselves) while there is a greasy pole to climb up. In order to make the greasy pole a bit less greasy, Hearthstone has a complicated step up process: It is impossible to lose ranks from 25 to 20. From 20 to 5 win streaks grant a bonus rank, From 5 to 1, going up is as easy as going down, above 1 (legendary) an actual MMR is shown.

Now, Hearthstone is a casual 1v1 game. I don’t think this would really work out in a 6v6 competitive game. Most significantly, it would exaggerate luck in climbing, which would lower match quality.

All that being said, there is one massive improvement that Blizzard can make: use quick play data to seed the first competitive game. I can’t think of any reason they don’t do this, except a misguided notion of purity, or plain laziness.

4 Likes

I know 99% of the people achieve level 25 via QuickPlay. But what if you gain level 25 through custom games? or LucioBall? It might be more difficult to seed.

1 Like

At least make an effort…

Some gold games there’s utter noobs playing.

It’d be like me playing in masters+. I’d pity the team with me on it…

Require N games of quick play before competitive, where N is large enough to get a measure of someone’s ability.

2 Likes

This is not a good idea and nobody should be planning this. You cannot think like some robot where numbers are always the true meaning you have to consider things that machines cannot, this is qualitative data and Blizz has hit the big stinker when it come to this.

Human behavior teaches us that quickplay are people not trying hard, they are testing new hero, practicing hero, practicing new strategy. I can roughly estimate that many player will play 1000 SR below the true skill level.

It is really not me trolling when I am saying that blizz should consult me before any decision in matchmaking, they are in desperate need of help. This matchmaker is like a science experiment, maybe I can compare to the frankenstein but maybe worse because we cannot burn it.

1 Like

It’s interesting how everyone is SO quick to just assume Blizzard’s matchmaking system is infallible and that anything said to the contrary must be “crazy talk”. People are also quick to start with that premise and work backwards to justify their belief rather than provide any fact-based evidence to get to that conclusion independently of their built-in bias.

I on the other hand have been play for four years and have seen this same pattern the entire time I’ve been playing. It’s personal experience and a massive amount of observation – coming from someone who writes video games for a living and has written their own matchmaking system for poker. Yes, I am a subject matter expert on this in many ways.

And yet even as I type that I can hear people responding to discount my expertise on these topics because they’re once again starting from that place of inherent bias – just flat out believing without any proof that blizzard’s algorithm MUST be infallible.

The scientific method is quite clear. Independent and unbiased verification is required to prove that the matchmaking system does in fact work the way everyone’s defending it. I have provided CLEAR evidence to the contrary for YEARS. More than enough data points to extrapolate. Where is any of your proof?

2 Likes

You lie pretty routinely, so your “data” is worthless.

For example:

It is impossible to go zero for thousands, no matter how bad you are, or how hard you throw.

3 Likes

Oh really? Well, I’ve done it and I’ve been trying my hardest every game. Just for kicks and giggles, assume the opposite of your bias for once and work backwards from that and tell me how that experience can be explained?

Twitch. 00:29:04. You won Mystery Heroes.

You also lie in really lame, easily checkable ways.

3 Likes

Statistical anomalies don’t count. Such as smurfs & leavers. I’m talking about real data with real matches.

Look at these videos. You can’t see how INSANELY OVERMATCHED my team is in all these games???

And then when you get caught in a lie, you make the lamest excuses.

Going zero for thousands means that there were no “statistical anomalies”, and that you had a “perfect” record.

We’re done here.

1 Like

Your inability to be realistic about the data is the problem. You are so blinded by your allegiance to blizzard that you can’t see truth. I pity you.